 |
|
| If you can't view the Datasheet, Please click here to try to view without PDF Reader . |
|
|
|
| If you can't view the Datasheet, Please click here to try to view without PDF Reader . |
|
|
| Datasheet File OCR Text: |
| MB87J2120 & mb87p2020 -a lavender & jasmine fujitsu semiconductor fujitsu microelectronics europe gmbh european mcu design centre (emdc) am siebenstein 6-10 d-63303 dreieich-buchschlag germany version : 1.7 file: mb87p2020.fm hardware manual colour lcd/crt/tv controller speci?ation
MB87J2120, mb87p2020-a hardware manual page 2 file: /usr/home/msed/gdc_dram/doc/manual/mb87p2020.fm copyright ?2001 by fujitsu microelectronics europe gmbh european mcu design centre (emdc) am siebenstein 6-10 d-63303 dreieich-buchschlag germany this document contains information considered proprietary by the publisher. no part of this document may be copied, or reproduced in any form or by any means, or transferred to any third party without the prior written consent of the publisher. the document is sub- ject to change without prior notice. revision history version date remark 0.8 05. apr. 2001 first release 0.9 27. apr. 2001 preliminary release 1.0 29. jun. 2001 overview section and register list reviewed sdc, pp, aaf, dipa and ulb descriptions reviewed 1.1 20. jul. 2001 register list improved, lavender pinning added, overall review 1.2 02. aug. 2001 apll spec included (cu) review: overview, functional descriptions, register/command lists preliminary ac spec for jasmine 1.3 05. oct. 2001 ac spec for both devices, lavender added/jasmine reviewed two pinning lists - sorted by name/pin number ulb dma limit description (dma fifo limits vs. ipa block size) sdc register description reviewed 1.4 11. oct 2001 clari?d ac spec output characteristics (20/50pf conditions) 1.5 27. mar 2002 pinning and additional registers for mb87p2020-a added design description for changes in mb87p2020-a added ac spec updated for mb87p2020-a 1.6 05. apr 2002 ac spec for MB87J2120 updated 1.7 22. jul 2002 description changed to mb87p2020-a table of contents table of contents page 3 table of contents part a - lavender and jasmine overview 1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.1 application overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.2 jasmine/lavender block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 5 2 features and functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 3 clock supply and generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 4 register and command overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4.1 register overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4.2 command overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 part b - functional descriptions b-1 clock unit (cu) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 1 functional description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 1.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 1.2 reset generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 1.3 register set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2 apll speci?ation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.1 definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.1.1 phase skew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.1.2 duty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.1.3 lock up time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.1.4 jitter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.1.5 variation in output cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34 2.1.6 maximum power consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.2 usage instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3 clock setup and con?uration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.1 configurable circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.2 clock unit programming sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.3 application notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 b-2 user logic bus controller (ulb) . . . . . . . . . . . . . . . . . . . . . . . . . . 39 MB87J2120, mb87p2020-a hardware manual page 4 1 functional description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 1.1 ulb functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 1.2 ulb overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 1.3 signal synchronisation between mcu and display controller . . . . . . . . . . . . . . . . . . . . . . . . . 44 1.3.1 write synchronization for lavender and jasmine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 1.3.2 read synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 1.3.3 dma and interrupt signal synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 1.3.4 output signal configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 1.4 address decoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 1.4.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 1.4.2 register space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 1.4.3 sdram space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 1.4.4 display controller bus access types (word, halfword, byte). . . . . . . . . . . . . . . . . . . . . . . 51 1.4.5 display controller data modes (32/16 bit interface) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 1.5 command decoding and execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 3 1.5.1 command and data interface to mcu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 1.5.2 command execution and programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 1.5.3 structure of command controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 1.5.4 display controller commands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 1.5.5 registers and flags regarding command execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 1.6 flag and interrupt handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 1.6.1 flag and interrupt registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61 1.6.2 interrupt controller configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 1.6.3 interrupt generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 1.6.4 interrupt configuration example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 1.6.5 display controller flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 1.7 dma handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 1.7.1 dma interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 1.7.2 dma modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 1.7.2.1 level triggered dma (demand mode) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 1.7.2.2 edge triggered dma (block-,step-, burstmode) . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 1.7.3 dma settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 1.7.4 dma programming examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 2 ulb register set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71 2.1 description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 2.2 ulb initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 b-3 sdram controller (sdc) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 table of contents table of contents page 5 1 function description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 1.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 1.2 arbitration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 1.3 sdram timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 1.4 sequencer for refresh and power down . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 1.5 address mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 1.5.1 elucidations regarding address mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 1.5.1.1 block structure of pixel data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 1.5.1.2 access methods and devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 1.5.1.3 program example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 1.5.1.4 address calculation - example with concrete {x,y} pixel. . . . . . . . . . . . . . . . . . . 90 2 sdram ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 2.1 external sdram i-/o-pads with configurable sampling time (lavender) . . . . . . . . . . . . . . 92 2.2 integrated sdram implementation (jasmine) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 3 con?uration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 3.1 register summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 3.2 core clock dependent timing configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 3.2.1 general setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 3.2.2 refresh configuration for integrated dram (jasmine). . . . . . . . . . . . . . . . . . . . . . . . . . 95 b-4 pixel processor (pp). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 1 functional description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 1.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 1.1.1 pp structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 1.1.2 function of submodules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 00 1.2 configuration registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 1.3 special command options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 1.3.1 bitmap mirror . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 1.3.2 bitmap direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 2 format de?itions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 2.1 legend of symbols. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 2.2 data formats at ulb interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 2.3 data formats for video ram / sdc interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 b-5 antialiasing filter (aaf) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 1 functional description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 MB87J2120, mb87p2020-a hardware manual page 6 1.1 aaf overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 1.1.1 top level structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 1.1.2 aau function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 1.1.2.1 super sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 1.1.2.2 drawing resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 1.1.2.3 data and address processing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 1.2 configuration registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 1.3 application notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 1.3.1 restrictions due to usage of aaf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 1.3.2 supported colour formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 1.3.3 related sdc configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 b-6 direct and indirect physical memory access unit (dipa) . . . . . 117 1 overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .119 2 con?uration registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .121 2.1 register list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 2.2 recommended settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 2.3 related settings and informations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 b-7 video interface controller (vic) . . . . . . . . . . . . . . . . . . . . . . . . . . 123 1 introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .125 1.1 video interface controller functions and features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 1.2 video data handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 2 vic description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .127 2.1 data input formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 2.2 data format in video ram (sdram). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 2.3 data input timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 2.3.1 videoscaler-mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 2.3.2 ccir-mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 2.3.3 external-timing-mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2 3 vic settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .135 3.1 register list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 3.2 register description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 b-8 graphic processing unit (gpu) . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 1 functional description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .151 table of contents table of contents page 7 1.1 gpu features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 1.2 gpu overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 1.2.1 top level structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 1.2.2 dfu function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 1.2.3 ccu function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 1.2.4 lsa function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 1.2.5 bsf function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 2 color space concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 2.1 background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 2.2 data flow for color space conversion within gpu. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 2.3 mapping from logical to intermediate color space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 2.4 mapping from intermediate to physical color space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 3 gpu control information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 3.1 layer description record . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 3.2 merging description record . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 3.3 display interface record . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 3.4 supported physical color space / bit stream format combinations . . . . . . . . . . . . . . . . . . 167 3.5 twin display mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 3.6 scan modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 3.7 yuv to rgb conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 3.7.1 yuv422 demultiplexing and chrominance interpolation . . . . . . . . . . . . . . . . . . . . . . . 171 3.7.2 matrix multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 3.7.3 inverse gamma correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 3.8 duty ratio modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 3.8.1 working principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 3.8.2 usage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 3.9 master timing information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 3.10 generation of sync signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 3.10.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 3.10.2 position matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 3.10.3 sequence matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 3.10.4 combining first-stage sync signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 3.10.5 sync signal delay adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 3.11 pixel clock gating. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 3.12 numerical mnemonic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 3.12.1 color space code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 3.12.2 bit stream format code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2 3.12.3 scan mode code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 MB87J2120, mb87p2020-a hardware manual page 8 3.13 bit to color channel assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 83 3.14 gpu signal to gdc pin assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 3.14.1 multi-purpose digital signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 3.14.2 analog pixel data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 3.14.3 dedicated sync signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 4 3.14.4 sync mixer connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 3.14.5 color key output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 4 gpu register set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .187 4.1 description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 4.2 determination of register contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 4.2.1 values derived from display specs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 4.2.2 values determined by application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 4.2.3 user preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 4.3 gpu initialization sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 5 bandwidth considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .198 5.1 processing bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 5.1.1 average bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 5.1.2 peak bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 5.2 memory bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 5.2.1 average bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 5.2.2 peak bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 5.3 recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 6 functional peculiarities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .204 6.1 configuration constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 6.2 bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 6.3 image processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 6.4 data output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 6.5 diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 7 supported displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .207 7.1 passive matrix lcd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 7.2 active matrix (tft) displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 7.3 electroluminescent displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 7.4 field emission displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 7.5 limitations for support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 7.5.1 no support due to vcom inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 7.5.2 problems due to 5v cmos interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 table of contents table of contents page 9 b-9 cold cathode fluorescence light driver (ccfl) . . . . . . . . . . . . 213 1 introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 2 signal waveform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 2.1 general description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 2.2 duration of the phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 2.3 pulse shape of fet1 and fet2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 18 3 register description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 3.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 3.2 control bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 4 application notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 4.1 ccfl setup example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 4.2 ccfl protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 part c - pinning and electrical speci?ation 1 pinning and buffer types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 1.1 pinning for mb87p2020-a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 1.1.1 pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 1.1.2 buffer types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 1.2 pinning for mb87p2020 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 1.2.1 pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 1.2.2 buffer types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 1.3 pinning for MB87J2120. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 1.3.1 pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 1.3.2 buffer types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268 2 electrical speci?ation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 2.1 maximum ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 2.1.1 power-on sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 2.1.2 external signal levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 2.1.3 apll power supply level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 2.1.4 dac supply. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 2.1.5 sdram supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 2.2 recommended operating conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 2.3 dc characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 2.4 mounting / soldering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 2.5 ac characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 MB87J2120, mb87p2020-a hardware manual page 10 2.5.1 measurement conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 2.5.2 definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 2.5.3 clock inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276 2.5.4 mcu user logic bus interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 2.5.5 interrupt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 2.5.6 dma control ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 2.5.7 display interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 2.5.8 video input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 2.5.9 ccfl fet driver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 2.5.10 serial peripheral bus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 2.5.11 special and mode pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 4 2.5.12 sdram ports (lavender) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 part d - appendix d-1 jasmine command and register description. . . . . . . . . . . . . . . . 289 1 register description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .291 2 flag description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .319 3 command description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .323 3.1 command list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 3.2 command and i/o control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330 d-2 hints and restrictions for lavender and jasmine . . . . . . . . . . . . 331 1 special hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .333 1.1 ipa resistance against wrong settings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 1.2 ulb_dreq pin timing to host mcu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 1.3 clkpdr master reset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334 1.4 mau (memory access unit) commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334 1.5 pixel processor (pp) double buffering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 35 1.6 robustness of ulb_rdy signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 6 1.7 robustness of command pipeline against software errors . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 1.8 dma resistance against wrong settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 2 restrictions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339 2.1 esd characteristics for i/o buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 2.2 command fsm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 2.3 gpu mastertiming synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1 table of contents table of contents page 11 2.4 read limitation for 16 bit data interface to mcu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342 2.5 sdc sequencer readback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 2.6 direct sdram access with 16bit and 8bit data mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 2.7 input fifo read in 16bit mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 2.8 ulb_dstp pin function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 2.9 software reset for command execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 2.10 aaf settings double buffering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 2.11 pixel engine (pe) commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 8 2.12 pixel read back commands (getpixel, xchpixel) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 2.13 display interface re-configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 1 d-3 abbreviations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 MB87J2120, mb87p2020-a hardware manual page 12 page 13 part a - lavender and jasmine overview MB87J2120, mb87p2020-a hardware manual page 14 graphic controller overview overview page 15 1 overview 1.1 application overview the MB87J2120 "lavender" and mb87p2020-a ?asmine are colour lcd/crt graphic display control- lers (gdcs) 1 interfacing to mb91xxxx micro controller family and support a wide range of display devices. the architecture is designed to meet the low cost, low power requirements in embedded and especially in automotive 2 applications. lavender and jasmine support almost all lcd panel types and crts or other progressive scanned 3 moni- tors/displays which can be connected via the digital or analog rgb output. products requiring video/camera input can take advantage of the supported digital video interface. the graphic instruction set is optimized for minimal traffic at the mcu interface because it? the most important performance issue of co-processing graphic acceleration systems. lavender uses external connected sdram, jasmine is a compatible gdc version with integrated sdram (1mbyte) and comes with additional features. lavender and jasmine support a set of 2d drawing functions with built in pixel processor, a video scaler interface, units for physical and direct video memory access and a powerful video output stream formatter for the greatest variety of connectable displays. figure 1-1 displays an application block diagram in order to show the connection possibilities of jasmine. for lavender external sdram connection is required in addition. 1.2 jasmine/lavender block diagram figure 1-2 shows all main components of jasmine/lavender graphic controllers. the user logic bus con- troller (ulb), clock unit (cu) and serial peripheral bus (spb) are connected to the user logic bus inter- face of 32 bit fujitsu risc microprocessors. 32 and 16 bit access modes are supported. 1. the general term ?raphic display controller or its abbreviation ?dc is used in this manual to identify both devices. mainly this is used to emphasize its common features. 2. both display controllers have an enhanced temperature range of -40 to 85 o c. 3. tv conform output (interlaced) is also possible with half the vertical resolution (line doubling). table 1-1: gdc components shortcut meaning main function ccfl cold cathode fluorescence lamp cold cathode driver for display backlight cu clock unit clock gearing and supply, power save dac digital analog converter digital to analog conversion for analog display dpa (part of dipa) direct physical memory access memory mapped sdram access with address decoding gpu graphics processing unit frame buffer reader which converts to video data format required by display dfu (part of gpu) data fetch unit graphic/video data acquisition ccu (part of gpu) colour conversion unit colour format conversion to common intermediate overlay format MB87J2120, mb87p2020-a hardware manual page 16 lsa (part of gpu) line segment accumulator layer overlay bsf (part of gpu) bitstream formatter intermediate format to physical display format converter, sync generation ipa (part of dipa) indirect physical memory access sdram access with command register and fifo mau (part of pp) memory access unit pixel access to video ram mcp (part of pp) memory copy memory to memory copying of rectan- gular areas pe (part of pp) pixel engine drawing of geometrical ?ures and bit- maps pp pixel processor graphic oriented functions sdc sdram controller sdram access and arbitration spb serial peripheral bus serial interface (master) ulb user logic bus (see mb91360 series speci?ation) address decoding, command control, ?g, interrupt and dma handling table 1-1: gdc components shortcut meaning main function host mcu e.g. vpx3220a, saa7111a mb91xxxx video scaler digital video rgb analog mb87p2020 or MB87J2120 (jasmine or lavender) figure 1-1: application overview graphic controller overview overview page 17 the ulb provides an interface to host mcu (mb91360 series). the main functions are mcu (user logic bus) control inclusive wait state handling, address decoding and device controls, data buffering / synchro- nisation between clock domains and command decoding. beside normal data and command read and write operation it supports dma flow control for full automatic data transfer from mcu to gdc and vice versa. also an interrupt controlled data flow is possible and various interrupt sources inside the graphics controller can be programmed. the clock unit (cu) provides all necessary clocks to module blocks of gdc and a fr compliant (ulb) interface to host mcu. main functions are clock source select (xtal, ulb clock, display clock or special pin), programmable clock multiplier/divider with apll, power management for all gdc devices and the generation of synchronous reset signal. for fujitsu internal purposes one independent macro is build in the gdc asic, the serial peripheral bus (spb). it? a single line serial interface. there is no interaction with other gdc components. all drawing functions are executed in pixel processor (pp). it consists of three main components pixel en- gine (pe), memory access unit (mau) and memory copy (mcp). all functions provided by these blocks are related to operations with pixel addresses {x, y} possibly enhanced with layer information. gdc sup- ports 16 layers by hardware, four of them can be visible at the same time. each layer is capable of storing vic video interface controller yuv-/rgb-interface to video grabber table 1-1: gdc components shortcut meaning main function user logic bus interface (ulb) command control video dacs clock unit ccfl spb (cu) anti aliasing filter (aaf) pixel engine (pe) ccu mau mcp dipa vic dfu lsa bsf sdram controller (sdc) pixel processor (pp) user logic bus xtal bus pix video scaler interface light back analog video digital video serial mb87p2020 (jasmine) graphic processing unit (gpu) MB87J2120 (lavender) embedded dram (1mbyte) or external sdram (8mbyte) figure 1-2: component overview for lavender and jasmine graphic controllers MB87J2120, mb87p2020-a hardware manual page 18 any data type (graphic or video data with various colour depths) only restricted by the bandwidth limitation of video memory at a given operating frequency. drawing functions are executed in the pe by writing commands and their dedicated parameter sets. all commands can be taken from the command list in section 4.2. writing of uncompressed and compressed bitmaps/textures, drawing of lines, poly-lines and rectangles are supported by the pe. there are many spe- cial modes such as duplicating data with a mirroring function. writing and reading of pixels in various modes is handled by mau. single transfers and block or burst transfers are possible. also an exchange pixel function is supported. with the mcp unit it is possible to transfer graphic blocks between layers of the same colour representation very fast. only size, source and destination points have to be given to duplicate some picture data. so it of- fers an easy and fast way to program moving objects or graphic libraries. all pp image manipulation functions can be fed through an antialiasing filter (aaf). this is as much faster than a software realisation. due to the algorithm which shrinks the graphic size by two this has to be com- pensated by doubling the drawing parameters i.e. the co-ordinates of line endpoints. dipa stands for direct/indirect physical access. this unit handles rough video data memory access with- out pixel interpretation (frame buffer access). depending on the colour depth (bpp, bit per pixel) one or more pixel are stored in one data word. dpa (direct pa) is a memory-mapped method of physical access. it is possible in word (32 bit), half word (16 bit) or byte mode. the whole video memory or partial window (page) can be accessed in a user definable address area of gdc. ipa (indirect pa) is controlled per ulb command interface and ipa access is buffered through the fifos to gain high access performance. it uses the command getpa and putpa, which are supporting burst accesses, possibly handled with interrupt and dma control. for displaying real-time video within the graphic environment both display controllers have a video inter- face for connection of video-scaler chips, e.g intermetall? ic vpx32xx series or phillips saa711x. addi- tionally the video input of jasmine can handle ccir standard conform digital video streams. several synchronisation modes are implemented in both controllers and work with frame buffering of one up to three pictures. with line doubling and frame repetition there exist a large amount of possibilities for frame rate synchronisation and interlaced to progressive conversion as well. due to the strict timing of most graphic displays the input video rate has to be independent from the output format. so video data is stored as same principles as for graphic data using up to three of the sixteen layers. the sdc is a memory controller, which arbitrates the internal modules and generates the required access timings for sdram devices. with a special address mapping and an optimized algorithm for generating control commands the controller can derive full benefit from internal sdram. this increases performance respective at random (non-linear) memory access. the most complex part of gdc is its graphic data processing unit (gpu). it reads the graphic/video data from up to four layers from video memory and converts it to the required video output streams for a great variety of connectable display types. it consists of data fetch unit (dfu), colour conversion unit (ccu) which comes with 512 words by 24-bit colour look up table, line segment accumulator (lsa) which does the layer overlay and finally the bitstream formatter (bsf). the gpu has such flexibility for generating the data streams, video timings and sync signals to be capable of driving the greatest variety of known dis- play types. additional to the digital outputs video dacs provide the ability to connect analog video destinations. a driver for the displays cold cathode fluorescence lamp (ccfl) makes the back light dimmable. it can be synchronized with the vertical frequency of the video output to avoid visible artefacts during modulating the lamp. graphic controller overview features and functions page 19 2 features and functions table 2-1: lavender and jasmine features in comparison MB87J2120 (lavender) mb87p2020-a (jasmine) general features 2m words x 32 bit external sdram (64 mbit) ?o internal sdram 256k words x 32 bit internal sdram (8 mbit) package: bga-256p-m01 package: fpt-208p-m06 chip select sharing for up to four gdc devices synchronized reset (needs applied clock) immediate asynchronous reset, synchronized reset release pixel manipulation functions 2d drawing and bitmap functions - lines and polygons - rectangular area - uncompressed bitmap/1bit pixelmask - compressed bitmap (tga format)/1bit pixelmask pixel memory access functions - put pixel - put pixel fc (fixed colour) - put pixel word (packed) - exchange pixel - get pixel layer register for text and bitmap functions layer register enhancement for drawing func- tions (simplifies pixel addressing) copy rectangular areas between layers anti aliasing filter (aaf) - resolution increase by factor 2 for each di- mension (2x2 filter operator size) additional 4x4 aaf operator size display free programmable bitstream formatter for a great variety of supported displays (single/ dual/alternate scan): - passive matrix lcd (single/dual scan) - active matrix (tft) displays - electroluminescent displays - field emission displays - tv compatible output - crts... additional twin display mode feature (simul- taneous digital and analog output without lim- itation of dis_d[23:16] that carry special sync signals). 24 bit digital video output (rgb) MB87J2120, mb87p2020-a hardware manual page 20 on-chip video dac, 50m samples/s (dot clock) flexible three-stage sync signal programming (trigger position/sequence, combining and delay) for up to 8 signal outputs colour keying between two limits brightness modulation for displays with a cold cathode fluorescence lamp back-light display resolution/drawing planes up to 16383 pixels for each dimension 4 layer + background colour simultaneous display and graphic overlay, programmable z-order blinking, transparency and background attributes free programmable display section of a layer separable colour lut with 256 entries x 24 bit colour lut expansion to 512 entries duty ratio modulation (drm) for pseudo hue/grey levels hardware support for 16 layers, usable for graphic/video without restrictions performance sharing with adjustable priorities and configurable block sizes for memory transfers en- able maximal throughput for a wide range of applications variable and display independent colour space concept: layers with 1, 2, 4, 8, 16, 24 bit per pixel can be mixed and converted to one dis- play specific format (logical-intermediate- physical format mapping) additional gpu a yuv to rgb converter in order to allow yuv coded layers additional gamma correction rams are in- cluded (3x256x8bit) physical sdram access memory mapped direct physical access for storage of non-graphics data or direct image access indirect physical memory access for high bandwidth multipurpose data/video memory access mcu interface 32/16 bit mcu interface, designed for direct connection of mb91xxxx family (8/16/32bit access) dma support (all mb91xxxx modes) interrupt support video interface video interface vpx32xx series by micronas intermetall, phillips saa711x and others additional ccir conform input mode video synchronization with up to 3 frame buffers clock generation flexible clocking concept with on-chip pll and up to 4 external clock sources: - xtal - ulb bus clock - pixel clock - additional external clock pin (mode[3]/rclk) separate power saving for each sub-module table 2-1: lavender and jasmine features in comparison MB87J2120 (lavender) mb87p2020-a (jasmine) graphic controller overview clock supply and generation page 21 3 clock supply and generation gdc has a flexible clocking concept where four input clocks (osc_in/out, dis_pixclk, ulb_clk, rclk) can be used as clock source for core clock (clkk) and display clock (clkd). the user can choose by software whether to take the direct clock input or the output of an apll independ- ent for core- and display clock. both output clocks have different dividers programmable by software (div x for clkd and div z for clkk). the clock gearing facilities offer the possibility to scale system performance and power consumption as needed. beside these two configurable clocks (clkk and clkd) gdc needs two additional internal clocks: clkm and clkv (see also figure 3-1). clkv is exclusively for video interface and is connected to input clock pin vsc_clkv. clkm is used for user logic bus (ulb) interface and is connected to input clock ulb_clk. as already mentioned ulb_clk can also be used to build clkk and/or clkd. table 3-1 shows all clocks used by gdc with their requirements. table 3-1: clock supply clock type symbol requirements unit min typ max xtal clock input osc_in, osc_out 12 a -64mhz reserve clock input rclk ulb_clk b -64mhz ulb clock input ulb_clk - - 64 mhz pixel clock input dis_pixclk - - 54 mhz video clock input vsc_clkv - - 54 c mhz core clock internal clkk ulb_clk - 64 mhz display clock internal clkd - - 54 mhz video clock internal clkv - - 54 c mhz ulb clock internal clkm - - 64 mhz (jasmine only) mul y div x div z inv inv osc_in/out dis_pixclk ulb_clk rclk pll clock direct clock vsc_clkv system clock prescaler pixel clock prescaler invert option invert option clkk clkm clkv clkd apll figure 3-1: clock gearing and distribution MB87J2120, mb87p2020-a hardware manual page 22 a. if used as pll input. apll input frequency has to be at minimum 12 mhz, regardless which clock is routed to apll. b. if used as direct clock source bypassing the apll, the user should take care that resulting core clock fre- quency is above or equal to mcu bus interface clock. be aware of tolerances! c. the video interface is designed to achieve 54 mhz but there is a side condition that video clock should be smaller than half of core clock. graphic controller overview register and command overview page 23 4 register and command overview 4.1 register overview the gdc device is mainly configurable by registers. these configuration registers are mapped in a 64 kbyte large address range from 0x0000 to 0xffff. it is possible to shift this register space in steps of 64 kbyte by the mode[1:0] pins in order to connect multiple gdc devices. above this 4*64 kbyte = 256 kbyte address range the sdram video memory could be made visible for direct physical access. at byte address 0x1f:ffff gdc memory map ends with a total size of 2 mbyte. 4.2 command overview the command register width is 32 bit. it is divided into command code and parameters: partial writing (halfword and byte) of command register is supported. command execution is triggered by writing byte 3 (code, bits [7:0]). thus parameters should be written before command code. not all commands need parameters. in these cases parameter section is ignored. in table 4-1 all commands are listed with mnemonic, command code and command parameters (if neces- sary. this is only a short command overview, a more detailed command list can be found in appendix. table 4-1: command list mnemonic code function addressed device bitmap and texture functions putbm 01 h store bitmap into video ram pixel processor putcp 02 h store compressed bitmap into video ram puttxtbm 05 h draw uncompressed texture with ?ed foreground and background colour puttxtcp 06 h draw compressed texture with ?ed foreground and background colour drawing functions (2d) dwline 03 h "draw a line" - calculate pixel position and store linecol into video ram pixel processor dwpoly 0f h "draw a polygon" - draws multiple lines between de?ed points, see dwline dwrect 04 h "draw an rectangle" - calculate pixel addresses and store rectcol into video ram pixel operations 0 7 code 31 parameters MB87J2120, mb87p2020-a hardware manual page 24 putpixel 07 h store single pixel data into video ram pixel processor putpxwd 08 h store word of packed pixels into video ram putpxfc 09 h store ?ed colour pixel data in video ram getpixel 0a h load pixel data from video ram xchpixel 0b h load old pixel in output fifo and store pixel from input fifo into video ram memory to memory operations memcopy 0c h memory copy of rectangular area. transfer of bit- maps from one layer to another or within one layer. pixel processor physical framebuffer access putpa 0d h store data in physical format into video ram, with physical address auto-increment dipa getpa page 25 part b - functional descriptions MB87J2120, mb87p2020-a hardware manual page 26 page 27 b-1 clock unit (cu) MB87J2120, mb87p2020-a hardware manual page 28 clock unit functional description page 29 1 functional description 1.1 overview the clock unit (cu) provides all necessary clocks to gdc modules and an own interface to host mcu (mb91360 series) in order to have durable access even if ulb clocks switched off. the main functions of cu are: clock source select (oscillator, mcu bus clock, display clock and a reserve clock input) programmable clock muliplier with apll separate dividers for master (core) clock and pixel clock power management for all gdc modules generation of synchronized reset signal mb91360 series compliant (ulb) bus interface for clock setup figure 1-1 shows the overview of the clock unit. osc_in, dis_pixclk, ulb_clk and rclk 1 are pos- sible to use as input sources. both clock outputs of the main unit (masterclk and pixelclk) and two directly used clock inputs (ulb_clk and vsc_clkv) driving the clock gates unit which distributes to all connected gdc sub-modules. 1. mode[3] pin is used for rclk at lavender ulb ulb register set clock gates main unit reset generator user logic bus address decoder clk[con|pd]r_[rd|wr] rstx rstx clkconr masterclk pixelclk rstx mcp_clk_out vis_clkv_out spb_clkm_out ulb_clk_out ulb_clkm_out spb_clkmx_out ulb_clkmx_out lock clkpdr sw_rst sync_rstx_cu [15] inverted clock outputs not inverted clock outputs sync_rstx [11] pe_clkk_out mau_clkk_out k pp_clkk_out aaf_clkk_out dipa_clkk_out vis_clkk_out sdc_clkk_out ccfl_clkk_out gpuf_clkk_out gpum_clkk_out gpub_clkk_out pix_clkd_out ulb_clkkx_out osc_in dis_pixclk rclk ulb_clk vsc_clkv osc_in ulb_clk resetx _a _rdx _wrx _csx _clk _d _a _wrx _clk figure 1-1: block diagram of clock unit MB87J2120, mb87p2020-a hardware manual page 30 the gdc device has four different clock domains, that means clocks derived from four different sources. the largest part of the design runs at core clock which operates at the highest frequency driven by the mas- terclk output. thus normally the apll is used to provide a higher internal operation frequency. the next domain is the display output interface which operates at pixel clock frequency. for most applications it is recommended that this is the clock from osc_in pin, divided by two 1 . so the crystal oscillator has to be choosen to have a whole-numbered multiple of the display clock frequency. preferred routing is the di- rect clock source channel since some displays require a small clock jitter which is not able to provide by the apll. the other clocks for mcu interface (ulb_clk) and video interface (vsc_clkv) are not de- rived by the clock routing and generating part and used directly from the appropriate input pin. finally the generated source clocks of the for domains go to the clock gating/distribution module. there are gated clock buffers and inverters for each gdc module implemented. each module has it? own clock en- able flag which can be programmed for modules needed by the application only. this method saves power of not used functional blocks of gdc (refere to table 3-1). the configuration of cu is stored in two registers, clkconr and clkpdr, which are connected to user log- ic bus for writing and reading. the bus interface consists of an address decoder and circuitry for different access types (word, halfword and byte access over a 16 or 32 bit bus connection). 1.2 reset generation gdc works with an internally synchronized, low active reset signal. the global chip reset can be triggered by an external asynchronous reset or internally by software reset (configuration bit in clkpdr). the external triggered resetx results in resetting all gdc components including the clock unit, however software reset has no influence on cu internal registers. lavender synchronizes its external reset (rstx pin). reset is delayed until 4 clock cycles of each ulb_clk and osc_in are executed. this gives stability against spikes on the rstx line but has the dis- advantage of delayed reset response of lavender. for jasmine internal reset is active immediately after tying resetx low plus a small spike filter delay. due to the synchronization of resetx the internal reset state ends after 4 clock periods of osc_in and 4 clock periods of ulb_clk after releasing resetx pin. reset output rstx_sync for all internal gdc register states are synchronized with osc_in, however internal clock unit registers are synchronized with ulb_clk in addition. thus a minimum recovery time of 4 clock cycles of osc_in plus 4 cycles of ulb_clk is needed before writing to clock unit configuration registers is possible after resetx be- comes inactive. the reset generator of jasmine has a spike filter implemented, which suppresses short low pulses, typical smaller than 9 ns. under best case operating conditions (-40 deg. c; 2.7v; fast) maximum suppressed spike width is specified to 5.5ns. this is the maximum reset pulse width which did not result in resetting the gdc device. minimum pulse width for guaranteed reset is specified to 1 clock cycle of osc_in (80 ns typical). 1.3 register set table 1-1 listst the clock setup registers. clkconr (clock configuration register) is mainly for generation of the base clocks and the routing/selection from one of the four input sources. it controls the clock dividers and the use of the apll. the possibility to use a second clock path, called direct clock source, gives a high flexibility for using the apll either for masterclk or pixelclk generation or both. also the pin function of dis_pixclk can be defined in this register. if dis_pixclk is selected as clock source the pin should be configured as an input. upper 8 bits of clkconr are used as identification of the different gdc types. lavender is identified with reading back a ?x00? jasmine with a ?x01? use of dis_pixclk as pixel clock output and selection of dis_pixclk for the clock source can result in unintentional feedbacks and has to be avoided. 1. preferred is an even divider value to achive 50% clock duty clock unit functional description page 31 clkpdr (clock power down register) is a set of enable bits for the clocks provided to the dedicated gdc modules. a bit set to ? means the clock is enabled. if a module requires multiple clocks (inverted ones or different domains) the enable bit switches all these lines. additional clkpdr controls the work of the pll and gives status information about it? lock-state. also a global gdc reset function can be executed by setting a configuration bit of this register. table 1-1: cu registers register bit function description reset value clkconr [31:30] direct clock source 00 crystal oszillator (reset default) 01 pixel clock 10 mcu bus clock 11 reserved clock input (rclk a ) "00" [29:24] system clock prescaler (div z) [5:0] system clock prescaler value 0 [23:22] pll clock source 00 crystal oszillator (reset default) 01 pixel clock 10 mcu bus clock 11 reserved clock input (rclk) "00" [21:16] pll feedback divider (div y) [5:0] pll multiplier value 0 [15] system clock select 0 direct 1 pll output ? [14] pixel clock select 0 direct 1 pll output ? [13] inverted pixel clock 0 not inverted 1 inverted ? [12] output disable dis_pixclk 0 internal pixelclock (output) 1 external pixelclock (input), high-z output ? [11] reserved test operation b 0 normal operation 1 core clock output on pin spb_tst ? [10:0] pixel clock prescaler (div x) [10:0] pixelclock prescaler value 0 MB87J2120, mb87p2020-a hardware manual page 32 clkpdr 0 pp/pe pixel engine clock enable 0 = disable, 1 = enable ? 1 pp/mau clock enable 0 = disable, 1 = enable ? 2 pp/mcp clock enable 0 = disable, 1 = enable ? 3 aaf clock enable 0 = disable, 1 = enable ? 4 dipa clock enable 0 = disable, 1 = enable ? 5 vis clock enable 0 = disable, 1 = enable ? 6 sdc clock enable 0 = disable, 1 = enable ? 7 ccfl clock enable 0 = disable, 1 = enable ? 8 spb clock enable 0 = disable, 1 = enable ? 9 ulb clock enable 0 = disable, 1 = enable ? 10 gpu clock enable 0 = disable, 1 = enable ? 11 pll enable 0 = power down , 1 = run mode ? 12 vic clock invert (jasmine) 0 = not inverted, 1 = inverted ? pll lock (lavender) 0 = unlocked, 1 = locked (read only) c 13 reserved - ? 14 pll lock (jasmine) 0 = unlocked, 1 = locked (read only) b ? 15 global hw reset 0 = run mode, 1 = reset (write only) d ? [31:24] chip id 0 = lavender, 1 = jasmine (read only) - a.rclk is mapped on mode[3] at lavender. b.only applicable for jasmine c.normally all register bits are read-write. as the pll lock bit is status information only, no write access is possible to it. the lock bit is for test operation only and should not be used. d.read access results always in a value of ?? writing ? starts global hw reset function, writ- ing ??releases reset. table 1-1: cu registers register bit function description reset value clock unit apll specification page 33 2 apll speci?ation this informations are for implemented apll - u1pn741a. output range is given for apll output direct- ly, not for the divider outputs. the apll macro has an operating supply voltage (vdd) of 2.5 0.25v. the oscillation guaranteed frequen- cy range, maximum output frequency range and operating junction temperature range of the apll are shown in the table below. table 2-1: apll speci?ations operating junction temperature (tj) -40 to 125 deg. c voltage supply (vddi) 2.5 v +/- 0.25 v oscillation guaranteed frequency range 120 to 208.4 mhz maximum output frequency range a a.range in which oscillation may be possible. 5.77 to 598.1 mhz input frequency range 12 to 160 mhz table 2-2: apll characteristics for guaranteed design range input [mhz] fb divider out- put [mhz] phase skew [ps] duty [%] lock up time [us] jitter [ps] variations in output cycle [ps] power con- sumption [mw] 25 8 200 444 -448 100.5 87.9 70 176 -142 +130 -134 2.45 20 a a.operating temperature -20 ... 90 deg. c, operating voltage 2.5 v +/- 0.15 v. 8 160 540 -520 102.6 94 100 232 -168 +64 -170 1.9 13.217 b b.operating temperature -40 ... 125 deg. c. 10 132.17 1200 -1360 104.4 99.33 500 420 -246 +234 -278 1.88 12 b 10 120 1334 -1466 111.1 100.7 25 760 -560 480 -560 1.793 40 b 4 160 640 -700 105.9 101.5 20 350 -190 238 -304 1.147 23.5 b 8 188 740 -940 110.6 97.2 65 208 -162 160 -182 2.387 33 b 6 198 560 -540 100.4 88 165 172 -140 148 -180 2.397 160 b 1 160 240 -150 101.2 98.9 12 190 -140 110 -140 1.147 50 b 4 200 340 -280 98.5 87.7 26 148 -140 96 -120 2.45 20 b 10 200 3600 -4200 109.5 87.5 88 560 -360 420 -480 2.45 MB87J2120, mb87p2020-a hardware manual page 34 2.1 de?itions 2.1.1 phase skew maximum phase differences between reference clock and feed back clock measured by the ck pin of the pll and the feedback clock measured by the fb pin. 2.1.2 duty duty is the maximum and minimum values indicated by the ratio of a high pulse width to a low pulse with (t low to t high ) in one cycle of the output clock measured by the x pin of the pll. 2.1.3 lock up time lock up time is the time period that starts when the s pin of the pll changes from 0 to 1 and ends when the pll is locked. 2.1.4 jitter jitter is the maximum and minimum values for cycle variations (t 2 -t 1 ,t 3 -t 2 and t 4 -t 3 ) in two continuous cycles measured by the x pin of the pll. 2.1.5 variation in output cycle the max./min. time periods from the rising edge of the output clock to thenext rising edge measured by the x pin of the pll. observation points: 1400. the max./min values in t 1 , t 2 , t 3 and t 4 in above figure. fb ck x tt high low lock up time s l x t1 t2 t3 t4 clock unit apll specification page 35 2.1.6 maximum power consumption this is the maximum power consumption of the pll when pll is in locked state. the power dissipated by extrenal dividers is not included into this amount. 2.2 usage instructions input the clock of crystal oscillator level into the ck pin of the pll. variations in the input clock directly affects the pll output, leading to unconformity to the speci?ations. in addition to the normal power supply, the pll has a power supply for vc0 (pin name: apll_avdd, apll_avss). the vc0 power supply should be separate from other power supplies. MB87J2120, mb87p2020-a hardware manual page 36 3 clock setup and con?uration 3.1 con?urable circuitry clock configuration can be easily done by setting up both registers clkconr (clock configuration regis- ter) and clkpdr (clock power down register). clkconr mainly controls the setup of multiplexers and clock dividers in the main part of cu, which is shown in figure 3-1. clkpdr decides which clocks should be enabled and distributed to the appropriate modules, listed in table 2-1. during change of clkconr all enable bits in clkpdr[10:0] have to be turned off to attain spike protec- tion. table 3-1: mapping of clock sources, outputs and their enable bits clkpdr control bit clock source clock output 0|1|2|3 master a pixel processor (pp) 0 master pp: pixel engine 1 master pp: memory access unit 2 master pp: memory copy 3 master anti aliasing filter 4 master direct/indirect physical access 5 master video interface 5 video scaler (vsc_clkv) video interface 6 master sdram controller 7 master cold cathode fluorescence light 8 mcu bus (ulb_clk) serial peripheral bus 9 master user logic bus interface and com- mand controller 9 mcu bus (ulb_clk) user logic bus interface (tristate) apll div y div z div x xor xz_cu_pixck [12] default path clkconr clkpdr [ bits ] { bits } [23:22] lock {12} l x s a ck fb [21:16] [29:24] masterclk tst [10:0] [14] pixelclk tst pll clock direct tst_mas_clk tst_pix_clk [31:30] {11} [15] [13] osc_in dis_pixclk ulb_clk rclk figure 3-1: clock routing and configuration bits clock unit clock setup and configuration page 37 all clocks except vsc_clkv can be used as master or pixel clock source. vsc_clkv is for video inter- face dedicated use only. there are no special requirements for the quartz crystal parameters. at the case of overtone oscillation ad- ditional external inductance l?5-10uh and capacitor c?10pf are needed. two capacitors c=22pf have to be connected from the osc pins to ground in any case. figure 3-2 shows recommendet circuit. without a crystal oscillator connected (e.g. extrnal oscillator) the clock has to fed in the osc_in pin. 3.2 clock unit programming sequence this section gives a recommendation for the sequence for gdc clock configuration. in general the clock unit registers should be configured before all other gdc setup information. apply stable clock and do a hardware reset. write clkconr for the required mode. clock gates are disabled per reset default. switch on pll and optionally apply software reset (set bits [11] and [15] in clkpdr). clear software reset bit (optional, only if set before). wait until apll has stabilized and locked (lock-up time) 1 polling of lock bit is optional and not suf?ient that pll locking process has ?ished. this signal is for fujitsu test purpose only. apll lock-up state is reached if lock bit becomes stable ?? this is guaranteed after speci?d pll lock-up time of 500us. set required clock enable bits to open the clock gates. 10 master graphic processing unit 10 pixel clock a graphic processing unit a.master and pixel clock could be derived from one of four possible clock inputs (osc_in, dis_pixclk, ulb_clk, rclk/mode[3]) with or without use of the pll. 1.the lock up time measured from pll start (clkpdr_run=1) to lock state (clkpdr_lck=fixed 1). table 3-1: mapping of clock sources, outputs and their enable bits clkpdr control bit clock source clock output osc_in osc_out xq cc c? l? figure 3-2: crystal connection between pins osc_in and osc_out MB87J2120, mb87p2020-a hardware manual page 38 an example sequence for this procedure is listed below: 3.3 application notes the clock unit provides an internally synchronized reset signal to all gdc components. therefore it? nec- essary to have a stable clock applied to the osc_in pin during resetx is low and/or at least after release of resetx. otherwise the internal circuitry is not initialized properly or clock unstability after reset release can cause malfunction. with the direct clock source it? possible to use a external ulb_clk from the mcu or rclk as clock source for almost all internal gdc components. the apll is not able to handle jitter/variations in input clock. if the gdc should operate in single clock mode over ulb_clk driven by the mcu, ulb_clk and osc_in have to be bridged. in any case a clock has to feed in osc_in pin, otherwise the reset state would not be left. if system or pixel clock divider are initialized with an even value tis results in an odd divider value (value interpreted +1). in this situation the duty of the output clock is not even 1:1. most important this is for low values. in case of not even duty the high duration is smaller than the low duration. following table lists clock divider and duty relationship. table 3-2: clock division and resulting duty setup value divider duty 0 1 1/1 1 2 1/1 2 3 2/1 3 4 1/1 4 5 3/2 5 6 1/1 6 7 4/3 ... /* cu control information */ g0clkpdr = 0x00008800; // sw reset and apll enable g0clkpdr = 0x00000800; // release sw reset, clock gates are still closed g0clkcr = 0x010e8001; // master=xtal*15/2, pixclk=xtal/2 not inverted output g0clkpdr = 0x00000ef1; // enable gpu, ulb, ccfl, sdc, vic, dipa, pe figure 3-3: clock configuration procedure with reset page 39 b-2 user logic bus controller (ulb) MB87J2120, mb87p2020-a hardware manual page 40 user logic bus functional description page 41 1 functional description 1.1 ulb functions the ?ser logic bus interface (ulb) provides an interface to host mcu (mb91360 series). it is respon- sible for data exchange between mcu and the graphic display controllers (gdc) lavender or jasmine 1 . the communication between mcu and the display controller is register based and all display controller reg- isters are mapped in the mcu address space. the task of ulb is to organise write- or read accesses to different display controller components, including ulb itself, depending on a given address. for read accesses the ulb multiplexes data streams from other components and has to control the amount of needed bus wait states using mcu? ulb_rdy pin. the ulb provides also a command- and data interface to so called ?xecution devices (pixel processor (pp) and indirect memory access unit (ipa)). these execution devices are responsible for drawing com- mand execution or for the handling of sdram access commands. the data transfer to and from execution devices is always fifo buffered. in order to ensure a rapid data transfer between mcu and display control- ler ulb contains one input and one output fifo which are mapped to certain memory addresses within the display controllers memory space. ulb controls the mcu port of these fifos (write for input fifo and read for output fifo) while the ports to execution devices is controlled by the device itself. the command interface has a two stage pipeline so command and register preparation is possible during command execution of previous command. most commands can have an infinite amount of processing data. the fifos help to reduce the number of bus wait states. additionally to fifo data exchange direct access to sdram and to initialisation registers is managed by ulb. this direct sdram access maps the sdram physical into mcus address space. drawing functions use a logical address mode for sdram access. due to this direct (and also indirect via fifos) physical access to sdram it can also be used to store user data and not only layer data (bitmaps, drawing results). for direct sdram access (frame buffer or video ram) the display controller internal sdram bus arbi- tration influences the mcu command execution time directly via ulb bus wait states via ulb_rdy signal. therefore longer access times should be calculated for this kind of memory access. ulb is able to handle memory or register access operations concurrently to normal command execution (fifo based). beside normal data and command read and write operation ulb supports also dma flow control for full automatic (without cpu activity) data transfer from mcu to display controller or vice versa. only one di- rection at one time is supported because only one mcu-dma channel is utilised. also an interrupt control- led data flow based on programmable fifo flags is possible. in both cases the ulb bus is used for data transfer. ulb offers a set of some special registers controlling the display controller behaviour or show the state of the controller with respect to mcu: flag-register flag-behaviour-register interrupt-mask-register interrupt-level-register dma-control-register command register sdram access settings debug registers ulb also provides an interrupt controller that can be programmed very flexible. every flag can cause an interrupt (controllable by interrupt-mask-register) and for jasmine it is selectable whether the interrupt for a certain flag is level or edge triggered. furthermore for every flag the programmer can determine whether the flag is allowed to be reset by hardware or not (static or dynamic flag behaviour). 1. the term display controller?is used as generic name for lavender and jasmine and covers both devices. MB87J2120, mb87p2020-a hardware manual page 42 the ulb internal dma controller is able to use all dma modes supported by mb91360 series. it operates together with input or output fifo and uses programmable limits. in demand mode the controller calculates the amount of data to transfer by its own. in other modes the programmer has to ensure that enough space is available in input fifo so that no data loss can occur. because of different clock frequencies for ulb bus and display controller core clock an important ulb function is the data synchronisation between these two clock domains. a asynchronous interface is offered by ulb which allows independent clocks for ulb and core as long as ulb clock is equal or slower than core clock. in order to adjust the display controller? operation mode so called ?ode pins?( mode[3:0] ) are used: the display controller can also act as an 16 bit device from mcu? point of view. in this case ulb con- verts write data from 16 to 32 bit and read data from 32 back to 16 bit in order to hide interface param- eters to internal display controller components. the data mode can be set by mode[2] . up to four display controllers can join one chip select signal. so it is necessary to set the controller ?umber by mode[1:0] . ulb takes care about correct address decoding depending on this ?umber? in order to allow flexible pcb layout some signals can be inverted and can be set to tristate. for jasmine the inversion of ulb_rdy is controlled by mode[3] . for more details about signal settings see chapter 1.3.4. in short terms the main functions of ulb are: mcu (user logic bus) bus control inclusive wait state handling for read access via ulb_rdy pin address decoding and control of other display controller components data buffering and synchronisation between different clock domains command decoding and execution control flag handling programmable interrupt handling programmable dma based input/output fifo control 16/32 bit conversion for writing and reading 1.2 ulb overview figure 1-1 shows the block diagram of ulb top level design. table 1-1 gives an overview on main functions of ulb top level modules. important modules will be ex- plained in more detail in the following chapters. table 1-1: top level modules of ulb name main functions input sync synchronization for write signals (mcu -> display controller) interrupt controller (ic) flag register management management of different interrupt sources user de?able interrupt source masking and trigger condition interrupt signal generation (jasmine: programmable length for edge request) address decoder (ad) handling of other display controllers using the same chip select signal address decoding and calculation for direct sdram access address segmentation management control of mcu data i/o as result of address decoding activation of command control, register control unit (ctrl) and direct physical memory access units (dpa) as result of address decoding user logic bus functional description page 43 command controller (cc) command decoding management of command execution condition decoding for control command execution i/o controller (ioc) read/write control for data part of user logic bus access control and status signal generation for data fifos and registers for mcu and display controller site clock domain synchronization for read data bus (display con- troller->mcu) bus multiplexing for display controller read data busses dma controller (dmac) dma ?w control for mcu site of input and output fifo register file (rf) storing of user de?able parameters for ulb table 1-1: top level modules of ulb name main functions interface ctrl data interface dpa data mb91f361 interrupt gdc i/o?ring ulb dma controller register file interrupt controller to dpa to ctrl device control ctrl command controller ififo read ofifo write i/o controller ififo mcu clock domain core clock domain ofifo input sync address decoder rdy dma data in / out csx/rdx/wrx[3:0] address figure 1-1: block diagram for top level of ulb MB87J2120, mb87p2020-a hardware manual page 44 1.3 signal synchronisation between mcu and display con- troller 1.3.1 write synchronization for lavender and jasmine the data flow inside the ulb is divided in write- and read-direction. these data directions are completely independent. that? why there are two synchronisation points for ulb bus signals; one for write direction and one for read direction. the first synchronisation point is located inside ?nput sync and is responsible for all ulb signals coming from mcu ( ulb_csx,ulb_wrx, ulb_rdx, ulb_a, d_in ). other incoming mcu signals for dma are used directly inside the dma controller which is partly clocked by ulb clock. note: for a correct operation of input synchronization ulb clock has to be equal or slower than display controller core clock. note that all tolerances for clocks should be taken into consideration. in jasmine a configurable sample behaviour for input signals has been introduced. in order to avoid inter- ferences from external bus different sample modes can be set. it can be chosen between four different sam- ple modes. each mode combines one or more out of three sample points distributed over one bus cycle. figure 1-2 shows these sample points. note that the sample points are only valid for control signals ( ulb_csx,ulb_wrx, ulb_rdx ); address and data bus are only sampled at point 2 (see figure 1-2). the sampling is equal for read and write accesses. the sample mode can be set in register ifctrl_smode (see also table 2-1); table 1-2 shows all possible settings and sample modes. table 1-2: control signal sample modes for jasmine mode ifctrl_smode[1:0] comment 3 point mode [default] 00 all three sample points have to have the same value 2 of 3 point mode 01 two out of three sample points have to have the same value (majority decision) ulb_cs ulb_clk ulb_a ulb_d ulb_wrx[n] ulb_rdx sample point: 1 02 figure 1-2: bus cycle sample points for jasmine user logic bus functional description page 45 beside different sample modes jasmine? input synchronisation circuit contains priority logic to distinguish between read or write access in the case that both control signals ( ulb_rdx and ulb_wrx[n] ) were de- tected. because the i/o controller (ulb read path) may detect a read access the ulb_rdx signal for read access has the highest priority. in lavender a different input synchronisation circuit is implemented were always one sample point is used 1 . 1.3.2 read synchronization for lavender and jasmine the ulb_rdy signal is gated by ulb_csx . this means that the ulb_rdy sig- nal goes immediately high after ulb_csx=? has been detected. it can not be ensured that correct data are transferred to mcu in this case. a protection against wrong tristate bus control signal switching is implemented in ulb. the bus driver is only valid if ulb_csx=? and ulb_rdx=? . in all other cases ulb data bus is not driven. additionally to the described safety mechanisms which are implemented in both devices (lavender and jas- mine) jasmine has a programmable timeout for ulb_rdy signal generation. therefore a counter is imple- mented which is loaded with the value set in register rdyto_rdyto[7:0] at the beginning of a bus read cycle 2 . if the read value does not arrive within the counter runtime 3 ulb_rdy signal is forced to ? and the mcu can finish the bus read cycle. note that in this case data transferred to mcu are not valid. the occurrence of a ulb_rdy timeout is reflected in the flag flnom_erdy ( ulb_rdy timeout error; see also flag description in appendix) which has been implemented in jasmine. additionally to the error flag the ad- dress where the timeout error occurred is stored in register rdyaddr_addr[20:0] in order to allow an application running on mcu an error handling. the ulb_rdy timeout counter can be disabled by turning rdyto_rdtoen off (set to ??. in regular operation mode a ulb_rdy timeout can only occur if no memory bandwidth can be allocated by the device handling the read request. because the command execution is fifo buffered (see also chapter 1.1) and a read access from fifo always returns a value within short time no timeout error can oc- cur in normal command execution. only a direct mapped memory read access (dpa read access; see also chapter 1.4.3) in conjunction with very limited bandwidth may cause a ulb_rdy timeout error in normal operation. beside this reason for timeout error in normal operation also disturbed bus transfers or bad signal integrity may cause a timeout error. 1.3.3 dma and interrupt signal synchronization the dma input and output signals ( ulb_dreq, ulb_dack, ulb_dstp ) are used/generated in the dma controller which is partly operating at ulb clock. because these signals are generated in this part no synchronisation is necessary. the signals influencing the dma signal generation have to be synchronized from lavender/jasmine core to ulb domain. for more details regarding dma see chapter 1.7. another signal which needs to be transferred from lavender/jasmine core to ulb clock domain is the in- terrupt request signal ( ulb_intrq ). in chapter 1.6 a detailed description of interrupt signal generation is given. the synchronisation of the interrupt request signal is different between lavender and jasmine. jas- 2 point mode 10 sample point 1 and 2 (see ?ure 1-2) have to have the same value 1 point mode 11 sample point 1 determines result value 1. for lavender sample point 1 is used to catch signals within ulb clock domain. afterwards the caught sig- nals will be sampled with core clock. as a result the real sample point depends on clock ration between ulb and core clock. 2. a ulb bus read cycle is detected when ulb_csx =0 and ulb_rdx =0. 3. the runtime ends when the counter reached value ?? table 1-2: control signal sample modes for jasmine mode ifctrl_smode[1:0] comment MB87J2120, mb87p2020-a hardware manual page 46 mine supports level and edge triggered interrupt requests with programmable edge length while lavender is only supporting level triggered interrupt requests. in level triggered interrupt mode the interrupt request is only reset if the flag which causes this interrupt is also reset. because of software flag reset the request signal is stable for a very long time and no internal handshake mechanism is needed. in difference to level triggered interrupt edge triggered interrupt reacts on a rising edge of a flag. this edge causes a pulse of programmable length on interrupt request signal. 1.3.4 output signal con?uration in order to allow flexible system integration both display controllers allow configuration for some output signals to mcu. this includes an option to signal inversion and a tristate control in order to allow external pull up resistors. if the tristate control is enabled the according pin drives tristate (?? instead of high level (?? while low level (?? is driven in any case (emulated open drain). table 1-3 contains the settings for all configurable signals ( ulb_dreq , ulb_dstp and ulb_intrq ) valid for both display controllers. in jasmine additionally ulb_rdy can be controlled by special pins 1 . note that there are differences in controlling the signal behaviour between lavender and jasmine. while in lavender the signals ulb_dreq , ulb_dstp and ulb_intrq are controlled together with two register bits for tristate and inversion ( dmaflag_tri and dmaflag_inv ) in jasmine every signal has its own control (see table 1-3). 1.4 address decoding 1.4.1 overview the useable address space for display controller chip select signal ( ulb_csx )is2 21 byte (2 mbyte) and the available space is divided into one configuration register space where all configuration registers are lo- cated and one sdram space were sdram windows can be established and accessed this space is con- figurable. figure 1-3 shows the address space with the default settings for sdram space of lavender and jasmine. table 1-3: control for feedback signals signal setting lavender jasmine ulb_dreq tristate register: dmaflag_tri register: ifctrl_drtri (1: tristate) invert register: dmaflag_inv register: ifctrl_drinv (1: invert) ulb_dstp tristate register: dmaflag_tri register: ifctrl_dstri (1: tristate) invert register: dmaflag_inv register: ifctrl_dsinv (1: invert) ulb_intrq tristate register: dmaflag_tri register: ifctrl_inttri (1: tristate) invert register: dmaflag_inv register: ifctrl_intinv (1: invert) ulb_rdy tristate no control possible pin: rdy_trien (1: tristate) invert no control possible pin: mode[3] (1: invert) 1. a register based control is not possible because read accesses would potentially not work in this case or the mcu may hang with a wrong signal. user logic bus functional description page 47 jasmine and lavender support up to four devices for one chip select ( ulb_csx ) signal. also a mixed en- vironment with different display controllers is possible. register and sdram space are used by all con- nected display controllers. figure 1-4 shows a possible scenario for four display controllers which treats only as an example. many other configurations for sdram space are possible while register space is fixed configured. 1.4.2 register space the size and location of configuration registers for every display controller is fixed. the size is set to 64 kbyte for every display controller and the location is specified by mode pins ( mode[1:0] ) as de- scribed in table 1-4. table 1-5 shows the register space for one display controller decoded by ulb address decoder. this decod- er has a built in priority for the case of overlapping address areas. one display controller reserves the register space for other controllers. because the address decoder has a decoding priority it is not possible to overlay the register space for other controllers with sdram windows. table 1-4: address ranges for register space of different display controllers controller number mode[1:0] address range 0 00 00000h...0ffffh 1 01 10000h...1ffffh 2 10 20000h...2ffffh 3 11 30000h...3ffffh sdram space 0k 64k 256k 1m 2m configurable 0x00040000 0x00100000 0x000c0000 register space for gdc0 0x00010000 0x001fffff not configured video memory window register space for gdc0 jasmine default configuration 768k not configured lavender default configuration video memory register space register space for other gdc register space for other gdc figure 1-3: address space for lavender and jasmine with default configuration MB87J2120, mb87p2020-a hardware manual page 48 within the upper part of register space (address range 0xfc00-0xffff) for one display controller a re- served area is located. this area is needed for internal and/or external devices with the same ulb_csx sig- nal as the display controller which have their own address decoders. internal devices using this area are currently the clock unit (cu) and the serial peripheral bus driver (spb) (see also table 1-5). note that lavender register space is compatible to jasmine register space. only new registers for new func- tions were added or not needed registers were removed but the same function can be found on the same reg- isters. there are only a few exceptions (see register list for more details). table 1-5: register address space for gdc priority address range ulb addressed component target component/register 1 0x0000 ulb command register 0x0004 input fifo 0x0008 output fifo 0x000c flag register (normal access) a 0x0010 flag register (reset access) a 0x0014 flag register (set access) a 0x0018 interrupt mask (normal access) a 0x001c interrupt mask (reset access) a 0x0020 interrupt mask (set access) a register space sdram space sdram gdc 0 sdram gdc 1 sdram gdc 2 sdram gdc 3 gdc 0: gdc_offset0 gdc 0: gdc_size0 gdc 3: gdc_offset1 gdc 3: gdc_size1 gdc 0: gdc_size1 gdc 0: gdc_offset1 gdc 1: gdc_offset0 gdc 1: gdc_size0 gdc 1: gdc_offset1 gdc 1: gdc_size1 64k 128k 192k 256k 2 21 gdc 0 (64k) gdc 1 (64k) gdc 2 (64k) gdc 3 (64k) csx mode[1:0] gdc 0: sdram_offset0 gdc 0: sdram_offset1 gdc 0: gdc_size1 gdc 0: gdc_size0 gdc 3: sdram_offset1 gdc 3: gdc_size1 gdc 1: sdram_offset1 gdc 1: gdc_size1 gdc 1: sdram_offset0 empty space gdc 1: gdc_size0 figure 1-4: address space example for four graphic display controller (gdc) devices user logic bus functional description page 49 1.4.3 sdram space for direct sdram access two independent windows can be set within ulb for one display controller. the term ?indow means that a part of the sdram memory can be blended into the mcu address space. win- dows can be set up by defining window parameters within ulb? register space. before an established win- dow is useable the sdram space has to be enabled. the following parameters can be used to set up sdram windows (see also figure 1-4): 2 0xfc00 - 0xfc07 reserved area cu 0xfc08 - 0xfcff empty c 0xfd00 - 0xfd0f spb 0xfd10 - 0xffff empty c 3 0x0024 - 0x009f ctrl ulb 0x0100 - 0x0217 sdc 0x1000 - 0x1243 gpu - ldr 0x1300 - 0x1383 gpu - mdr 0x1400 - 0x140f gpu - mtx b lavender: 0x2000 - 0x23ff jasmine: 0x2000 - 0x27ff gpu - clut 0x2800 - 0x2bff gpu - gamma b 0x3000 - 0x3263 gpu - dir 0x3270 - 0x3273 gpu - sdc 0x4000 - 0x403b vic 0x4100 - 0x4133 pp 0x4200 - 0x420b dipa 0x4400 - 0x4407 ccfl 0x4500 - 0x450f aaf 4 other display controllers (see table 1-4) - empty c 5 sdram window1 range dpa sdram 6 sdram window0 range dpa sdram 7 empty area within sdram space - empty c a. refer to section 1.6 for an explanation of register access types. b. jasmine only c. a special ?mpty signal is generated because the ulb is not allowed to drive the data bus for a read access inside an empty area while the i/o controller detects a valid read access. table 1-5: register address space for gdc priority address range ulb addressed component target component/register MB87J2120, mb87p2020-a hardware manual page 50 wndof0_off , wndof1_off to define an offset in mcu address space wndsz0_size , wndsz1_size to define sdram window size for mcu and sdram address space wndsd0_off , wndsd1_off to define sdram offset all these parameters are explained in detail in table 2-1. as displayed in figure 1-3 2 mbyte (2 21 byte) - 256 kbyte (register space) = 1.75 mbyte can be used for both windows within mcu address space. this address space is equal in lavender and jasmine while the available sdram memory differs according to table 1-6. for jasmine it is possible to map the entire sdram memory into ulb? sdram space in order to have linear access to sdram. for lavender only parts of sdram can be mapped to mcu address space at one given time but a dynamic reconfiguration is possible. all to one chip select signal connected display controllers share the available sdram space (1.75 mbyte). there is no additional restriction about the size or order of sdram windows of different controllers. figure 1-4 gives just one example but many more configurations are possible. gaps between sdram windows for a particular display controller are handled by ulb as well as sdram windows of other controllers. both possibilities can not be distinguished by the address decoder and they produce an empty space hit which means that no data are driven for read access and a write access is simply ignored. within one display controller overlapping sdram windows are allowed; this is controlled by address de- coder priority according to table 1-5. this is not true for sdram windows from different controllers. write access is possible; the value is written to all sdram windows mapped to this address. read access is not possible and can damage display controller or mcu because no ulb bus driving control is available be- tween different display controllers. note: there is no control for overlapping windows of more than one display controller within sdram space. reading from such an area can damage display controller or mcu. with help of sdram windows the sdram memory can be written or read with physical sdram ad- dresses via a display controller component called ?pa (direct physical memory access). in difference to this addressing mode all drawing functions use a logical address (pixel coordinates with (layer, x, y)). the sdram controller (sdc) within the display controller maps the logical pixel coordinates to a physical sdram address. this mapping is block based 1 and differs between lavender and jasmine since it depends on sdram architecture. a picture previously generated by pixel processor (pp) with help of drawing/bit- map commands can not be read back in a linear manner because the logical to physical address mapping has to be performed in order to get the right physical address for a given logical address (layer, x, y). also for writing graphics which should be displayed by the graphic processing unit (gpu) of a display control- ler the logical to physical mapping has to be taken into account. the easiest and most portable way (between lavender and jasmine) to write or read to/from logical address space is to use pp commands. in this case the mapping is done automatically and controller independent in hardware. the physical sdram windows can be used for any kind of data as long they do not present a picture which should be displayed by gpu. within one sdram window a linear access to the data is possible. because the start address of each layer can be set in physical sdram addresses it is possible to divide the sdram memory between layer and user data. note that there is no layer overrun control implemented in hardware within the display controller. while the sdram memory windows are mapped into the mcu address space the access is basically or- ganised as normal register access. but there are some important differences compared to normal register table 1-6: sdram memory for lavender and jasmine display controller sdram type sdram size lavender external 8 mbyte (64 mbit) jasmine internal 1 mbyte (8 mbit) 1. see sdc speci?ation for details. user logic bus functional description page 51 access. in difference to a register which can be exclusively accessed via its address in mcu address space the sdram is a resource that has to be shared between different display controller components. for one special sdram address many different ways to access it 1 can be found. therefore the sdram controller (sdc) arbitrates the accesses with different priority. for the direct sdram access does this mean that an undefined access time depending on current system load occurs. for read accesses the ulb_rdy signal is used for synchronisation between mcu and display controller while for write accesses a flag ( rdpa ) in flag register is used. the application has to poll this flag (wait as long as flag is zero 2 ) in order to synchronize write accesses to sdram window. as a side effect no dma access is possible for writing to sdram window. since for reading the ulb_rdy signal is used dma access is possible. while for jasmine every kind of access (word (32 bit), halfword (16 bit) and byte (8 bit)) is possible for reading and writing to/from sdram windows for lavender the read access is limited to word access (32 bit). table 1-7 sums up the access types and synchronization methods for both display controllers. because of the sdram access arbitration and write restrictions (flag polling) direct physical sdram ac- cess is not the best method to access sdram memory. for logical pixel access it is recommended to use command based and fifo buffered drawing and access functions (see chapter 1.5). a better way to access the sdram memory in physical addressing mode is the indirect sdram memory access via ipa component of display controller. this access type has some advantages compared to direct access: linear access to sdram without window limits fifo buffered transfer for effective sdram access address auto increment for reading and writing (burst sdram access) dma is possible for reading and writing 1.4.4 display controller bus access types (word, halfword, byte) the mb91xxxx mcus support three different bus access types for writing data from mcu to display con- troller. word access: write 32bit halfword access: write 16bit byte access: write 8bit for a more detailed description see mb91360 series hardware manual. 1. beside the described direct physical access the pixel processor can access with logical addresses and also an indirect physical access (command based) via ipa is possible. 2. make sure this ?g is set to dynamic behaviour. table 1-7: access type and synchronisation for lavender and jasmine access item lavender jasmine read synchronisation ulb_rdy signal ulb_rdy signal access word word, halfword, byte dma possible possible write synchronisation flag: flnom_rdpa flag: flnom_rdpa access word, halfword, byte word, halfword, byte dma - - MB87J2120, mb87p2020-a hardware manual page 52 table 1-8 lists the supported bus access types for different address areas and data modes for lavender and jasmine. note that for input and output fifo only word access is allowed. partial access can not be sup- ported because every write/read access increments internal pointers. figure 1-5 shows the address to register mapping within display controller. note that mb91xxxx mcus use big endian byte order which means that byte0 is the msb byte of a 32bit word. every byte has its own valid signal ( ulb_wrx[3:0] ) and a combination of valid signals determines the access type for a write access. for reading always the whole bus width depending on data mode (32- or 16bit) is transferred from display controller to mcu. the mcu selects the right part for current read instruction internally. 1.4.5 display controller data modes (32/16 bit interface) mb91xxxx mcus are able to communicate with devices with different bit sizes for data bus (32-, 16- and 8bit). normally lavender and jasmine were designed to act as 32bit devices. in this mode optimal perform- ance can be achieved and no internal data mapping is necessary. for special purposes both display controller can act as 16bit devices. with help of this mode parts of data bus can be made available as general purpose i/o pins or it may be possible to connect the display controller to a 16bit mcu 1 . table 1-8: bus access types for lavender and jasmine address area lavender jasmine 32bit data mode (mode[2]=1) 16bit data mode (mode[2]=0) 32bit data mode (mode[2]=1) 16bit data mode (mode[2]=0) input fifo word word word word output fifo word word word word register space a a. without input and output fifo word, halfword, byte word, halfword, byte word, halfword, byte word, halfword, byte sdram space word word word, halfword, byte word, halfword, byte wrx[2] wrx[1] wrx[0] address mcu address space display controller register wrx[3] 0 31 0 byte 0 byte 1 byte 2 byte 3 7 figure 1-5: mcu address to display controller register mapping user logic bus functional description page 53 beside the chip select signal configuration the 16 bit mode has to be selected for the address space of gdc(s) within mcu (see mb91360 series hardware manual) and with mode[2] pin (1: 32bit mode; 0: 16bit mode) at the display controller. all data transfer is done at msb side of data bus (pins: ulb_d[31:16] ). for write accesses does this mean that the bus signals ulb_wrx[2:3] are not used and should be set to ? in 16bit mode. table 1-9 gives an overview on used parts of data bus and used con- trol signals. in 16bit mode display controller address space (register and sdram space) is accessible as in 32bit mode. all bus access types described in chapter 1.4.4 can be used transparently. for word write accesses a mb91xxxx mcu splits the word access into two half word accesses with con- secutive addresses (increment two). since the display controller is able to handle half word and byte access- es (see chapter 1.4.4) 16bit mode can be supported. ulb contains a 16bit to 32bit converter that is responsible for adapting the 16bit bus access to internal 32bit bus structure. for the special case of fifo write access, where only word access is allowed not only in 16bit data mode, a special circuit was included which collects data for fifo write access. for read accesses ulb contains an additional converter which is responsible for converting the internal 32bit bus structure to external 16bit bus structure. for reading from output fifo a special circuit is in charge of reading only once from fifo and store the value for second read. 1.5 command decoding and execution 1.5.1 command and data interface to mcu the display controller command interface to mcu consists of the following main parts: command register where command instruction and command coded parameters can be written and read input- and output fifo for data exchange between mcu and display controller command dependent flags set by command controller debug register (read only) in order to watch command controller behaviour writing to command register has to be synchronised with command execution within display controller. in chapter 1.5.3 the structure and function of ulb? command decoder will be explained in detail. from pro- grammers point of view a flag ( flnom_cwen ) shows when command controller is able to receive a new command. an application should poll flnom_cwen flag before writing a new command or if interrupt controlled flow is implemented it should write commands only after flnom_cwen causes an interrupt. if the display controller ?pplication programming interface (api) is used, flag polling is done automati- cally before writing a new command. figure 1-6 shows the draw line command within api as an example. word gdc_cmd_dwln(dword line_col) { while (!g0flnom_cwen); /* wait until command write enable flag is set */ 1. this mcu should have the same bus interface as mb91xxxx or it has to be adapted with help of glue logic. table 1-9: signal connection for data modes signal data mode 32 bit data mode (mode[2]=1) 16 bit data mode (mode[2]=0) data bus ulb_d[31:0] ulb_d[31:16] write control signals ulb_wrx[3:0] ulb_wrx[1:0] set ulb_wrx[3:2] to ? (pull up) read control signals ulb_rdx ulb_rdx MB87J2120, mb87p2020-a hardware manual page 54 g0linecol = line_col; g0cmd = gdc_cmd_dwline; return (0); }; figure 1-6: draw line function as an example for command synchronisation the function in figure 1-6 writes only the command and command dependent register settings (in this case the line colour) to display controller, data transfer is done afterwards with help of fifos. word gdc_fifo_inp(dword *p_arr, word pcnt, byte dma_ena) { if (dma_ena == 0) { for (;pcnt>0;pcnt--) { while (g0flnom_iff != 0) g0flrst_iff = 1; g0ififo = *p_arr++; } } else { while ((dmaca0 & 0x80000000) != 0); dmasa0 = (io_lword)p_arr; // source address dmaca0 = dmaca0 | 0x80000000 | 0x0e100000 | pcnt; // dma operation enable g0dmaflag_en = 1; } return (0); }; figure 1-7: function to put data to display controllers input fifo to write data to input fifo another api function ( gdc_fifo_inp ) shown in figure 1-7 is used. gdc_fifo_inp takes a pointer to an array with values 1 which should be written to fifo together with data count. note that data count is not limited to fifo size, it can have any size. before data can be written to fifo (register: g0ififo ) the api function polls the full flag of input fifo in order to avoid fifo over- flow. the general flow for command execution and programming is discussed in chapter 1.5.2. optional the transfer can be controlled by dma (parameter dma_ena ). see chapter 1.7 for more details about dma and its handling within an application. both display controllers (lavender and jasmine) contains one input and one output fifo. the sizes of these fifos are different in display controllers and listed in table 1-10. every fifo has a set of flags which allow a flow control by an application. a detailed description of fifo flags can be found in flag description located in appendix. beside full and empty flag also two programma- ble limits (one lower and one upper limit) is implemented in both display controllers. these limits can be used to perform an action within an application based on a certain fifo load. this is often more efficient than polling the full or empty flag for every single data word. note that every flag can cause an interrupt (for details see chapter 1.6) so that fifo flags can be used for interrupt based flow control. a general rule for input fifo should be to check whether the free space in fifo is large enough for the amount of data words which have to be written. before reading from output fifo the application should check whether enough data are available in output fifo. this can be done by polling the fifo flags or by generating an interrupt. as a replacement of application based flow control (polling or interrupt) dma can be used to write data to input fifo or read data from output fifo. for dma the hardware takes care about fifo load but an ap- 1. dword ?means 32 bit unsigned. table 1-10: fifo sizes for lavender and jasmine fifo size for lavender size for jasmine input fifo 128 words 64 words output fifo 128 words 64 words user logic bus functional description page 55 plication has to prepare the data first and can not generate them ?n the fly? the ulb dma controller is described in detail in chapter 1.7. command dependent flags and debug registers are listed and described in chapter 1.5.5. 1.5.2 command execution and programming figure 1-8 shows a flow chart of display controller command execution and a c-example of a display con- troller command (getpixel). for this example the c-api for lavender and jasmine is used. it is described in a separate manual. before a new command can be written to command register the flag flnom_cwen should be checked. therefore the flag register can simply be polled or an interrupt can be generated as result from the rising edge of this flag. see chapter 1.6 for more details about flag and interrupt handling. afterwards command buffered registers can be written as well as the new command code itself. note that at this time the previous command may be still running (see also chapter 1.5.3 for a detailed discussion) and also the previously buffered register contents is used. command buffered registers contain settings for a certain command (e.g. line colour for the dwline com- mand) which have to be synchronized to command change. the time of command change is determined by hardware so that it is necessary to store values for next command in a separate register (e.g. new line colour for next dwline command right after the first one). a list of command buffered registers can be found in the command description located in appendix. also a detailed register description is placed there. the next step within command execution is to send data to input fifo. the application has to take care that no fifo overrun occurs. therefore it should watch the fifo flags either by polling or via interrupt (see also chapter 1.5.1). an application can compute input data with its own speed, the display controller waits for new data if input fifo runs empty. flnom_if* write data to input fifo flnom_cwen write command 1 *) (read commands only) flnom_cwen flnom_of* read data from output fifo (read commands only) dependent registers *) this command flushes input fifo. usually noop can be used . set command write command 0 // ---------------------------------------- // write command to display controller and // set command registers for getpixel // (no registers required) // ---------------------------------------- gdc_cmd_gtpx(); // ---------------------------------------- // write data to input fifo // ---------------------------------------- for (jj=0;jj user logic bus functional description page 57 for every bitmap dimension. the reading of data from output fifo is only done if the calculated package size ( pkg_size ) has been reached or if the bitmap has been completely finished (last package). as flush command for input fifo after writing a whole package the next getpixel command is sent to display controller. // struct for bitmap data struct s_bm { byte layer; // layer to operate on word x,y,dx,dy; // coordinates: offsets x,y; length dx,dy dword *data; // bitmap data from x,y to (x+dx-1,y+dy-1) // amount: dx*dy }; /* read back function with optimal package sizes */ dword getbm(struct s_bm *bm) { dword amount; word x, y; byte pkg_cnt, pkg_size, reqcnt, of_size; // calculate optimal package size for given request count reqcnt = g0reqcnt + 1; // minimum data amount for block transfer of_size = g0clkpdr_id? 64: 128; // fifo size for jasmine : lavender pkg_size = (of_size / reqcnt) * reqcnt; // initialize data counters amount = 0; // bitmap pixels accumulative pkg_cnt = 0; // intra package count gdc_cmd_gtpx(); // gdc mode: get pixel while (!g0flnom_ife); /* ififo should be empty that packet fits into */ // bitmap region, picture processing loop for (y = 0; y < bm->dy; y++) { for (x = 0; x < bm->dx; x++) { // write address to input fifo (relative to bm start point) g0ififo = pix_address(bm->layer, x + bm->x, y + bm->y); pkg_cnt++; // get data if pkg_size completed (or even smaller last package) if (pkg_cnt == pkg_size || (y == bm->dy - 1 && x == bm->dx - 1)) { gdc_cmd_gtpx(); // flush by sending new command g0oful_ul = pkg_cnt; // initialize block size fifo limit while (g0flnom_ofh == 0); // wait for all data avoids of empty polling while (pkg_cnt) { // receive data from ofifo bm->data[amount++] = g0ofifo;) // set pixel in bitmap array pkg_cnt--; } // while pkg_cnt } // if pkg_cnt == pkg_size } // x-loop } // y-loop return amount; // return number of data words in array } figure 1-9: c-example for reading large data amounts from display controller note that for writing data to input fifo in example from figure 1-9 no flag polling is necessary because it is known that the amount of data is not larger than fifo size (output fifo and input fifo have the same size) and the input fifo is empty at package start. normally flag polling is necessary before writing data to input fifo. the api function gdc_fifo_inp which was already described in chapter 1.5.1 automat- ically takes care of this issue. in the example in figure 1-9 the package size is calculated dynamically in order to experiment with different values for reqcnt . normally it is possible to set up reqcnt according to application needs and calculate the maximal package size offline. if an application is not dependent on reading data from output fifo at MB87J2120, mb87p2020-a hardware manual page 58 once (e.g. per dma or interrupt controlled) it may be easier to read data from fifo as soon as they appear. figure 1-10 shows a code example where this is demonstrated. // data array dword data[size]; // loop over package size for (k=0; k < size;k++){ // wait as long as ofifo is empty // make sure g0flnom_ofe is to dynamic!!! while (g0flnom_ofe); // read data into array data[k] = g0ofifo; } figure 1-10: c-example for reading data continuously 1.5.3 structure of command controller the ulb contains a command controller which is responsible for controlling so called ?xecution devices (ed)?within display controller. these eds are responsible for command execution and data processing. in current implementation of lavender and jasmine four execution devices are handled by ulb command controller. an overview on execution devices and their functions is given in table 1-11. see specifications of these devices for a detailed description. most of display controller commands are so called ?nfinite commands which means that these commands have an unlimited number of processing data. the stop condition for infinite commands is writing a new command. therefore a second register is needed which contains the currently executed command. this reg- ister is controlled by hardware and not writeable by mcu. jasmine has the possibility to watch currently executed command with a read only debug register ( cmddeb ; see chapter 1.5.5 for details). between the two command registers the command decoder is located. the command write time from com- mand to shadow command register is determined by hardware because it depends on the execution state of previous command. the structure of command execution unit within ulb is shown in figure 1-11. in order to avoid command pipeline overflow and to implement a command flow control between display controller and mcu a flag ?ommand write enable ( flnom_cwen ) is implemented. this flag signals that a new command can be written into command register ( flnom_cwen=1 ). by writing the new command the old command is still executed and the pipeline is filled with two commands which can be watched from mcu by cmd_wr_en=0 1 . table 1-11: execution devices within lavender and jasmine execution device function pixel processor (pp) pixel engine (pe) drawing of graphical primitives drawing and rle decompression of bitmaps memory access unit (mau) writing and reading of pixel-addressed data memory copy (mcp) copying of pixel-addressed blocks within sdram memory physical memory access unit (dipa) ipa writing and reading of word-addressed data via input- and output fifo 1. this is only true when this ?g is set to dynamic behaviour which is the reset value. see section 1.6 for a detailed explanation. user logic bus functional description page 59 additionally the write event of a new command triggers a mechanism that is responsible for dividing data between different commands. this allows to write data for next command into input fifo while the current command is still running and the selected execution device reads data from fifo. note: for the programmer it is important not to write data for currently executed com- mand after writing a new command because these data would be interpreted as data for new command. after the currently running command has finished and all data in input fifo have been processed the ulb command controller writes next command into shadow register and sets cmd_wr_en=1 ? lavender differs between read and write commands. for commands reading data from display controller via output fifo for lavender an additional condition has to be met before execution of a new command is started. note: for a lavender read command output fifo has to be empty before a new com- mand is started. in jasmine this additional condition need not be respected and the output fifo can collect data from different commands. two different error cases regarding the ulb command controller can be observed by mcu via special error flags ( flnom_ecode , flnom_edata ). a detailed description of flags regarding the command execution can be found in chapter 1.5.5. 1.5.4 display controller commands display controller commands for lavender and jasmine are divided per execution type into infinite, finite and special commands. an additional subdivision can be made into write and read commands independent from execution type (finite or infinite). for a detailed command list see appendix. infinite command execution is processed as described in section 1.5.3. the stop condition for infinite com- mands is the writing of next command. infinite commands are: putpa , dwline , dwrect , putpixel , putpxwd , putpxfc , getpixel , xchpixel , memcp and dwpoly instruction command ready command stop command start command reset for execution devices fifo read data code fifo space fifo space for execution devices fifo read from execution devices fifo empty for execution devices fifo read fifo empty parameter command mcu write_valid mcu read command cmddeb waiting command executed command shadow_valid fifo write data mcu write command fifo write_valid controller command command shadow register command decoder command register (cmd) input fifo (ififo) figure 1-11: command execution within ulb MB87J2120, mb87p2020-a hardware manual page 60 in difference to infinite commands for finite commands the amount of data to be processed is fixed. it is defined by various register settings inside the execution devices or command coded parameters in case of getpa ( cmd_par ). finite commands are: putbm , putcp , puttxtbm , puttxtcp , getpa special commands are control commands influencing the command execution itself or execution devices but they do not process data. there are two special commands: software reset ( swreset ) and no operation ( noop ) the noop command is included in the normal command execution pipeline as described in section 1.5.3 with the exception that no execution device is activated and no data are read from input fifo or written to output fifo. this command can be used to force the previous command to end data processing. note that all data send during noop command is active are kept for next command. the swreset command treats as a synchronous reset for command execution. this command is not in- cluded in the normal command pipeline and is executed immediately. the command controller inside ulb and also all execution devices (complete pp, aaf and ipa inside dipa) go in its initial state, the command pipeline will be emptied and the fifos will be reset so that all data will be lost. note: during swreset input- and output fifo will be reset so that data loss will occur. not affected by software reset are display controller parts not responsible for command execution (sdc, vic, gpu, dpa inside dipa, cu, ccfl and parts of ulb). these devices continue running and process- ing data. to reset these devices a hardware reset is necessary 1 . due to a not interruptible sdram access software reset is not completed in one clock and needs execution time depending on running sdram access for pp or ipa. therefore the command flow control has also to be used by an application after software reset. note: also after swreset the flag flnom_cwen has to be polled in order to ensure a save reset operation and execution of following commands. 1.5.5 registers and ?gs regarding command execution in order to allow an application controlling and watching the command execution ulb contains some flags (within flag register) to provide the following functions: flag: flnom_cwen watching the command execution state as already described in section 1.5.3 flags: flnom_ripa , flnom_rmcp , flnom_rmau , flnom_rpe , flnom_bipa , flnom_bmcp , flnom_bmau , flnom_bpe watching the state (busy or ready) of a specific execution device. because all flags are high active both variants are offered by display controller to capture the needed event (busy or ready). flag: flnom_ecode a wrong command code was sent to display controller. the wrong code is treated internally as a noop command which means that the command controller simply waits for a new command and no data processing is performed. all data sent to input fifo are kept for next command as an exception for noop command behaviour. flag: flnom_edata this error flag is set when an execution device tries to read data from an empty input fifo. this may indicate a malfunction of execution device but the interpretation of this flag heavily depends on execu- tion device implementation. all flags are handled as described in section 1.6 and all are able to cause an interrupt if required. a detailed flag description of all display controller flags can be found in flag description located in appendix. 1. a hardware reset can also be triggered by software by set register clkpdr_mrst to ?? user logic bus functional description page 61 additionally to flags inside flag register debug registers are implemented in lavender and jasmine in order to watch command controller status. table 1-12 lists these debug registers. note that some registers are only implemented in jasmine. 1.6 flag and interrupt handling 1.6.1 flag and interrupt registers the interrupt controller inside ulb contains one 32 bit flagregister ( flnom ; address 0x000c) and one interrupt-mask-register ( intnom ; address 0x0018) which allows a very flexible flag handling and inter- rupt generation control. in order to avoid data inconsistencies during bit masking within flag- or interrupt-mask-register the mask process is implemented in hardware for these two registers. this helps to avoid flag changes by hardware between a read and a write access (read->mask->write back). to distinguish between set-, reset- and direct write access different addresses are used: address ( flnom , intnom ): normal write operation address + 4 ( flrst , intrst ): reset operation (1: reset flag on specified position, 0: don? touch) address + 8 ( flset , intset ): set operation (1: set flag on specified position; 0: don? touch) all of these three addresses write physically to one register with three different methods. for reading all addresses return the value of the assigned register ( flnom or intnom ). for writing all bus access types as described in chapter 1.4.4 are possible for each of these addresses. 1.6.2 interrupt controller con?uration figure 1-12 shows the basic structure of interrupt generation circuit for one flag. the flagregister itself is set and reset able by hard- or software. a set event by hard- or software sets the flag to ? and a reset event sets the flag to ?? software flag access has a higher priority than hardware events but hardware events may be present some clock cycles around software access which is only one clock active after synchronisation. note: despite the higher priority of software access the hardware event may overwrite the software settings after mcu write access if the set- or reset condition for the desired flag is still true. table 1-12: ulb debug registers register bits name description device name address ulbdeb 0x0098 7/6 a :0 a. lavender/jasmine value due to different fifo sizes if current input fifo load (command independent) all 15/14 a :8 of current output fifo load all 23/22 a :16 iflc command dependent input fifo load all cmddeb 0x009c 7:0 cmd currently executed command (see chapter 1.5.3) jasmine 31:8 par command coded parameter for cur- rent command ( getpa only) jasmine MB87J2120, mb87p2020-a hardware manual page 62 a flag set by hardware is always possible; the hardware reset can be switched off in order to avoid dynamic flag changing. this flag behaviour is referred to as ?tatic flag behaviour. a forbidden hardware reset is important for a handshake implementation between display controller and mcu for instance in connection with interrupts. if flag set and reset is allowed the flag behaviour is called ?ynamic behaviour. in this case the desired flag simply follows the input signal. note that flag changes may occur with core or display clock which may be higher than ulb bus clock 1 . therefore it is not possible to trace some hardware events because you can not achieve a suitable sample rate. if the toggle rate for a real hardware flag is slow enough you can of course set the flag behaviour to dynamic. many flags represent a state which can be manipulated by software so that dynamic behaviour makes sense for these flags because they can be indirectly influenced by software. for instance the full flag for input fifo can only change its value after writing data to input fifo. the flag behaviour can be set with register flagres . flag hardware reset can be turned on (1: dynamic flag behaviour) or off (0: static flag behaviour) for each flag separately. 1.6.3 interrupt generation the first operation for interrupt generation is the masking of flagregister by interrupt-mask-register. with this mechanism an application can determine which flags can cause an interrupt by simply set (?? at the same bitposition as the flag in interrupt-mask-register. every flag can be source for an interrupt because an or combination of flags is implemented in interrupt controller. after the flagregister a level detection circuit is implemented (see figure 1-12) which detects the rising edge of a flag. with the intlvl register the programmer can choose for every flag whether to take the flag itself (level interrupt) or the edge detection signal with one core clock length (edge interrupt). in level triggered interrupt mode an interrupt handshake should normally be used between mcu and lav- ender/jasmine. this means that the flag responsible for interrupt will be reset inside interrupt service routine (isr). the isr is only called when mcu detects an interrupt request. as a result the interrupt request is only taken back from lavender/jasmine after flag reset. so it is ensured that the interrupt signal is stable for many ulb clocks in level triggered mode. be careful with dynamic flags in this context . in edge triggered interrupt mode no handshake between mcu and display controller is necessary. the dis- play controller signals the mcu with a pulse on interrupt request signal (pin ulb_intrq ) that a certain event occurred within display controller. the mcu can call its isr and does not need to reset the flag which caused the interrupt if the flag behaviour is set to dynamic. if the flag behaviour is set to static and a reset access from mcu is missing no more interrupts can be generated because no more rising edges for the in- teresting flag occur. 1. the real sample rate is again lower since it is the time between two bus read cycles. intre q intc (jasmine only) generation (flnom) interrupt edge . . . . . interrupt mask (intnom) flagregister . . flagres s r flag_reset(hw) flag_set(hw) mcu d mcu mcu address 1 0 1 0 intlvl flag0 flag31 interrupt . figure 1-12: interrupt generation within interrupt controller for one flag user logic bus functional description page 63 jasmine contains an edge generation circuit which is responsible for a prolongation of an impulse in case of an edge interrupt. the impulse length can be set in intreq_intc within a range from 0 to 63 ulb clocks. for every edge impulse at the input of the edge generation circuit a pulse with the programmed length will be generated at output. if a level interrupt occurs the output signal follows the input signal syn- chronized to ulb clock domain. lavender does not contain an edge generation circuit. therefore no edge interrupt is possible. for lavender the default value for intlvl register is edge trigger for interrupt for all flags. make sure to set the register intlvl to 0x00000000 during lavender initialisation. for mcu interrupt programming ??level should be used for display controller interrupt. 1.6.4 interrupt con?uration example in figure 1-13 an example configuration for display controller and mcu is given. in this example an inter- rupt should be activated when the input fifo load is equal or lower than ? (register: g0iful ). to acti- vate the interrupt generation the bit 3 of interrupt-mask-register is set to ? via the set address for this register. the interrupt trigger for display controller is set to level. for mcu first all interrupts are turned off, global interrupt level and level for gdc-interrupt is set, the mcu interrupt trigger is also set to level to ensure a save detection. at the end the port for external interrupts is enabled, pending interrupt requests will be deleted and interrupt execution is turned on again in order to enable gdc interrupt execution. in mb91xxxx hardware manual the interrupt initialisation is described in more detail. ;; --------------------------------- ;; init gdc writereg g0flagres, 0x3f401fff; set fifo flags to dynamic writereg g0iful, 0x00000001; set input fifo limits for interrupt writereg g0intlvl, 0x0 ; level triggered writereg g0intset, 0x8 ; if <= if-low(=1) ;; --------------------------------- ;; init mcu andccr #0xef ; disable all interrupts ;; set interrupt level for ext. int0 ldi #0x14,r0; set interrupt level to 20 ldi #icr00, r1; load address for ext. int0 stb r0, @r1 ;; set global interrupt level to 30 stilm #0x1e ;; initialize external interrupt ldi #0x1, r0; enable only int0 ldi #enir, r1; load address for int. enable register stb r0, @r1 ;; set interrupt request level ldi #0b01, r0; set ??level for int0 ldi #elvr, r1; load address for external level register sth r0, @r1 ;; enable interrupt ports ldi #0b00000001, r0; enable int0 ldi #pfrk, r1; port function register for interrupt 0 stb r0, @r1 ;; clear all interrupt requests ldi #0, r0 ldi #eirr, r1 stb r0, @r1 nop nop nop ;; enable interrupts orccr #0x10; set i-bit in ccr register MB87J2120, mb87p2020-a hardware manual page 64 ;; --------------------------------- figure 1-13: interrupt display controller and mcu initialisation example 1.6.5 display controller ?gs all display controller flags are located in the flagregister ( flnom ) inside ulb interrupt controller and handled as described in section 1.6.1. all flags are explained in appendix. note that some flags are only available for jasmine. 1.7 dma handling 1.7.1 dma interface in order to improve data transfer speed and to automate fifo load controlling during command execution lavender and jasmine contain a dma controller which operates together with dma-controller (dmac) integrated in mb91xxxx series mcus. it is located inside ulb? i/o-controller and handles the display controller dma interface (gdc-dmac). this interface consists of additional control signals; data transfer is handled by i/o controller as for normal mcu accesses. the dma connection between display controller and mcu is shown in figure 1-14. the gdc-dmac requests a dma transfer by setting ulb_dreq to ?? the mcu acknowledges this re- quest by set ulb_dack to ? during a valid bus cycle. ulb_dack -pulses for other devices connected to mcu are ignored by gdc-dmac because the ulb_dack signal is gated with ulb_csx for display con- troller. in order to stop the mcu-dmac externally by display controller the dma stop signal ( ulb_dstp ) exist. this signal creates an error condition inside mcu-dmac that can also cause an interrupt (see mb91xxxx manual for more details). a better solution than using ulb_dstp signal for disabling mcu-dmac is to disable dma in mcu first by writing dmacax_paus=0 and dmacax_denb=0 and turn off gdc-dmac afterwards.the dstp pin at mcu is not needed in this case and can be used as general purpose i/o. note that the ulb_dstp pin at display controller may not be supported in future display controller releases. 1.7.2 dma modes the mcu dma-controller can deliver/get data in two different ways: 1. ulb_dreq level triggered (demand mode) 2. ulb_dreq edge triggered (block-, step- and burstmode) for a detailed description of supported dma modes see mb91360 series hardware manual. lavender/ jasmine dma interface ulb mcu ulb_dack ulb_dstp ulb_dreq figure 1-14: dma connection between display controller and mb91xxxx user logic bus functional description page 65 1.7.2.1 level triggered dma (demand mode) in case 1. the length of the dreq signal defines the amount of data to be transferred. from mcu point of view an external device has to control the length of dreq impulses according to inter- nal buffer sizes. it is responsible for the division of data stream while the mcu is only controlling the total amount of data to be transferred. in the special case of lavender/jasmine does this mean that the gdc- dmac counts the amount of free words for input fifo (write dma) or the number of words in output fifo (read dma). table 1-14 gives an overview on transfer sizes in different modes for display controller. before starting a demand transfer the gdc-dmac tests for dma start condition 1 , detects the number of words to be transferred, loads a counter with this value and counts this counter to zero. during counting ulb_dreq is set to active. this procedure is repeated until dma within display controller is disabled. the gdc-dmac does not know the total amount of words to be transferred. it only tries to fill (write dma) or to flush (read dma) its fifos. at the end of a complete dma transfer ulb_dreq could still be active because from display controller? point of view dma is enabled and input fifo needs data or output fifo has to deliver data. after disabling dma for display controller the ulb_dreq signal goes inactive. figure 1-15 shows the start of a write dma demand transfer. display controller requests one input fifo fill cycle 2 . during this time a display controller command is active and reads data from input fifo concur- rently. after a short break the second fill cycle is requested. 1.7.2.2 edge triggered dma (block-,step-, burstmode) in case 2. (edge triggered dma transfer) only the rising edge of ulb_dreq signal is important. the amount of data to be transferred is set within mcu (see also table 1-14). a mcu peripheral device has to ensure that the ulb_dreq impulse is long enough to be recognized by mcu. in case of gdc the ulb_dreq signal goes inactive after the mcu has acknowledged the dma re- quest 3 . depending on mcu mode (block-, step- or burstmode) a mcu defined amount of data words is transferred to or from display controller fifos. the programmer has to ensure that no fifo overflow can occur by setting up the appropriate value for input fifo lower limit ( ifdma_ll ) or output fifo upper limit ( ofdma_ul ) (see chapter 1.7.3 for a detailed description). figure 1-16 shows a write dma transfer in block mode. the block size is set to 10 words. despite of this jasmine 4 toggles the ulb_dreq signal after every falling edge of ulb_dack signal because it does not know the mcu settings. it can not distinguish between block-, step- or burst mode. 1. for write dma: ifdma_ll >= input fifo load; for read dma: ofdma_ul <= output fifo load. 2. for jasmine one complete ?l cycle contains 64 words. 3. this is the ?st high to low edge of the ulb_dack signal combined with a valid chip select signal. 4. lavender shows the same behaviour but this example was made with jasmine. figure 1-15: write dma in demand mode MB87J2120, mb87p2020-a hardware manual page 66 1.7.3 dma settings in order to use dma feature for display controller it is necessary to set up dma according to table 1-13. dmaflag_en is the general dma enable flag; if this bit is set to ? all dma operations are stopped. ad- ditionally the falling edge of this flag during a running dma transfer causes a reset of gdc-dmac and mcu-dmac via ulb_dstp signal in order to stop dma transfer completely. this is important because the transferred data belong to a command and a running dma transfer influences next command and its data stream which is not necessarily controlled by dma. for dma operation only one dma channel is available between display controller and mcu. therefore only one fifo can be written or read per dma at a given time. the programmer can select the fifo that should be read or written with help of dma by set dmaflag_io according to table 1-13. an additional gate with ulb_wrx or ulb_rdx ensures that only accesses for the selected mode are accepted. the selected dma mode can be selected with dmaflag_mode. see chapter 1.7.2 for more details about dma modes. the trigger condition for dma start can be set separately for input ( ifdma_ll ) and output fifo ( ofdma_ul ). it represents a fifo load and is completely independent from flag settings according to flag table 1-13: dma register settings register bits flag name description default value name address ifdma 0x0088 7/6 a :0 a. lavender/jasmine value due to different fifo sizes ll lower limit for dma access to input fifo 10 ofdma 0x008c 23/22 a :16 ul upper limit for dma access from output fifo 60 dmaflag 0x0090 12:8 dstp duration of ulb_dstp signal in ulb clocks. 7 2 mode ?? dma demand mode ?? dma block/step- or burst mode 0 1 en ?? enable dma 0 0io ?? use dma for input fifo ?? use dma for output fifo 1 figure 1-16: write dma in block mode user logic bus functional description page 67 description table in appendix. figure 1-17 shows the meaning of these limits for input and output fifo while in table 1-14 the calculation for transfer sizes for fifos is listed. in case of input fifo the fifo load has to be equal or smaller to meet trigger condition for dma. for output fifo the load has to be equal or greater to trigger dma transfer (see also figure 1-17). because the fifos can be accessed from gdc side (read from input fifo and write to output fifo) when trigger condition becomes true the fifo load itself is taken for calculation. in figure 1-17 this situation is drawn. 1 (2) (3) for block/step- or burst mode condition (2) should be fulfilled for input fifo in order to avoid fifo over- flow. for output fifo at least one block should be available so that the condition (3) should be met. oth- erwise wrong data may be delivered because blocksize is transfered at once and not all data are available at transfer time. for setup of dma trigger levels the gdc internal packet sizes for data processing have to be considered. this are ipa block transfers and reqcnt for pixel data packetizing in pixel processor. it depends on com- mand how many words are read from input fifo or written to output fifo at once. the amount of data words is determined by type of command data stream (address informations, colour data with different col- our depths). see command description in appendix for a detailed command description. in general the at once processed information must be available in ififo or has to fit into ofifo. if this state is not reached the execution devices wait for data transfer by mcu until the requirements are fulfilled. if the dma trigger levels are not set up accordingly deadlock situations can occur (data lack or jam). table 1-14: transfer count calculation for dma mode input fifo output fifo demand mode trans = fifosize a - fifo load a. see table 1-10 for fifo sizes for lavender and jasmine trans = fifo load block/step- or burst mode trans = MB87J2120, mb87p2020-a hardware manual page 68 for dma demand mode settings according to (4) (0x3f for jasmine, 0x7f for lavender) to keep the input fifo full all the time and according to (5) to keep output fifo flushed all the time avoid problems with internal packetizing. data will be transferred immediately if possible. (4) (5) for dma block/step/burst mode other trigger levels are required due to additional restriction for mcu block transfer sizes (see equation (2) and (3) with their description). also at demand mode it is possible to set up other trigger levels in order to transfer more words at once. in this cases special care should be taken to avoid the described deadlock situations. reserve for required data amount (ififo) or space (ofifo) for packetized procession must be guaranteed all the time. there are two possibilities to stop dma transfer. the first is the falling edge of dmaflag _ en as already mentioned. the second possibility is to interrupt the running command by a swreset command. because the dma controlled stream is normally coupled to the currently executed command 1 interruption of this command should also cause dma dropout. because of fifo reset during swreset already transferred data will also be deleted. 1.7.4 dma programming examples in figure 1-18 an example for a mcu- and lavender dma initialization is given. the mcu dma channel ??is used for dma connection. in dmacb0 and dmaca0 mcu registers the parameters for mcu-dmac are set; with set of bit 31 in dmacr dma operation is enabled inside mcu. ;;; dma variables blk_dma1: equ 1 ; block size dtc_dma1: equ 0 ; transfer count ofhigh_dma1: equ 1 ; dma limit dma1a: equ 0x8e100000 + (blk_dma1 << 16) + dtc_dma1) ; build data word for dmaca0 dma1of: equ (ofhigh_dma1 << 16) ;; --------------------------------- ;; init dma for data output ;; --------------------------------- writereg dmasa0, g0ofifo ; source address writereg dmada0, 0x001c0000 ; destination address writereg dmacb0, 0x28000004 ; type=00,md=10(demand),ws=10(word),inc.destination writereg dmacr, 0x80000000 ; enable mcu-dmac writereg dmaca0, dma1a writereg g0ofdma, dma1of writereg g0dmaflag, 0x00000206 ; of, en_dma=1, demand mode, dstp=2 ; writereg g0dmaflag, 0x0000021e ; of, en_dma=1, demand mode, dstp=2, inv, tri ; wait for dma to finish waitdma2: readreg dmaca0 ;nach r2 ldi #0x80000000, r3 and r3, r2 bne waitdma2 figure 1-18: mcu- and gdc-dma initialization example 1. for input fifo it is also possible to deliver data for waiting command (see section 1.5). but the swreset command ?shes the command pipeline completely so that also the waiting command will be deleted. dma transfer has to stop anyway. ifdma_ll fifosize 1 = ofdma_ul 1 = user logic bus functional description page 69 the chosen dma mode is demand mode which should be set inside mcu and display controller. addi- tionally for display controller the output fifo is selected for dma operation and signal inversion and tristate behaviour is not selected according to board implementation (see chapter 1.3.4). the dma trigger limit for output fifo is set to 1 (register: g0ofdma ) which means that every new data word in output fifo causes a dma transfer. in demand mode this ensures an empty fifo after dma operation but this causes also a lot of protocol overhead because a handshake between display controller and mcu is necessary for every data word. therefore transfer performance may decrease slightly. the loop labelled with ?aitdma2 at the end of figure 1-18 stops program execution until the end of dma transfer. the following code can assume that data are transferred from display controller to mcu also as result of this low fifo limit. figure 1-19 shows an example where a rle compressed bitmap is transferred to display controller with help of dma. this example uses c-api functions which are described in detail in a separate manual. see c-comments for a short explanation. /*****************************************************************************/ /* constant declarations */ /*****************************************************************************/ // picture infos: // - rle compressed. // - 16bpp. // - 111 x 43 pixel, origin: 0, 0 // picture dimensions const hd_fujitsu_x = 111; const hd_fujitsu_y = 43; // number of data words for ?d_fujitsu?array const hd_fujitsu_num = 800; // array definition const dword hd_fujitsu_array[] = { 0xeeffdfb8, 0xffdf04fd, // ..data.. 0xdf000000}; void main(void) { // use dma (demand mode) // --------------------------------------------- // set up mcu- and gdc-dmac // --------------------------------------------- // ulb_dma_hdg(dummy,dummy,mode,block,direction) // mode: 00: block/step, 01: burst, 10: demand // block: block size (1 in demand mode) // direction: 1: input fifo; 0: output fifo ulb_dma_hdg(0, 0, 2, 1, 1); // --------------------------------------------- // write parameter and command (see api description) // --------------------------------------------- gdc_cmd_ptcp(0x0,hd_fujitsu_x-1,0x0,hd_fujitsu_y-1,0x0,0x0,0x0,0x0,0); // --------------------------------------------- // activate dma transfer // --------------------------------------------- gdc_fifo_inp((dword *)hd_fujitsu_array, (word)hd_fujitsu_num, 1); // --------------------------------------------- // wait for end of transfer // --------------------------------------------- while ((dmaca0 & 0x80000000) != 0); // --------------------------------------------- // send noop in order to flush input fifo // --------------------------------------------- MB87J2120, mb87p2020-a hardware manual page 70 gdc_cmd_nop(); } figure 1-19: dma programming example with help of c-api the first step is to initialize mcu-dmac as well as gdc-dmac with help of api function ulb_dma_hdg ? this function does not start the transfer; it sets only dma parameters for the following transfer. afterwards the command for writing compressed bitmaps to display controller is sent to command register together with the dimensions of the bitmap. see chapter 1.5 for more details about command execution. with help of api function gdc_fifo_inp dma transfer is started. note that the last function param- eter is ? which means that dma transfer is enabled (see chapter 1.5.1 for a discussion about this function). in order to synchronize dma data flow with program flow a wait loop for end of dma transfer is included into the source code. this is not necessary in any case since the api functions ulb_dma_hdg and gdc_fifo_inp ?wait for the end of previous dma transfer before they perform any action. in the example in figure 1-19 the dma synchronization is necessary in order to make sure that all data are transferred when a noop command is sent to display controller to force a input fifo flush. without this synchronization not sent data would be kept for the next command after noop . user logic bus ulb register set page 71 2 ulb register set 2.1 description some ulb registers are controlled by ulb itself and some are handled by ctrl in the same manner as for all other gdc components. for the programmer there is no difference accessing these registers. an overview on ulb- and ctrl controlled registers has already been given in table 1-5, section 1.4. in table 2-1 an overview on all ulb registers is given. all addresses are relative to start of register space for given gdc; see section 1.4 for details. the address values are byte addresses and can be accessed in word (32 bit), halfword (16 bit) or byte (8 bit) mode from mcu, except the fifos which can only be ac- cessed in word mode. flag and interrupt mask register handling: as already mentioned in section 1.6 the ulb contains one flag- and one interrupt mask register with special access modes; therefore in table 2-1 flag- and interrupt mask register have each three addresses. table 2-1: ulb register description register bits group name description default value name address cmd 0x0000 31:8 par command parameter 0 7:0 code command code 0xff (noop) ififo 0x0004 31:0 - input fifo - ofifo 0x0008 31:0 - output fifo - flnom 0x000c 31:0 - flag register (normal write access) a 0x20400000 flrst 0x0010 31:0 - flag register (reset write access) a 0x20400000 flset 0x0014 31:0 - flag register (set write access) a 0x20400000 intnom 0x0018 31:0 - interrupt mask register (normal write access) ?? use ?g for interrupt 0 intrst 0x001c 31:0 - interrupt mask register (reset write access) ?? use ?g for interrupt 0 intset 0x0020 31:0 - interrupt mask register (set write access) 0 intlvl 0x0024 31:0 interrupt level/edge settings ?? positive edge of ?g triggers interrupt b ?? high level of ?g triggers inter- rupt 0xffffffff wndof0 0x0040 20:0 off mcu offset for sdram window 0 0x10000 wndsz0 0x0044 20:0 size size of sdram window 0 0x20000 MB87J2120, mb87p2020-a hardware manual page 72 wndof1 0x0048 20:0 off mcu offset for sdram window 1 0x50000 wndsz1 0x004c 20:0 size size of sdram window 1 0x00001 wndsd0 0x0050 23:0 off sdram offset for sdram window 0 0x000000 wndsd1 0x0054 23:0 off sdram offset for sdram window 1 0x100000 sdflag 0x0058 0 en ?? enable sdram space for gdc c ?? any access to sdram space is ignored by gdc c 0 iful 0x0080 23:16 ul input fifo upper limit for ?g- or interrupt controlled ?w control flag ifh=1 if ifload d >= iful:ul 0x0c 7:0 ll input fifo lower limit for ?g- or interrupt controlled ?w control flag ifl=1 if ifload d <= iful:ll 0x03 oful 0x0084 23:16 ul output fifo upper limit for ?g- or interrupt controlled ?w control flag ofh=1 if ofload e >= oful:ul 0x3c 7:0 ll output fifo lower limit for ?g- or interrupt controlled ?w control flag ofl=1 if ofload e <= iful:ll 0x0f ifdma 0x0088 7:0 ll lower limit for dma access to input fifo 0x0a ofdma 0x008c 23:16 ul upper limit for dma access from output fifo 0x3c table 2-1: ulb register description register bits group name description default value name address user logic bus ulb register set page 73 dmaflag 0x0090 12:8 dstp duration of ulb_dstp signal. this value can be set in order to ensure a save mcu-dmac reset. normally the default value should work. 7 4 tri ?? set ??to tristate (?? for ulb_dreq, ulb_dstp and intrq 0 3 inv ?? invert ulb_dreq, ulb_dstp and intrq 0 2 mode ?? dma demand mode ?? dma block/step- or burst mode 0 1 en ?? enable dma 0 0 io ?? use dma for input fifo ?? use dma for output fifo 1 flagres 0x0094 31:0 - ?? set ?g to dynamic behaviour f ?? set ?g to static behaviour f 0x20400000 ulbdeb 0x0098 23:16 iflc input fifo load for current com- mand attention : this value changes with gdc core clock; correct sampling by mcu can t be ensured. value is read-only; writing is ignored. 0x00 15:8 of output fifo load attention : this value changes with gdc core clock; correct sampling by mcu can t be ensured. value is read-only; writing is ignored. 0x00 7:0 if input fifo load independent from current command attention : this value changes with gdc core clock; correct sampling by mcu can t be ensured. value is read-only; writing is ignored. 0x00 a. for meaning of ?gs and default value see section 1.6. b. attention : this is only allowed when gdc core clock is equal to ulb bus clock (see section 1.6) c. see section 1.4. d. ifload: input fifo load e. ofload: output fifo load f. for a description of ?g handling see section 1.6. table 2-1: ulb register description register bits group name description default value name address MB87J2120, mb87p2020-a hardware manual page 74 2.2 ulb initialization ulb contains no lockable registers; so it is possible to write to every register at any time. there is no initialization order for ulb but some general rules should be followed: for lavender intlvl register should normally be set to 0x00000000 in order to trigger interrupt on high level (see section 1.6 for more details). sdram space for direct memory access has to be initialized and enabled for use. write valid values to wndofx , wndszx , wndsdx and 0x00000001 to sdflag . be careful about overlapping windows when more than one gdc is connected to mb91xxxx (see section 1.4 for more details). initialize iful or oful with valid limits before using fifo limit flags ( ofh,ofl,ifh,ifl ) for in- terrupt or polling. otherwise default values will be taken as valid limits. initialize mcu, ifdma , ofdma and dmaflag in this sequence with valid values in order to use dma for data transfer (see section 1.7 for details). note that dmaflag_en should be written at last because it starts dma transfer triggered by gdc. flagres register should be initialized correctly before interrupt is enabled inside mcu or flags are polled within user program in order to meet applications need. read-only ulbdeb register exists only for debugging purpose; in normal applications flags in connec- tion with fifo limits should be preferred. page 75 b-3 sdram controller (sdc) MB87J2120, mb87p2020-a hardware manual page 76 sdram controller function description page 77 1 function description 1.1 overview this module is part of a graphic display controller (gdc) especially for automotive applications. the gdc supports a set of 2d drawing functions (pixel processor) a video scaler interface, units for physical and di- rect video memory access and a powerful video output stream formatter for a great variety of connectable displays. inside the gdc there is a memory controller which arbitrates the internal modules and generates the re- quired access timings for sdram devices. with a special address mapping and an algorithm for generating optimized control commands the controller can derive full benefit from 4-bank-interleaving supported by the sdrams. so the row activation time is hidden if switching to another memory bank in most cases. this increases performance respective at random (non-linear) memory access. power down modes with clock suspend (csus) with and without self-refresh are supported. 1 the sdram controller arbitrates gdc internal device requests for data transfers from/to the video mem- ory. important is also the address calculation which controls the mapping from a given logical address (in layer, x- and y-pixel format) to the physical bank, row and column address. thus the other devices are in- dependent from the physical implementation of the memory structure. if the application for gdc needs physical video memory (frame buffer) access, knowledge is needed how a logical address (layer, x, y) is converted to the physical address in video memory and how this address maps into physical host mcu address space. there is also a buffered access method without mapping into mcu address space possible, then only internal video memory address is of interest. 1.jasmine implementation (gdc with integrated dram) makes use of an integrated single-bank sdram. therefore special features as 4 bank interleaving and power suspend/self refresh are not supported by the device. pp dipa gpu vic d d intra word pixel addressing d cmd write dqm arbiter address calculation a layer information command controller refresh timer 4 x 512k x 32 bit 64m sdram d figure 1-1: sdc block diagram embedded into gdc MB87J2120, mb87p2020-a hardware manual page 78 main functionality is to provide an arbitrated video memory access for gdc components such as pixel processor (pp), direct/indirect memory access (dipa), the video interface controller (vic) and finally the graphic processing unit (gpu) which reads pixel data from memory and formats the output stream. because of the different requirements of the various components there are to support various access types, such as burst and block modes with adjustable transfer sizes. 1.2 arbitration the arbitration of the four main gdc parts works priority based. the setup of priority values can be decided by the requester component itself and is signalized to the sdram controller. the benefit is that the setup can vary for different applications. there is also the possibility to change priority on the fly, e.g. if buffer state changes. the connected component can decide about the urgency of the transfer. video ram arbitration is done in principle of cooperative multitasking. this did not waste bandwidth if a requester device uses only a part of its dedicated bandwidth as if time slicing would be used. all devices share the commonly available bandwidth resource. main advantage is that system performance could be scaled and optimized for a wide range of different applications. a decision about granting the next device is done priority based at the end of a currently processed device request. the currently processed device is excluded from priority based selection for the next one. this re- sults in granting requests for the two devices with given highest priorities alternately, if requested. only idling between these two devices could be used for the other ones. therefore the devices with the two high- est priorities could be considered as real-time (normally this should be gpu and vic). 1.3 sdram timing configurable options for the appropriate sdram timings listed in table 1-2. defaults are listed for 100 mhz sdram types of mb811643242a for lavender. values for integrated dram version for jasmine are given in a separate column. the configuration value is a number of wait states. that means additional table 1-1: gdc modules and its priority registers with recommended con?uration device register comment gpu sdcp_lp = 3 sdcp_hp = 7 low and high priority, real time device with automatic priority scaling vic sdram_lp = 2 sdram_hp = 6 low and high priority, real time device with automatic priority scaling pp/aaf sdcprio = 1 no automatic priority scaling sup- ported dipa dipactrl_pdpa = 5 dipactrl_pipa = 0 priorities for dpa and ipa access table 1-2: sdram command timings parameter default jasmine description trp (ras precharge time) 30 ns 22.5 ns time from same bank pre to row actv com- mand trrd (ras to ras bank active delay time 20 ns - time from actv to opposite bank actv com- mand tras (ras active time) 60 ns 37.5 ns time from actv to same bank pre command sdram controller function description page 79 clocks of idling before the next sdram access command. so the configuration value is lower by one than the required timing from the sdram data sheet. if an absolute minimum time is given it? necessary to evaluate the corresponding number of clock periods for configuration. this depends on and should be op- timized for the required core clock frequency. following procedure should be used: 1. divide given timing by the core clock period 2. round up to next integer 3. subtract one to have the right wait-state value cas latency can be setup to values of 2 or 3 for lavender. jasmine is not programmable for different cl values. additional configurable options are the refresh period (normally 16 us for one row) and the power on sta- bilization timer (200 us) before the first initialization sequence begins to run. the refresh counter is reset after each execution of a single row refresh job (not considering if it runs as time-out or idle task) and it causes an time-out if the counter value reaches zero. the configuration values are given in a number of sys- tem clocks. 1.4 sequencer for refresh and power down fixed command sequences such as sdram initialization, auto refresh, power down or wake-up and trans- fers of special data structures are easier to implement in a fixed and preprogrammed manner. these tasks are assigned to the sequencer unit of sdc. to keep the amount of memory low and guarantee a defined device shutdown the power down sequences are not a part inside the standard micro program for refresh and initialization. however special power down sequences are loaded into memory when needed. if the sdc currently processes a transfer controlled by the address/command generator unit these sequence will be finished normally and then the new loaded routine inside the micro program storage is executed. one word of micro program code consists of an address argument (bits [12:6] for lavender, bits [11:6] for jasmine 1 ), flow control instruction and a container command (sdram command). 2 figure 1-2 shows the format of one micro program entry. bits [3:1] of sdram command coding the ras, cas and we signal. the internal representation is inverted compared with the sdram ports. bit [0] for controlling the auto precharge (ap) feature is not controllable by the sequencer and internally fixed to ?? table 1-3 lists the trcd (ras to cas delay time) 30 ns 22.5 ns time from same row actv to read or writ command trw (read to write recovery time) a 10 (8) t 7 (5) t pipeline recovery time from each read to writ command a.this setup is not regarding dram timing, but required to avoid bus collision on internal busses or external sdram tri-state busses due to pipelined operation. values in parenthesis are possible if anti aliasing filter is switched off and then no read-modify-write access is required. 1.jasmine has reduces sequencer size of 32 words. thus address argument is 5 bit only. 2.logical address operations and data validation flag are not needed in this application without preprogrammed data structures. table 1-2: sdram command timings parameter default jasmine description 12 0 31 3 5 6 13 4 1 address sdram_cmd instr not used figure 1-2: micro program entry MB87J2120, mb87p2020-a hardware manual page 80 possible entries. in this application there is no preprogrammed data transfer implemented. so the com- mands actv, writ, read and bst should not be used. sequence programming is done with special flow control instructions, sub program calls, loops, power down entry and exit are supported. 1 table 1-4 lists possible flow control instructions and their coding. in general it? recommended to use the ?kctrl tool for gdc setup. it optimizes sdram timing based on core clock frequency, calls the asmseq sequencer program and generates valid code for the sequencer. an example of the micro code is provided with the assembler tools. there is also a compression tool which gen- erates smaller programs with sub-routine calls from a linear coded micro program ( asmseq_delay ). size of sequencer memory is 64 entries for lavender and 32 entries for jasmine. table 1-3: sdram control commands mnemonic description representation {ras, cas, we} mrs mode register set 111 ref auto/self refresh 110 pall precharge all banks 101 actv activate row 100 writ write 011 read read 010 bst burst stop 001 nop no operation 000 1.read-write control, supported by the assembler (srw, rrw) is not implemented. there are no data structures programmable for special transfers. table 1-4: flow control instructions mnemonic description representation run run linear program ?w 000 ret return from sub rou- tine 001 call call subroutine on address argument (no nested calls possible) 010 loop repeat program at address argument if loop counter not reached 011 end alias for loop #0 when loop counter is ? 011 pde power down entry 110 pdx power down exit 111 sdram controller function description page 81 1.5 address mapping this section describes the relationship between logical and physical addresses and how to map from a given logical address to the physical memory position. at the begin we have to introduce the meaning of the used layer description record (ldr) information. the address unit uses the parameters of the 16 ldr entries: pha(i) physical address offset dsz_x(i) domain size x cspc_csc(i) color space code. the address offset pha , stored in the ldr, describes the start address of a layers position. this is where the most upper left pixel of a picture is located. due to the block structure of the picture data, only the part of the row address is valid for the physical start address offset entry (bits [22:12] for lavender or [19:10] for jasmine, see figure 1-3). lower bits are fixed to ?? that restriction applies because of the same row can? be used for different layers. domain size in x-dimension dsz_x is given in logical pixels. it is needed to calculate the pixels per line. the y-dimension is not needed, there is no automatic layer size limitation im- plemented. please note that there is no dsz_y register implemented. color space code cspc_csc is a rep- resentation of the appropriate color format. it is converted internally to calculate the bits per pixel (bpp) by power of two, which is equal to the number of bits the pixel address has to be shifted to get the right word address. the physical address is a combination of sdram bank, row and column address. the significance is pre- defined in following order from row over bank to column address. thus the picture data is stored in a block- ing structure drawn in section 1.5.1, figures 1-6/1-7. additional to physical word addressing there can be distinguished between several byte addresses. the complete physical address format is shown in figures 1- table 1-5: color space codes code color space type bpp shift 0x0 1bpp 1 0 0x1 2bpp 2 1 0x2 4bpp 4 2 0x3 8bpp 8 3 0x4 rgb555 16 4 0x5 rgb565 16 4 0x6 rgb888 32 5 0x7 yuv422 16 4 0x8 yuv444 32 5 0x9-0xd reserved for gpu intermediate color space 0xe yuv555 a a.jasmine only 16 5 0xf yuv656 a 16 5 MB87J2120, mb87p2020-a hardware manual page 82 3 and 1-4 as it is used for ipa and dpa access methods.for ldr entries of pha only row bits are valid, lower bits are fixed to ?? this is due to the physical start address is aligned on the row grid. for comparison, logical address format is shown in figure 1-5. now back to the relationship between logical pixel address containing layer, x and y location and the phys- ical mapping of the pixel data. the needed bits per pixel line of the layer can be calculated as 1 xbits = domszx << shift = domszx * bpp remark: expression << shift ?interpretable as * bpp ?with the restriction that bpp has a value range of power of two { } for determining shift value from its color space code see table 1-5. it depends on the layer number and how cspace is configured for it. layer memory can only be divided into whole numbered parts of rows in each dimension. each row seg- ment has a width of 8 words. lavender uses 2 adjacent banks with same row number, thus a virtual row of 16 words is formed. from the number of bits the needed number of horizontal memory rows is xrows = xbits[18:9] + (xbits[8:0]? 1: 0) (for lavender) xrows = xbits[18:8] + (xbits[7:0]? 1: 0) (for jasmine) finally the row, bank and column addresses can be derived from the logical address components and this temporary values. con catenation of row, bank and column address enhanced with 2 bits for byte addressing results in the physical address. ra = y[13:6] * xrows + (x << shift)[18:9] (for lavender) ra = y[13:5] * xrows + (x << shift)[18:8] (for jasmine) ba = {y[5], (x << shift)[8]} (for lavender only) ca = {y[4:0], (x << shift)[7:5]} 1.squared brackets stand for vector slices, curly braces are vector combinations. 12 22 11 10 9 2 1 0 31 23 not used row bank column byte figure 1-3: physical address format (lavender) 10 9 2 1 0 31 column byte 19 20 not used row figure 1-4: physical address format (jasmine) 0 31 15 14 30 29 16 13 layer [3:2] x layer[1:0] y figure 1-5: logical address format 2 0 2 1 2 2 2 3 2 4 2 ,, 5 ,,, sdram controller function description page 83 1.5.1 elucidations regarding address mapping 1.5.1.1 block structure of pixel data drams have not equal access timings if randomly accessed. if a row address is already activated, faster access can be done. each row consists of 256 col addresses. as compromise between horizontal and vertical operation a block oriented access scheme is implemented. a block is identical with one row, each 256 col addresses with faster access. disadvantage of this block structure is a more complicated pixel addressing over direct physical access methods. block size is defined to 8 words horizontal 1 and 32 lines vertical 2 . for example, this results to 32x32 pixel block size at 8 bpp. if bank interleaving is used (on lavender chip), same row address for each of the 4 banks are combined to a macro block with double size in horizontal and vertical dimension. this gives the chance to activate a row in another bank before reading from it during access is running on another bank. this hides row ac- cess time in most cases. for access by pixel address the block structure is not relevant. it is mapped automatically by hardware to the right physical address. if non-picture data or data which should not be displayed is stored via physical access, address can be interpreted as linear space without rows, banks and columns. only if physical access on graphic data is required, the block based philosophy should be considered. 1.number of pixels depending on color depth (bpp) 2.word and pixel have same meaning MB87J2120, mb87p2020-a hardware manual page 84 1.5.1.2 access methods and devices this section is not sdc relevant but the reader can have benefit in better architectural understanding of the gdc device. it depicts setup of other gdc macros which are related to addressing data in memory. pixel addressing for drawing commands there are two main sections of command regarding addressing method. first group uses pixel address in- formation over input fifo, second group makes use of registers for pixel coordinates. commands grp1 (fifo): putpixel, putpxwd, putpxfc, getpixel, xchpixel, dwline, dwpoly, dwrect, memcp if ppcmd_ulay register set to 0, complete logical address information is fed through input fifo. format consists of layer, x and y position. the 32 bit pixel address word is combined to (from msb to lsb) {l[3:2], x[13:0], l[1:0], y[13:0]}. if ppcmd_ulay register set to 1, layer information is used from layer register ppcmd_lay and is not evaluated from input fifo. the appropriate layer bits of input fifo data are don? care. address format consists of x and y position. the 32 bit pixel address word is combined to {l[- -], x[13:0], l[- -], y[13:0]}. commands grp2 (regs): putbm, putcp, puttxtbm, puttxtcp ... column 0 7 0 00 f8 f8 08 00 ... 7 ... ... row 2 word row 0 row 1 ... bank 0 bank 2 bank 3 bank 0 bank 1 bank 2 bank 3 bank 0 bank 1 bank 1 bank 2 bank 3 figure 1-6: memory mapping of bank, row and column address (lavender) sdram controller function description page 85 these commands have it? dedicated coordinate registers. no address information is fed through input fifo. they are using xymin, xymax, ppcmd_lay registers. direct memory mapped physical access (dpa) command interface is not necessary for this access method. however some specialities should kept in con- sideration while using the dpa interface. dipa clock is enabled dpa should be enabled by setting sdflag to ? two windows are possible to map into mcu address space (shares gdc chip select) window address offset wndof and size wndsz are mapped to required address space wndsd is set to the section start address of video ram which appears in the window mapped address calculates to phy_map = cs_region + wndof - wndsd. wndsz limits size of accessible address range. if exceeded no write permission is granted and tri-state buffers kept close at reading. dpa runs completely unbuffered. additional there are no real-time or preferred data channels to the video ram available, the normal sdc requesting and arbitration procedures apply. the normal case is that dpa row row 3 row 2 ... ... column 0 7 0 00 f8 f8 08 00 ... 7 row row n row ... word ... row 0 row 1 n+1 n+2 2n ...... ...... figure 1-7: memory mapping of row and column address (jasmine) MB87J2120, mb87p2020-a hardware manual page 86 has to wait for higher prioritzed jobs an the currently running task for completion. thus only a slow access and difficult predictable timing results from this behaviour. the not predictable access time requires dpa_rdy polling at writing due to the rdy line pull down feature is in general only supported for read access in gdc. with setting higher priority for dpa access than gpu (display output) this situation can be improved. this can help when high bandwidth components are running continuously, i.e. gpu, pp and vic. the risk of interrupting the real-time streams of gpu and vic increases only negligible, but beware of changing de- fault priority when working at the upper limit of bandwidth consumption. indirect physical access (ipa) commands: putpa, getpa ipa makes full benefit of physical access while using burst transfer techniques. additional no restrictions apply with address range limitation. physical address is transferred to the input fifo. data packets are also routed through input or output fifos. to achieve maximum data throughput physical address auto-incre- ment is implemented for getpa function. if logical pixel data is transferred via physical access, be aware of physical address incrementing method. due to the fact that single transfers with converted addresses (logical to physical) are not effective over this device, the user should check if block based transfers are possible. pixel data have to be divided into seg- ments even to 8 data words in x-dimension and then next line of block can be transferred. burst transfer should start aligned on the block grid (8x32 words). with this method only one start address has to be con- verted and sent followed by a data block transfer of up to 256 words is possible. another way is the transfer of 8-words line segments with a start address with only moderate amount of address overhead. this has the advantage that there is no need for restricting pixel position to the block grid in y-dimension. problematic in any case is random access on pixels over physical addresses. command and address calcu- lation effort is too high. dedicated pixel commands should be used then. if packetized block transfer is used, priority of ipa device has not that much influence on data rate com- pared to dpa. but increasing of ipa priority may cause interruption of real-time processes of vic and gpu. only two devices with highest priority setting are kept as real-time due to the arbitration scheme. important parameters for ipa are input and output fifo data amount min/max thresholds (dipaif_ifmax, dipaif_ifmin, dipaof_ofmax, dipaof_ofmin). this thresholds control when a transfer is started/stopped (max) and adjust block size of memory transfers (min). higher block size im- proves performance but increases risk of data stream interruption for other devices. 1.5.1.3 program example following example demonstrates address mapping and physical access with relationship to pixel coordi- nates. intention is to draw a rectangular area with radiancies and finally copy the drawn object per physical access. sdram controller function description page 87 the example given in figure 1-8 did not use polling of dpa_rdy flag. the information that the dpa de- vice is ready for writing is implicitly given by the finished read access before. dpa has a two-stage buffer and read access is synchronized by using the hardware flow control over ulb_rdy line. the following write access has at least one free buffer available. /* ... before this section of code init_gdc from mkctrl tool was included */ clrlayer(0x008080, 0); clrlayer(0x008080, 1); /* ... here initialization of window handles w0...w2 is normally included */ xoff = 30; yoff = 45; aafen = 1; /* demonstration of physical access - draw original pattern */ puts (&w0, "[1] drawing original pattern\n"); drawrect (0xd0d000, 0, xoff, yoff, 100, 100); gdc_cmd_nop(); /* cancel dwrect before aaf enable */ pxp_aaf_the(aafen); for (ang=0; ang<6.2828; ang+=0.15707) { /* draw radiancies */ recx = (word) (50*sin(ang)); recy = (word) (50*cos(ang)); drawline(0x000000, 0, (xoff+50)< sdram controller function description page 89 /* build physical address from pixel address, function uses dram address */ /* window 0 only, pls ensure that dram mapping to mcu address is enabled */ dword phy_address (volatile byte layer, word x, word y) { byte bpp; /* color depth lookup, 1/2/4/8/16/32 */ byte ca; /* dram column address (8 bit) */ word domszx; /* layer size x-dimension (14 bit) */ dword xbits; /* number of bits for one line (19 bit) */ word xrows; /* number of row blocks in x-dimension (10 bit) */ word ra; /* dram row address (without layer offset) */ byte ba; /* dram bank address, lavender only (2 bit) */ dword physdc; /* physical address sdram controller view */ dword phymcu; /* physical address mcu view */ dword layoffs; /* layer offset */ /* determine layer parameters and number of bits per line of pixels (x)*/ /* formula: xbits = domszx << shift = domszx * bpp */ bpp = bpp_lookup(layer); /* layer color depth */ domszx = g0dsz_x(layer); /* layer pixel width */ layoffs = g0pha(layer); /* layer start address */ xbits = domszx * bpp; /* layer bitsize width */ /* column address, formula: ca = {y[4:0], (x << shift)[7:5]} */ /* y: 32 lines, x: 8 words each block */ ca = ((y & 0x1f) << 3) + (((x*bpp)&0xff)/32); switch (g0clkpdr_id) { case 1: /* ----------- id: jasmine, gdc-dram, single bank ------------ */ /* number of memory rows (grid of pixel blocks) in x-dimension */ /* each partially used row has to be considered (add 1 if remainder) */ /* formula: xrows = xbits[18:8] + (xbits[7:0]? 1: 0) */ xrows = (xbits>>8) + ((xbits & 0xff)? 1: 0); /* dram row address relative to layer start address, each row has a */ /* block size of 256 bit in x-dimension and 32 lines in y-dimension */ /* formula: ra = y[13:5] * xrows + (x << shift)[18:8] */ ra = y/32*xrows + (x*bpp/256); /* combination of address bits: [19:10]ra, [9:2]ca, [1:0]byte */ physdc = layoffs + (ra<<10) + (ca<<2); break; case 0: /* -------- id: lavender, gdc with external dram, 4 bank ----- */ /* formula: xrows = xbits[18:9] + (xbits[8:0]? 1: 0) */ xrows = (xbits>>9) + ((xbits & 0x1ff)? 1: 0); /* row block size is 512 bit in x-dim, 64 lines in y-dim */ /* formula: ra = y[13:6] * xrows + (x << shift)[18:9] */ ra = y/64*xrows + (x*bpp/512); /* there exist 4 banks with same row address, thus each row block */ /* consits of 4 bannk parts */ /* formula: ba = {y[5], (x << shift)[8]} */ ba = ((y>>5) & 0x1)*2 + (((x*bpp)>>8) & 0x1); /* combination of address: [22:12]ra, [11:10]ba, [9:2]ca, [1:0]byte */ physdc = layoffs + (ra<<12) + (ba<<10) + (ca<<2); break; default: return -1; /* err: wrong chip id */ } /* add gdc0 offset (g0cmd is first gdc address) and ulb settings */ /* such as mcu window 0 offset and subtract sdram offset */ /* check consistency if memory window is mapped and reachable */ if (! g0sdflag_en ) return -2; /* err: sdram mapping not enabled */ if (physdc < g0wndsd0 || physdc >= g0wndsd0+g0wndsz0) return -3; /* err: pixel address outside mapped video ram space */ /* dram address gdc base address */ phymcu = (dword) (char *) physdc + (dword) (char *) &g0cmd /* mcu offset sdram offset */ + (char *) g0wndof0 - (char *) g0wndsd0; return phymcu; } figure 1-12: logical to physical address conversion routine MB87J2120, mb87p2020-a hardware manual page 90 last example is an intelligent clearlayer function, which determines the start address of the next layer au- tomatically. in that way layer size in y-dimension is calculated before drawing a monochrome rectangle filling the complete layer. 1.5.1.4 address calculation - example with concrete {x,y} pixel this example calculation is for the lavender chip. first calculation returns number of row-blocks in x-di- mension. we need 20 (0x14) rows side by side for storing 640 pixels with color depth 16 bit. xbits = domszx * bpp = 640 * 16 = 10240 = 0x2800 xrows = xbits[18:9] + (xbits[8:0]? 1: 0) = 10240/512 + 0 = 20 = 0x14 mkctrl tcl-gui name in gdc_reg.h value gdc_offset0 wndof0 0x40000 gdc_size0 wndsz0 0x80000 sdram_offset0 wndsd0 0x0 physical address layer 0 pha(0) 0x20000 domain size x dsz_x(0) 640 color space code cspc_csc(0) rgb555 (16bpp) void clrlayer(dword color, byte layer) { dword r0[2]; int j; byte ppw; dword next_layer_phy, physsz; word domszxwrd, domszxblk, domszylin, domszy; /* --------------- calculation of domain size y --------------- */ next_layer_phy = 0x100000; /* max memory size */ /* search next layer start address and calculate physical size from actual to next layer */ for (j=0; j<16; j++) { if ((layer!=j) && (g0pha(j)>g0pha(layer)) && (next_layer_phy>g0pha(j))) { next_layer_phy = g0pha(j); } } /* calculate max domain size y to fit in physical layer size */ physsz = next_layer_phy - g0pha(layer); ppw = 32/bpp_lookup(layer); /* pixel per word */ domszxwrd = g0dsz_x(layer)/ppw + (g0dsz_x(layer)%ppw? 1: 0); if (g0clkpdr_id == 1) { /* jasmine 8x32 row blocks */ domszxblk = domszxwrd/8 + ((domszxwrd & 0x7)? 1: 0); domszylin = physsz/4/domszxblk/8; domszy = domszylin - (domszylin & 0x1f); } if (g0clkpdr_id == 0) { /* lavender 16x64 row blocks */ domszxblk = domszxwrd/16 + ((domszxwrd & 0xf)? 1: 0); domszylin = physsz/4/domszxblk/16; domszy = domszylin - (domszylin & 0x3f); } /* ----------------- clea layer function ------------------- */ r0[0] = pix_address(layer, 0, 0); r0[1] = pix_address(layer, g0dsz_x(layer)-1, domszy-1); gdc_cmd_dwrt(color); gdc_fifo_inp(r0, 2, 0); } figure 1-13: clearlayer with automatic layer size detection sdram controller function description page 91 assumed first pixel {0,0} address is searched for, physical address is layer 0 start address directly. phymap = gdc_cs_area + wndof0 - wndsd0 = 0x30000000 + 0x40000 - 0 phymap = 0x30040000 phyaddr = phymap + pha(0) = 0x30060000 now for pixel position of layer 0, x=115, y= 250 physical address should be transformed. ra = y[13:6] * xrows + (x << shift)[18:9] ra = 250/64*20 + 115*16/512 = 3*20 + 1840/512 = 63 = 0x3f = 0b11 1111 ba = {y[5], (x << shift)[8]} = {bit 5 of 0xfa, bit 8 of 0x730} = 0b11 ca = {y[4:0], (x << shift)[7:5]} = {0b11010, 0b001} = 0b1101 0001 con catenation of ra, ba and ca results in the physical sdram word address. additional con catenation of two bits 0b00 convert it to a byte address (x 4). physdram = {ra, ba, ca, 0b00} = 0b11 1111. 11.11 0100 01.00 = 0x3ff44 finally physical address offset for layer 0 has to be added. phyaddr = 0x20000 + 0x3ff44 = 0x5ff44 this address can be used for ipa directly. for dpa gdc_cs_area, wndof0 and wndsd0 have to be considered. phyaddrdpa = phymap + phyaddr = 0x30040000 + 0x5ff44 = 0x3009ff44. MB87J2120, mb87p2020-a hardware manual page 92 2 sdram ports from timing point of view critical is specially read accesses due to the fact, that delays of clock to the mem- ory device and data back from memory to gdc have to be summarized relative to the internal rising clock edge. figure 2-3 shows when read data from sdram are valid after feeding back to gdc. there is the additional restriction for keeping the setup and hold times of the data input registers, which reduces the valid region of sampling time additionally. that? why active clock edge of the input flip-flop has to be inside the gray marked region. best decision is the mid of this to have enough space for deviations. there are different possibilities to compensate the part of total delay which is larger than one clock period. one is to insert a delay buffer in the clock lines of the input registers, another is to sample data on falling clock edge of input register. additional to the input register clock delay there are to satisfy the hold time requirements of all sdram input signals (outputs from gdc). due to the fact that the input signal timing has to be relative to the receiving clock at the sdram, this depends much on the delay of output clock buffer and clock wire delay. 2.1 external sdram i-/o-pads with con?urable sampling time (lavender) additional to the tri state control there is an special timing feature implemented at the sdram ports. inside the sdc there is a configurable hold time adjustment for the outputs and sampling time adjustment for the data input. the implemented solution uses configurable delays for each signal group (addresses and control signals, tri state control, write data and read data). in this way it? possible to compensate the influence of the board layout in a comfortable way. registers for the sdram interface control the timing of the ports. four different timings have to be satis- fied: hold time for sdram input pins address, command and dqm (tdcbtaout) hold time for sdram data input pins (tdcbtdout) sdram: d core: sdram: d core: clk clk tdclk tac toh tdd tdd tsd thd sampling area figure 2-1: sdram interface timing sdram controller sdram ports page 93 sampling time for sdram data outputs (tdcbtdin) delay for switching the tri state buffer enable signal (tdcbtoe) figure 2-2 shows the schematic of the implemented sdram interface circuitry. variable clock line delays are implemented as buffer chains with multiplexed taps. the multiplexers respective the resulting delays are controlled by a two-bit value for each signal group in the delay configuration byte. under typical con- ditions a programmable range from nearly 1ns up to 4ns is possible in steps of 1ns. recommended values for the interface setup are tdcbtaout=2ns, tdcbtdout=2ns, tdcbtdin=3ns and tdcbtoe=1ns. 2.2 integrated sdram implementation (jasmine) jasmine has no interface to external sdram. delay adjustment is not needed and not implemented for the integrated solution. d q dq dq qd dq din_delay sdram clock delay configurable interface registers signal and external wire delay clock driver ras cas we dqm a dq addr oe wdata_sdram rdata_sdram clkk addr_delay tristate_delay dout_delay clk chip border ras, cas, we, dqm, cke figure 2-2: design of sdram interface MB87J2120, mb87p2020-a hardware manual page 94 3 con?uration 3.1 register summary a summary of initialization control registers is given in table 3-1. defaults are for 100 mhz operation fre- quency. address definitions for symbolic word addresses are in the file ?bp_const.v? table 3-1: con?uration information of sdc symbol bits description reset value sdwait_opt [20] interleave opt a : execute precharge and acti- vate during running bursts of previous access. 1 unused jasmine sdwait_trp [19:16] trp: ras precharge time (pre -> actv, default 2 wait states) 2 sdwait_trrd [15:12] trrd b : ras to ras bank active delay time (actv -> actv, default 1 wait state) 1 unused jasmine sdwait_tras [11:8] tras: ras active time (actv -> pre, default 5 wait states) 5 sdwait_trcd [7:4] trcd: ras to cas delay time (actv -> read|writ, default 2 wait states) 2 sdwait_trw [3:0] trw: read to write recovery time (read -> writ, default 7 wait states) wrong reset value, jasmine requires 7/5, lav- ender 10/8, depending on used aaf or not. 3 ! sdinit [15:0] init period: power on stabilization time (default 20000) 20000 sdrfsh [15:0] refresh period: single row refresh period (default 1600) 1600 sdseqram[] [0:63] lavender [0:31] jasmine [13:0] sequencer ram [13:0], 64 words [13:7]addr, [6:4]instr, [3:0]{ras,cas,we,ap} (jasmine has [12:7] address argument) unde?ed sdmode [12:0] mrs c : sdram mode register [9] burst write enable, [6:4] cl, [3] interleave burst, [2:0] burst length (0:3)=1,2,4,8 / 7=full (default 0x033) 0x0033 lavender only sdif_tao [7:6] tdcbtaout d : address output register clock delay (controls also cke, dqm, ras, cas, we outputs) 0 lavender only sdif_tdo [5:4] tdcbtdout: data output register clock delay 0 lavender only sdif_tdi [3:2] tdcbtdin: data input register clock delay 0 lavender only sdif_toe [1:0] tdcbtoe: output enable register clock delay 0 lavender only sdram controller configuration page 95 3.2 core clock dependent timing con?uration 3.2.1 general setup sdram access wait states and refresh periods are configurable to support a wide range of scalability in matter of system performance and power consumption. additional to the row refresh time out value the re- fresh sequence in the micro program sequencer is adaptable for its dedicated core frequency. the configuration tool generates optimized settings for a given core clock frequency to met best perform- ance result. if a fixed setting should be used over a certain frequency range (i.e. if clock scaling is used with- out the effort spending for re-configuration) the minimum frequency is required to calculate the refresh period and the highest frequency should be used to calculate the values for the wait state timers. thus refresh condition and minimum access timing is always satisfied. 3.2.2 refresh con?uration for integrated dram (jasmine) setup of minimum refresh rate is based on core clock cycles. thus the value for the refresh period has to be configured for its dedicated core clock frequency. additional to the core frequency, maximum junction tem- perature has significant influence on refresh period. tjmax = up to 100 degc : tref = 16.4ms tjmax = 101 to 110 degc : tref = 8.2ms tjmax = 111 to 125 degc : tref = 4.1ms assumed row refresh duration for all 1024 rows is evenly distributed to refresh all rows within above spec- ification, refresh timing of 16/8/4 us have to be set up. for automotive temperature range 4 us are required. sdcflag_busy [0] cbpbusy: set these ?g before changing the power on initialization period or the sequencer ram. reset it after the access make changes take affect. 0 sdcflag_dqmen [1] dqm partial write feature e : optimize byte/ 8bpp and half word/16bpp access if set to 1. 0 jasmine only pha[0:15] [22:12] [19:10] layer start addresses: row address offset for start position of each layer. first bit positions for lavender, second for jasmine. [22/19:0] can be handled as byte address, but only shown bits are stored. unde?ed dsz_x[0:15] [29:16] layer widths: x component of layer size in pixel unde?ed cspc_csc[0:15] [3:0] color depth table: color de?ition code for evaluation of the number of bits per pixel (bpp) unde?ed a.lavender only. jasmine works with single bank. b.has no effect for jasmine. c.fixed and not accessible for jasmine. d.dcbt interface timing adjustable for lavender only. e.jasmine only. table 3-1: con?uration information of sdc symbol bits description reset value MB87J2120, mb87p2020-a hardware manual page 96 page 97 b-4 pixel processor (pp) MB87J2120, mb87p2020-a hardware manual page 98 pixel processor functional description page 99 1 functional description the pixel processor (pp) is a component of the graphics display controller (gdc), which realizes the main functions ?rawing of geometrical functions and ?riting and reading single pixel to and from video ram. it is implemented in the gdc design between user logic bus interface (ulb) and the sdram con- troller (sdc)/anti aliasing filter (aaf), in reference to block diagram in gdc specification. the pp con- sists of three separate devices (execution devices) and other submodules. the execution devices are the pixel engine (pe) for drawing function, the memory access unit (mau) for single pixel access and the memory copy unit (mcp) for copying rectangular areas. the submodules are control interface, ulb in- terface, the sdc interface and the fifos for pixel addresses and pixel data. more information about the internal module can be found in the corresponding chapters and sections. 1.1 overview 1.1.1 pp structure the internal structure of pp is shown in figure 1-1. it consists of the execution devices (pp, mau, mcp) and the submodules control interface, ulb interface and sdc interface. the ulb block in the figure 1-1 sdc bus command control and data pixel processor sdc/aaf ulb control bus sdc-inter- ulb-interface ctrl-interface pe mau mcp fifo addrdata figure 1-1: pp block diagram MB87J2120, mb87p2020-a hardware manual page 100 is only for understanding pp in context of gdc and represents the ulb functionality, such as command control signals and data bus for reading and writing. 1.1.2 function of submodules the pe is a unit, which realizes the main function of pp, which are drawing of geometrical figures and writ- ing compressed and uncompressed pictures into the video ram. the supported commands are: drawing commands dwline (draw lines) dwrect (draw arectangular areas) dwpoly (draw a polygon) bitmap commands putbm (write an uncompressed bitmap into video ram) putcp (write an rle compressed bitmap into video ram) puttxtbm (write an uncompressed 1 bit pixel mask as a bitmap with a higher colour depth into video ram) puttxtcp (write an rle compressed 1 bit pixel mask as a bitmap with a higher colour depth into video ram). the second unit in pp is the mau. it realizes the single pixel access for reading and writing. following commands can be used by the programmer: pixel commands putpixel (set one pixel on the display) putpxfc (set one pixel with a ?ed colour on the display) xchpixel (read-modify-write of a single pixel) getpixel (read one pixel from the videoram) putpxwd (write a 32bit data word with a number of pixels into the video ram; the number of pixels depends on the colour depth of the target address). the third execution device is the mcp. it gives a simple way for the programmer to copy rectangular areas from the source layer to the target layer. the only command, which can be used is: memory to memory commands memcp (copy rectangular area from one layer to another) the sdci is a component, which manages the access to the sdc. the sdci collects pixel adresses and data to packages, which should be transferred to the video ram with one request. the addresses and data are collected in the fifos (address and data fifo), every one of them has a size of 32bit*(64+2) words. the control interface is connected to the internal control bus system of the gdc, which allows the pro- grammer to have access to the configuration registers of all sub modules or macros. all configuration reg- isters of the pp are connected to the control bus system. pixel processor functional description page 101 1.2 con?uration registers the following table 1-1 shows the configuration registers of pp and their initial values. table 1-1: register list of the pp address name width bits comment initial 1240h 1244h 1248h 124ch 1250h 1254h 1258h 125ch 1260h 1264h 1268h 126ch 1270h 1274h 1278h 127ch cspc_csc[0:15] 4 [3:0] color space code (to calculate color depth) for layer[0:15] color de?ition of layer 0 color de?ition of layer 1 color de?ition of layer 2 color de?ition of layer 3 color de?ition of layer 4 color de?ition of layer 5 color de?ition of layer 6 color de?ition of layer 7 color de?ition of layer 8 color de?ition of layer 9 color de?ition of layer 10 color de?ition of layer 11 color de?ition of layer 12 color de?ition of layer 13 color de?ition of layer 14 color de?ition of layer 15 0h 4100h bgcol_en bgcol_col a 1 24 [24] [23:0] enable background color background color data (depends on colour depth) background pixel colour data is used by commands puttxtbm and puttxtcp 0h 4104h fgcol 24 [23:0] foreground colour data (depends on colour depth) 0h 4108h ignorcol_en ignorcol_col 1 24 [24] [23:0] enable ignore colour ignored colour data (depends on colour depth) ignored pixel colour data is used by commands putbm and putcp 0h 410ch linecol 24 [23:0] line colour data (depends on colour depth) line pixel colour data is used by command dwline 0h 4110h pixcol 24 [23:0] pixel colour data (depends on col- our depth) pixel colour data is used by com- mand putpxfc 0h 4114h plcol 24 [23:0] polygon colour data (depends on colour depth) line pixel colour data is used by command dwpoly 0h MB87J2120, mb87p2020-a hardware manual page 102 4118h rectcol 24 [23:0] rectangle colour data (depends on colour depth) line pixel colour data is used by command dwrect 0h 411ch xymax_xmax xymax_ymax 14 14 [29:16] [13:0] endpoint x-dimension endpoint y-dimension bitmap shape description (stop address incrementation) for the commands putbm, putcp, puttxtbm and puttxtcp 0h 4120h xymin_xmin xymin_ymin 14 14 [29:16] [13:0] startpoint x-dimension startpoint y-dimension bitmap shape description (start address incrementation) for the commands putbm, putcp, puttxtbm and puttxtcp 0h 4124h ppcmd_ulay ppcmd_lay ppcmd_en ppcmd_dir ppcmd_mir 1 4 1 1 2 [28] [27:24] [16] [8] [1:0] use target layer register target layer enable anti aliasing ?ter bitmap direction bitmap mirror target layer information is used by putbm, putcp, puttxtbm, puttxtcp any time. it is used by dwline, dwrect, dwpoly if ulay enabled. direction is relevant for putbm, putcp, puttxtbm and puttxtcp (0=horizontal, 1=vertical). mirror b for putbm, putcp, puttxtbm, puttxtcp (0h=with- out mirroring, 1h=x-axis, 2h=y- axis, 3h=x- and y-axis). 0h 4128 sdcprio 3 [2:0] pp priority for sdc-request 0h 412ch reqcnt 8 [7:0] maximum number of data words, which should be tranferred in one video ram access cycle (packet size). c word count = reqcnt+1 0h 4130h readinit 1 [0] read con? register, double buffer read access behavior: 0=internal buffered register, 1=con? register. d 0h a. note! all colour data in the registers is right aligned (lsb) and depends on the colour depth/resolution. b. the mirror funcionality means doubling at the limitation lines, which are de?ed by xmax or ymax (see ?ure 1-2). table 1-1: register list of the pp address name width bits comment initial pixel processor functional description page 103 1.3 special command options 1.3.1 bitmap mirror bitmap mirror option is for enhancement of symmetric objects. original object is drawn in any case. it is possible in one or both x- and y-direction. thus only the half or the quarter of an bitmap needs to be trans- ferred. mirroring is usable for bitmap commands putbm, putcp, puttxtbm and puttxtcp. mirroring axis are defined by the end point coordinates. 1.3.2 bitmap direction bitmap direction option swaps x- and y- coordinates relative to the start point. original bitmap is not drawn. it could be understood as flipping the object at an axis through the start point with an angle of 45 degree. if ppcmd_dir = 1 the new end point calculates to xmax?= xmin + ymax - ymin ymax?= ymin + xmax - xmin c. reqcnt is important for bandwith considerations. large values give fast pp data throughput but increasing risk of shortage for real-time devices gpu/vic. if reqcnt is different from 0, data can remain in pp fifos (not suf?ient amount for packetized transfer). the remaining data is ?shed with start of the next command, i.e. if the drawing is ?ished with a noop command. d. to avoid consistency problems during read-modify-write access it is recommended to set readinit behav- ior to ?? thus read and write accesses use the same register. the copy procedure from con?uration register to the internal buffered register is controlled by the pixel processor. the internal value is ?ed during running commands. xmin xmax ymin ymax direction mirror in x-direction mirror in x-direction and y-direction mirror in y-direction figure 1-2: mirroring function ymax? ymax ymin xmin xmax xmax? ppcmd_dir = 0 ppcmd_dir = 1 figure 1-3: direction option MB87J2120, mb87p2020-a hardware manual page 104 2 format de?itions this chapter describes the colour and data formats of all pp specific interfaces, such as ulb interface or sdc interface. you can also find explanation of all used symbols. 2.1 legend of symbols the table describes the mnemonics of data formats. table 2-1: mnemonics and symbols mnemonic & symbols description [...] data double word (32bit): [single colour data]: one pixel [colour data]: more than one pixel in one word (uncompressed) [coded data]: rle compressed pixel data [colour enable data]: 1bpp pixel mask (uncompressed) [coded colour enable data]: rle compressed 1bpp pixel mask <...> data byte (8bit): [l, x, y] or [l high , x, l low , y] pixel address (32bit): {...} {x, y}: pixel coordinates (2*14bit) {l high , l low }: layer number (2*2bit) {reg}: value of register, e.g. rcolor 31 23 15 7 0 byte0 byte1 byte2 byte3 70 byte 31 29 15 13 0 layer number x-coordinate y-coordinate 32 10 pixel processor format definitions page 105 2.2 data formats at ulb interface color data at ulb interface (ififo/ofifo) and for configuration registers is lsb aligned. only data for putpxwd is an exceptional case, it delivers data in physical form as stored in video ram. the ulb delivers data in three formats: pixel addresses single pixel data: all single colour data is rigth aligned in the data word; the number of used bits depends on the colour depth. pixel data stream (compressed or decompressed): (...) data package: it consists of one or more 32bit data words, which contain all necessary in- formation for the current command. e.g. ([x, y], [l high , x, l low , y]) this is the data (start point, end point and layer) for the command dwline, two data words are used. table 2-2: symbols of rle compression resolution mode command byte colour bytes bpp1 compressed <1nnn nnnc> colour information in command byte (bit[0]) uncompressed <0nnn nnnn> (nnnn nnn+1 mod 8) bytes with colour data MB87J2120, mb87p2020-a hardware manual page 106 2.3 data formats for video ram / sdc interface at the sdc interface all colour formats are possible and supported. color data is transferred and stored in msb aligned form. this is opposite to the ififo/ofifo data at ulb interface. rgb888 compressed <1nnn nnnn> 3 bytes colour data ( pixel processor format definitions page 107 all pixel data in block (single pixel) or burst mode are left aligned. in burst mode more than one pixel data can be in one data word. table 2-3: color formats on the sdc interface colour format comment rgb888 true colour format single pixel ([single colour data]) pixel burst ([colour data]) rgb565 high colour format single pixel ([single colour data]) pixel burst ([colour data], one data word of the burst and maximum of two pixel per word) rgb555 high colour format single pixel ([single colour data]) pixel burst ([colour data], one data word of the burst and maximum of two pixel per word) 31 23 15 7 0 red green blue dummy 31 23 15 7 0 red green blue dummy 31 20 15 7 0 red green blue dummy 26 31 20 15 7 0 red green blue 26 red green blue first pixel second pixel 31 20 15 7 0 red green blue dummy 26 21 dummy 31 20 15 7 0 red green blue 26 21 dummy red green blue first pixel second pixel dummy MB87J2120, mb87p2020-a hardware manual page 108 bpp8, bpp4, bpp2, bpp1 colour index format these colour formats represents index information for the clut (colour lock-up table) and not colour information in rgb format. the maximum of numbers in one word depends on the colour resolution (bits per pixel). single pixel ([single colour data]) pixel burst ([colour data]) table 2-3: color formats on the sdc interface colour format comment 31 23 15 7 0 dummy colour index byte, depends on colour resolu- tion; the left aligned bits are used 31 23 15 7 0 31-(bpp-1) first pixel page 109 b-5 antialiasing filter (aaf) MB87J2120, mb87p2020-a hardware manual page 110 aaf antialiasing filter functional description page 111 1 functional description the anti aliasing filter (aaf) is a component of the graphics display controller (gdc), which realizes the anti aliasing filter function for some pixel commands. it is implemented in the gdc design between the pixel processor (pp) and the sdram controller (sdc), in reference to block diagram in gdc specifica- tion. the aaf processes data on the fly which should be stored in the video ram. only for true or high colour resolution pixel data an applicable result could be expected. averaging of indexed colours may lead to unexpected effects. 1.1 aaf overview 1.1.1 top level structure the internal structure of aaf is shown in figure 1-1. it consists of a bypass multiplexor, a control block, the antialiasing unit (aau) itself and two compatible interfaces to the sdc and pp modules. the bypass is needed to switch the aaf in a transparent (disabled) mode. if aau is disabled and the bypass is active the clock supply of the whole aaf module can be switched off for power saving purposes. addi- tional if aaf is not used the sdram timing for read-to-write recovery can be made faster. the bypass sdc bus sdc bus aaf control signals aaf_en anti aliasing filter figure 1-1: aaf block diagram sdc-interface (sdc side) sdc pixel processor ulb sdc-interface (pp side) all signals of sdc bus, such as req, mctrl, addr ... aaf enable signal register set data, such as thresh- old or opti_en ... control bus ctrl-interface anti aliasing unit MB87J2120, mb87p2020-a hardware manual page 112 mode guarantees a direct connection between pp an sdc without pixel data and pixel address manipulation by aaf (aau is off). the control block realizes reading or writing of configuration registers. the anti aliasing unit manipulates the pixel data and manages the access to pp and sdc. 1.1.2 aau function the main function of the anti aliasing algorithm is to reduce the staircasing effect. this effect is a result of projecting (drawing) objects onto the physical pixel resolution of the display. to get a higher image quality the implemented super sampling algorithm could be used to avoid stair effects on the edges of objects, e.g. lines or circles. in this case the anti aliasing is done on the fly, that means that during the drawing of objects the object itself is antialiased with the current background. this results in smoother edges by averaging fore- ground and background color information. 1.1.2.1 super sampling the objects are drawn at a higher resolution, which per definition leads to less staircasing. this virtual res- olution is doubled in x and y direction for 2x2 super sampling or four times in both dimensions for 4x4 super sampling. the image is then converted to the normal resolution by averageing the sub pixels to a new resulting pixel, which is stored in video ram. foreground pixels are generated by pixelprocessor or written through its interface. background pixels are read from memory. 1.1.2.2 drawing resolution the super sampling method requires doubled or four times the drawing resolution. if antialiasing is enabled, coordinates have to be doubled or multiplied by four. the pixel processor did not enhance the resolution automatically. it is the task of the application to work on the finer resolution of the sub-pixel grid. a side-effect of the high er resolution is that the amount of generated pixel data is four or sixteen times high- er compared to the original picture without antialiasing. additional each write access converts to a read- modify-write access du to the fact that background information is required. this has to be considered in terms of drawing performance. to reduce the effect of the high number of read-modify-write (rmw) ac- cesses to the same pixel coordinate write-back optimization is implemented for subsequent writes to the same pixel address. it is recommended to switch on optimization by setting opti_en to ?? 1.1.2.3 data and address processing the algorithm is based on single pixel processing due to sequential averaging in video memory. therefore pp is forced to block mode, each pixel data requires its own address information. packet data formats are not supported (e.g. putpxwd). incoming pp requests of variable block size were divided into single rmw accesses to the sdc. thats why some restrictions apply during aaf usage, which will be discussed in sec- tion 1.3.1. figure 1-2: antialiasing with super sampling. from left to right: without antialiasing, doubled resolu- tion with sub-pixels, averaging of sub-pixels. aaf antialiasing filter functional description page 113 formulas for sequential averaging of foreground and background data are shown in table 1-1. pixel data from pp is fg, background from memory is bg and fg is the new resulting pixel, which is written to video memory. the low precision of the 2x2 super sampling can be compensated by adaptive coefficient selec- tion, controlled by color difference thresholds. the value of the thresholds depends on human eye sensitiv- ity on chrominance contrast and differs for various types of graphics. right setup values could be evaluated empirically. addresses of pixel coordinates are divided by 2 in 2x2 super sampling mode for each x/y dimension. in 4x4 super sampling mode pixel address of resulting pixel is divided by 4. thus 4 or 16 sub-pixels of the generated high-resolution object are mapped to the same pixel address in video memory. the sequential av- eraging is the result of the multiple applied formula on the same memory address. 1.2 con?uration registers the following table shows the configuration registers of aaf and their initial values. note! the number of the used bits of threshold registers depends on colour depth. that meens: rgb888: for all colour channels all bits of threshold registers are used by aaf algorithm rgb565: for the red and blue colour channel only the bit[7:3] are used for calculation and for the green channel the bit[7:2] table 1-1: formulas for sequential averaging formula condition fg?= (2 fg + 2 bg) / 4 2x2 super sampling size. if difference between fg and bg color is below threshold. threshold1 applies if fg aaf antialiasing filter functional description page 115 1.3.2 supported colour formats the aaf supports only pixel data manipulation for true or high colour resolutions like rgb888, rgb565 and rgb555. 1.3.3 related sdc con?uration the information in this section is very important for the programmer, because wrong timing at the sdc can hang up the aaf. the ?ead to write recovery time (trw) in the sdc should be set to the minimum value of 10 for lavender or 7 for jasmine. this timing parameter selects a delay between read and write access to the sdram and thus it influences the difference between read data and write data appear on the internal table 1-4: supported colour formats colour format comment rgb888 true colour format (single pixel a ) a. a single pixel means one pixel in a 32bit data word at the sdc bus (sdc format in block mode; pixel data is msb aligned). rgb565 high colour format (single pixel) rgb555 high colour format (single pixel) bpp8, bpp4, bpp2, bpp1 colour index format (single pixel) this colour formats can not be correctly manipulated by the aaf, because the pixel data represents index information for the clut (colour lock-up table) and not colour information in rgb format. if the aaf is used and one of this colour formats should be transferred to the sdc, the aaf calculates a new colour index, which is stored in the sdc. the visible result at the display may not be the intended one. note! to avoid such problems, the programmer should disable the aaf for this col- our format access. 31 23 15 7 0 red green blue dummy 31 20 15 7 0 red green blue dummy 26 31 20 15 7 0 red green blue dummy 26 21 dummy 31 23 15 7 0 dummy colour index byte, depend on colourcolour resolution the left aligned bits are used MB87J2120, mb87p2020-a hardware manual page 116 bus connected to the aaf. aaf requires at minimum two extra clock cycles which is needed for internal data processing. description of events at clock cycles from figure 1-3: 1. at this edge the aaf takes the pixel data of the old pixel from rdata bus (sdc side). 2. at this event the aaf sets the result at the wdata bus. 3. sdc takes the pixel data of new pixel from the wdata bus. note! the delay between rvalid (1.) and wvalid (2.) can be bigger, but not smaller! to gain more system performance, trw is possible to reduce to 8 (lavender) or 5(jasmine) if antialiasing filter is bypassed. no other module has any requirements to this read-write timing. the minimum values of 8/5 guarantee that there will be no bus conflict at the sdram interface itself. the setting of trw is inde- pendent from core clock frequency. there is no requiremnt to reduce trw setting to its minimum. it is only an aspect of performance tuniung. additional it did not that much influence system performance, it is save to setup a value of 10 or 7 all the time. table 1-5: minimum settings for trw jasmine lavender aaf on 710 aaf off 58 2 clock cycles clkk rvalid rdata(31:0) wvalid wdata(31:0) 3 clock cycles 1. 2. 3. set by trw register (minimum 10/7) figure 1-3: timing for read-modify-write on sdc bus for aaf access page 117 b-6 direct and indirect physical memory access unit (dipa) MB87J2120, mb87p2020-a hardware manual page 118 dipa direct/indirect physical access overview page 119 1 overview the purpose of the dipa device is to provide video memory (frame buffer) access methods for the mcu without the need of the pixel processor inside the graphic display controller (gdc) to be used. this is per- formed by using physical sdram write and read access. the data words in video memory can be accessed with raw physical addresses. the function divides into two blocks, dpa (direct physical memory access) and ipa (indirect physical memory access). dpa is for direct memory mapped video ram access, address range of video ram has a certain offset regarding gdc internal mapping and gdc chip select address mapping of mcu address space. data trans- fer over dpa are single word, half word or byte accesses. each single access has to be arbitrated with com- peting gdc devices gpu, vic and pp by the sdram controller (sdc), therefore this method of direct frame buffer access is relative slow. ipa offers also the possibility to have physical access to the video ram. main advantage is that the transfer is executed in a buffered manner over the input and output fifos. addressing is done without any address offset calculation required. physical address is transferred as parameter from 0 to the upper bounds of video memory size. data blocks can be transferred with a given block size and start address. the slow-down due ipa prio dpa prio ifmax ifmin ofmax ofmin config sdram controller ififo ofifo sdc interface ipa dpa dipa ulb figure 1-1: dipa block diagram within gdc environment MB87J2120, mb87p2020-a hardware manual page 120 to the needed sdc arbitration time has much less influence when multiple data words are transferred as packet at once. additional performance increase is reached because of fifo buffering. ipa is used if the following commands are executed: putpa, write address and data ?ld of variable size via input fifo getpa (n), read n data words via output fifo the sdc interface multiplexes between ipa and dpa devices and controls its concurrence. dpa has prec- edence over ipa. there is no necessity for physical access of video data only. not used video memory, i.e. if the memory is not assigned in the layer description record (ldr, see gdc registers), can be used for general purposes. if pixel data should be accessed the logical to physical address mapping has to be considered (see sdram controller specification). for non-image data addressing can be in a linear way. dipa direct/indirect physical access configuration registers page 121 2 con?uration registers 2.1 register list configuration of dipa internal registers is needed for sdram arbitration priority information and to con- trol ipas fifo buffering. this influences system performance and bandwidth balance only. initial values (reset-state) for ipa configuration are not default values an some of them restricted for oper- ation. an configuration of block sizes is required before ipa operation is possible. 2.2 recommended settings figure 2-1 shows the relationship between ipa settings for input or output fifo and the current fifo load. if the load is smaller than ifmin or ofmin no data transfer is performed. in the load range between ifmin/ofmin and ifmax/ofmax a single transfer with at least ifmin/ofmin data words is triggered. above ifmax/ofmax fifo load the transfer is packetized into packages of ifmax/ofmax size. the configuration with respect to system behaviour depends on the bandwidth reserve for video ram ac- cess (transfer rate and core clock dependent). normally gpu and vic have highest access arbitration pri- ority due to the fact that this are real-time applications and its bandwidth violations would lead to data loss (drop of display or video data may be the result of wrong configuration). table 2-1: dipa performance adjustment registers register address bits function initial dipactrl_pdpa 0x4200 [18:16] dpa arbitration priority [0...7] 0x0 dipactrl_pipa 0x4200 [2:0] ipa arbitration priority [0...7] 0x0 dipaif_ifmin 0x4204 [6:0] minimum block size for data transfer to the video ram (>= 2) 0x02 dipaif_ifmax 0x4204 [22:16] maximum block size for data transfer to the video ram (>= ipaif_min) 0x02 dipaof_ofmin 0x4208 [6:0] minimum block size for data transfer from the video ram (>= 1) 0x01 dipaof_ofmax 0x4208 [22:16] maximum block size for data transfer to the video ram (>= ipaof_min) 0x01 0 ifmin/ofmin fifosize ifmax/ofmax no transfer transfer possible ifmax/ofmax size packetized transfer with figure 2-1: dipa settings for input and output fifo MB87J2120, mb87p2020-a hardware manual page 122 dpa with its single word accesses has normally not that important influence on system performance. the granted access duration is very short. if the bandwidth requirements are not close to the edge, highest pri- ority should be not problematic if assigned (i.e. dpa=7, gpu=6, vic=5). if considered that bandwidth is very critical for gpu or vic a setting of gpu =7, vic=6 and dpa=5 should be assigned. ipa can lock the video ram interface relative long time, even if a large maximum block size is specified. therefore a very low priority should be assigned to ipa, at least lower than the real-time devices gpu and vic. a low value did not necessarily reduce ipa performance. the data stream is buffered through the input and output fifos. only reaction time increases slightly for low priority requests. if bandwidth violations occur for gpu or vic at ipa usage at lowest priority, the maximum block transfer size should be reduced. it may be possible that gpu fifo under-run or vic fifo overflow occurs while ipa locks the ram access with large packet transfers. just at the usage of high resolution displays with large video bandwidth require- ment this aspect should be considered. now some considerations for minimum block size settings. in general the minimum values must be less or equal to the maximum settings. these values are for controlling efficiency of ipa requests to the sdc. if the amount of data in the fifos is lower the given value, no action is initiated. the ipa device waits for a required, worthwhile packet size. the greater the packet size, the higher the throughput, but additional high- er risk of gpu or vic interruption due to video memory locking during longer transfer times. 2.3 related settings and informations from application point of view the setup of appropriate control information to the dipa configuration reg- isters is not sufficient for accessing video memory in physical address format. additional settings are re- quired for a complete configuration. detailed information are in ulb and sdc documentation. ulb handles the mapping of the video ram address used for dpa (memory mapped). other commands use the in- and output fifos too. additional command processing is controlled by ulb. sdc did the logical to physical address conversion. if picture data should be accessed, sdc calculates based on layer description record (ldr) information the physical video memory address. layer start off- sets and bit size of pixel data influences logical to physical address mapping. this knowledge is necessary to locate a given picture coordinate {layer, x, y} at its dedicated physical memory position. if ipa is used in dma mode both control mechanisms for dma buffer sizes (dma demand mode, see ulb description) and for ipa fifo buffering should be considered to stay not in conflict with each other. oth- erwise this could lead to deadlock situations in data flow. page 123 b-7 video interface controller (vic) MB87J2120, mb87p2020-a hardware manual page 124 video interface controller introduction page 125 1 introduction 1.1 video interface controller functions and features the purpose of the video interface controller (vic) is to receive video data from an external video con- troller. the vic operates only in synchronous mode, which means that the vic clock should be provided by external video pixel decoder. asynchronous modes where no clock is needed are not supported. an external video controller converts an analog video signal into a digital component stream. some of the controllers have additional features like video resizing, picture quality control (contrast, brightness etc.), anti-aliasing-filtering and color format conversion (yuv to rgb). vic reads digital video signals from video controller output, assorts data and writes them into graphic con- troller? video memory (sdram) area reserved for external video data. due to a new bus shuffler 1 , byteswap and clock invert functions the video interface gets a wider range of flexibility. the following points list all vic features briefly: data port with 8 bit (port a only) or 16 bit (port a and port b) different color formats: rgb555, rgb565, rgb888, yuv444, yuv655 2 , yuv555 2 , yuv422 single clock and double clock mode (port a only) different input timing (videoscaler-mode and ccir) with a wide range of variations due to bus shuffler (jasmine only) programmable polarity for control signals alpha key mapping with programmable color windowing function (jasmine only) progressive and interleaced input skip function to reduce data rate frame management unit to synchronize video input and displayed video output programmable layer addresses picture start flag in flag register (jasmine only) 1.2 video data handling vic reads real-time video data from a special display controller interface and writes them into video mem- ory (sdram) with help of sdram controller (sdc). as video target a normal layer can be used. within the display controller no special marking for video layers is necessary, it is treated as all other layer which contain drawings and bitmaps. in order to synchronize input video frame rate with display frame rate vic contains a frame synchronization controller that works together with gpu. for synchronization a two layer and a three layer mode is availa- ble. in gpu only one of the used layers needs to be set up. which of these two or three layers is actually used is decided by frame synchronization controller in real time. in two layer synchronization mode one write and one read layer is used for synchronization and the frame controller compares vertical read/write positions for actual layer. in three layer synchronization mode only a complete written layer is displayed by gpu. it depends on the frame rate difference between video input and display output which layer is displayed how often. if gpu is 1. this shuf?r is only available for jasmine. lavender does not contain this feature. 2. jasmine only; see chapter 2.2 MB87J2120, mb87p2020-a hardware manual page 126 much faster than vic it can happen that one layer is displayed more than one time. on the other hand, if vic is much faster than gpu the display of some layers may be skipped. in any case no artefacts within one layer are visible. to decrease memory size needed for video input all needed layers for synchronization (two or three) may be set to one layer number. in this case vic and gpu uses only one layer. of course no frame rate synchro- nization is done and artefacts may be visible on display within video. as an additional feature vic supports a so called ?lpha channel? via a special pin ( vsc_alpha )anex- ternal video scaler signals the vic not to take the pixel color for the current coordinate but a special color previously defined in a special register ( vicalpha ). this color is written to video memory instead of vid- eo color. together with other features of display controller such as transparent color and blinking inside gpu various effects can be achieved. jasmine contains a window function where a rectangular area (window) can be defined which treats as clip- ping rectangle for video input data. only data within this window are written to video memeory. this function can be used to cut out a part of the video frame in order to save memory. it is also used to remove blanking information in special input timing modes (see chapter 2.3 for details). video interface controller vic description page 127 2 vic description 2.1 data input formats video interface controller supports different color formats displayed in figure 2-1 but does not convert color format into each other. color depth for video target layers has to match color depth of incoming video data. color conversion and mapping to display color depth will be done in output stage within gpu. the video interface controller has two data ports (port a and b), each 8 bit wide. in single port mode only port a will be used. double port mode means that data will be received on both ports (a and b, 16 bit). additionally vic supports two different clock modes. single clock mode uses only one clock phase (edge) while double clock mode uses both phases. in double clock mode, data on port a changes with every edge of clock. port b (if enabled) and control signal only works in single clock mode. different color modes de- pending on clock mode are shown in figure 2-1. MB87J2120, mb87p2020-a hardware manual page 128 figure 2-1: color assignment in single - and double port mode single port (port a only) single clock t1 t2 t3 t4 u a 7 ... u a 0 y a 7 ... y a 0 v a 7 ... v a 0 y b 7 ... y b 0 yuv422 t1 t2 r7 ... r3 g7,g6 g5 ... g3 b7 ... b3 rgb555 t1 t2 v7 ... v3 y7,y6 y5 ... y3 u7 ... u3 yuv555 t1 t2 r7 ... r3 g7...g5 g4 ... g2 b7 ... b3 rgb565 t1 t2 v7 ... v3 y7...y5 y4 ... y2 u7 ... u3 yuv655 double clock t1 t2 u a 7 ... u a 0 y a 7 ... y a 0 v a 7 ... v a 0 y b 7 ... y b 0 yuv422 r7 ... r3 g7,g6 g5 ... g3 b7 ... b3 rgb555 v7 ... v3 y7,y6 y5 ... y3 u7 ... u3 yuv555 r7 ... r3 g7...g5 g4 ... g2 b7 ... b3 rgb565 v7 ... v3 y7...y5 y4 ... y2 u7 ... u3 yuv655 1 2 1 2 1 2 1 2 1 2 1 2 double port single clock port a port b y a 7 ... y a 0 u a 7 ... u a 0 v a 7 ... v a 0 y b 7 ... y b 0 t1 t2 yuv422 r7 ... r3 g7,g6 g5 ... g3 b7 ... b3 rgb555 v7 ... v3 y7,y6 y5 ... y3 u7 ... u3 yuv555 r7 ... r3 g7...g5 g4 ... g2 b7 ... b3 rgb565 v7 ... v3 y7...y5 y4 ... y2 u7 ... u3 yuv655 double clock port a port b y7 ... y0 u7 ... u0 u7 ... u0 yuv444 1 2 r7 ... r0 g7 ... g0 g7 ... g0 rgb888 1 2 b7 ... b0 v7 ... v0 video interface controller vic description page 129 2.2 data format in video ram (sdram) figure 2-2 shows the physical placement of all supported video color formats within a layer. these formats are equal to those of graphic processing unit (gpu) and pixel processor (pp) so that video input pp draw- ings can be mixed within one layer. of course a drawing will be overwritten with the input video picture but nonoverlapping merging of drawing and video is possible. be also aware of layer synchronization be- tween vic and gpu; either only one layer is taken for synchronization or the drawing is copied on all two or three layers with help of memcp command. if only one of two/three layers contains the drawing you will see a flicker on display output. vic performs a format mapping from video input format according to figure 2-1 to the lavender/jasmine internal format according to figure 2-2 for a given color depth. note that some color formats are not sup- ported by lavender. table 2-1 gives an overview on available formats for both display controllers. figure 2-2: different color formats in video ram table 2-1: supported video color formats for lavender and jasmine format lavender jasmine comment yuv422 grayscale yes lavender does not contain a yuv to rgb conversation therefore this for- mat is converted to grayscale with 256 steps. yuv444 grayscale yes lavender does not contain a yuv to rgb conversation therefore this for- mat is converted to grayscale with 256 steps. rgb888 yes yes rgb565 yes yes y 0 u 0 y 1 v 0 y 2 u 2 y 3 yuv 422 v 0 y 0 u 0 v 1 y 1 r 0 g 0 b 0 r 1 g 1 r 0 g 0 b 0 g 1 r 2 g 2 b 2 yuv 444 24 bit 16 bit --- --- rgb 888 24 bit r 1 b 1 r 3 g 3 rgb 565 16 bit r 0 g 0 b 0 g 1 r 2 g 2 b 2 r 1 b 1 r 3 g 3 rgb 555 16 bit x x xx x xx x x x 32 bit xx x xx x x x u 1 b 1 x v 0 y 0 u 0 y 1 v 2 y 2 u 2 v 1 u 1 v 3 y 3 yuv 555 16 bit x x x v 0 y 0 u 0 y 1 v 2 y 2 u 2 v 1 u 1 v 3 y 3 yuv 655 16 bit note: rgb --> vyu MB87J2120, mb87p2020-a hardware manual page 130 2.3 data input timing the vic component supports three different input timing modes which can be selected using register vicvisyn_sel . the preferred and default mode is denoted "videoscaler-mode" in the following. 2.3.1 videoscaler-mode videoscaler-mode has been implemented for vpx video pixel decoder family (micronas intermetall), which provides up to 16 data bits (data; pin: vsc_d[15:0] ), a video clock (clkv; pin: vsc_clkv ), a vertical synchronization signal (vref; pin: vsc_vref ), a field identification signal (field; pin: vsc_ident ), a horizontal video active signal (vact; pin: vsc_vact ) and additional an optional alpha key signal (alpha; pin: vsc_alpha ). other video deocoders are supported if they provide a sim- ilar interface and timing. see timing diagram (figure 2-3) for detailed information. figure 2-3: videoscaler timing (general) vref pulse indicates the start of a new field. field changes should happen at least one clock cycle before negative edge of vref (see figure 2-3). vact marks an active video line. due to downscale function of videoscaler inactive lines are possible. positive edge of vact indicates the start of an active line, negative edge denotes the end of line. during vact is active, vic samples data received on data ports and writes them into video memory (sdram). polarities of control signals are adjustable, see register vicpctrl . rgb555 yes yes yuv555 - yes lavender does not contain a yuv to rgb conversation (see gpu description) yuv655 - yes lavender does not contain a yuv to rgb conversation (see gpu description) table 2-1: supported video color formats for lavender and jasmine format lavender jasmine comment 2..9 lines >1 clock cycle 1 line vref field vact data video interface controller vic description page 131 figure 2-4: videoscaler timing (detailed) 2.3.2 ccir-mode in ccir-mode no explicit control signals are used. control informations are directly merged into data stream (data port a). vic component extracts control information from data and maps it to internal signals similar to videoscaler-mode. the ccir-mode is only available for mb87p2020-a (jasmine). figure 2-5: definitions for ccir mode there are two timing reference codes, one at the beginning of each video data block (sav: start of active video) and one at the end of video data block (eav: end of active video). each code consists of a four-byte- sequence. the first three bytes are fixed preamble. the fourth word carries information about field infor- mation (f), field blanking (v) and line blanking (h). table 2-2: timing reference code sequence bit number 76543210 first 1 1 1 11111 second 0 0 0 00000 third 0 0 0 00000 fourth 1 f v h p3 p2 p1 p0 clkv vact data_a (double clock) data_a data_b (single clock) first sample last samples eav sav field 2 field 1 frame MB87J2120, mb87p2020-a hardware manual page 132 f-bit: '0' during field number 1 '1' during field number 2 v-bit: '1' during field blanking '0' else h-bit: '0' in sequence sav '1' in sequence eav bits [3:0] (p3..p0) are protection bits. the value of those bits depends on bits f, v and h. the vic implements an error detection and correction based on protection bits. it is possible to detect two- bit-errors and to correct one-bit-errors. in case of a two-bit-error the controller takes f- and v-bit directly from current trc (without error correc- tion) and inverts h-bit received from previous trc. 2.3.3 external-timing-mode the external-timing-mode is only available for mb87p2020-a (jasmine). external-timing-mode is together with vic windowing function (register: viclimen ) a generalized videoscaler-mode. it is able to receive a horizontal synchronization signal (hsync) and a vertical syn- chronization signal (vsync) which is provided by many video devices. note that the data has to be one of the formats described in chapter 2.1. polarities can be controlled by register extpctrl. control signal (hsync, [pin vsc_vact ], vsync, [pin vsc_vref ] and parity [pin vsc_ident ]) will be taken directly from input. the hsync -pulse tags the begin of a new line and the vsync -pulse the begin of a new field/frame. active video needs not to be indicated additionally. the start and length of the visible window should be controlled by windowing function for both dimensions. table 2-3: de?ition of sav, eav and protection bits ?ed sav/eav protection bits f v h p3p2p1p0 10000000 10011101 10101011 10110110 11000111 11011010 11101100 11110001 video interface controller vic description page 133 figure 2-6: external-timing related registers: viclimen, viclimh, viclimv, extpctrl hsync 1 line (blanking and active video) hsync^ extpctl[1] internal vact after windowing offset length (active video line) MB87J2120, mb87p2020-a hardware manual page 134 video interface controller vic settings page 135 3 vic settings 3.1 register list table 3-1: vic register description symbol address. description/de?ition vicstart h4000 input layer start coordinates x [29:16] : x coordinate ( default: 0 ) y [13:0] : y coordinate (default: 0) vicalpha h4004 color to be mapped when alpha pin is active and alpha_mode=1 (vicctrl[6]). bit occupancy depending on color mode (lsb alligned). col [23:0]: [7:0] : value for blue (u) [15:8] : value for green (y) [23:16] : value for red (v) default: 24?0 vicctrl h4008 mode [3:0]: color_mode 0100: rgb555 0101: rgb565 0110: rgb888 0111: yuv422 1000: yuv444 1110: yuv555 1111: yuv655 clock [4]: clock_mode 0: single clock 1: double clock port [5]: port_mode 0: single port 1: double port alen [6]: alpha_mode 0: don? use alpha information 1: use alpha bswap [7]: byte_swap 0: {a[15:8],b[7:0]} 1: {b[15:8],a[7:0]} MB87J2120, mb87p2020-a hardware manual page 136 vicfctrl h400c fst [3:0]: primary_layer_address sec [7:4]: secondary_layer_address trd [11:8]: third_layer_address odden [16]: enable_odd_?lds 0: disabled 1: enabled evenen [17]: enable_even_?lds 0: disabled 1: enabled vicen [18]: vic_enable 0: disabled 1: enabled skip [20]: skip_?lds_enable 0: use every ?ld 1: skip every 2nd ?ld of each type frame [21]: frame/?ld 0: ?ld mode (store one ?ld in each layer) 1: frame mode (interlocked storage of two ?lds in each layer) oddfst [22]: odd_?ld_?st 0: even ?ld is top ?ld 1: odd ?ld is top ?ld vicpctrl h4010 polarity control for vpx video synchronization signals. those settings only effects if "vpx video source" is selected ( vicvisyn [1:0] = 2'b00) vref [0]: vref polarity 0: high active 1: low active vact [1]: vact polarity 0: high active 1: low active field [2]: ?ld polarity 0: polarity unchanged 1: invert polarity alpha [3]: alpha polarity 0: high active 1: low active table 3-1: vic register description symbol address. description/de?ition video interface controller vic settings page 137 vicfsync h4014 swl [14:0]: switch level (2-layer mode only) sync [15]: layer sync mode 0: 2-layer mode 1: 3-layer mode rel [16]: frame rate relation (2-layer mode only) 0: gpu is faster than vic 1: vic is faster than gpu sdram h401c lp [2:0]: low priority (default: 3?010) hp [6:4]: high priority (default: 3?110) vicbsta h4020 load [6:0]: ?o load req [8:7]: sdc request status b00: idle b01: req1 b10: wait1 clr [9]: clear ?o add [12:10]: address register load bx00: empty bx01: 1 burst bx10: 2 burst bx11: 3 burst bx00: 4 burst (full) b0xx : norm b1xx : full, but no error (error ?g will be set with next new address) fsm [15:13]: ?o_fsm state b000 : norm b001 : wait b101 : wait1 b010 : error b111 : reset b110 : error1 b011 : reset1 aerr [16] : address error ff [17] : ?o full fe [18] : ?o empty table 3-1: vic register description symbol address. description/de?ition MB87J2120, mb87p2020-a hardware manual page 138 vicrlay h4024 aivl [3:0]: actual written video layer livl [11:8]: anteriorly written video layer aovl [23:16]: actual output video layer vicvisyn h4028 sel [1:0] : video source select b00 : vpx b01 : ccir - trc b10 : ccir - ext. sync start [8] : selector for picture start ?g (to ulb) 0 : vic - start writing picture 1 : gpu - start reading picture shuff [20:16] : data bus shuf?r default: 5'b00000 for details see chapter 3.2. del [25:24] : data delay b00: delay data 0 cycle respective vsc_vact b01: delay data -1 cycles respective vsc_vact b10: delay data -2 cycles respective vsc_vact b11: delay data -3 cycles respective vsc_vact viclimen h402c limena [0] : limit size enable 0: limit size disabled 1: limit size enabled viclimh h4030 hoff [10:0]: horizontal offset hen [26:16]: horizontal window length (including offset!) only effects if viclimen[0] = 1! viclimv h4034 voff [10:0]: vertical offset ven [26:16]: vertical window length (including offset!) only effects if viclimen[0] = 1! table 3-1: vic register description symbol address. description/de?ition video interface controller vic settings page 139 3.2 register description vicstart (4000h): gdc stores video fields or frames in different memory areas called layers. the vicstart register defines layer start coordinates in x- and y-direction. note that coordinates are equal for all layers used for video input (one, two or three layer). related registers: vicfctrl [11:8], vicfctrl [7:4], vicfctrl [3:0] figure 3-1: start coordinate definitions vicalpha (4004h): the vicalpha register contains the value of color which will be mapped if alpha key signal is active and alpha_mode is enabled (see register vicctrl [6]). note that bit occupancy depends on color mode (see figure 3-2). related register: vicctrl [6] extpctl h4038 polarity control for external video synchronization signals. those set- tings only effects if "external video source" is selected ( vicvisyn[1:0] = 0b10) vref [0]: vsync 0: high active 1: low active href [1]: hsync 0: high active 1: low active parity [2]: parity 0: polarity unchanged 1: invert polarity alpha [3]: alpha 0: high active 1: low active clkpdr hfc04 vii [12]: video clock invert 0: don t invert clock 1: invert clock table 3-1: vic register description symbol address. description/de?ition vicstart[13:0] (y) vicstart[29:16] (x) layer field or frame MB87J2120, mb87p2020-a hardware manual page 140 figure 3-2: possible color occupancy in register vicalpha vicctrl (4008h): the vicctrl register contains settings for color mode, clock mode, port mode and byte swap. those set- tings should be done according to settings of external video decoder. please use only suggestive combina- tions of color-, clock- and port-mode. vicfctrl (400ch): bit [3:0] ( fst ), [7:4] ( sec ), [11:8] ( trd ): gdc? memory space is divided in up to 16 areas (layer). each of them has a logical layer address and can contain video data. vic implies a frame synchronization controller which mangages two or three layers needed for frame synchronization (see register vicfsync ). layer addresses for vic could be defined in register vicfctrl . note that same addresses are allowed, but field/frame synchronization won't work cor- rectly. related registers: vicstart , vicfsync bit 16 and 17 ( evenen and odden ): if those flags are set to zero, every field of the specified type (odd/even) will be skipped. so evenen (en- able even fields) and odden (enable odd fields) submit selection of field type. it is recommended to enable only one field type to supress field jumping during displaying video layers. function is to be seen in context with vicfctrl [20] ( skip ). those flags have no relevance if vicfctrl [21] ( frame ) is set. related registers: vicfctrl [18], vicfctrl [22:20] bit 18 ( vicen ): the vicen (vic enable) switches video interface unit on (1) or off (0), (?ain switch?. it does also enable or disable synchronization of video in- and output. related registers: vicfctrl [17:16], vicfctrl [22:20] bit 20 ( skip ): this flag controls the skip function of vic. if skip is set to 1, every second field of each field type is skipped. skip enable function is to be seen in closed connection with vicfctrl [16] ( evenen ) and vicfctrl [17] ( odden ). if both of them and skip are set to one, every second frame is skipped. so it is possible to reduce data rate between vic and sdram to a half of the output data rate from external video pixel decoder. a quarter of input data rate can be achieved be enabling only one field and skip. an example is shown in the following figure (figure 3-3). related registers: vicfctrl[18:16], vicfctrl[22:21] rgb888 rgb yuv444 vyu rgb555 rgb rgb565 rgb yuv555 vy yuv655 vyu u yuv422 vyu video interface controller vic settings page 141 figure 3-3: reducing data rate by skipping fields (examples) bit 21 ( frame ): the frame flag controls the storage format of input video in video ram. if frame is set to 1, a complete frame (even and odd field) will be saved interlocked in each video layer. in case of frame=0 every layer contains one field and interleaced to progressive format conversion can be done in output process (line dou- bling). related registers: vicfctrl [18:16], vicfctrl [20], vicfctrl [22], vicpctrl [2] or extpctrl [2] in association with vicvisyn [1:0] figure 3-4: frame mode vs. field mode bit 22 : oddfst (odd field first) determines the position of the fields in a frame related to time. the following fig- ure illustrates the issue. this flag has no relevance if vicfctrl [21]is set to 0. vref field input data odd even enable_odd_fieds enable_even_fields skip_enable 111 100 011 vicstart[13:0] vicstart[29:16] frame mode field mode layer layer field1 field2 MB87J2120, mb87p2020-a hardware manual page 142 figure 3-5: odd_field_first flag: mode of action vicpctrl (4010h): the flags in this word set the polarity of input control signals (vact, vref, alpha, field). note that those bits only influence the behaviour of vic if videoscaler-mode is enabled as video source ( vicvisyn[1:0]=0b00 ). related register: vicvisyn [1:0] vicfsync (4014h): synchronization of video input and video output is frame/field based. frame synchronization unit is to en- sures correct order of pictures and avoids overtaking of read and write pointer in the same active layer. two modes are possible and can be selected by vicfsync [15] ( sync ). bit 15 ( sync ) : if layer sync mode ( sync ) is set to 1, the 3 layer synchronization mode will be used. the used memory space is reduced by selecting three layer mode and setting all layer addresses ( vicfctrl [3:0], vicfc- trl [7:4], vicfctrl [11:8]) to the same layer number. then video input and output are not synchronized. in two layer mode ( sync=0 ) the synchronization will be performed by comparing vertical read/write po- sition (gpu/vic) in actual layer. there are two layers for video i/o (primary layer: vicfctrl [3:0], sec- ondary layer vicfctrl [7:4]). the third layer address ( vicfctrl [11:8]) has no effect. releated registers: vicfsync [16], vicsync [14:0], vicfctrl [3:0], vicfctrl [7:4], vicfc- trl [11:8] bit 16 ( rel ) : this bit has relevance if vicfsync [15] is set to 0 (2-layer-mode). the frame rate relation (rel) flag indicates the relation between input (vic) and output (gpu) frame rate. the faster unit toggles its layer each time a layer is fully processed. the other unit compares its line pointer with the switch level vicfsync [14:0] ( swl ) to decide, which layer using next. for example, if rel is set to 0, frame output rate is higher than frame input rate. the current active vic layer toggles with vref field input data odd even one frame if odd_field_first =0 one frame if odd_field_first =1 layer lines video interface controller vic settings page 143 every new input frame. the next gpu layer is synchronized depending on current vic layer and vertical position of write pointer (in this layer) which will be compared with swl . if current vertical write position is above the programmable threshold swl , the current vic write layer will be allocated to gpu because an overtaking of read and write pointer in the same layer is excluded. otherwise ( swl is less than vic write pointer) the frame synchronization unit denotes the inactive vic layer to be the next active gpu layer. releated registers: vicfsync [15], vicsync [14:0], vicfctrl [3:0], vicfctrl [7: 4] figure 3-6: example of frame synchronization: video output is faster than video input bits [14:0] ( swl ) : this bit has only effect if vicfsync [15] is set to 0 (2-layer-mode). for explanation of swl in context with vicfsync[16] (rel) see previous point. swl will be compared with the vertical address. note that this address (and swl too) includes an offset ( vicstart [13:0]). if frame mode is enabled ( vicfctrl [21] = 1), swl indicates the field (0: even, 1: odd) where swl is located. in field mode ( vicfctrl [21] = 0) swl [14] has no function (should be set to 0). releated registers: vicfsync [16], vicsync [15], vicfctrl [3:0], vicfctrl [7:4] vicrlay (4024h): vicrlay contains some layer information, which can be read from ulb (i.e. after a picture start was in- dicated). three different informations will be given: vicrlay [3:0] ( aivl ) - this is the layer video data will be written in currently vicrlay [11:8] ( livl ) - this is the previous written video layer vicrlay [23:16] ( aovl ) - this is the current displayed video layer vicrlay is a read only register. related registers: vicvisyn[8], vicfctrl[11:8], vicfctrl[7:4], vicfctrl[3:0] vicvisyn (4028h): bit [1:0] ( sel ): switches to select an input timing scheme. in videoscaler-mode (b'00) vic interface uses explicit synchro- nization signals. polarity of those signals can be selected by vicpctrl . sel=0b10 is a similar timing scheme as videoscaler-mode, polarity can be controlled by extpctrl . ccir-mode uses timing reference codes merged into the data stream. no explicit control signals are used. bit[8] ( start ): both, vic and gpu, generate their own picture start signal. it indicates the begin of writing (vic) or read- ing (gpu) a new picture into or from data ram. one flag ( flnom_vicsyn ) indicates the picture start to 0 even 0 odd 1 even 1 odd 2 even 2 odd 3 even 3 odd 4 even 4 odd 5 even 5 odd input frames (interleaced) time abab ab input layer vic 0 1 2 3 4 5 output frames (progressive) time a b ab a b output layer gpu 0 0 1 3 a b ab MB87J2120, mb87p2020-a hardware manual page 144 ulb. start selects one of the two signals (vic-start or gpu-start) and allocates it to flnom_vicsyn . this is a useful feature to synchronize anything on picture start controlled by software. related register: vicrlay bits[20:16] bus shuffler ( shuf ) figure 3-7: definition of port - internal bus - assignment vic owns two data ports which are splitted into four internal busses (figure 3-7). to get more flexibility (i.e. connect another external video controller to vic without changing the board) vic includes various functions: clock invert function ( clkpdr [12]) byte swap to exchange ports a and b bus shuffler to exchange internal busses (all combination are possible, see table 3-2) a schematic overview about data path are given in figure 3-8. note that contradictory settings can abrogate each other (i.e. setting shuf=2 abrogates setting vicctrl_bswap=1 ). figure 3-8: schematic overview about busshuffler function a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 b0 b2 b4 b6 b8 clock a b a0 a2 a4 a6 a8 ap a-1 a0 a2 a4 a6 ad a1 a3 a5 a7 a9 an b0 b2 b4 b6 b8 b ports internal buses port a port b clkv vic_dcu clock_unit vic clkv clkvx byteswap vii ap ad an b ap ad an b shuff video interface controller vic settings page 145 related registers: vicctrl [7], clkpdr [12] table 3-2: vic bus shuf?r settings bus shuf?r value bus a, pos. edge (ap) bus a, delayed (ad) bus a, neg. edge (an) bus b (b) 0 (default) ap ad an b 1adapan b 2 b ad an ap 3 ap b an ad 4apadban 5anadap b 6apanad b 7 ap b ad an 8adbanap 9 ad b ap an 10 ad an ap b 11 ad an b ap 12 ap an b ad 13 ad ap b an 14 b ad ap an 15 b an ad ap 16 b an ap ad 17 b ap ad an 18 b ap an ad 19 an ad b ap 20 an ap ad b 21 an ap b ad 22 an b ap ad 23 an b ad ap others ap ad an b MB87J2120, mb87p2020-a hardware manual page 146 bits [25:24] ( del ) : del provides an option to shift data relative to pin vsc_vact with up to -3 cycles. figure 3-9: del definitions (example: del=3 ) viclimen (402ch): the vic-subunit vic_limit is enabled or disabled by this switch. so a pan function can be realized. an- other reason to enable vic_limit could be the compensation of different input sync timing. relating con- trol registers are viclimh (horizontal offset and horizontal window length [video clocks per step]) and viclimv (vertical offset and vertical window length [lines per step]). for horizontal parameters the context of input pixel per clock which depends on video colour depth and win- dow offset/length should be considered. since some colour formats transport less than one pixel per clock and the offset and length is set up in clock units it is necessary to calculate the setup value depending on colour depth. for each colour depth a factor ? for horizontal offset and length can be calculated as follows: registers viclimh_hoff and viclimh_hen have to be multiplied by ?? note that for most colour for- mats ? can be set to one so that offset and length can be set directly in pixel dimensions. table 3-3 lists all colour formats with their factor. additionally some colour formats need a special alignment for horizontal offset and length. otherwise col- our distortions may occur. table 3-3 gives also an overview on necessary alignments. table 3-3: horizontal offset and length parameters for different colour depths format clock port factor ? alignment yuv422 single single 2 4 double 1 2 single double 1 2 rgb555 single single 2 2 double 1 1 single double 1 1 yuv555 single single 2 2 double 1 1 single double 1 1 rgb565 single single 2 2 double 1 1 single double 1 1 vact data del=3 f clocks pixel ---------------- = video interface controller vic settings page 147 related registers: viclimh , viclimv , vicvisyn [25:24] figure 3-10: window definitions extpctrl (4038h): the flags in this word set the polarity of input control signals (hsync, vsync, alpha, parity). note that those bits are only an effective if ?xt. sync?is selected as video source ( vicvisyn [1:0]=0b10). related registers: vicvisyn [1:0] yuv655 single single 2 2 double 1 1 single double 1 1 yuv444 double double 1 1 rgb888 double double 1 1 table 3-3: horizontal offset and length parameters for different colour depths format clock port factor ? alignment h_win_length h_win_offset v_win_offset v_win_length vref_edge vact_edge MB87J2120, mb87p2020-a hardware manual page 148 page 149 b-8 graphic processing unit (gpu) MB87J2120, mb87p2020-a hardware manual page 150 graphic processing unit functional description page 151 1 functional description 1.1 gpu features the gpu (graphics processing unit) is a part of fujitsu? display controllers MB87J2120 (?avender? and mb87p2020-a (?asmine?. in the following these two controllers are referred to as gdc (graphics display controller). the gpu is responsible for producing visible images from pixel data stored in videoram (sdram). its features include: maximum ?xibility by con?uration registers for most functions wide range of supported display types and sizes (see chapter 7) twin display mode (digital and analog displays in parallel) (jasmine only) up to four layers from a whole of 16 displayed simultaneously and overlapping in four planes size, color resolution and position on the display con?urable on a per-layer basis blinking and transparency individual for each layer separate setting for display background color yuv to rgb matrix with con?urable coef?ients as well as a loadable gamma table included (jas- mine only) synchronization with the video interface controller (vic) allowing frame rate conversion ?xible generation of almost arbitrary sync signals individual settings for clamping values at digital and analog outputs (lavender can only handle one clamping color) color key output internal and external pixel clock masking and division by 2 for internal pixel clock programmable interrupts at speci?d display location for insuf?ient memory bandwidth diagnostic read back current display location current blink state of layers current input fifo load the image on display is produced in a three-step approach: fetching pixel data, transforming and merging the layers from their respective color space to one common, and finally formatting the bit stream appropriate for the actual display used. MB87J2120, mb87p2020-a hardware manual page 152 1.2 gpu overview 1.2.1 top level structure figure 1-1 shows the gpu top level structure, which consists of the four main components dfu, ccu, lsa, and bsf. the cbp is actually only an auxiliary unit to couple the other units to the system-wide con- trol bus. control registers are assigned their respective units, so there is no gpu-global register set (al- though it appears as such for the programmer). the functions of each unit will be discussed briefly in the following sections. 1.2.2 dfu function the dfu (data fetching unit) interacts with the sdc (sdram controller) to transport pixel data from videoram into the gpu pipeline. since videoram is a shared resource within gdc, the dfu is equipped with 4kbits of fifo to balance data rate and optimize ram usage by enabling burst accesses. dfu structure is shown in figure 1-2. pixels from layers displayed simultaneously are fetched sequentially according to their respective start points, display offsets, color resolutions, and vertical stacking. this is car- ried out by the fetch fsm. the data obtained from sdc is put into the fifo, from where it is read out later by the pixelpump and fed through the ccu into the lsa, where the vertical (z-)order is actually real- ized.the pixelpunp is also used to carry out the chrominance demultiplexing for yuv422 layers (cf. 3.7.1). figure 1-1: gpu top level structure. thick black lines show the flow of pixel data, thin black lines the control flow. thick gray lines represent register set data. dfu data unit ccu color conversion unit fetching control bus port lsa line segment cbp bsf bit accu formatter dac i / o buffer stream sdc videoram ctrl bus display graphic processing unit functional description page 153 1.2.3 ccu function the ccu (color conversion unit) handles all transformations necessary to convert the layers displayed from their respective color space to one common color space. further, special colors for each layer can be defined that trigger transparency and blinking. these colors are detected by the ccu and the appropriate action is carried out. the ccu contains seven functional sub-units plus its own register set. this register set stores all layer-spe- cific color information which control the transformations. figure 1-3 shows the structure of the ccu. the blinkers component provides a pulse-width modulated blink signal for each layer. in the colorconv com- ponent blink and transparent colors are detected. the clut (color look-up table) is used to convert lay- ers with color resolutions of less than or equal to 8 bits to higher resolutions. there is only one clut for figure 1-2: dfu structure. pixel data is transported from sdc via the fetchfsm to the input fifo. from there it is read out by the pixelpump and fed into ccu. the register set contains all necessary geometric and color space values to fetch the pixels. some of the registers are double-buffered and updated on a per-frame basis. fetch fsm to sdc from sdc input fifo register set w/ framelock pixelpump to cbp to ccu figure 1-3: ccu structure. pixel data is fed in by dfu and travels through colorconv, optionally through clut, and is prepared by spacemap for output to lsa. matrix gamtbl colorconv blinkers spacemap register set duty ratio modulator from dfu to cbp clut to lsa MB87J2120, mb87p2020-a hardware manual page 154 all layers, however, each layer may have its own offset for look-up. the matrix is used for transforming yuv (ycbcr) video data into rgb colorspace (cf. 3.7.2). it is supplemented with a gamma table (gamtbl) to carry out a non-linear inverse gamma correction (cf. 3.7.3). the spacemap component merges the indi- vidual layer color spaces to one common intermediate color space and further to one physical color space. the duty ratio modulator is used to provide pseudo gray levels (or colors) for displays with a low number of bits per pixel. color space conversion is explained more detailed in chapter 2. 1.2.4 lsa function the lsa (line segment accumulator) is used to realize the vertical layer order in up to four planes. it acts also as internal fifo between gpu front end and back end, where clock domain transition between gdc core clock and pixel clock takes place. figure 1-4 shows the lsa structure. the lsa contains two ram blocks controlled by the bankmanager component. these components per- form the actual fifo functionality. three adjustment components (one for writing, two for reading) com- plete the unit. they are used to efficiently employ the rams. there are two rams to allow for parallel read-out when interfacing dual-scan displays. pixel data is fed in sequentially via writeadjust to one of the rams. read-out is performed in parallel through the respective readadjust component. z-ordering (vertical layer order) is achieved by writing pix- els plane by plane sequentially into the lsa from bottom plane to topmost plane, thus overwriting pixels hidden underneath others in higher planes. 1.2.5 bsf function the bsf (bit stream formatter) unit performs the necessary preparations to actually output the picture on a physical display (either with analog or digital interface). the bsf contains components for signal prepa- ration and for synchronization. the structure is shown in figure 1-5. the timing component produces the synchronization frame for image data (see 3.9). it controls the lsa read-out and provides positional information to the syncsigs component. the xsync component supports timing when gdc is running with external synchronization. the syncsigs component generates synchro- nization signals for the physical display in a very flexible way (see 3.10). further, it can produce a program- mable interrupt. the components digswitch and dac sigs prepare pixel data for output on displays with digital or analog interface, respectively. they select necessary signals and perform a delay compensation between pixel and synchronization signals. the colorkey component allows the application of chroma-keying with gdc by detecting whether the pix- el lie within a programmable color range (see 3.14.5). figure 1-4: lsa structure. pixel data is prepared by writeadjust and then stored in one of the rams. during read-out data is adjusted once again by the respective readadjust. write adjust to bsf from dfu from ccu bank manager upper lsa ram lower lsa ram read adjust for lower read adjust for upper graphic processing unit functional description page 155 figure 1-5: bsf structure. pixel data is fetched from lsa and prepared for output by digswitch and dacsigs. timing controls this fetching and establishs a position reference for syncsigs, which generates synchronization signals. to lsa register set timing xsync key color switch dig to cbp sync sigs from lsa key digital out hsync, vsync, vref analog out dac sigs external sync in MB87J2120, mb87p2020-a hardware manual page 156 2 color space concept 2.1 background the gdc-asic allows to use several numbers of bits per pixel to represent pictures in its videoram (1, 2, 4, 8, 16, and 24 bits per pixel). they may even be different in different planes. the color 1 space defined by these bits shall be referred to as logical color space. actual displays, however, might have color resolutions differing from those in the logical space. the number of bits per pixel as supported by the display actually wired to the gdc-asic shall be referred to as physical color space. obviously, there is the need to map the logical onto the physical color space. additional internal units such as a color look-up table (clut), a yuv to rgb matrix (jasmine only), and a duty ratio modulator (drm) lead to another internal representation which shall be referred to as intermediate color space. this internal representation is laid out as to provide some kind of common ground between logical and physical color space. 2.2 data flow for color space conversion within gpu figure 2-1 gives an overview on the logical data flow for color space conversion. this does not represent actual gpu-pipeline stages but conceptual fields where the different color spaces apply. pixel data held in videoram belongs to logical color space. it is mapped to intermediate color space through one of the ways listed in table 2-1. mapping to physical color space is carried out according to table 2-2. signals actually wired to the physical display are interpreted in this physical color space. how- ever, there is one additional stage of conversion: bit stream formatting. displays may require data in a stream with bit groups not necessarily equal to one pixel. therefore, an appropriate postprocessing is indis- pensable (see 1.2.5 and 3.3). 1. ?olor?here refers to different hues as well as to different shades of monochrome physical color space intermediate color space logical color space display interface signals video ram clut yuv gtab drm bsf dac matrix display figure 2-1: basic data flow for color space conversion. graphic processing unit color space concept page 157 2.3 mapping from logical to intermediate color space table 2-1 lists the available mappings from logical to intermediate color space. the term direct designates an immediate mapping wire by wire with more significant bits being clamped to zero as needed. for in- stance, logical 2bpp is mapped directly onto intermediate rgb111 this way: i[3:0] = ??& l[1:0]. the term lookup means using the look-up table for mapping. the term aligned designates a mapping in which wires are connected in groups according to color channels with appropriate value scaling. for in- stance: logical rgb555 to intermediate rgb666 (i.e. expansion of color channel bit with): i[17:0] = l[14:10] & l[14] & l[9:5] & l[9] & l[4:0] & l[4] logical rgb555 to intermediate rgb333 (i.e. reduction of color channel bit with): i[8:0] = l[14:12] & l[9:7] & l[4:2] the term matrix designates application of a matrix multiplication. when transforming logical yuv422 to an rgb intermediate format an optional chrominance interpolation may be applied (see 3.7). the term ach- romatic describes a mapping from yuv to rgb color spaces by dropping chrominance information and transforming the luminance to gray levels. all matrix and achromatic mappings are auto aligned to their target bit with 1 . the mapping from logical to intermediate color space is controlled by the settings for logical color space in the layer? respective layer description record and the setting of intermediate color space, matrix parame- ters, and transfer bit in the merging description record. 2.4 mapping from intermediate to physical color space table 2-2 lists the mappings from intermediate to physical color space. the terms direct and aligned keep their meaning. the term modulated designates the usage of the duty ratio modulator (see 3.8). the purpose of this unit is to provide additional (pseudo-) levels (shades of hue or gray) for output, i.e. virtually expand the color resolution in the physical color space. this is achieved by tuning the duty ratio of actually existing physical bits. accordingly, the term gamma means employing the gamma tables for mapping. the mapping is controlled by the settings of intermediate color space in the merging description record and the setting of physical color space in the display interface record.the resulting mapping from logical to physical color spaces is shown by table 2-3. it is contains the possible path of transformation from pixel in videoram to those on the physical display. 1. since matrix results are always in rgb888 color space this auto aligning implies at most a reduction of bit- width whereas achromatic mapping can result in both, bit width reduction as well as expansion (see also 3.7.2). MB87J2120, mb87p2020-a hardware manual page 158 table 2-1: mapping from logical to intermediate color space. mappings in normal face are default, those in italics designate alternative m appings. logical color space intermediate color space codes bits per pixel code 1bpp rgb111 4bpp rgb222 rgb333 rgb444 rgb666 rgb888 1bpp lookup / direct lookup 10 2bpp lookup / direct lookup / aligned lookup 21 4bpp lookup lookup / direct lookup 42 8bpp lookup lookup / direct lookup / aligned lookup 83 rgb555 n/a aligned aligned / direct a a. direct mapping is not available when rgb gamma forcing is applied. 16 4 rgb565 aligned aligned / direct a 16 5 rgb888 aligned direct 24 6 yuv422 matrix / achromatic 16 7 yuv444 matrix / achromatic 24 8 yuv555 b b. this format is not available for lavender. matrix / achromatic 16 14 yuv655 b matrix / achromatic 16 15 bits per pixel 1 3 4 6 9 12 18 24 code 0 9 2 10 11 12 13 6 graphic processing unit color space concept page 159 . table 2-2: mapping from intermediate to physical color space. intermediate color space physical color space bits per pixel code 1bpp rgb111 4bpp rgb333 rgb444 rgb666 rgb888 1bpp direct n/a 10 rgb111 n/a direct n/a 39 4bpp modulated direct n/a 42 rgb222 n/a modulated n/a 610 rgb333 n/a direct / gamma a a. selectable for intermediate color space stemming from one of the yuv logical color spaces, or if rgb gamma forcing is applied . n/a 911 rgb444 n/a direct / gamma a n/a 12 12 rgb666 n/a direct / gamma a n/a 18 13 rgb888 n/a direct / gamma a 24 6 bits per pixel 1 3 4 6 12 18 24 code 0 9 2 11 12 13 6 MB87J2120, mb87p2020-a hardware manual page 160 table 2-3: resulting mapping from logical to physical color space logical color space physical color space bits per pixel code 1bpp rgb111 4bpp rgb333 rgb444 rgb666 rgb888 1bpp dd, ld, lm ld, lm ld ld ld ld ld 10 2bpp dd, ld,am, lm dd, ld, ad, lm ad, ld ld ld, la ld, la ld 21 4bpp ld, dm , lm dd, ld, lm dd, ld ld ld, la ld, la ld 42 8bpp ld, lm ld, dm, lm ld ad, ld ld ld ld 83 rgb555 ad ad dd, ad dd, ad 16 4 rgb565 ad ad, aa dd, ad dd, ad 16 5 rgb888 ad ad ad dd 24 6 yuv422 xd, yd xd, yd xd, yd xd, yd 16 7 yuv444 xd, yd xd, yd xd, yd xd, yd 24 8 yuv555 a a. this format is not available for lavender. xd, yd xd, yd xd, yd xd, yd 16 14 yuv655 a xd, yd xd, yd xd, yd xd, yd 16 15 bits per pixel 1 3 4 9 12 18 24 code 0 9 2 11 12 13 6 graphic processing unit color space concept page 161 meaning of abbreviations: d: direct mapping l: look-up mapping a: aligned mapping m: mapping by duty modulation x: matrix multiplication y: conversion to achromatic MB87J2120, mb87p2020-a hardware manual page 162 3 gpu control information the following sections describe the information necessary for output of video and picture data on a display wired to gdc. the tables give only the type of information and its respective size, they do not imply any assignment to gdc-registers. the intention is to sort information according to their conceptual place of ap- plication and to group data belonging together. 3.1 layer description record this group of data describes all information belonging to a single layer l (graphics or video) in videoram (i.e. within logical color space). table 3-1: layer description record information size (bits) 1 physical address in vram a the address of the word containing the ?st pixel of the layer according to mcu address space. a. this information is held in sdc 32 2 domain size dx l a , (dy l) logical area covered by the layer in videoram. dy only pro forma, the user is responsible for providing enough ram space per layer. 2 x 14 unsigned 3 first pixel to be displayed fx l , fy l upper left corner of area in vram actually to be displayed 2 x 14 unsigned 4 display window size wx l , wy l size of area to be displayed 2 x 14 unsigned 5 display window offset ox l , oy l offset of layer area from upper left corner of physical display. 2 x 14 signed see ?ure 3-1 for an illustration of geometry. 6 color space code see section 3.12.1 4 7 transparent color the color (in logical color space) to be ignored for display 1?4 8 transparency enable 1 9 blink color (con?ming to 6) 1?4 10 blink alternative color (con?ming to 6) if a pixel in video ram has a logical color equal to blink color , then it is displayed alternating in blink color / alternative color . either color may also be the transparent color , in this case the pixel and the layer below it are displayed alternatively. 1?4 11 blink rate # of frames on, # of frames off 2 x 8 12 blink enable 1 graphic processing unit gpu control information page 163 31 0 phys. addr. a phys. addr. b wx a wy a wy b wx b dx b dx a px py dy b dy a dx, dy = domain size wx, wy = window size px, py = physical display size fx b, fy b fx a, fy a ox a, oy, a ox b, oy b fx, fy = first pixel ox, oy = offset ms b ms a ms = minimal memory size = dx * dy / (pixelperword) memory physical memory logical display figure 3-1: illustration for geometric relations between videoram, layer coordinates and display coordinates. MB87J2120, mb87p2020-a hardware manual page 164 3.2 merging description record this record contains all data needed to convert each logical layer into intermediate and physical color space, together with appropriate handling of z-order. table 3-2: merging description record. information size (bits) 1 color space code for intermediate color space 4 2 layer to intermediate transfer codes (one per layer) flag to decide between two ways of transforming the layer s logical color space into the intermediate (common) color space. maximal two legal ways are de?ed for all combinations of logical to intermediate color space mapping (see section 2.3). 16 x 1 3 z-order register contains a place for each layer concurrently displayed. this de?es the z-order, i.e. the ?st place contains the number of the topmost layer, the second place the number of the layer immediately below and so on (see ?. 3-2). 4 x 4 4 plane enable flag to switch display of the respective plane on or off. 4 5 clut offsets (one per layer) this offset is added to the pixel value when forming the address for look-up. 16 x 9 6 clut contents 256 x 1..24 a 512 x 1..24 b 7 background color (con?ming to intermediate color space) the color to be displayed on locations where there is no pixel from any layer of the video ram. this allows to keep the visible layers as small as possible to save ram bandwidth yet still be able to control the appearance of the display back- ground. 1?4 8 background enable this ?g can be used to improve gpu performance when there is at least one layer that covers the whole display area. 1 9 pseudo level duty ratios these give the values used by the duty ratio modulator to produce pseudo levels. since the modulator produces either 16 (4 bit to 1 conversion) or 3 x 4 (6 bit to 3 conversion) pseudo levels there is a maximum of 14 really modulated levels, two are simply on and off level. 14 x 6 10 pre-matrix biases (jasmine only) these values are added to y, u, v before matrix multiplication 3 x 8 11 matrix coefficients (jasmine only) factors for matrix multiplication, all treated as unsigned integer in the range of 0? (see 3.7.2). 9 x 8 12 gamma correction table contents (jasmine only) values used for gamma correction after multiplication. correction is done by table look-up. 3 x 256 x 8 13 gamma table enable (jasmine only) flag to directly map the matrix result into physical color space. 1 graphic processing unit gpu control information page 165 remark : the purpose of the background color and z-order register is a clearer concept of visibility, a more user-friendly programming interface and a unified handling of all layers. if there was no z-order register the user still would be able to swap layers by changing their respective videoram start addresses. how- ever, this would require more accesses than the single access to the z-order register. 3.3 display interface record this group of data contains information necessary to describe the geometric and timing behaviour of the physical display as well as timing data to generate various synchronization signals. 14 rgb gamma forcing (jasmine only) flag to force pixels from rgb555, rgb565, rgb888 through the gamma table, valid only when gamma table enable is set. 1 15 yuv422 chroma interpolation enable (jasmine only) since in logical color space yuv422 two pixels share their chrominance infor- mation linear interpolation between pairs of adjacent pixels is used to smooth the appearance of the picture. 1 a. for MB87J2120 (lavender). b. for mb87p2020-a (jasmine). table 3-3: display interface record. information size (bits) 1 display physical size px, py 2 x 14 2 color space code for physical color space, see 3.12.1 4 3 bit stream format code de?es the number of bits written per video clock period (see section 3.12.2) 4 4 scan mode distinction between single / dual scan, special ?ptrex-mode 2 table 3-2: merging description record. (continued) information size (bits) plane 3 [ layer a ] [ layer b ] foreground background plane 2 plane 1 plane 0 layer b [layer c ] z-order register background color [ empty ] layer a layer c figure 3-2: illustration for z-order register. MB87J2120, mb87p2020-a hardware manual page 166 5 display mode distinction between single and twin display mode. 1 6 r/b swap flag to allow for a swap between red and blue channel for rgb111 color space, only useful for bit stream formats s4 and s8 1 7 physical width in scan dots = px*bitsperpixel / bitsperscanclock 14 8 dual scan offset distance between the two lines which are concurrently output. 14 9 external sync control control bits to de?e reactions on external sync signals, see 3.9 4 10 master timing definition positional parameters to control output and sync signal generation, see 3.9 6 x 15 11 sync pulse generator control position parameters for pulse generators, see 3.10.2 6 x 4 x 31 12 sync sequencer control parameters for sequence-based signal generation, see 3.10.3 64 x 32 + 5 13 sync mixer control parameters that control the ?al generation of sync signals from merged pulse generator and sync sequencer outputs, see 3.10.4 8 x 47 14 sync switch half cycle delay on/off ?g for each sync mixer output 8 15 pixel clock gate control bits to determine clock gating, divider and gating type, see 3.11 4 16 color key limits upper and lower color limit for which a match signal is generated, con?ming to color space 2 x 3 x 1? 17 color key polarity selects whether match signal is low or high active 1 18 blank clamping values values that are output at analog and digital outputs when blanking is active (no pixel data is output) 1 x 1..24 a 2 x 1..24 b 19 main output enable flags bits to control tristate drivers for display signals 29 a. lavender contains one clamping value for digital and analog output. b. jasmine contains two different clamping values for digital and analog output. table 3-3: display interface record. (continued) information size (bits) graphic processing unit gpu control information page 167 3.4 supported physical color space / bit stream format combinations table 3-4 gives an overview on the legal physical color space / bits stream format combinations for the pri- mary display. these are derived from the features of the supported displays. an analog output for s9?12 is also possible in twin display mode (see next section). 3.5 twin display mode twin display mode is only available for mb87p2020-a (jasmine). MB87J2120 (lavender) does not sup- port this feature. the gpu offers the opportunity to run a digital and an analog display in parallel (to output according data at digital and analog output pins, respectively). this is referred to as twin display mode and may be used to concurrently output the picture and post-processing it (e.g. do some color keying). the implication, how- ever, is that both data streams use identical resolutions and sync timing. some sync signals may not be avail- able at the multipurpose pins as well (see 3.14.1). in twin display mode one display becomes the primary display , the other the secondary display . the distinction which display is primary depends on the bit stream format as shown in table 3-5. when the analog display is secondary pixel data are auto scaled to 8 bit per color channel for dac outputs to maintain full voltage swing (bit with at the digital outputs remains according to selected bit stream for- mat). this auto scaling works as explained for bit expansion in 2.3. table 3-4: supported combinations of physical color space to bit stream format physical color space codes bit stream format code s1 s2 s4 s6 s8 s9 s12 s18 s24 an 1bpp x x x rgb111 x x x 4bpp x x rgb333 x rgb444 x rgb666 x rgb888 x x table 3-5: primary / secondary display distinction in twin display mode bit stream format primary display secondary display an analog digital s9?24 digital analog others digital no twin display mode available. MB87J2120, mb87p2020-a hardware manual page 168 3.6 scan modes three scan modes are supported and determine the sequence pixel data that form the bit streams being out- put. figure 3-3 gives an illustration on geometrical reference points on the display used later to describe the according bit streams. 4. single scan: pixels are written in one continuous stream, starting at upper left corner, proceeding left to right, line by line. 5. dual scan: pixels are written in two parallel streams. the ?st stream starts at the upper left corner, proceeding left to right, line by line. the second starts at the left margin of the line de?ed by dualscanoffset (usually py/2), proceeding left to write, line by line. each stream contains dualscanoffset lines, i.e. the stream for the upper display segment determines the number of lines 6. zigzag scan: aka ?ptrex-mode? pixels are written in one stream, starting at the upper left corner, proceeding from left to right, however, line number sequence is peculiar: every other line number is biased with dualscanoffset, i.e. the ?st line written is line #0, the second line is line number dualscanoffset, the third line written is line #1, the fourth line written is line #(dualscanoffset+1) etc. table 3-6 shows the resulting bit streams according to the different scan modes. please note that these streams, as well as the reference points in figure 3-3 count in scan dots, which are not necessarily pixels (this is determined by the bit stream format). a scan dot consists of all bits output during on display clock cycle and may contain the bits of more than one pixel or even bits from fractions of adjacent pixels. [1, dualscanoffs?1] [pxscandots?1, dualscanoffs?1] [pxscandots?1, physsizey?1] [pxscandots?2, physsizey?1] [pxscandots?2, dualscanoffs?1] [0, dualscanoffs?1] [0, 0] [1, 0] [pxscandots?2, 0] [pxscandots?1, 0] [1, 1] [1, 0] [pxscandots?2, 1] [pxscandots?1, 1] [1, physsizey?1] [0, physsizey?1] [1, dualscanoffs] [0, dualscanoffs] [pxscandots?2, dualscanoffs] [pxscandots?1, dualscanoffs] figure 3-3: geometric reference points on the display. graphic processing unit gpu control information page 169 table 3-6: bit stream output sequence according to scan mode, given as [x, y] pairs. where not stated differently, one row is equivalent t o one display clock cycle. single scan dual scan zigzag scan note upper stream lower stream 0?n cycles optional blanking (vertical and horizontal) a [0, 0] [0, 0] [0, dualscanoffs] [0, 0] b [1, 0]] [1, 0] [1, dualscanoffs] [1, 0] ? [pxscandots-2, 0] [pxscandots-2, 0] [pxscandots-2, dualscanoffs] [pxscandots-2, 0] [pxscandots-1, 0] [pxscandots-1, 0] [pxscandots-1, dualscanoffs] [pxscandots-1, 0] c 0?n cycles optional blanking (horizontal) [0, dualscanoffs] d [0, 1] [0, 1] [0, dualscanoffs+1] [1, dualscanoffs] e [1, 1] [1, 1] [1, dualscanoffs+1] [2, dualscanoffs] ? [pxscandots-1, 1] [pxscandots-1, 1] [pxscandots-1, dualscanoffs+1] [pxscandots-1, dualscanoffs] f 0?n cycles optional blanking (horizontal) g [0, 2] [0, 2] [0, dualscanoffs+2] [0, 1] [1, 2] [1, 2] [1, dualscanoffs+2] [1, 1] ? [pxscandots-2, 2] [pxscandots-2, 2] [pxscandots-2, dualscanoffs+2] [pxscandots-2, 1] MB87J2120, mb87p2020-a hardware manual page 170 please refer to section 3.9 for a more detailed explanation of display timing, especially blanking insertion. [pxscandots-1, 2] [pxscandots-1, 2] [pxscandots-1, dualscanoffs+2] [pxscandots-1, 1] 0?n cycles optional blanking (horizontal) [0, dualscanoffs+1] [0, 3] [0, 3] [0, dualscanoffs+3] [1, dualscanoffs+1] [1, 3] [1, 3] [1, dualscanoffs+3] [2, dualscanoffs+1] ? ? [pxscandots-2, physsizey-1] [pxscandots-2, dualscanoffs-1] [pxscandots-2, 2*dualscanoffs-1] [pxscandots-2, dualscanoffs-1] [pxscandots-1, physsizey-1] [pxscandots-1, dualscanoffs-1] [pxscandots-1, 2*dualscanoffs-1] [pxscandots-1, dualscanoffs-1] h 0?n cycles optional blanking (horizontal and vertical) i a. leading blanking before start of active frame area b. leftmost scan dot of topmost line (start of active display area) c. rightmost scan dot on ?rst line, last scan dot of horizontal cycle for single and dual scan d. trailing horizontal blanking after ?rst line for single and dual scan, for zigzag scan there is no blanking between rightmost scan dot of topmost line and leftmost scan dot of line number dualscanoffs e. start of second line (leftmost scan dot) for single and dual scan f. rightmost scan dot on second line for single and dual scan, for zigzag scan this is rightmost scan dot on line number dualscanoffs, i.e. last scan dot of horizontal cycle for zigzag scan g. trailing horizontal blanking h. last scan dot of last line, end of active display area i. trailing horizontal and vertical blanking table 3-6: bit stream output sequence according to scan mode, given as [x, y] pairs. where not stated differently, one row is equivalent t o one display clock cycle. single scan dual scan zigzag scan note upper stream lower stream graphic processing unit gpu control information page 171 3.7 yuv to rgb conversion full support for yuv to rgb conversion as described in this section is only available for mb87p2020-a (jasmine). MB87J2120 (lavender) is able to handle yuv444 and yuv422 in a limited manner. it simply ignores the color components of the color value and displays it in 256 step grayscale format also called ?achromatic mapping?. this behaviour is also available for jasmine as alternative mapping for yuv422, yuv444, yuv555 and yuv655. 3.7.1 yuv422 demultiplexing and chrominance interpolation the yuv422 format is special in the respect that it implies chroma sub-sampling and distribution of u(cb) and v(cr) values on every other pixel (see figure 3-4 below). therefore, pairs of pixels have to be fetched from memory to obtain complete color information. this re- sults in the constraints that layer?s horizontal first pixels (fx) as well as their horizontal window size (wx) must not be odd. due to internal processing of pixels in portions of display segments the horizontal display offset (ox) is also limited to even positions. these restrictions are enforced automatically by gpu by ig- noring the lsbs of the parameters mentioned before. in order to process the yuv422 format in the matrix (see next section) the pixelpump within the dfu does the necessary demultiplexing to assign each pixel the complete chrominance information, i.e. pixels arriv- ing at the matrix have the structure shown in figure 3-5. the effect of chrominance sub sampling can be decreased and hence the appearance of the image improved by applying a linear interpolation between consecutive chrominance values as illustrated in figure 3-6. this interpolation is controlled by a configuration register. luminance ? ? 8 bits each chrominance ? ? 8 bits each figure 3-4: memory map for pixels in yuv422 format. pixels on even positions carry u-chrominance, those on odd positions v-chrominance, luminance ? ? 8 bits each chrominance ? ? 8 bits each ? ? 8 bits each figure 3-5: pixel structure after yuv422 demultiplexing. although now each pixel is assigned complete chrominance information there is still a chrominance sub sampling y 2 i 2 ? y 2 i 1 ? y 2 i y 2 i 1 + y 2 i 2 + y 2 i 3 + u 2 i 2 ? v 2 i 2 ? u 2 i v 2 i u 2 i 2 + v 2 i 2 + chroma 2 i 2 ? chroma 2 i chroma 2 i 2 + i 012 ,,, = y 2 i 2 ? y 2 i 1 ? y 2 i y 2 i 1 + y 2 i 2 + y 2 i 3 + u 2 i 2 ? u 2 i u 2 i 2 + v 2 i 2 ? v 2 i v 2 i 2 + MB87J2120, mb87p2020-a hardware manual page 172 3.7.2 matrix multiplication gpu incorporates a yuv to rgb matrix with prebiasing to convert pixels in yuv (ycbcr) color spaces to rgb color spaces. to flexibly adjust to different input standards, all matrix coefficients can be config- ured. the following equation is calculated for the yuv to rgb conversion: (6) all matrix coefficients are derived from their respective configuration registers in either of the fol- lowing ways: a) chroma coef?cients for green channel and : (7) b) all others: (8) that is, all configuration registers are treated as 8 bit unsigned binary numbers, thus resulting in a range of ?2?0 for coefficients and and 0?2 for all other coefficients. all prebias values are stored as 8 bit signed binary numbers in their con?guration registers (cf. table 4-1). the calculated intensity values for r, g and b are limited to 8 bit positive binary numbers with values out- side the legal range being clipped to the respective margins, i.e. underflows (negative values) are mapped to zero, overflows (values >255) to 255. note: there is only one set of matrix parameters (pre biases and coefficients) for all layers. yuv color spac- es with less than 8 bits per channel (i.e. yuv555 and yuv655) are therefore auto scaled to yuv444 in advance to fit into a unified value range for all yuv color spaces. this scaling works similar to that ex- plained for the mapping from logical to intermediate color spaces in section 2.3. 3.7.3 inverse gamma correction when processing yuv material originally produced for display on crts color channel intensities are cal- culated according to equation (6) may be distorted by a gamma function according to (9) to cope with the non linear electro-optical characteristics of the crt?s phosphorus. (9) with i mtx being the intensity of one color channel according to equation (6), i act the actual (linear) intensity, c and i 0 constants and . however, a linear representation (i.e. no gamma distortion) might be de- manded. therefore, gpu includes three gamma look-up tables (one per color channel) to accomplish the inverse transformation to obtain i act from i mtx as well as any other (possibly non linear) post-matrix trans- luminance ? ? chrominance ? ? ? ? figure 3-6: pixel structure after linear chrominance interpolation y 2 i 2 ? y 2 i 1 ? y 2 i y 2 i 1 + y 2 i 2 + y 2 i 3 + u 2 i 2 ? u 2 i 2 ? u 2 i + 2 ----------------------------- u 2 i u 2 i u 2 i 2 + + 2 ----------------------------- u 2 i 2 + u 2 i 2 + u 2 i 4 + + 2 ----------------------------------- - v 2 i 2 ? v 2 i 2 ? v 2 i + 2 ---------------------------- v 2 i v 2 i v 2 i 2 + + 2 ---------------------------- v 2 i 2 + v 2 i 2 + v 2 i 4 + + 2 ---------------------------------- - r mtx g mtx b mtx ?? ?? ?? ?? ?? c yr c ur c vr c yg c ug c vg c yb c ub c vb ?? ?? ?? ?? ?? y prebias y + u prebias u + v prebias v + ?? ?? ?? ?? ?? = c ij p ij c ug c vg c ij p ij 128 ? () ? = c ij p ij 128 ? () = c ug c vg i mtx ci act i 0 + ? = 2.2 graphic processing unit gpu control information page 173 formation. this is done by loading the gamma tables with the appropriate contents. for instance, the inverse gamma correction would require the contents of address i mtx to be: (10) in a different application the gamma look-up tables might be used to adopt rgb color spaces to special dis- play characteristics. this is referred to as rgb gamma forcing . in this case, the gamma tables are not avail- able for yuv gamma correction anymore, although the matrix still can be used for yuv to rgb conversion. 3.8 duty ratio modulation 3.8.1 working principle this paragraph briefly describes the duty ratio modulator integrated. the purpose of this unit is to provide additional (pseudo-) levels (shades of hue or gray) for output, i.e. virtually expand the color resolution in the physical color space. this is achieved by tuning the duty ratio of actually existing physical bits. assume there is a frame rate of 100 hz on a black and white display (1 bit physical color space). pseudo gray levels can be obtained when pixels are not set black all 100 frames per second but say 80 frames only. thus, the pixel is seen in dark gray instead of black. the less time a pixel is set to black the lighter the gray becomes. this is equivalent to modulating the duty ratio of the pixel signal. vertical sync (i.e. the frame rate) is used to modulate the pixels, since this signal must be common for all pixels on display because otherwise artefacts may occur. if the decision whether to set a pixel to black or white to obtain the desired pseudo gray level is made e.g. on a line-by-line basis it depends on the ratio be- tween the number of lines and the set pseudo gray level where (geometrically on the display) ?gray? pixels are set to black or white. unsuitable settings could then cause actually ?gray? pixels to stay either black or white or visibly flicker. therefore, all ?gray? pixels of the frame of the same pseudo gray level are set to the same physical value of either black or white to achieve a unified optical impression. the number of frames constituting a basic modulation period depends on the number of pseudo gray levels desired and on the linearity of the display characteristics. for instance, if just black, white and gray is in- tended the basic modulation period might be limited to two frames. black pixels are set black every frame (100% duty ratio), white pixels are not set black at all (0% duty ratio) and ?gray? pixels are set black every other frame (50% duty ratio). however, the ?gray? optically perceived might not be 50% black. this is due to a possibly nonlinear electro-optical characteristics of the display. in order to cope with this adjustments must be made possible. an optically 50% black might be achieved with a 60% duty ratio, for instance. one should be able to adjust the duty ratio with an accuracy of some percent (5 or 6 bits precision). hence, longer periods i.e. more frames (30?100) are needed. when using a basic period of e.g. 100 frames a 50% duty ratio is equivalent to an overall 50 frames on and 50 frames off. this implies no statement about when to display the pixel as active and inactive, respectively. setting the pixel first for 50 frames active and afterwards the next 50 frames inactive seems to be equivalent to setting the pixel active every other frame. unfortunately, this is not the case in general. most displays feature a frame rate in the magnitude of 100hz which means that the modulation period of 100 frames lasts for one second. both methods are equivalent only for displays slow enough to integrate over this rather long period. on faster displays the first method (simple pulse width modulation with 50 consecutive frames activity) will cause a blinking of 2 hz clearly visible for the human eye. consequently, a simple pulse width modulation is inadequate and a distribution of on and off values as even as possible is required. this is achieved using a derivation of bresenham?s line algorithm. this algorithm was originally designed to draw a line from point to point on a lattice of discrete pixels with minimal deviation from the ideal graph . the basic idea is to scan the distance step by step and draw pixels at appropriate y -values using a so-called decision variable. this variable tells for a given whether to draw the pixel on or . i act i mtx [] i mtx i 0 ? () c ? () 1 ? == ax a y a , () bx b y b , () ymxn + ? = ? xx b x a ? = xk 1 + () xk 1 + () yk () (,) xk 1 + () yk () 1 + (,) MB87J2120, mb87p2020-a hardware manual page 174 for the application within the gpu there are some simplifications: the basic equation is reduced to , with y representing the activity, x the modulation period and slope m the desired duty ratio. fur- ther, the distance can be limited to a power of 2. the value of the decision variable from bresenham?s algorithm distinguishing between and immediately tells when to set the signal active and when inactive. 3.8.2 usage as listed in table 2-2, there are two mappings for which the duty ratio modulator is applied: from inter- mediate 4bpp to physical 1bpp and for intermediate rgb222 to physical rgb111. usage differs for either case since a different number of pseudo levels is required. mapping from intermediate 4bpp to physical 1bpp in this case there is a total of 16 different pixel values in intermediate color space which need be mapped to a total of only two different pixel values in physical color space. the 16 intermediate values are interpreted as gray scale, thus representing black, white and 14 levels of gray. these 14 levels have to be produced by the drm. the duty ratios used for this 14 pseudo levels are given in gpu registers (see chapter 4). table 3-7 gives the assignment between pixel value and pseudo levels. please note that for a given drm its output is either one or zero at any given time. what differs is the duty ratio, which is adjustable via gpu register. mapping from intermediate rgb222 to physical rgb111 in this case the pixel value is interpreted as three color channels r, g, and b which are mapped individually from intermediate to physical color space. each intermediate color channel has a width of 2 bits, thus al- lowing four different values. as a whole, intermediate rgb222 can represent 64 different colors. however, since each color channel is regarded separately, there are only two pseudo levels per channel, since the other two are constant on and off, respectively. thus, a whole of six modulators is needed. table 3-8 gives the assignment for this case. table 3-7: assignment between pixel values in intermediate and physical color space when using duty ratio modulation for mapping from intermediate 4bpp to physical 1bpp. pixel value pixel value interm. phys. interm. phys. 0 constant off 8 drm 7 output 1 drm 0 output 9 drm 8 output 2 drm 1 output 10 drm 9 output 3 drm 2 output 11 drm 10 output 4 drm 3 output 12 drm 11output 5 drm 4 output 13 drm 12 output 6 drm 5 output 14 drm 13 output 7 drm 6 output 15 constant on ymx ? = ? x yk () yk () 1 + graphic processing unit gpu control information page 175 3.9 master timing information the generation of the correct timing is a rather complex issue, due to the great variety of displays supported. therefore, a coherent, flexible and easy-to-use approach is extremely important. this is achieved by sepa- rating the gpu internal timing (for fetching and bit stream generation) from external timing (for sync sig- nals). the main timing frame for all display types is completely defined by four or six values. fig. 3-7 gives a geometrical interpretation of the four primary values. the general idea is to assign every scan dot (i.e. the bits written during one display clock cycle) a pair of integer numbers, one for the horizontal and one for vertical timing, thus representing display timing as co- ordinates within a two-dimensional scan area. x-values smaller than zero represent horizontal blanking be- fore visible pixels, x-values greater than or equal to pxscandots horizontal blanking after visible pixels. active display area (i.e. visible pixels, either background or layer) is designated by values from 0 to ( px- scandots - 1). the same applies toy-values. those below zero designate leading vertical blanking, those greater than or equal to py trailing vertical blanking, and those between zero and ( py - 1) visible lines. table 3-8: assignment between pixel values in intermediate and physical color space when using duty ratio modulation for mapping from intermediate rgb222to physical rgb111. pixel value channel r channel g channel b interm., bits [5:4] phys., bit [2] interm., bits [3:2] phys., bit [1] interm., bits [1:0] phys., bit [0] 0 constant off 0 constant off 0 constant off 1 drm 4 output 1 drm 2 output 1 drm 0 output 2 drm 5 output 2 drm 3 output 2 drm 1 output 3 constant on 3 constant on 3 constant on xstart single scan dual scan py = segment y ystop+dualscanoffs ystart+dualscanoffs xstop xstart xstop ystart ystop dualscanoffs py px scandots px scandots ystop ystart active display data active display data blanking blanking blanking figure 3-7: interpretation of main timing parameters in relation to geometrical parameters of the physical display. MB87J2120, mb87p2020-a hardware manual page 176 there is a special case: tv-conform timing requires that every other frame consists of an odd number of lines to achieve an odd sum of lines for two consecutive frames (representing the two fields of an interlaced tv picture). therefore, an additional coordinate had to be introduced, referred to as field bit , thus actually expanding the coordinate space to three dimensions. even for this case the number of visible pixels remain constant, what differs is the number of blanking lines. these ?timing coordinates? form the basis for all timing-related signal generation. for a complete descrip- tion a maximum six timing and three geometrical parameters are necessary. the geometrical parameters are the number of scan dots in each direction, i.e. the area with visible pixels on the physical display, and the extra dualscanoffs parameter for dual scan displays (i.e. the number of lines of each display segment). the number of necessary timing parameters is either four or six, depending whether or not field toggling is enabled (i.e. frames with differing number of blanking lines are used). for normal operation a set of four timing parameters is sufficient: xstart , xstop , ystart , and ystop . the ?timing coordinates? men- tioned above cover a range from [ xstart , ystart ] to and including [ xstop , ystop ]. if field toggling is used, then there is a pair of ( ystart , ystop ) values for every frame type, which is used alternatively. this explanation applies also to dual scan displays with the distinction that the y values describe only the upper display segment, the lower segment simply is assigned the same timing (since its respective pixel stream is output in parallel to the upper segment). 3.10 generation of sync signals 3.10.1 overview to achieve maximal flexibility, generation of sync signals is a three stage approach. in a first stage, signals are generated which carry positional timing information (according to the timing coordinates described in the section before). there are two methods to obtain these signals. the second stage combines them to form more complex waveforms. the third stage is used for a programmable delay of half a pixel clock cycle. during display operation the sync generating components are provided with the current timing position as integer numbers in 2?s complement representation. 3.10.2 position matching one way to form first-stage signals is a simple position matching to trigger an rs-flip-flop. this is done by an array of six identical sync pulse generators (spgs). fig. 3-8 shows the working principle. graphic processing unit gpu control information page 177 the output of a sync pulse generator is set or reset if the current position equals the respective programma- ble position in all bits for which its don?t-care-vector (which is also programmable) contains zeros.the off- matching is dominant, i.e. when both, on and off position are matched at the same time, the output of the sync pulse generator is reset. 3.10.3 sequence matching a more sophisticated yet more powerful approach to form first-stage signals is the application of a sequenc- er ram to match a whole sequence of positions. fig. 3-9 shows the principle of operation. there is one sync sequencer (syncseq) which is able to follow an arbitrary sequence of timing positions and generate an appropriate output signal. the length of the sequence as well as the contents of the ram, consisting of position and the assigned output value are programmable. fy x 14 0 14 0 fy x 14 0 14 0 fy x 14 0 14 0 fy x 14 0 14 0 fy x 14 0 14 0 == == match position off don ? t-care vector off don ? t-care vector on match position on s r spg out current position timing coordinate space xstart xstop ystart ystop figure 3-8: matching positions with sync pulse generators. MB87J2120, mb87p2020-a hardware manual page 178 the operation is as follows: at the begin, the address counter is reset to zero and the ram outputs the first position to match and the output value for this position. if the comparator signals a match, the ram address is incremented, the preset output value is propagated and the ram now outputs the next position to match. this match/address increment cycle continues until the programmed sequence length is reached. if the last position is matched, the address counter is reset to zero again and the cycle starts anew. it is thus possible to generate arbitrarily complex waveforms with up to 64 edges (which is the maximum sequence length). 3.10.4 combining first-stage sync signals as shown above, there are six sync pulse generator outputs and one sync sequencer output. to obtain more complex waveforms, these signals can be combined in a second stage. here, an array of eight sync mixers (smx) is used to calculate boolean functions of first-stage signals. each sync mixer can form any boolean function of up to five inputs. the basic structure of one such mixer is depicted in fig. 3-10 fy x 14 0 14 0 == current position addr incr. ram sequencer out enable xstart xstop ystart ystop timing coordinate space sequencer next position to match figure 3-9: matching whole sequences of positions with the sync sequencer. graphic processing unit gpu control information page 179 the concept of the sync mixers needs some explanation. in a first step the signals to be combined are se- lected. these are referred to then as s0?s4 and form the address for the function table. this function table is used to look up the result of the boolean operation the five selected signals shall be subject to. an example may help understand the topic. assuming the outputs of three sync pulse generators shall form a combined signal with the function , one would proceed as follows. at first, the sync mixer signals s0?s4 are assigned the sync pulse generator outputs or constant zero by programming the respective multiplexers. the next step is to build the function?s truth table, as shown in table 3-9. since the intended function has only three inputs, only eight entries need be specified. s3 s2 s1 s0 s4 syncmixer out ctrl reg mux 8 to 1 ctrl reg mux 8 to 1 ctrl mux 8 to 1 ctrl reg mux 8 to 1 reg mux 0 8 16 24 31 8 to 1 const 0 syncseq spg0 spg1 spg2 spg3 spg4 spg5 fcttable mux 32 to 1 reg ctrl figure 3-10: basic structure of a sync mixer. each of the five address lines of the 32 to 1 multiplexer can be individually selected from any of the seven (plus one constant zero) first-stage signals. the out- put is the result of a table look-up. the register fcttable contains the truth table of the boolean function calculated. smx syncseq spg 0 spg 1 ? = MB87J2120, mb87p2020-a hardware manual page 180 it is recommended that s4?s0 are listed in order of binary number representation. this allows to take the function result row immediately as register contents for the sync mixer function table, i.e. the last row is interpreted as binary 32 bit number with the lsb in the first row and the msb in the last. for the example this would be [xxxx xxxx xxxx xxxx xxxx xxxx 0000 1000] binary, with x?s denoting arbitrarily set or reset bits, since these will never be read out of the function table. 3.10.5 sync signal delay adjustment before the outputs of the eight sync mixers are connected to actual gdc pins, they are fed through a pro- grammable delay stage. this allows the signals either to be left untouched or delayed for half a pixel clock cycle. this delay can be set for each of the eight sync mixer output signals individually with the sync switch register. 3.11 pixel clock gating the gpu allows to flexibly shape the pixel clock output at the respective pin. this is done with a program- mable control register that defines how the output of sync mixer 7 affects the output pixel clock. fig. 3-11 shows the resulting waveforms according to gate control settings. it should be mentioned that the control by the mixer output is the only way to determine the position of clock edges when using the divided clock. if the output clock is not gated, but divided, then the first rising clock edge occurs exactly at the same time with the first rising edge of the pure internal pixel clock after reset. table 3-9: function table for the sync mixer example. selected first-stage signals desired output s4 = 0 s3 = 0 s2 = spg1 s1 = spg0 s0 = syncseq smx = f(s0?s4) 00 0 0 0 0 00 0 0 1 0 00 0 1 0 0 00 0 1 1 1 00 1 0 0 0 00 1 0 1 0 00 1 1 0 0 00 1 1 1 0 combinations [s4?s0] = 10000?11111 can never occur since s4 and s3 are selected constantly zero need not be speci?ed graphic processing unit gpu control information page 181 3.12 numerical mnemonic de?itions 3.12.1 color space code this code is intended to designate color resolutions in the different color spaces uniquely. it provides infor- mation on the current color space as well as for transformations between these spaces. table 3-10: color space code de?nitions code color space code color space mnemonic bits occupied mnemonic bits occupied 0 1bpp 1 8 yuv444 32 1 2bpp 2 9 rgb111 2 2 4bpp 4 10 rgb222 6 3 8bpp 8 11 rgb333 9 true clock polarity gating enable gating type clock divider or or or or and and and and x x x x off off off off inv inv true true 1:2 1:1 1:2 1:1 gating singal internal display clock inv inv true 1:1 true on on 1:1 true 1:1 1:2 inv on 1:2 on 1:1 inv 1:2 1:2 on on on on MB87J2120, mb87p2020-a hardware manual page 182 3.12.2 bit stream format code this code describes number of bits actually written to the display (some displays require more than one pix- el written at once). 3.12.3 scan mode code this code determines the scan mode of the display, i.e. the number and layout of pixel streams written in parallel to the physical display. 4 rgb555 16 12 rgb444 12 5 rgb565 16 13 rgb666 18 6 rgb888 32 14 yuv555 a 16 7 yuv422 16 15 yuv655 a 16 a. this format is only available for mb87p2020-a (jasmine). table 3-11: bit stream format code de?nitions code bit stream code bit stream mnemonic bits written mnemonic bits written 0 s1 1 6 s12 12 1 s2 2 7 s18 18 2 s4 4 8 s24 24 3 s6 6 9 an analog output 4 s8 8 10-15 reserved 5s9 9 table 3-12: scan mode de?nitions. code 0231 mnemonic sglscan dualscan zzagscan illegal table 3-10: color space code de?nitions code color space code color space mnemonic bits occupied mnemonic bits occupied graphic processing unit gpu control information page 183 3.13 bit to color channel assignment table 3-13 gives the assignment of the bits to color channels (rgb or yuv) corresponding to bit width (i.e. the number of bits occupied by one pixel) as used gpu internal for chromatic intermediate and physical color spaces. this assignment is important for direct mapping (to determine which gdc pins carry the de- sired signal) and for clut and gamma table loading. table 3-13: bit to rgb/yuv assignment for chromatic color spaces. color space bits assigned to r g b y u(cb) v(cr) rgb111 a a. default assignment with no r/b swap [2] [1] [0] rgb111 b b. alternative assignment with active r/b swap [0] [1] [2] rgb222 [5:4] [3:2] [1:0] 8bpp c c. this assignment is valid only when 8bpp is mapped aligned to intermedi- ate rgb333 [7:5] [4:2] [1:0] rgb333 [8:6] [5:3] [2:0] rgb444 [11:8] [7:4] [3:0] rgb555 [15:11] [10:6] [4:0] rgb565 [15:11] [10:5] [4:0] rgb666 [17:12] [11:6] [5:0] rgb888 [23:16] [15:8] [7:0] yuv422 d d. bits [7:0] contain u and v alternating for every other pixel. [15:8] [7:0] yuv444 [15:8] [7:0] [23:16] yuv555 e e. this format is only available for mb87p2020-a (jasmine). [10:6] [4:0] [15:11] yuv655 e [10:5] [4:0] [15:11] MB87J2120, mb87p2020-a hardware manual page 184 3.14 gpu signal to gdc pin assignment 3.14.1 multi-purpose digital signals some gdc pins are shared by gpu signals, since pins are a limited package resource. this sharing depends on the physical display connected to gdc. table 3-14 shows the assignments. 3.14.2 analog pixel data if scan mode is set to sglscan and bit stream format is an (analog display is primary) or twin display mode (jasmine only) is active and bit stream format is s9?s24 (analog display is secondary), pins a_red, a_green, and a_blue carry analog pixel data, i.e. colors or gray scales as well as the blanking clamp value, otherwise they are high-z. they are high-z as well when the dac output enable bit is reset. 3.14.3 dedicated sync signals three pins for dedicated sync signals are available. table 3-15 gives the assignment. table 3-14: gpu multi-purpose digital signal to gdc pin assignment. pixel data designates actual colors or gray scales as well as the blanking clamp value. gdc pin gpu signal usage a) display with scan mode sglscan dis_d[23:19] digital pixel data [23:19] when bit stream format is either s24, or an and twin dis- play mode active (digital display is secondary), otherwise delay adjusted outputs of sync mixer 7?3 dis_d[18:0] digital pixel data b) display with scan mode dualscan dis_d[23:19] delay adjusted outputs of sync mixer 7?3 dis_d[18:16] constant zero dis_d[15:8] lower display segment digital pixel data dis_d[7:0] upper display segment digital pixel data c) display with scan mode zzagscan dis_d[23:19] delay adjusted outputs of sync mixer 7?3 dis_d[18:8] constant zero dis_d[7:0] digital pixel data note 1 : for cases with dis_d[23:19] being not connected to the delay adjusted outputs of sync mixers 3 to 7 these signals are not available at gdc pins. note 2 : independent of the gpu signals being connected to the pins their output is further controlled by the respective output enable settings. dis_pixclk when used as output the pin carries the pixel clock as formed due to gating and divider settings (cf. 3.11) table 3-15: gpu sync signal to gdc pin assignment. gdc pin gpu signal usage dis_hsync this pin is always connected to the delay adjusted output of sync mixer 0. graphic processing unit gpu control information page 185 3.14.4 sync mixer connections table 3-16 shows the mapping of sync mixer outputs to gdc pins or its internal connections. sync mixer 7 is used for pixel clock gating as already described in chapter 3.11. sync mixer 6 is connected to the flag flnom_gsync which can be used by an application to trigger on an event regarding display output. for example with help of this flag an application can synchronize its draw- ing on display frame rate. it is also possible to generate a mcu interrupt with this flag (see ulb description for details about flags and interrupt handling). note that this connection is only available for mb87p2020 and mb87p2020-a. sync mixer 5 can be taken as synchronization input for the ccfl driver. this allows a flexible ccfl flash rate synchronization with display output refresh rate. see ccfl description for further details about syn- chronization settings. note that this connection is only available for mb87p2020-a. internal connections for sync mixers are always established independent from display settings. 3.14.5 color key output the gdc pin dis_ck carries the output of the color key unit. it is high-z when the respective output enable bit is reset. behaviour: the output becomes active with a user-controlled polarity when data for visible pixels is output that lies within limits that are also defined by the user. it is inactive during blanking periods. the behaviour further depends on physical color space / bit stream format combinations: for combinations that lead to an output of one pixel per pixel clock, the output is updated on a per- pixel basis. dis_vsync this pin is always connected to the delay adjusted output of sync mixer 1. dis_vref this pin is always connected to the delay adjusted output of sync mixer 2. note : although the pin names seem to imply a certain signal function this is not the case. in fact, any of the pins above can output any arbitrary waveform generated with the respective sync mixer. table 3-16: sync mixer connections sync mixer gdc pin internal connection sync mixer 0 dis_hsync sync mixer 1 dis_vsync sync mixer 2 dis_vref sync mixer 3 dis_d[19] sync mixer 4 dis_d[20] sync mixer 5 dis_d[21] ccfl synchronization (mb87p2020-a only) sync mixer 6 dis_d[22] flag flnom_gsync (mb87p2020(-a) only) sync mixer 7 dis_d[23] clock gating table 3-15: gpu sync signal to gdc pin assignment. gdc pin gpu signal usage MB87J2120, mb87p2020-a hardware manual page 186 for combinations with more than one pixel per pixel clock cycle (either all complete pixels or with partial remainders, i.e. not all bits of one pixel are included) the output corresponds to the geometri- cally leftmost complete pixel. for combinations with pixel clock cycles for which no complete pixel is output (instead remaining bits from the previous pixel clock and some bits of the next pixel) the output corresponds to the geo- metrically leftmost complete previous pixel. the matching is done separately for each color channel. a pixel matches the key when the following con- dition holds: and and the pixel?s color channels are extracted automatically corresponding to the physical color space. limit red, lower pixel red limit red, upper ? limit green, lower pixel green limit green, upper ? limit blue / mono, lower pixel blue / mono limit blue / mono, upper ? graphic processing unit gpu register set page 187 4 gpu register set 4.1 description addresses are hexadecimal byte addresses in mcu address space, initial values are given as mnemonic or hexadecimal, as appropriate. bit slices are according to the word starting at address addr . registers marked in lock are read-only during dir_mtimingon set to 1. table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value layer description record information ldr0_phaddr ldr15_phaddr pha(i) 1000 ? 103c layer 0 to layer 15 physical address in videoram (these are actually sdc registers, included here for completeness of layer description records only.) undef. ldr0_domsz ldr15_domsz dsz_x(i) 1040 ? 107c layer 0 to layer 15 domain size, i.e. size of whole area covered by the layer [29:16] = dx = width (unsigned integer) (these are actually sdc registers, included here for completeness of layer description records only. there is no dy stored, it is the user?s responsibility to provide suf?cient ram space for each layer.) undef. ldr0_1stpxl ldr15_1stpxl dp1_x(i) dp1_y(i) 1080 ? 10bc layer 0 to layer 15 ?rst pixel to be displayed, i.e. offset within logical layer domain [29:16] = fx (unsigned integer) [13:0] = fy (unsigned integer) undef. ldr0_winsz ldr15_winsz wsz_x(i) wsz_y(i) 10c0 ? 10fc layer 0 to layer 15 window size, i.e. size of visible area [29:16] = wx = width (unsigned integer) [13:0] = wy = height (unsigned integer) undef. ldr0_offs ldr15_offs wof_x(i) wof_y(i) 1100 ? 113c layer 0 to layer 15 display offset, i.e. offset from upper left corner of physical display [29:16] = ox (signed integer) [13:0] = oy (signed integer) undef. ldr0_trcol ldr15_trcol ltc(i) 1140 ? 117c layer 0 to layer 15 transparent color, i.e. color for which a pixel is not to be displayed (= transparent) [23:0] = color (according to color space) undef. ldr0_blcol ldr15_blcol lbc(i) 1180 ? 11bc layer 0 to layer 15 blink color, i.e. color for which a pixel is displayed either in this or blink alternative color [23:0] = color (according to color space) undef. ldr0_blalt ldr15_blalt lac(i) 11c0 ? 11fc layer 0 to layer 15 blink alternative color [23:0] = color (according to color space) undef. MB87J2120, mb87p2020-a hardware manual page 188 ldr0_blrate ... ldr15_blrate lbr_off(i) lbr_on(i) 1200 ? 123c layer 0 to layer 15 blink rate de?nitions [15:8] = (#-1) of frames for blink alternative color [7:0] = (#-1) of frames for blink color undef. ldr0_cspace ldr15_cspace cspc_csc(i) cspc_te(i) cspc_lde(i) 1240 ? 127c layer 0 to layer 15 color space & format de?ni- tions [3:0] = color space code [8] = transparency enable [9] = line doubling enable (for video ?eld/frame conversion) undef. merging description record information mdr_format cformat_csc cformat_gforce cformat_ipolen cformat_gamen cformat_litc 1300 intermediate color space & conversion de?nition [3:0] = intermediate color space [5] = rgb gamma forcing (valid only when gamma table enable is set) [6] = yuv422 chroma interpolation enable [7] = gamma table enable [31:16] = layer to intermediate transfer codes (0 = normal transfer, 1 = alternative transfer) 1bpp off off off 0000 mdr_bgnd backcol_col backcol_en 1304 background color, i.e. color to display for pixels not covered by any layer [23:0] = color (according to intermediate color space) [24] = background enable 0000 0 mdr_blinkctrl mbc_en mbc_cbs 1308 blink enable bits & blink state read-back [15:0] = blink enable (one bit per layer) [31:16] = current blink state (read-only) 0000 mdr_zorder zorder_en0 zorder_en1 zorder_en2 zorder_en3 zorder_tm0 zorder_tm1 zorder_tm2 zorder_tm3 130c z-order register, i.e. four places to describe stack- ing of layers [3:0] = # of downmost layer (plane 0) [7:4] = # of next layer (plane 1) [11:8] = # of top but one layer (plane 2) [15:12] = # of topmost layer (plane 3) [16] = enable plane 0 [17] = enable plane 1 [18] = enable plane 2 [19] = enable plane 3 0 0 0 0 off off off off mdr_clutoffs0 mdr_clutoffs15 clutof(i) 1340 ? 137c table of 16 color table offsets [8:0] = offset into look-up table for layer i undef. mdr_drm0 mdr_drm13 drm(i) 1380 ? 13b4 table of 14 de?nitions for pseudo gray levels [5:0] = duty ratio de?nition undef. yuv to rgb matrix parameters (jasmine only) table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value graphic processing unit gpu register set page 189 mtx_prebias prebias_y prebias_cb prebias_cr 1400 pre matrix bias values [23:16] = prebiasy [15:8] = prebiascb [7:0] = prebiascr f0 80 80 mtx_coeffr coeffr_yxr coeffr_cbxr coeffr_crxr 1404 matrix coef?cients to red channel [23:16] = yxr [15:8] = cbxr [7:0] = crxr 80 0 b3 mtx_coeffg coeffg_yxg coeffg_cbxg coeffg_crxg 1408 matrix coef?cients to green channel [23:16] = yxg [15:8] = cbxg [7:0] = crxg 80 2c 5b mtx_coeffb coeffb_yxb coeffb_cbxb coeffb_crxb 140c matrix coef?cients to blue channel [23:16] = yxb [15:8] = cbxb [7:0] = crxb 80 e2 0 clut clut(i) 2000 ? 27fc color look-up table with 512 (jasmine) or 256 (lavender) entries [23:0] = color (according to intermediate color space) undef. gamma correction tables (jasmine only) gamma gamma_r gamma_g gamma_b 2800 ? 2bfc look-up tables for inverse gamma correction [23:16] = r [15:8] = g [7:0] = b undef undef undef display interface record information dir_phsize phsize_x phsize_y 3000 x display physical size, i.e. visible pixel area [29:16] = px = width [13:0] = py = height 0 0 dir_format phfrm_csc phfrm_rbsw phfrm_bsc phfrm_fte phfrm_sm phfrm_tdm phfrm_ies phfrm_hsyae phfrm_vsyae phfrm_pol 3004 x output format de?nitions [3:0] = physical color space [4] = r/b swap [11:8] = bit stream format code [12] = ?eld toggle enable (distinguishes between odd & even ?elds) [17:16] = scan mode [18] = twin display mode (1 = enable) [24] = int./ext. sync (0 = int. sync, i.e. hsync,vsync, vref are outputs) [25] = xhsync active edge (0 = low/high edge) [26] = xvsync active edge [26] [27] = xfref polarity (0 = odd ?eld ref. is low active) 1bpp off s1 off sglsca n off int 0 0 0 table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value MB87J2120, mb87p2020-a hardware manual page 190 dir_pxscandots phscan_sclk 3008 x physical size in scan dots (px*bitsperpixel/bit- sperscanclock) [29:16] = sx 0 dir_dualscanoffs dualscof 300c x dual scan y offset (usually py/2) [13:0] = offset 0 dir_mtimoddstart mtimodd_x(0) mtimodd_y(0) 3010 x master timing: odd/only ?eld start (in scan clock cycles, with respect to visible area) [30:16] = x (2?s complement) [14:0] = y (2?s complement) 0 0 dir_mtimoddstop mtimodd_x(1) mtimodd_y(1) 3014 x master timing: odd/only ?eld stop (in scan clock cycles, with respect to visible area) [30:16] = x (2?s complement) [14:0] = y (2?s complement) 0 0 dir_mtimevenstart mtimeven_y(0) 3018 x master timing: even ?eld start (in scan clock cycles, with respect to visible area) [14:0] = y (2?s complement) 0 dir_mtimevenstop mtimeven_y(1) 301c x master timing: even ?eld stop (in scan clock cycles, with respect to visible area) [14:0] = y (2?s complement) 0 dir_mtimingon mtimon 3020 master timing switch [0] = switch (0 = off, 1 = on) off dir_timingdiag timdiag_x timdiag_field timdiag_y 3024 diagnostic register, contains current timing posi- tion read- only dir_spg0poson spgpson_x(i) spgpson_field(i) spgpson_y(i) 3030 sync pulse generator 0, ?switch-on? position [30:16] = x scan position (2?s complement) [14:0] = y scan position (2?s complement) [15] = ?eld (0=odd, 1=even ?eld) 0 0 0 dir_spg0maskon spgmkon_x(i) spgmkon_field(i) spgmkon_y(i) 3034 sync pulse generator 0, ?switch-on? don?t care vec- tor [30:0] = mask bits (1= do not include this bit into position matching) 0 dir_spg0posoff spgpsof_x(i) spgpso_field(i) spgpso_y(i) 3038 sync pulse generator 0, ?switch-off? position [30:16] = x scan position (2?s complement) [14:0] = y scan position (2?s complement) [15] = ?eld (0 = odd, 1 = even ?eld) 0 0 0 dir_spg0maskoff spgmkof_x(i) spgmkof_field(i) spgmkof_y(i) 303c sync pulse generator 0, ?switch-off? don?t care vector [30:0] = mask bits (1= do not include this bit into position matching) 0 dir_spg1poson 3040 sync pulse generator 1, ?switch-on? position (cf. dir_spg0poson) table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value graphic processing unit gpu register set page 191 dir_spg1maskon 3044 sync pulse generator 1, ?switch-on? don?t care vec- tor (cf. dir_spg0poson) dir_spg1posoff 3048 sync pulse generator 1, ?switch-off? position (cf. dir_spg0poson) dir_spg1maskoff 304c sync pulse generator 1, ?switch-off? don?t care vector (cf. dir_spg0poson) dir_spg2poson 3050 sync pulse generator 2, ?switch-on? position (cf. dir_spg0poson) dir_spg2maskon 3054 sync pulse generator 2, ?switch-on? don?t care vec- tor (cf. dir_spg0poson) dir_spg2posoff 3058 sync pulse generator 2, ?switch-off? position (cf. dir_spg0poson) dir_spg2maskoff 305c sync pulse generator 2, ?switch-off? don?t care vector (cf. dir_spg0poson) dir_spg3poson 3060 sync pulse generator 3, ?switch-on? position (cf. dir_spg0poson) dir_spg3maskon 3064 sync pulse generator 3, ?switch-on? don?t care vec- tor (cf. dir_spg0poson) dir_spg3posoff 3068 sync pulse generator 3, ?switch-off? position (cf. dir_spg0poson) dir_spg3maskoff 306c sync pulse generator 3, ?switch-off? don?t care vector (cf. dir_spg0poson) dir_spg4poson 3070 sync pulse generator 4, ?switch-on? position (cf. dir_spg0poson) dir_spg4maskon 3074 sync pulse generator 4, ?switch-on? don?t care vec- tor (cf. dir_spg0poson) dir_spg4posoff 3078 sync pulse generator 4, ?switch-off? position (cf. dir_spg0poson) dir_spg4maskoff 307c sync pulse generator 4, ?switch-off? don?t care vector (cf. dir_spg0poson) dir_spg5poson 3080 sync pulse generator 5, ?switch-on? position (cf. dir_spg0poson) dir_spg5maskon 3084 sync pulse generator 5, ?switch-on? don?t care vec- tor (cf. dir_spg0poson) dir_spg5posoff 3088 sync pulse generator 5, ?switch-off? position (cf. dir_spg0poson) dir_spg5maskoff 308c sync pulse generator 5, ?switch-off? don?t care vector (cf. dir_spg0poson) dir_ssqcycle ssqcycle 30fc sequencer cycle length [4:0] = (#-1) of sequencer cycles 0 table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value MB87J2120, mb87p2020-a hardware manual page 192 dir_ssqcnts ssqcnts_out(i) ssqcnts_seqx(i) ssqcnts_field(i) ssqcnts_seqy(i) 3100 ? 31fc sequencer position de?nitions [30:16] = x scan position (2?s complement) [14:0] = y scan position (2?s complement) [15] = ?eld (0=odd, 1=even ?eld) [31] = output value, when position is reached undef. dir_smx0sigs smxsigs_s4(i) smxsigs_s3(i) smxsigs_s2(i) smxsigs_s1(i) smxsigs_s0(i) 3200 sync mixer 0 signal select [14:12] = s4 [11:9] = s3 [8:6] = s2 [5:3] = s1 [2:0] = s0 s i == 0: const. zero s i == 1 sync sequencer output s i == 2?7 sync pulse generator 0?5 output 0 0 0 0 0 dir_smx0fcttable smxfct(i) 3204 sync mixer 0 function table [31:0] = output value mixer output = function table [a] a = s4*2 4 +s3*2 3 +s2*2 2 +s1*2 1 +s0*2 0 0 dir_smx1sigs 3208 sync mixer 1 signal select (cf. dir_smx0sigs) dir_smx1fcttable 320c sync mixer 1 function table (cf. dir_smx0fcttable) dir_smx2sigs 3210 sync mixer 2 signal select (cf. dir_smx0sigs) dir_smx2fcttable 3214 sync mixer 2 function table (cf. dir_smx0fcttable) dir_smx3sigs 3218 sync mixer 3 signal select (cf. dir_smx0sigs) dir_smx3fcttable 321c sync mixer 3 function table (cf. dir_smx0fcttable) dir_smx4sigs 3220 sync mixer 4 signal select (cf. dir_smx0sigs) dir_smx4fcttable 3224 sync mixer 4 function table (cf. dir_smx0fcttable) dir_smx5sigs 3228 sync mixer 5 signal select (cf. dir_smx0sigs) dir_smx5fcttable 322c sync mixer 5 function table (cf. dir_smx0fcttable) dir_smx6sigs 3230 sync mixer 6 signal select (cf. dir_smx0sigs) dir_smx6fcttable 3234 sync mixer 6 function table (cf. dir_smx0fcttable) dir_smx7sigs 3238 sync mixer 7 signal select (cf. dir_smx0sigs) dir_smx7fcttable 323c sync mixer 7 function table (cf. dir_smx0fcttable) dir_sswitch sswitch 3240 x sync switch [7:0] = delay select (0=no, 1=0.5 cycle delay) 0 table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value graphic processing unit gpu register set page 193 dir_pixclkgate pixclkgt_gt pixclkgt_gon pixclkgt_cp pixclkgt_hc 3248 x pixel clock gate control (by output of sync mixer 7) [0] = gate type (0=and, 1=or) [1] = gate enable (0=off) [2] = clock polarity (0=true, 1=inverted) [3] = divider (0=1:1, 1=1:2) 0 0 0 0 dir_cklow cklow_llr cklow_llg cklow_llb cklow_op 3250 color key lower limits (according to physical color space) [23:16] = red channel [15:8] = green channel [7:0] = blue/monochrome channel [24] = key out polarity (0=active high, 1=active low) 0 0 0 0 dir_ckup ckup_ulr ckup_ulg ckup_ulb 3254 color key upper limit (according to physical color space) [23:16] = red channel [15:8] = green channel [7:0] = blue/monochrome channel pin xo_ckey is activated, when all pixel channels lie within (including limits) their respective limits 0 0 0 dir_aclamp aclamp_aclr aclamp_aclg aclamp_aclb 3258 jasmine: clamping value for analog outputs lavender: clamping value for analog and digital output [23:16] = red channel [15:8] = green channel [7:0] = blue channel 0 0 0 dir_dclamp dclamp 325c clamping value for digital outputs (jasmine only) [23:0] = value output during blanking 0 dir_mainenable mainen_doe mainen_hsoe mainen_vsoe mainen_vroe mainen_ckoe mainen_dacoe mainen_refoe 3260 main display output enable ?ags [23:0] = digital data output enable [24] = hsync output enable (when int. sync) [25] = vsync output enable (when int. sync) [26] = vref output enable (when int. sync) [27] = ckey output enable [28] = dacs output enable [29] = reference voltage enable 0 0 0 0 0 0 0 sdram controller interaction gpu_sdcprio sdcp_lp sdcp_hp sdcp_ifl 3270 sdram controller request priorities [2:0] = low priority value (not used) [6:4] = high priority value [15:8] = input fifo load (read-only) 3 7 table 4-1: gpu register set (continued) . mnemonic (api-names) addr lock function initial value MB87J2120, mb87p2020-a hardware manual page 194 4.2 determination of register contents 4.2.1 values derived from display specs resolution and formats the display spec states the number of physically visible pixels in each direction. these numbers are the val- ues used in dir_phsize , i.e. the number of visible pixels per line is the px value, the number of lines the py value. the code for physical color space is derived from the number of bits per pixel (color resolution). after phys- ical color space is defined, intermediate color space has to be chosen from the legal mappings, given in table 2-2, section 2.4. the number of pixels the display expects per pixel clock cycle (or the number of bits expected per pixel clock in relation to the color resolution, as actually stated), together with the display interface type (digital or analog) determine the bit stream format. once derived, the values for px and bit stream format are used to determine the physical size in scan dots, i.e. pxscandots . this value is calculated this way: . that means that the value of pxscandots is rounded towards the nearest integer that is less than or equal to the fractional value. this does not prevent a fractional number of pixels per scan clock being output as re- quired by some displays. timing and sync-signals after entry of resolution and formats timing is the next item to specify. this has to be done in accordance to the pixel clock set up in the clock unit (cu). as was elaborated in section 3.9, all timing is described in terms of the timing coordinate space. the first task is therefore to establish its basic unit, i.e. the time needed per scan dot. this value can be obtained either directly from the display?s spec sheet or need to be calculated from other values given there. due to the fact that only discrete pixel clock rates are possible, this unit time has to be rounded to the nearest value allowed by the actual pixel clock. hence, now the values for xstart and xstop (bits [30:16], 2?s complement, addresses 0x3010 and 0x3014) can be calculated. the first of these contains the start time of a line in relation to the begin of active display (i.e. the first visible pixel) expressed in timing coordinates, i.e. the number of pixel clock cycles. this time is obtained from the display?s spec sheet either as time or number of clocks. if given as time, the value is de- termined this way: . please note that for leading blanking this value is negative, so if the spec sheet gives the number of clock cycles directly, it has to be entered as neg- ative number as well. depending on how the length of a line is stated in the display spec, the xstop value can be obtained in two ways: from a time value: from the number of clock cycles: the corresponding ystart and ystop values are calculated in a similar manner. if no tv-conform tim- ing is needed then the respective ?odd? values are sufficient (bits [14:0], 2?s complement, addresses 0x3010 and 0x3014). the value for ystart is basically the number of leading blanking lines entered as negative value (cf. note for xstart ). if only a period of time for leading blanking is given in the spec the value can be calculated as follows: depending on whether the global number of display lines (active and blanking) or the frame period is given in the spec, the ystop value can be calculated these ways: using a global number of lines: using the frame period: pxscandots px bitsperpixel bitsperscanclock ------------------------------------------------ - = t scandot 1 pixelclock ? xstart t linestart t activedisplay ? () t scandot ? = xstop xstart timeperline t scandot ? () 1 ? + = xstop xstart cyclesperline 1 ? + = ystart vertleadblank timeperline ? = vertleadblank xstop xstart ?1 + () pixelclock ? () ? = ystop numberoflines ystart 1 ? + = ystop t frame timeperline ? () ystart 1 ? + = graphic processing unit gpu register set page 195 in case of tv-conform timing being required two different pairs of ystart and ystop values are needed, referred to as ?odd? and ?even?. they will be used alternating every other frame. usually, the ystart val- ues will be the same, whilst the ystop values differ by one. thus it is possible to achieve an odd number of lines for two consecutive frames. note: the complete entry of all necessary start and stop values is indispensable regardless of whether in- ternal or external generation of sync signals is specified. after determination of start and stop values sync signals can be specified as obtained from the display spec if internal generation of sync signals is requested. for the majority of cases usage of sync pulse generators as described in 3.10.2 should be sufficient. the basic task for their application is to determine the values for start and duration of the required sync signals. these values are then converted to the ?timing coordinates? and entered into the appropriate gpu registers (addresses 0x3030 to 0x308c). for instance, a horizontal sync signal that starts a display line and lasts for would result in a value for its ?switch-on? position of and a ?switch-off? position of , whilst the respective y-components set to ?don?t care?. hence, this signal will provide a pulse of length at the beginning of each line. position calculations are similar when the sync sequencer shall be used. the required polarity and optional half cycle delay at the respective gdc pin is produced by the appropriate sync mixer and sync switch settings as explained in 3.10.4 and 3.10.5. external pixel clock depending on the specified gdc clock source for pixel clock (received from pin or derived from internal clocks) the gpu can produce a tailored pixel clock signal at the clock pin for the display. some displays may require a gated clock that has periodical gaps. this behaviour is stated in the display spec. if necessary, it can be produced using the clock gating described in 3.11 in conjunction with the settings for sync mixer 7 that controls this gate. the gate control signal is produced in the same manner as described above for the sync signals. size, timing, and sync dependencies although the values for physical display size, timing, and sync generation are programmable within wide ranges, they are strictly interlocked for a certain display. therefore, it is important to enter consistent values. the previous paragraphs gave the rules to determine the values, here, table 4-2 shall conclude them as a quick overview to check for correct magnitudes in correspondence to different scan modes for a display with a given pixel resolution of (see also section 3.6).the timing values are approximative and have to be tuned for blanking. table 4-2: dependencies for size and timing according to scan mode for a display of given resolution. values scan mode sglscan dualscan zzagscan size px x py y pxscandots timing basic horizontal (hsync) cycle, i.e. pxscandots basic vertical (vsync) cycle, i.e. y resulting frame period t hsync p xon , xstart = p xoff , p xon , t hsync t scandot ? () + = t hsync xy px bitsperpixel bitsperscanclock ------------------------------------------------ - xstop xstart ? 2 pxscandots ystop ystart ? y 2 ? pxscandots y pxscandots y 2 ------------------------------------------- pxscandots y MB87J2120, mb87p2020-a hardware manual page 196 interface to physical display the electrical interface of the actual display wired to the gdc determines the number, type and possibly the direction of signals at the pins. direction information is needed for sync signals, as the gpu supports external and internal synchronization. in case of external synchronization the pins dis_hsync, dis_vsync, and dis_vref are inputs, otherwise outputs. the digital multi-purpose pins dis_d[23:0] carry either pixel data or sync signals as given in table 3-14, section 3.14. pins not necessary for operation can be switched to high-z by resetting their corresponding output enable bit in the dir_mainenable word (address 0x3260). this applies also to the color key output pin dis_ck as well as to the sync signal pins dis_hsync, dis_vsync, and dis_vref when in inter- nal synchronization mode. in case of a display with analog inputs connected to gdc the pins a_red, a_green, and a_blue carry the output if the internal dacs when the following conditions hold (otherwise they are high-z): scan mode (bits [17:16] of dir_format ) is set to sglscan and dac output enable as well as reference voltage enable (bits [28] and [29] of dir_mainenable , address 0x3260) is set and either : ? twin display mode (bit [18] of address 0x3004) is off ? physical color space (bits [3:0] of dir_format , address 0x3004) is set to rgb888, ? bit stream format (bits [11:8] of dir_format ) is set to an (analog display is primary), or (jasmine only) ? twin display mode is on ? physical color space is rgb333?rgb888 ? bit stream format is set to s9?s24 (analog display is secondary) when connecting analog displays to gdc via capacitors a clamping value for defined levels is necessary. this value is entered into the dir_aclamp register (address 0x3258) and is output during blanking periods. the similar value dir_dclamp (address 0x325c) exists for the digital pins and may there be useful for external glue circuitry. 4.2.2 values determined by application matrix settings (jasmine only) if image data is processed in one of the yuv formats appropriate matrix parameters and possibly gamma table settings are necessary. first, the pre-matrix biases (bits [23:0] of address 0x1400) need be determined. these are signed 8 bit numbers in 2?s complement which are used to adjust the value ranges of luminance and chrominance before matrix multiplication. adjustments should be made in a manner that: ? luminance (y) values are mapped to a monotonous range of 0?0xff (zero to full scale) ? chrominance (u, v) values are mapped to a monotonous range of 0x80?0?0x7f (most nega- tive?neutral gray?most positive) if the image data comply to the itur-601 standard then the default values need not be changed. next, the matrix coefficients have to be calculated, i.e. their values are mapped to 8 bit binary numbers. this is accomplished using the following formula: (11) please be aware that (cf. 3.7.2) ? the absolute value of any coef?cient must lie in the range of 0?2 ? the values of coef?cients cbxg and crxg (bits [15:8] and [7:0] of address 0x1408)are inter- nally treated as negative. if the image data comply to the itur-601 standard then the default values need not be changed. configreg abs coeff ( ) 128 ? mod 256 = graphic processing unit gpu register set page 197 gamma settings (jasmine only) if there is any post processing required after matrix multiplication then this can be accomplished for any function by loading appropriate values into the gamma tables (bits [23:0] of addresses 0x2800?0x2bfc). in order for the gamma conversion to take affect the gamma enable flag (bit [7] of ad- dress 0x1300) must be set. 4.2.3 user preferences in general, all layer specific settings (positions, color resolution, blink/transparency color, blink rate, line doubling) are unlimited at the user?s command 1 . further, blink enables, z-order, and clut offsets can also be chosen arbitrarily. when displaying yuv422 image data chroma interpolation can be switched on to im- prove image quality. it is also up to the user to switch on line doubling, especially for video data fetched by vic. the contents of the color look-up table is also free to choose, with the only limitation that the effective bit width of the contents is determined by the intermediate color space. this limitation applies also to the back- ground color. a similar limitation holds for the color key limits, but with the physical color space. duty ratio modulation (cf. 3.8) may be applied to improve the impression of the image displayed. its usage depends on the physical color space and intermediate color space. for the mappings [intermediate 4bbp to physical 1bpp] and [intermediate rgb222 to physical rgb111] duty ratio modulation is used, hence, the values for the pseudo levels to be generated have to be entered into the respective registers (addresses 0x1380 to 0x13b4). 4.3 gpu initialization sequence in order to ensure a correct operation of the connected display as well as for the gpu itself control data has to be entered in a certain sequence. further, not every control register may be written at any given time. basically, information that defines display timing can not be changed after mastertimingon has been set to one. initialization should be carried out as follows: 7. determine parameters of the display physically connected to the gdc and derive timing characteris- tics. load the values into the appropriate registers. 8. determine what interface signals are needed and enable them by setting the values in dir_mainenable. 9. load the contents of the clut, if needed. 10. derive the necessary intermediate color space (dependent on the physical color space), using tables 2-2 and 2-3. 11. determine whether matrix multiplication is necessary. if so, load pre matrix biases and matrix coef?- cients (when differing from defaults). if post processing of matrix results is required load gamma tables with appropriate contents. 12. load the duty ratio modulator registers, if needed. 13. set layer geometry by loading the respective layer description record registers. 14. now, mastertimingon may be set, thus starting signal generation at the output pins. if the z-order register remained unchanged in reset state, all pixels are displayed in background color. during normal operation, color attributes (layer, blinking, transparent, background, clut contents) may be set at any time and take effect immediately. changing layer color spaces, layer position parameters or z-order register contents affects the display at the start of the next frame being fetched from video ram 2 1. there is a limitation concerning the color space in that there has to be a legal mapping to the intermediate color space, as given in table 2-1. i gam fi mtx () = MB87J2120, mb87p2020-a hardware manual page 198 5 bandwidth considerations 5.1 processing bandwidth 5.1.1 average bandwidth for a first go/no go decision an estimation of average gpu bandwidth demands might be sufficient. it is done on the basis of the number of pixels processed per frame and the frame period. in general, the condition (12) must hold for the gpu to work properly. these two periods are calculated as follows: a) frame period: (13) (14) (15) b) processing period: (16) note 1: displayarea counts only when background is enabled. note 2: layerarea counts only for potentially visible pixels, i.e. parts of layers outside the physical display area need not be processed and therefore do not count here. however, pixels hidden underneath others do count. 5.1.2 peak bandwidth to have a more reliable statement about whether or not confirming to bandwidth restrictions the peak band- width has to be calculated, since this is virtually the limiting constraint. therefore, a more detailed elabo- ration into gpu function is necessary. pixels are processed in portions of one line segment . the size of one such line segment is a function of the visible layers logical color spaces, the physical color space, an internal burst limit (512 pixels) for video- ram accesses, and the actual number of pixels per line. (17) the values for the number of pixels per word (either per videoram word or per lsa word) according to color spaces are given in tables 5-1 and 5-2. in the special case that no layer is active defaults to 1. 2. due to internal fifos the physical display is affected with a delay dependent on picture size, visible layers and their color resolution. t proc t frame t frame t displayclock x ? ? y ? ? = x ? xstop xstart ? = y ? yoddstop yoddstart ?1 + for disabled field-toggling 1 2 -- - yoddstop yoddstart ?1 + () yevenstop yevenstart ?1 + () + () ? for enabled field-toggling ? ? ? ? ? = t proc t background t layers + = t coreclock displayarea layerarea + () ? = t coreclock physx physy ? () wx i wy i ? () visible layers + ?? ?? ? = seglen min 32 min ppw log l , () ? 32 ppw phys ? 512 physsizex ,,, () = min ppw log l , () graphic processing unit bandwidth considerations page 199 the limiting constraint (12) can now be refined to (18) since every line segment has to be fully processed during a time shorter or equal to the time a segment is output. these two periods of time are calculated as follows: a) time to output a line segment: (19) (20) (21) the values for bits per scan dot (bpsd) and bits per physical pixel (bpp phys ) are given in tables 5-3 and 5-4. please note that only certain combinations of bit stream formats and physical color spaces are supported (cf. table 3-4, p. 165). table 5-1: number of pixels per videoram word according to logical color space. logical color space code 1bpp 2bpp 4bpp 8bpp rgb555 rgb565 rgb888 yuv422 yuv444 yuv555 a a. this format is only available for mb87p2020-a (jasmine). yuv655 a ppw log 3216842212122 table 5-2: number of pixels per lsa word according to physical color space. physical color space code 1bpp rgb111 4bpp rgb333 rgb444 rgb666 rgb888 ppw phys 24862211 table 5-3: number of bits per scan dot according to bit stream format. bit stream format code s1 s2 s4 s6 s8 s9 s12 s18 s24 an bpsd 1 2 4 6 8 9 12 18 24 24 table 5-4: number of bits per physical pixel according to physical color space. physical color space code 1bpp rgb111 4bpp rgb333 rgb444 rgb666 rgb888 bpp phys 1 3 4 9 12 18 24 max t seg proc , () t seg out , t seg out , t displayclock seglen streamfactor ? ? () = streamfactor physsizex pxscandots ? () scanfactor ? = bpsd bpp phys ? () scanfactor ? = scanfactor 1 for scanmode dualscan 2 for scanmode = dualscan ? ? ? = MB87J2120, mb87p2020-a hardware manual page 200 b) maximum time to process a segment: (22) to obtain the value maxsegpixel the segment with the maximum number of pixels processed has to be found. this is achieved following the algorithm shown in figure 5-1. as a result of this the minimal frequency ratio of core clock to display clock is derived as: (23) 5.2 memory bandwidth although memory access is more unlikely to become the bottleneck in contrast to processing it may turn out as limiting factor when memory traffic of other gdc units (e.g. pixel processor or direct access) in- creases. therefore, estimations for memory bandwidth are given here. max t seg proc , () t coreclock maxsegpixel ? = maxsegpixel := 0 for y := 0 to (physsizey-1) do begin act_layers:= {} foreach l in vis_layers do begin if y >= offsy(l) and y < (offsy(l)+wy(l)) then act_layers := {act_layers, l} end if act_layers == {} then continue segm_start := 0 while (segm_start < physsizex) do begin segm_stop := min((segm_start + seglen), physsizex) seg_pixels := 0 foreach l in act_layers do begin if ox(l) >= segm_stop or (ox(l) + wx(l)) < segm_start then continue if ox(l) < segm_start then layer_start := segm_start else layer_start := ox(l) if (ox(l) + wx(l)) >= segm_stop then layer_stop := segm_stop else layer_stop := ox(l) + wx(l) seg_pixels := seg_pixels + (layer_stop -layer_start) end if logical_color_space(l) == yuv422 then seg_pixels := seg_pixels + 2 if background_enabled then seg_pixels := seg_pixels + (segm_stop - segm_start) if seg_pixels > maxsegpixel then maxsegpixel := seg_pixels end figure 5-1: algorithm to determine the maximum number of processed pixels per line segment. f coreclock f displayclock -------------------------- t displayclock t coreclock -------------------------- - = maxsegpixel seglen ------------------------------- scanfactor ? graphic processing unit bandwidth considerations page 201 5.2.1 average bandwidth this estimation is done on the basis of the visible 1 layer area, i.e. the area of active layers within the geo- metrical borders of the physical display, since all pixels displayed had to be fetched. the mean number of words (which is equal to the number of core clock cycles) transferred from memory is calculated as follows: (24) the number of pixels per videoram word was already given in table 5-1. for correct gdc function con- dition (25) must hold: (25) 5.2.2 peak bandwidth as already mentioned in section 5.1.2, gpu processes data in portions of a line segment. this implies that the segment with the maximum number of words fetched has to be found (it should be noted that this is not necessarily the segment with the maximum number of pixels). the algorithm used to find this segment is 1. the area counts also as visible if a layer is (partially) covered by another layer. w gpu area l 1 ppw log l , ------------------ ? ?? ?? visible layers = t frame t coreclock w gpu w other units + () ? MB87J2120, mb87p2020-a hardware manual page 202 shown in figure 5-2. it is quite similar to that in figure 5-1. segment length is determined using equation (17). with this result, condition (25) is refined to (26) where t seg, out is the time to output one segment as calculated in equation (19), and w peak, gpu is simply maxsegwords . 5.3 recommendations in order to ensure correct gpu function it is important to estimate the necessary bandwidth for pixel processing and output. however, one has to bear in mind that not all influences are known in advance, es- pecially the behaviour of other units the gpu shares the sdc and videoram with. therefore, one should provide for sufficient safety range when determining gdc setup. there are some issues that can help save memory and processing bandwidth. following the recommenda- tions given below it may be possible to operate at lower core clock / display clock ratios, thus improving emc and reducing power consumption. when the whole display area is covered by non-transparent layer pixels background painting can be switched off. multiple layers do not increase bandwidth when they do not cover each other. maxsegwords := 0 for y := 0 to (physsizey-1) do begin act_layers:= {} foreach l in vis_layers do begin if y >= offsy(l) and y < (offsy(l)+wy(l)) then act_layers := {act_layers, l} end if act_layers == {} then continue segm_start := 0 while (segm_start < physsizex) do begin segm_stop := min((segm_start + seglen), physsizex) seg_words := 0 foreach l in act_layers do begin if ox(l) >= segm_stop or (ox(l) + wx(l)) < segm_start then continue if ox(l) < segm_start then layer_start := segm_start else layer_start := ox(l) if (ox(l) + wx(l)) >= segm_stop then layer_stop := segm_stop else layer_stop := ox(l) + wx(l) if logical_color_space(l) == yuv422 then layer_stop := layer_stop + 1 seg_words := seg_words + truncate((layer_stop - layer_start + ppw log (l)-1) / ppw log (l)) end if seg_words > maxsegwords then maxsegwords := seg_words end figure 5-2: algorithm to determine the maximum number of words fetched per segment. t seg out , t coreclock w peak gpu , w p eak other units , + () ? graphic processing unit bandwidth considerations page 203 large transparent areas should be avoided, instead the layer?s window size should be limited to the smallest rectangle including all non-transparent pixels. color resolution (logical color space) should be reduced to the minimal number of colors needed simultaneously, a more colourful display can be achieved with useful clut entries as well. this approach decreases the necessary memory bandwidth. longer blanking periods decrease average bandwidth demands. for a given size and display clock dual scan displays demand a higher memory and processing band- width. displays with greater streamfactors according to equation (20) demand higher bandwidth. video display should be used with care, since it produces high memory traf?c (from vic to ram and from ram to gpu). vic provides some means to reduce to necessary bandwidth (dropping cer- tain frames and ?elds). MB87J2120, mb87p2020-a hardware manual page 204 6 functional peculiarities 6.1 con?uration constraints due to the great flexibility provided by the extensive programmability there is no built-in precaution against illegal, not supported or absurd register settings . programming has to be carried out with a thor- ough understanding of the features explained in this document. items to pay attention to include (but are not limited to): window size does not exceed domain size in every dimension window size greater than zero first pixel lies within window size offset within physical display dimensions clut offset may cause wrap-around for look-up color space settings con?rming to tables 2-1 and 2-2, chapter 2 non-transparent layers covering others layers multiply exposed (i.e. identical settings for z-order register entries, this is a waste of memory and processing bandwidth) disabled background will lead to garbage pixels in areas not covered by non-transparent layer con- tents values for master timing x stop must be greater than or equal to pxscandots , master timing y stop values must be greater than or equal to physsizey , otherwise internal fsms will hang sync pulse generator and sync sequencer entries should lie within timing range determined by the master timing registers combination of physical color space and bit stream format con?rming to table 3-4, chapter 3 all necessary output pins enabled for connected physical display color key limits according to physical color space register set entries that in?uence bit stream timing (master timing values, physical color space code, bit stream format code, display dimensions etc.) are locked with mastertimingon to prevent internal fsms from hanging. these values may only be set or changed when mastertimingon is reset. these registers are marked in table 4-1. 6.2 bandwidth bandwidth is an important issue when it comes to usage at performance limits. unfortunately, it is hard to estimate in advance whether certain settings will cause constraint violations, since other gdc-units, the gpu shares the videoram with, have an in?uence hard to predict. the rules given in chapter 5 should help checking for possible bottlenecks. as a support for software development gpu can trigger an interrupt when it suffers from critical data shortage. the interrupt is actually raised when the lsa runs empty (an empty dfu input fifo does not constitute a data shortage as such, since this state can only be temporary and recoverable). in the graphic processing unit functional peculiarities page 205 interrupt case the gpu ceases data fetching for the remaining frame period and triggers the ulb ?ag bwvio (bit 18 of ulb ?ag word). however, display output continues with blank pixel data to pre- vent possible damage at the connected display. this state is left either by soft-reset (i.e. switching mastertimingon off and back on again) or at the begin of the next frame period. if bandwidth is still too small the interrupt will be raised again. the priority with which gpu requests data from sdc can be controlled with the settings in the gpu_sdcprio register (bits [6:4] and [2:0] 1 , address 0x3270). per default the gpu has the highest priority amongst all gdc units. any changes of this value should be done with utmost care , since a continuous display timing requires a continuous data stream from ram be maintained. 6.3 image processing there are some points to be aware of when processing image data in yuv color space: matrix coef?cients cbxg and crxg are treated as having negative sign yuv555 and yuv655 color spaces are auto scaled to yuv444 before matrix multiplication processing yuv422 data restricts horizontal ?rst pixel, window size and offset to even values there is only either a mapping through matrix or achromatic mapping from yuv color spaces to intermediate and eventually to physical color space. therefore, if a direct mapping from yuv to gpu pins is desired a special approach is necessary. it consists of a re-labelling of layers containing yuv data to corresponding rgb color spaces with an equal number of bits per pixel (table 6-1) and then directly mapping these to the physical color space and thus to the pins (cf. table 3-14). note also that yuv422 uses chrominance multiplexing. image quality for pictures in yuv422 may be improved by using chrominance interpolation. to improve image appearance for video pictures grabbed by vic, gpu offers line doubling . with this feature (which can be switched on individually for each layer) vertical resolution is reduced to the half by displaying each line in videoram twice on the display (e.g. line #0 from ram is dis- played on display lines #0 and #1, ram line #1 on display lines #2 and #3 and so on). this approach eliminates inter-?eld jitter. when used in conjunction with ?eld dropping in vic it additionally reduces memory bandwidth. pixels of layers in color space rgb555, rgb565 and rgb888 may be fed through the gamma tables to apply further color processing. this mode is enabled when both ?ags, gamma enable and rgb gamma forcing, are set. similar arbitrary color processing is achieved for color spaces with less than 16 bits per pixel by simply using the clut. 1. bits [2:0] for low priority are not used in the current implementation, i.e. gpu always uses high priority requests. table 6-1: compatibility list for yuv to rgb re-labelling. yuv spaces corresponding rgb spaces yuv422 rgb565 yuv444 rgb888 yuv555 rgb555 yuv655 rgb565 MB87J2120, mb87p2020-a hardware manual page 206 6.4 data output gpu pixel output is always lsb-aligned, i.e. the geometrically leftmost pixel is represented in the least signi?cant bit(s). for cases where displays demand msb-alignment this has to be produced by appropriate wiring on the pcb. there are two exceptional cases: physical color space rgb111 and bit stream format s4 or s8. then fractions of a pixel are output and the alignment swap can not be obtained by wiring only. therefore, gpu provides the ?ag ?r/b-swap? which exchanges the red with the blue color channel (exchange of a pixel?s bit 0 and 2). since gpu is designed for lcds and equivalent progressive scan 1 display types it can not produce tv-conform interlaced output streams. however, it can produce tv-conform timing signals (hsync and vsync, cf. section 3.10) for displays which mimic tv behaviour. for bit stream formats other than s24 it is possible to control the values at the pins dis_d[23:19] directly by the user (aka port expansion mode). this is achieved by setting the respective sync mixer signal selects all to constant zero. then bit 0 of the mixer?s function table represents the pin value directly. this direct control works for all bit stream formats on pins dis_hsync, dis_vsync, and dis_vref, since these are always available. it is possible to run both, an analog and a digital display, concurrently with the same timing (only valid for mb87p2020-a (jasmine)). this is accomplished using the twin display mode (cf. 3.5). however, this works only for single scan displays and pins dis_d[23:19] may not be available for sync signal output then. 6.5 diagnostics gpu provides some diagnostic information in read-only registers that might help during software develop- ment. the following data are available: current blink state (bits[31:16], register mdr_blinkctrl, address 0x1308). a ?1? at a bit slice desig- nates that the corresponding layer would use its alternative blink color. the information may be used to synchronize different layers or to synchronize drawing and display. a simultaneous start of blink- ing for all layers can be achieved by simultaneous setting of the blink enable bits (bits [15:0] of mdr_blinkctrl). current timing position (register dir_timingdiag, address 0x3024) in ?timing coordinates? as explained in section 3.9. the information is useful to check for correct register settings. current input fifo load (bits [15:8], register gpu_sdcprio, address 0x3270). this information allows to check for potential bandwidth bottlenecks. during normal operation the fifo will be full ( ) most of the time. however, it does not constitute a critical state as such when it runs empty. only if there is not enough memory bandwidth left to re?ll it for longer periods of time the lsa will consequently run empty and raise an interrupt. 1. the term progressive scan means that a complete frame is output line by line, whilst interlaced means that a complete frame is output as two ?lds , where one ?eld contains all lines with odd and the other ?eld all lines with even numbers. load 130 graphic processing unit supported displays page 207 7 supported displays the tables in the following sections list the result of a display survey done to derive necessary and useful gpu features. they are divided in sections for passive matrix lcd, active matrix (tft) lcd, electrolu- minescent displays and field emission displays. some potential problems for support encountered during this survey are explained in section 7.5. the tables in sections 7.1 to 7.4 list only those analog inputs necessary for normal display operation. some of the displays feature extra inputs e.g. for dimming. these pins usually require potentiometers (or current d/a converters). the type of digital interface signals is abbreviated as follows: c5 denotes 5v cmos level, c3 denotes 3.3v cmos level, and t5 denotes 5v ttl level. color resolution is abbreviated this way: m denotes monochrome, c denotes low color resolution (less than 24 bits per pixel), and t denotes true color (16 million colors). MB87J2120, mb87p2020-a hardware manual page 208 7.1 passive matrix lcd none of the passive matrix displays includes an inverter to produce the high voltage needed for the ccfl backlight. vendor type res. x x y bits per pixel colors scan # of pixels written at once digital i/f signals max. pixclk (mhz) framerate (hz) power supply (v) # type min max optrex dmf-50970nc 160 x 113 3 c single a a. special pixel order: pixels 1?.160,y and 1?.160, y+58 are written as one line 2 2/3 13 c3/c5 b b. derived from used column driver ics 10?20 b ??? sharp lm32p10 320 x 240 1 m single 4 7 c5 6.5 70 80 5; -20 sharp lm64p89 640 x 480 1 m dual parallel 2 x 4 12 c5 6.5 59 125 5; -19 sharp lm64c142 640 x 480 3 c dual parallel 2 x (2 2/3) 20 t5 12.2 59 120 5; 27 sanyo dg24320c5pc 320 x 240 3 c dual parallel 2 x (1 1/3) 2 x (2 2/3) 21 c5 12.2 73 73 5; 26 graphic processing unit supported displays page 209 7.2 active matrix (tft) displays all active matrix displays work in single scan mode. vendor type res. x x y bits per pixel colors # of pixels written at once i/f signals max. pixclk (mhz) framerate (hz) power supply (v) ccfl inv. incl. analog # digital min max # type fpd lde052t-02 320 x 240 12 t 1 - 20 c5 26.8 55 65 5; 12 no fpd lde052t-12 320 x 240 12 t 1 - 20 c5 26.8 55 65 5; 12 yes fpd lde052t-32 320 x 240 24 t 1 3 8 c5 26.8 55 65 5; 12 yes fpd lde052t-52 320 x 240 24 t 1 3 8 c5 26.8 55 65 5; 12 no sharp lq5aw116 320 x 234 24 t 1 3 9 c5 7.6 55 65 8; -5 no hosiden hld0909 640 x 480 3 m 2 - 11 c5 13.3 50 70 5 no nec nl3224ac35-01 320 x 240 24 t 1 3 9 c5 8.4 48 52 9.5 yes nec nl6448ac33-18 640 x 480 18 c 1 - 22 c3/c5 29 58 62 3.3/5; 12 yes nec nl6448ac20-06 640 x 480 12 / 18 c 1 - 17 / 23 c5 29 58 62 5 yes sanyo alp401rdd 320 x 240 24 t 1 3 5 t5 20.2 51 71 12 yes sanyo alp401rdv 320 x 240 24 t 1 - 29 t5 6.6 56 63 12 no MB87J2120, mb87p2020-a hardware manual page 210 7.3 electroluminescent displays vendor type res. x x y bits per pixel colors # of pixels written at once digital i/f signals max. pixclk (mhz) framerate (hz) power sup- ply (v) # type min max sharp lj32h028 320 x 240 1 m 4 7 c5 6,6 60 120 5; 12 planar el160.80.50 160 x 80 1 m 4 8 c5 7.1 0 240 5; 12 planar el240.64 / el240.64-sd 240 x 64 1 m 1 6 c5 4.0 0 240 (5) a ; 12 planar el4737hb / el4737hb-ice 320 x 128 1 m 1 5 t5 5.1 0 120 (5); 12 planar el320.240.36 320 x 240 1 m 4 9 c5 7.1 0 120 (5); 12 planar el320.256-f6 el320.256-fd6 320 x 256 1 m 1 / 2 9 t5 25 75 5; 11?.30 b planar el320.256-fd7 320 x 256 1 m 1 / 2 8 t5 25 120 5; 11?.30 planar el512.256-h 512 x 256 1 m 1 / 2 7 t5 30 75 5; 11?.30 planar el640.400-cb1/ cd4 640 x 400 1 m 1 / 2 6 t5 30 70 5; 12 planar el640.400-cb3 640 x 400 1 m 1 / 2 6 t5 30 70 5; 24 planar el640.400-c2, - c3, -c4 640 x 400 1 m 1 / 2 7 t5 30 70 5; 11?.30 planar el640.400-cex 640 x 400 1 m 1 / 2 x 4 16 c5 6.6 160 5; 12 graphic processing unit supported displays page 211 except the planar el640.400-cex, which has an optional dual scan mode, all electroluminescent displays work in single scan mode . 7.4 field emission displays both field emission displays work in single scan mode. planar el640.480-af1, - ag1 640 x 480 1 m 2 x 4 11 c5 6.5 120 5; 12 planar el640.480-am8 640 x 480 1 m 2 x 4 11 c5 6.5 120 5; 24 planar el640.480-aa1 640 x 480 4 c 1 10 c5 30 80 5; 12 planar el640.480-asb 640 x 480 4 m 1 / 2 16 t5 30 70 12 a. optional b. arbitrary within range vendor type res. x x y bits per pixel colors # of pixels written at once digital i/f signals max. pixclk (mhz) framerate (hz) power sup- ply (v) # type min max motorola d07spd310 128 x 160 9 c 1 14 c3 1.6 60 75 3.3; 7.2 pixtech fe532s-m1 320 x 240 1 m 4 9 t5 6.0 50 350 5; 12 vendor type res. x x y bits per pixel colors # of pixels written at once digital i/f signals max. pixclk (mhz) framerate (hz) power sup- ply (v) # type min max MB87J2120, mb87p2020-a hardware manual page 212 7.5 limitations for support 7.5.1 no support due to vcom inversion there were another two displays in the survey, the fujitsu flc15adcaw and the sharp lq7bw506, which suffer from a rather ?substrate-oriented? interface. these displays do not have the necessary circuitry to prevent dc from the liquid crystal. therefore, they require rgb input voltage and common electrode voltage (v com ) be reversed every line. this inversion works as follows: for one line a voltage of e.g. -1.9v is applied to the common electrode. a voltage of 0.6v on the rgb inputs produces then a white, a voltage of 4.6v a black display. the next line, the common electrode is clamped to 5.9v. now, a voltage of 0.6v on rgb produces black and a voltage of 4.6v produces white. in other words: common electrode voltage is alternated between -1.9 and 5.9v, while the voltage difference for rgb remains at 4v but with alternating result (color) for every line. this causes the following problems: there is no way to produce the voltages mentioned above on-chip within gdc, since they exceed the limits of supply voltage by far. therefore, additional external circuitry is indispensable (sharp rec- ommend their dedicated video signal polarity reversing ic ir3y29a). as the spec sheets state, the voltages actually used need to be ?ne-tuned for every individual display to obtain optimum contrast (i.e. v com dc component (mean level), v com ac component (voltage difference), as well as rgb dc and ac components). this implies external potentiometers to pro- duce the bias voltages. 7.5.2 problems due to 5v cmos interfaces there is a number of displays, which demand 5v cmos levels for their digital interface signals (marked with ?c5? in the respective column). this is problematic since the gdc-asic is not able to guarantee the correct ?high?-level at its outputs since the asic runs at 3.3v. work-around: either external pull-up resistors or special level shifter ics are used as e.g. the philips 74alvc164245 dual supply translating transceiver. page 213 b-9 cold cathode fluorescence light driver (ccfl) MB87J2120, mb87p2020-a hardware manual page 214 ccfl driver introduction page 215 1 introduction the cold cathode fluorescence light driver is used to generate the needed signals for controlling a switched power supply and ionisation circuit of a cold cathode lamp. the intention was that brightness of the lamp should be adjustable in a wide range. following figure shows a block diagram of the circuit driving the cold cathode lamp which can be used for generating high voltage supplying the displays back light. in general ccfl is independent from other gdc components. it is initialized via the control bus port (cbp) interface which is used to setup almost all gdc control registers. after initialization ccfl works as stand-alone circuit. to avoid interferences between back light flashing frequency and picture refresh rate of the display a syn- chronization input is provided which internally is connected with the video frame synchronization signal of the graphic processing unit (gpu). another synchronization method is possible by the connected mcu over the control interface when writing to the synchronization flag of the control register. mb87p2020-a has a connection to gpu sync mixer 5 as third synchronization possibility. because a sync mixer can match many display coordinates within one gpu frame (see gpu description for details about sync signal generation) a higher ccfl frequency compared to simple frame synchronization can be achieved without losing synchronization between ccfl and display output. in short terms the ccfl has the following synchronization possibilities: internal gpu frame synchronization software synchronization internal sync mixer synchronization (mb87p2020-a only) simply spoken the ccfl circuit is a kind of pulse width modulation of lightning duration of the fluores- cence lamp within one frame period of the display. if the lamp is switched on ccfl provides bi-phase pulse signals (pins ccfl_fet1, ccfl_fet2) for direct controlling the complementary fet gates of primary inductance of the voltage transverter. additional to controlling the lightning and off durations the ignition of the lamp is handled (pins: ccfl_off, ccfl_ignit). ccfl ctrl gpu ignit fet1 fet2 off video cbp sync display inverter supply gdc figure 1-1: application overview for the ccfl circuit MB87J2120, mb87p2020-a hardware manual page 216 if a simple pwm output is needed (for ccfl supplies without direct fet connection or different voltage converter implementations) a low active pwm signal for brightness control is provided by the ccfl_off pin. ccfl driver signal waveform page 217 2 signal waveform 2.1 general description the following figure shows a waveform example for the external signals and the synchronisation signal. one cycle consists of five phases, a start-up phase, ionisation phase, pause, flash phase and the remaining time until the rising edge of the synchronisation signal. the duration of the start-up phase is fixed, the duration of the following three phases is determined by the values of the registers ignite, pause and flash. normally ignite and pause values are fixed setup to be optimal for the lamp characteristics. brightness modulation is done by changing the duration of the flash phase. table 2-1 describes the values of the external signals in each phase. while fet1 and fet2 directly control- ling the voltage converter, ignit and off are for controlling the height of input voltage or to disable it completely. 2.2 duration of the phases the duration of the certain phases is determined by the value of the prescaler register scale and the values of the registers ignite, pause and flash. the value of the prescaler register defines a base period, which is used to derive pulse shapes and duration of the phases. t bper = t clk scale table 2-1: phases of one ccfl cycle signal value start-up phase ionisation phase pause flash phase remaining time fet1/fet2 low pulsed a a.see section 2.3 low pulsed a low ignit high high low low low off low low low low high fet2 ignite ignition phase start ? up phase off time fet1 off flash phase sync 1 sync period (normally display frame) pause fixed setup flash duration modulation figure 2-1: principle waveform of output signals relative to sync MB87J2120, mb87p2020-a hardware manual page 218 the duration of the appropriate phases calculates as follows: 15. start-up phase:t start = t bper 8 16. ionization phase:t ion = t bper 4 ignite 17. pause:t pause = t bper 4 pause 18. flash phase:t ?ash = t bper 4 flash the fifth phase equals to the remaining off time until the next rising edge of the synchronisation signal note: if flash duration exceeds display frame duration the next synchronization will be dropped and lamp is off the next period after previous flash duration is ended. this could be result in flickering or malfunc- tion of the lamp. 2.3 pulse shape of fet1 and fet2 the following figure shows the pulse shape of pulses, which have to be generated during ionisation and flash phase. if the end of a phase is reached during a high period of a pulse, a shortened pulse has to be generated as shown in the following figure. fet1 fet2 1 cycle 7t 7t bper bper 1t bper 1t bper figure 2-2: timing definition of fet control signals fet1 fet2 bper 4t 1t bper 3/4 cycle 7t bper figure 2-3: timing definition of fet control signals for fractured number of cycles ccfl driver register description page 219 3 register description 3.1 overview 3.2 control bits for mb87p2020-a an additional control bit for selecting sync mixer synchronization was necessary. note that this new control bit (ccfl1_ssel) takes precedence over sncs flag. for other devices than mb87p2020-a writing to this control bit is ignored while reading returns always ?0?. table 3-1: control register ccfl1 name description position reset value access ccfl1_ssel a a. mb87p2020-a only sync select 1 [28] 0 r/w ccfl1_en ccfl enable [27] 0 r/w ccfl1_prot protect settings [26] 0 r/w ccfl1_sncs sync select 0 [25] 0 r/w ccfl1_sync sync software trigger [24] 0 r/w - reserved [23:8] ccfl1_scl prescaler register [7:0] 0 r/w table 3-2: duration register ccfl2 name description position reset value access ccfl2_fls flash duration [31:16] 0 r/w ccfl2_pse pause duration [15:8] 0 r/w ccfl2_ignt ionisation duration (ignition) [7:0] 0 r/w table 3-3: control bit description bit name description 28 a ccfl1_ssel syncsel. if this bit is not set (0), synchronization is controlled by sncs. otherwise (1) the output of gpu sync mixer 5 is used for synchro- nization. 27 ccfl1_en cocaden. if this bit is set, the cold cathode driver is enabled. other- wise the pins have its inactive state (fet1/2=0, ignit=0, off=1). 26 ccfl1_prot protect. if this bit is set, write access to the duration register changes the value of the register, but the durations of the phases are not in?uenced until this bit will be cleared. if this bit is not set, a write access changes the durations immediately. MB87J2120, mb87p2020-a hardware manual page 220 25 ccfl1_sncs syncsel. if this bit is not set (0), vsync display synchronisation is enabled at each ?rst active line. otherwise (1) synchronisation by the extysync ?ag (software) will be used. 24 ccfl1_sync extsync. setting these bit is interpreted as synchronization pulse if syncsel=1. the reset has to be done by software too. a. mb87p2020-a only table 3-3: control bit description bit name description ccfl driver application notes page 221 4 application notes 4.1 ccfl setup example the time base of ccfl driver should be near 800 ns. assumed we have a system clock frequency clkk of 60 mhz (16.6 ns). t bper = 800 ns ccfl1_scl = t clkk / t bper = 800 ns / 16.6 ns = 48 = 0x30 we have to setup a prescaler value of 0x30 in the scale register. possible values for ignition duration and pause can be 0x10 and 0x80 for instance. this depends mainly on the lamp characteristics and is not too critical for setting up the right value. if we have a frame rate of 60 hz so we get nearly 1 / 60hz / 800ns = 20833 t bper = 0x5161 t bper cycles. due to the fact that setup duration values are interpreted as macro cycles of 4 times t bper one period has a duration of 0x5161 t bper cycles / 4 = 0x1458 macro cycles. after subtraction of 0x10 ignition time and 0x80 pause duration and 1 for the start-up phase we got the re- sulting dimming range of 0x0000 to 0x13c7 macro cycles from dark to full luminescence. be careful not to setup a larger value for flash duration, then the lamp ignition may happen each other frame only and is flickering then. 4.2 ccfl protection during initialization of setup values there is no need to switch off the ccfl driver. to avoid inconsistency of configuration data a protection mechanism can be used. this method offers a way that no forbidden out- put timing is generated and the lamp or the supply circuitry would not be damaged. to do a secure config- uration following sequence should be used: 1. switch ccfl_prot to ?1? 2. con?gure ccfl2 duration registers. 3. release protection by setting ccfl_prot = ?0? this gives the possibility to modulate the durations without switching the lamp off. MB87J2120, mb87p2020-a hardware manual page 222 page 223 part c - pinning and electrical speci?ation MB87J2120, mb87p2020-a hardware manual page 224 pinning for lavender and jasmine pinning and buffer types page 225 1 pinning and buffer types 1.1 pinning for mb87p2020-a 1.1.1 pinning table 1-1 shows the pinning for mb87p2020-a (redesigned jasmine) sorted by pin number while table 1- 2 is sorted by names. both tables refer to a standard fujitsu 208-pin (l)qfp package fpt-208p-mxx. table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description 1 mode[0] bfnnqlx mode pin 2 mode[1] bfnnqlx mode pin 3 gnd gnd 4 place holder for analog area 5 a_green otamx analog green 6 dac2_vssa1 itavsx dac ground 7 dac2_vdda1 itavdx dac supply 2.5v 8 a_blue otamx analog blue 9 dac1_vdda1 itavdx dac supply 2.5v 10 dac1_vssa1 itavsx dac ground 11 a_vro otamx dac full scale adjust 12 dac1_vssa itavsx dac ground 13 dac1_vdda itavdx dac supply 2.5v 14 a_red otamx analog red 15 dac3_vdda1 itavdx dac supply 2.5v 16 dac3_vssa1 itavsx dac ground 17 vref itamx dac test pin vref 18 place holder for analog area 19 vddi vddi# core supply 2.5 v 20 test itchx fujitsu test pin 21 vsc_clkv itfhx video scaler clock 22 resetx itfhx reset 23 osc_out bx4mfsr2x xtal output 24 vdde[0] io supply 3.3v 25 osc_in ixn3x xtal input MB87J2120, mb87p2020-a hardware manual page 226 26 gnd gnd 27 vddi vddi# core supply 2.5 v 28 rdy_trien b3nnlmr2x control rdy pin behaviour 29 apll_avdd apll supply 2.5v 30 apll_avss apll gnd 31 rclk itfhx reserved clock 32 ulb_a[20] bfnnqlx ulb interface address 33 ulb_a[19] bfnnqlx ulb interface address 34 vddi vddi# core supply 2.5 v 35 ulb_a[18] bfnnqlx ulb interface address 36 ulb_a[17] bfnnqlx ulb interface address 37 ulb_a[16] bfnnqlx ulb interface address 38 gnd gnd gnd 39 ulb_a[15] bfnnqlx ulb interface address 40 ulb_a[14] bfnnqlx ulb interface address 41 ulb_a[13] bfnnqlx ulb interface address 42 ulb_a[12] bfnnqlx ulb interface address 43 vdde vdde# io supply 3.3v 44 ulb_clk itfhx ulb interface clock 45 ulb_a[11] bfnnqlx ulb interface address 46 ulb_a[10] bfnnqlx ulb interface address 47 ulb_a[9] bfnnqlx ulb interface address 48 ulb_a[8] bfnnqlx ulb interface address 49 ulb_a[7] bfnnqlx ulb interface address 50 gnd gnd gnd 51 ulb_a[6] bfnnqlx ulb interface address 52 ulb_a[5] bfnnqlx ulb interface address 53 ulb_a[4] bfnnqlx ulb interface address 54 ulb_a[3] bfnnqlx ulb interface address 55 ulb_a[2] bfnnqlx ulb interface address 56 ulb_a[1] bfnnqlx ulb interface address 57 ulb_a[0] bfnnqlx ulb interface address 58 ulb_cs bfnnqlx ulb interface chip select table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 227 59 ulb_rdx bfnnqlx ulb interface read 60 gnd gnd gnd 61 vdde vdde# io supply 3.3v 62 ulb_dack bfnnqlx ulb interface dma acknowledge 63 ulb_d[31] bfnnqmx ulb interface data 64 ulb_d[30] bfnnqmx ulb interface data 65 ulb_d[29] bfnnqmx ulb interface data 66 gnd[1] gnd 67 vdde[1] io supply 3.3v 68 ulb_d[28] bfnnqmx ulb interface data 69 ulb_d[27] bfnnqmx ulb interface data 70 ulb_d[26] bfnnqmx ulb interface data 71 ulb_d[25] bfnnqmx ulb interface data 72 gnd gnd gnd 73 vddi vddi# core supply 2.5 v 74 vdde[2] io supply 3.3v 75 ulb_d[24] bfnnqmx ulb interface data 76 ulb_d[23] bfnnqmx ulb interface data 77 ulb_d[22] bfnnqmx ulb interface data 78 ulb_d[21] bfnnqmx ulb interface data 79 vddi vddi# core supply 2.5 v 80 gnd[2] gnd 81 ulb_d[20] bfnnqmx ulb interface data 82 ulb_d[19] bfnnqmx ulb interface data 83 ulb_d[18] bfnnqmx ulb interface data 84 ulb_d[17] bfnnqmx ulb interface data 85 gnd gnd gnd 86 vddi vddi# core supply 2.5 v 87 ulb_d[16] bfnnqmx ulb interface data 88 ulb_d[15] bfnnqmx ulb interface data 89 ulb_d[14] bfnnqmx ulb interface data 90 ulb_d[13] bfnnqmx ulb interface data 91 gnd[3] gnd table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 228 92 vdde[3] io supply 3.3v 93 ulb_d[12] bfnnqmx ulb interface data 94 ulb_d[11] bfnnqmx ulb interface data 95 ulb_d[10] bfnnqmx ulb interface data 96 gnd gnd gnd 97 vdde vdde# io supply 3.3v 98 ulb_d[9] bfnnqmx ulb interface data 99 ulb_d[8] bfnnqmx ulb interface data 100 ulb_d[7] bfnnqmx ulb interface data 101 ulb_d[6] bfnnqmx ulb interface data 102 vdde[4] io supply 3.3v 103 gnd[4] gnd 104 ulb_d[5] bfnnqmx ulb interface data 105 ulb_d[4] bfnnqmx ulb interface data 106 ulb_d[3] bfnnqmx ulb interface data 107 gnd gnd gnd 108 ulb_d[2] bfnnqmx ulb interface data 109 vdde[5] io supply 3.3v 110 sdram_vcc[0] sdram supply 2.5v 111 ulb_d[1] bfnnqmx ulb interface data 112 ulb_d[0] bfnnqmx ulb interface data 113 gnd[5] gnd 114 vdde vdde# io supply 3.3v 115 sdram_vcc[1] sdram supply 2.5v 116 ulb_rdy otftqmx ulb interface ready 117 ulb_dstp bfnnqmx ulb interface dma stop 118 sdram_vcc[2] sdram supply 2.5v 119 gnd gnd gnd 120 ulb_dreq otftqmx ulb interface dma request 121 ulb_intrq otftqmx ulb interface interrupt request 122 ulb_wrx[0] itfhx ulb interface write enable (d[31:24]) 123 vddi vddi# core supply 2.5 v 124 ulb_wrx[1] itfhx ulb interface write enable (d[23:16]) table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 229 125 sdram_vcc[3] sdram supply 2.5v 126 ulb_wrx[2] itfhx ulb interface write enable (d[15:8]) 127 ulb_wrx[3] itfhx ulb interface write enable (d[7:0]) 128 sdram_pbi ipbix sdram test mode 129 vdde[7] io supply 3.3v 130 gnd gnd gnd 131 vddi vddi# core supply 2.5 v 132 sdram_tbst itbstx sdram test mode 133 sdram_ttst ittstx sdram test mode 134 vpd vpdx fujitsu tester pin 135 dis_d[0] b3nnlmr2x display data 136 dis_d[1] b3nnlmr2x display data 137 dis_d[2] b3nnlmr2x display data 138 vddi vddi# core supply 2.5 v 139 dis_d[3] b3nnlmr2x display data 140 dis_d[4] b3nnlmr2x display data 141 dis_d[5] b3nnlmr2x display data 142 gnd gnd gnd 143 dis_d[6] b3nnlmr2x display data 144 dis_d[7] b3nnlmr2x display data 145 dis_d[8] b3nnlmr2x display data 146 dis_d[9] b3nnlmr2x display data 147 vdde vdde# io supply 3.3v 148 dis_d[10] b3nnlmr2x display data 149 dis_d[11] b3nnlmr2x display data 150 dis_d[12] b3nnlmr2x display data 151 dis_d[13] b3nnlmr2x display data 152 dis_d[14] b3nnlmr2x display data 153 dis_d[15] b3nnlmr2x display data 154 gnd gnd gnd 155 vdde[8] io supply 3.3v 156 dis_d[16] b3nnlmr2x display data 157 dis_d[17] b3nnlmr2x display data table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 230 158 dis_d[18] b3nnlmr2x display data 159 dis_d[19] b3nnlmr2x display data 160 dis_d[20] b3nnlmr2x display data 161 dis_d[21] b3nnlmr2x display data 162 dis_d[22] b3nnlmr2x display data 163 dis_d[23] b3nnlmr2x display data 164 gnd gnd gnd 165 vdde vdde# io supply 3.3v 166 dis_ckey b3nnlmr2x display colour key 167 dis_pixclk b3nnnmr2x display pixel clock (programmable in/out) 168 dis_vsync b3nnlmr2x display programmable sync 169 dis_hsync b3nnlmr2x display programmable sync 170 dis_vref b3nnlmr2x display programmable sync 171 vsc_d[0] bfnnqlx video scaler data input 172 vsc_d[1] bfnnqlx video scaler data input 173 vsc_d[2] bfnnqlx video scaler data input 174 vsc_d[3] bfnnqlx video scaler data input 175 vsc_d[4] bfnnqlx video scaler data input 176 gnd gnd gnd 177 vddi vddi# core supply 2.5 v 178 vdde[9] io supply 3.3v 179 mode[2] bfnnqlx mode pin 180 mode[3] bfnnqlx mode pin 181 vsc_d[5] bfnnqlx video scaler data input 182 vsc_d[6] bfnnqlx video scaler data input 183 vddi vddi# core supply 2.5 v 184 vsc_d[7] bfnnqlx video scaler data input 185 vsc_d[8] bfnnqlx video scaler data input 186 vsc_d[9] bfnnqlx video scaler data input 187 vsc_d[10] bfnnqlx video scaler data input 188 vsc_d[11] bfnnqlx video scaler data input 189 gnd gnd gnd 190 vddi vddi# core supply 2.5 v table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 231 191 vsc_d[12] bfnnqlx video scaler data input 192 vsc_d[13] bfnnqlx video scaler data input 193 vsc_d[14] bfnnqlx video scaler data input 194 vsc_d[15] bfnnqlx video scaler data input 195 vsc_vref bfnnqlx video scaler vertical reference 196 vsc_vact bfnnqlx video scaler vact 197 vsc_alpha bfnnqlx video scaler alpha 198 vsc_ident bfnnqlx video scaler field identi?cation 199 spb_bus bfnnqhx spb interface 200 gnd gnd gnd 201 vdde vdde# io supply 3.3v 202 spb_tst bfnnqhx spb test 203 ccfl_fet2 otftqmx ccfl fet driver 204 ccfl_fet1 otftqmx ccfl fet driver 205 ccfl_ignit otftqmx ccfl supply control ignition 206 gnd[6] gnd 207 vdde[10] io supply 3.3v 208 ccfl_off otftqmx ccfl supply control off table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description 4 place holder for analog area 18 place holder for analog area 8 a_blue otamx analog blue 5 a_green otamx analog green 14 a_red otamx analog red 11 a_vro otamx dac full scale adjust 29 apll_avdd apll supply 2.5v 30 apll_avss apll gnd 204 ccfl_fet1 otftqmx ccfl fet driver 203 ccfl_fet2 otftqmx ccfl fet driver 205 ccfl_ignit otftqmx ccfl supply control ignition table 1-1: mb87p2020-a pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 232 208 ccfl_off otftqmx ccfl supply control off 13 dac1_vdda itavdx dac supply 2.5v 9 dac1_vdda1 itavdx dac supply 2.5v 12 dac1_vssa itavsx dac ground 10 dac1_vssa1 itavsx dac ground 7 dac2_vdda1 itavdx dac supply 2.5v 6 dac2_vssa1 itavsx dac ground 15 dac3_vdda1 itavdx dac supply 2.5v 16 dac3_vssa1 itavsx dac ground 166 dis_ckey b3nnlmr2x display colour key 135 dis_d[0] b3nnlmr2x display data 148 dis_d[10] b3nnlmr2x display data 149 dis_d[11] b3nnlmr2x display data 150 dis_d[12] b3nnlmr2x display data 151 dis_d[13] b3nnlmr2x display data 152 dis_d[14] b3nnlmr2x display data 153 dis_d[15] b3nnlmr2x display data 156 dis_d[16] b3nnlmr2x display data 157 dis_d[17] b3nnlmr2x display data 158 dis_d[18] b3nnlmr2x display data 159 dis_d[19] b3nnlmr2x display data 136 dis_d[1] b3nnlmr2x display data 160 dis_d[20] b3nnlmr2x display data 161 dis_d[21] b3nnlmr2x display data 162 dis_d[22] b3nnlmr2x display data 163 dis_d[23] b3nnlmr2x display data 137 dis_d[2] b3nnlmr2x display data 139 dis_d[3] b3nnlmr2x display data 140 dis_d[4] b3nnlmr2x display data 141 dis_d[5] b3nnlmr2x display data 143 dis_d[6] b3nnlmr2x display data 144 dis_d[7] b3nnlmr2x display data 145 dis_d[8] b3nnlmr2x display data table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 233 146 dis_d[9] b3nnlmr2x display data 169 dis_hsync b3nnlmr2x display programmable sync 167 dis_pixclk b3nnnmr2x display pixel clock (programmable in/out) 170 dis_vref b3nnlmr2x display programmable sync 168 dis_vsync b3nnlmr2x display programmable sync 3 gnd gnd 26 gnd gnd 38 gnd gnd gnd 50 gnd gnd gnd 60 gnd gnd gnd 72 gnd gnd gnd 85 gnd gnd gnd 96 gnd gnd gnd 107 gnd gnd gnd 119 gnd gnd gnd 130 gnd gnd gnd 142 gnd gnd gnd 154 gnd gnd gnd 164 gnd gnd gnd 176 gnd gnd gnd 189 gnd gnd gnd 200 gnd gnd gnd 66 gnd[1] gnd 80 gnd[2] gnd 91 gnd[3] gnd 103 gnd[4] gnd 113 gnd[5] gnd 206 gnd[6] gnd 1 mode[0] bfnnqlx mode pin 2 mode[1] bfnnqlx mode pin 179 mode[2] bfnnqlx mode pin 180 mode[3] bfnnqlx mode pin 25 osc_in ixn3x xtal input table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 234 23 osc_out bx4mfsr2x xtal output 31 rclk itfhx reserved clock 28 rdy_trien b3nnlmr2x control rdy pin behaviour 22 resetx itfhx reset 128 sdram_pbi ipbix sdram test mode 132 sdram_tbst itbstx sdram test mode 133 sdram_ttst ittstx sdram test mode 110 sdram_vcc[0] sdram supply 2.5v 115 sdram_vcc[1] sdram supply 2.5v 118 sdram_vcc[2] sdram supply 2.5v 125 sdram_vcc[3] sdram supply 2.5v 199 spb_bus bfnnqhx spb interface 202 spb_tst bfnnqhx spb test 20 test itchx fujitsu test pin 57 ulb_a[0] bfnnqlx ulb interface address 46 ulb_a[10] bfnnqlx ulb interface address 45 ulb_a[11] bfnnqlx ulb interface address 42 ulb_a[12] bfnnqlx ulb interface address 41 ulb_a[13] bfnnqlx ulb interface address 40 ulb_a[14] bfnnqlx ulb interface address 39 ulb_a[15] bfnnqlx ulb interface address 37 ulb_a[16] bfnnqlx ulb interface address 36 ulb_a[17] bfnnqlx ulb interface address 35 ulb_a[18] bfnnqlx ulb interface address 33 ulb_a[19] bfnnqlx ulb interface address 56 ulb_a[1] bfnnqlx ulb interface address 32 ulb_a[20] bfnnqlx ulb interface address 55 ulb_a[2] bfnnqlx ulb interface address 54 ulb_a[3] bfnnqlx ulb interface address 53 ulb_a[4] bfnnqlx ulb interface address 52 ulb_a[5] bfnnqlx ulb interface address 51 ulb_a[6] bfnnqlx ulb interface address 49 ulb_a[7] bfnnqlx ulb interface address table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 235 48 ulb_a[8] bfnnqlx ulb interface address 47 ulb_a[9] bfnnqlx ulb interface address 44 ulb_clk itfhx ulb interface clock 58 ulb_cs bfnnqlx ulb interface chip select 112 ulb_d[0] bfnnqmx ulb interface data 95 ulb_d[10] bfnnqmx ulb interface data 94 ulb_d[11] bfnnqmx ulb interface data 93 ulb_d[12] bfnnqmx ulb interface data 90 ulb_d[13] bfnnqmx ulb interface data 89 ulb_d[14] bfnnqmx ulb interface data 88 ulb_d[15] bfnnqmx ulb interface data 87 ulb_d[16] bfnnqmx ulb interface data 84 ulb_d[17] bfnnqmx ulb interface data 83 ulb_d[18] bfnnqmx ulb interface data 82 ulb_d[19] bfnnqmx ulb interface data 111 ulb_d[1] bfnnqmx ulb interface data 81 ulb_d[20] bfnnqmx ulb interface data 78 ulb_d[21] bfnnqmx ulb interface data 77 ulb_d[22] bfnnqmx ulb interface data 76 ulb_d[23] bfnnqmx ulb interface data 75 ulb_d[24] bfnnqmx ulb interface data 71 ulb_d[25] bfnnqmx ulb interface data 70 ulb_d[26] bfnnqmx ulb interface data 69 ulb_d[27] bfnnqmx ulb interface data 68 ulb_d[28] bfnnqmx ulb interface data 65 ulb_d[29] bfnnqmx ulb interface data 108 ulb_d[2] bfnnqmx ulb interface data 64 ulb_d[30] bfnnqmx ulb interface data 63 ulb_d[31] bfnnqmx ulb interface data 106 ulb_d[3] bfnnqmx ulb interface data 105 ulb_d[4] bfnnqmx ulb interface data 104 ulb_d[5] bfnnqmx ulb interface data 101 ulb_d[6] bfnnqmx ulb interface data table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 236 100 ulb_d[7] bfnnqmx ulb interface data 99 ulb_d[8] bfnnqmx ulb interface data 98 ulb_d[9] bfnnqmx ulb interface data 62 ulb_dack bfnnqlx ulb interface dma acknowledge 120 ulb_dreq otftqmx ulb interface dma request 117 ulb_dstp bfnnqmx ulb interface dma stop 121 ulb_intrq otftqmx ulb interface interrupt request 59 ulb_rdx bfnnqlx ulb interface read 116 ulb_rdy otftqmx ulb interface ready 122 ulb_wrx[0] itfhx ulb interface write enable (d[31:24]) 124 ulb_wrx[1] itfhx ulb interface write enable (d[23:16]) 126 ulb_wrx[2] itfhx ulb interface write enable (d[15:8]) 127 ulb_wrx[3] itfhx ulb interface write enable (d[7:0]) 43 vdde vdde# io supply 3.3v 61 vdde vdde# io supply 3.3v 97 vdde vdde# io supply 3.3v 114 vdde vdde# io supply 3.3v 147 vdde vdde# io supply 3.3v 165 vdde vdde# io supply 3.3v 201 vdde vdde# io supply 3.3v 24 vdde[0] io supply 3.3v 207 vdde[10] io supply 3.3v 67 vdde[1] io supply 3.3v 74 vdde[2] io supply 3.3v 92 vdde[3] io supply 3.3v 102 vdde[4] io supply 3.3v 109 vdde[5] io supply 3.3v 129 vdde[7] io supply 3.3v 155 vdde[8] io supply 3.3v 178 vdde[9] io supply 3.3v 19 vddi vddi# core supply 2.5 v 27 vddi vddi# core supply 2.5 v 34 vddi vddi# core supply 2.5 v table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 237 73 vddi vddi# core supply 2.5 v 79 vddi vddi# core supply 2.5 v 86 vddi vddi# core supply 2.5 v 123 vddi vddi# core supply 2.5 v 131 vddi vddi# core supply 2.5 v 138 vddi vddi# core supply 2.5 v 177 vddi vddi# core supply 2.5 v 183 vddi vddi# core supply 2.5 v 190 vddi vddi# core supply 2.5 v 134 vpd vpdx fujitsu tester pin 17 vref itamx dac test pin vref 197 vsc_alpha bfnnqlx video scaler alpha 21 vsc_clkv itfhx video scaler clock 171 vsc_d[0] bfnnqlx video scaler data input 187 vsc_d[10] bfnnqlx video scaler data input 188 vsc_d[11] bfnnqlx video scaler data input 191 vsc_d[12] bfnnqlx video scaler data input 192 vsc_d[13] bfnnqlx video scaler data input 193 vsc_d[14] bfnnqlx video scaler data input 194 vsc_d[15] bfnnqlx video scaler data input 172 vsc_d[1] bfnnqlx video scaler data input 173 vsc_d[2] bfnnqlx video scaler data input 174 vsc_d[3] bfnnqlx video scaler data input 175 vsc_d[4] bfnnqlx video scaler data input 181 vsc_d[5] bfnnqlx video scaler data input 182 vsc_d[6] bfnnqlx video scaler data input 184 vsc_d[7] bfnnqlx video scaler data input 185 vsc_d[8] bfnnqlx video scaler data input 186 vsc_d[9] bfnnqlx video scaler data input 198 vsc_ident bfnnqlx video scaler field identi?cation 196 vsc_vact bfnnqlx video scaler vact 195 vsc_vref bfnnqlx video scaler vertical reference table 1-2: mb87p2020-a pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 238 1.1.2 buffer types table 1-3 shows all used buffers for mb87p2020-a. 1.2 pinning for mb87p2020 1.2.1 pinning table 1-4 shows the pinning for mb87p2020 (jasmine) sorted by pin number while table 1-5 is sorted by names. both tables refer to a standard fujitsu 208-pin (l)qfp package fpt-208p-mxx. table 1-3: buffer types for mb87p2020-a buffer type description b3nnlmr2x bidirectional true buffer (3.3v cmos, iol=4ma,low noise type, esd improved) b3nnnmr2x bidirectional true buffer (3.3v cmos, iol=4ma, esd improved) bfnnqhx bidirectional true buffer (5v tolerant, iol=8ma, high speed type) bfnnqlx bidirectional true buffer (5v tolerant, iol=2ma, high speed type) bfnnqmx 5v tolerant, bidirectional true buffer 3.3v cmos, iol/ioh=4ma bx4mfsr2x oscillator output (esd improved) ipbix input true buffer for dram test (2.5v cmos with 25k pull-up) (sdram test only) itamx analog input buffer itavdx analog power supply itavsx analog gnd itbstx input true buffer for dram test (2.5v cmos with 25k pull-down) (sdram test only) itchx input true buffer (2.5v cmos) itfhx 5v tolerant 3.3v cmos input ittstx input true buffer for dram test control (2.5v cmos with 25k pull-down) ixn3x oscillator input (esd improved) otamx analog output otftqmx 5 v tolerant 3.3v tri-state output, iol/ioh=4ma vpdx 3.3v cmos input, disable input for pull up/down resistors, connect to gnd table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description 1 mode[0] bfnnqlx mode pin 2 mode[1] bfnnqlx mode pin 3 gnd gnd pinning for lavender and jasmine pinning and buffer types page 239 4 place holder for analog area 5 a_green otamx analog green 6 dac2_vssa1 itavsx dac ground 7 dac2_vdda1 itavdx dac supply 2.5v 8 a_blue otamx analog blue 9 dac1_vdda1 itavdx dac supply 2.5v 10 dac1_vssa1 itavsx dac ground 11 a_vro otamx dac full scale adjust 12 dac1_vssa itavsx dac ground 13 dac1_vdda itavdx dac supply 2.5v 14 a_red otamx analog red 15 dac3_vdda1 itavdx dac supply 2.5v 16 dac3_vssa1 itavsx dac ground 17 vref itamx dac test pin vref 18 place holder for analog area 19 vddi vddi# core supply 2.5 v 20 test itchx fujitsu test pin 21 vsc_clkv itfhx video scaler clock 22 resetx itfuhx reset (pull up) 23 osc_out yb002aax xtal output 24 vdde[0] io supply 3.3v 25 osc_in yi002aex xtal input 26 gnd gnd 27 vddi vddi# core supply 2.5 v 28 rdy_trien b3nnlmx control rdy pin behaviour 29 apll_avdd apll supply 2.5v 30 apll_avss apll gnd 31 rclk itfhx reserved clock 32 ulb_a[20] bfnnqlx ulb interface address 33 ulb_a[19] bfnnqlx ulb interface address 34 vddi vddi# core supply 2.5 v 35 ulb_a[18] bfnnqlx ulb interface address 36 ulb_a[17] bfnnqlx ulb interface address table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 240 37 ulb_a[16] bfnnqlx ulb interface address 38 gnd gnd gnd 39 ulb_a[15] bfnnqlx ulb interface address 40 ulb_a[14] bfnnqlx ulb interface address 41 ulb_a[13] bfnnqlx ulb interface address 42 ulb_a[12] bfnnqlx ulb interface address 43 vdde vdde# io supply 3.3v 44 ulb_clk itfhx ulb interface clock 45 ulb_a[11] bfnnqlx ulb interface address 46 ulb_a[10] bfnnqlx ulb interface address 47 ulb_a[9] bfnnqlx ulb interface address 48 ulb_a[8] bfnnqlx ulb interface address 49 ulb_a[7] bfnnqlx ulb interface address 50 gnd gnd gnd 51 ulb_a[6] bfnnqlx ulb interface address 52 ulb_a[5] bfnnqlx ulb interface address 53 ulb_a[4] bfnnqlx ulb interface address 54 ulb_a[3] bfnnqlx ulb interface address 55 ulb_a[2] bfnnqlx ulb interface address 56 ulb_a[1] bfnnqlx ulb interface address 57 ulb_a[0] bfnnqlx ulb interface address 58 ulb_cs bfnnqlx ulb interface chip select 59 ulb_rdx bfnnqlx ulb interface read 60 gnd gnd gnd 61 vdde vdde# io supply 3.3v 62 ulb_dack bfnnqlx ulb interface dma acknowledge 63 ulb_d[31] bfnnqmx ulb interface data 64 ulb_d[30] bfnnqmx ulb interface data 65 ulb_d[29] bfnnqmx ulb interface data 66 gnd[1] gnd 67 vdde[1] io supply 3.3v 68 ulb_d[28] bfnnqmx ulb interface data 69 ulb_d[27] bfnnqmx ulb interface data table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 241 70 ulb_d[26] bfnnqmx ulb interface data 71 ulb_d[25] bfnnqmx ulb interface data 72 gnd gnd gnd 73 vddi vddi# core supply 2.5 v 74 vdde[2] io supply 3.3v 75 ulb_d[24] bfnnqmx ulb interface data 76 ulb_d[23] bfnnqmx ulb interface data 77 ulb_d[22] bfnnqmx ulb interface data 78 ulb_d[21] bfnnqmx ulb interface data 79 vddi vddi# core supply 2.5 v 80 gnd[2] gnd 81 ulb_d[20] bfnnqmx ulb interface data 82 ulb_d[19] bfnnqmx ulb interface data 83 ulb_d[18] bfnnqmx ulb interface data 84 ulb_d[17] bfnnqmx ulb interface data 85 gnd gnd gnd 86 vddi vddi# core supply 2.5 v 87 ulb_d[16] bfnnqmx ulb interface data 88 ulb_d[15] bfnnqmx ulb interface data 89 ulb_d[14] bfnnqmx ulb interface data 90 ulb_d[13] bfnnqmx ulb interface data 91 gnd[3] gnd 92 vdde[3] io supply 3.3v 93 ulb_d[12] bfnnqmx ulb interface data 94 ulb_d[11] bfnnqmx ulb interface data 95 ulb_d[10] bfnnqmx ulb interface data 96 gnd gnd gnd 97 vdde vdde# io supply 3.3v 98 ulb_d[9] bfnnqmx ulb interface data 99 ulb_d[8] bfnnqmx ulb interface data 100 ulb_d[7] bfnnqmx ulb interface data 101 ulb_d[6] bfnnqmx ulb interface data 102 vdde[4] io supply 3.3v table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 242 103 gnd[4] gnd 104 ulb_d[5] bfnnqmx ulb interface data 105 ulb_d[4] bfnnqmx ulb interface data 106 ulb_d[3] bfnnqmx ulb interface data 107 gnd gnd gnd 108 ulb_d[2] bfnnqmx ulb interface data 109 vdde[5] io supply 3.3v 110 sdram_vcc[0] sdram supply 2.5v 111 ulb_d[1] bfnnqmx ulb interface data 112 ulb_d[0] bfnnqmx ulb interface data 113 gnd[5] gnd 114 vdde vdde# io supply 3.3v 115 sdram_vcc[1] sdram supply 2.5v 116 ulb_rdy otftqmx ulb interface ready 117 ulb_dstp bfnnqmx ulb interface dma stop 118 sdram_vcc[2] sdram supply 2.5v 119 gnd gnd gnd 120 ulb_dreq otftqmx ulb interface dma request 121 ulb_intrq otftqmx ulb interface interrupt request 122 ulb_wrx[0] itfhx ulb interface write enable (d[31:24]) 123 vddi vddi# core supply 2.5 v 124 ulb_wrx[1] itfhx ulb interface write enable (d[23:16]) 125 sdram_vcc[3] sdram supply 2.5v 126 ulb_wrx[2] itfhx ulb interface write enable (d[15:8]) 127 ulb_wrx[3] itfhx ulb interface write enable (d[7:0]) 128 sdram_pbi ipbix sdram test mode 129 vdde[7] io supply 3.3v 130 gnd gnd gnd 131 vddi vddi# core supply 2.5 v 132 sdram_tbst itbstx sdram test mode 133 sdram_ttst ittstx sdram test mode 134 vpd vpdx fujitsu tester pin 135 dis_d[0] b3nnlmx display data table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 243 136 dis_d[1] b3nnlmx display data 137 dis_d[2] b3nnlmx display data 138 vddi vddi# core supply 2.5 v 139 dis_d[3] b3nnlmx display data 140 dis_d[4] b3nnlmx display data 141 dis_d[5] b3nnlmx display data 142 gnd gnd gnd 143 dis_d[6] b3nnlmx display data 144 dis_d[7] b3nnlmx display data 145 dis_d[8] b3nnlmx display data 146 dis_d[9] b3nnlmx display data 147 vdde vdde# io supply 3.3v 148 dis_d[10] b3nnlmx display data 149 dis_d[11] b3nnlmx display data 150 dis_d[12] b3nnlmx display data 151 dis_d[13] b3nnlmx display data 152 dis_d[14] b3nnlmx display data 153 dis_d[15] b3nnlmx display data 154 gnd gnd gnd 155 vdde[8] io supply 3.3v 156 dis_d[16] b3nnlmx display data 157 dis_d[17] b3nnlmx display data 158 dis_d[18] b3nnlmx display data 159 dis_d[19] b3nnlmx display data 160 dis_d[20] b3nnlmx display data 161 dis_d[21] b3nnlmx display data 162 dis_d[22] b3nnlmx display data 163 dis_d[23] b3nnlmx display data 164 gnd gnd gnd 165 vdde vdde# io supply 3.3v 166 dis_ckey b3nnlmx display colour key 167 dis_pixclk b3nnnmx display pixel clock (programmable in/out) 168 dis_vsync b3nnlmx display programmable sync table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 244 169 dis_hsync b3nnlmx display programmable sync 170 dis_vref b3nnlmx display programmable sync 171 vsc_d[0] bfnnqlx video scaler data input 172 vsc_d[1] bfnnqlx video scaler data input 173 vsc_d[2] bfnnqlx video scaler data input 174 vsc_d[3] bfnnqlx video scaler data input 175 vsc_d[4] bfnnqlx video scaler data input 176 gnd gnd gnd 177 vddi vddi# core supply 2.5 v 178 vdde[9] io supply 3.3v 179 mode[2] bfnnqlx mode pin 180 mode[3] bfnnqlx mode pin 181 vsc_d[5] bfnnqlx video scaler data input 182 vsc_d[6] bfnnqlx video scaler data input 183 vddi vddi# core supply 2.5 v 184 vsc_d[7] bfnnqlx video scaler data input 185 vsc_d[8] bfnnqlx video scaler data input 186 vsc_d[9] bfnnqlx video scaler data input 187 vsc_d[10] bfnnqlx video scaler data input 188 vsc_d[11] bfnnqlx video scaler data input 189 gnd gnd gnd 190 vddi vddi# core supply 2.5 v 191 vsc_d[12] bfnnqlx video scaler data input 192 vsc_d[13] bfnnqlx video scaler data input 193 vsc_d[14] bfnnqlx video scaler data input 194 vsc_d[15] bfnnqlx video scaler data input 195 vsc_vref bfnnqlx video scaler vertical reference 196 vsc_vact bfnnqlx video scaler vact 197 vsc_alpha bfnnqlx video scaler alpha 198 vsc_ident bfnnqlx video scaler field identi?cation 199 spb_bus bfnnqhx spb interface 200 gnd gnd gnd 201 vdde vdde# io supply 3.3v table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 245 202 spb_tst bfnnqhx spb test 203 ccfl_fet2 otftqmx ccfl fet driver 204 ccfl_fet1 otftqmx ccfl fet driver 205 ccfl_ignit otftqmx ccfl supply control ignition 206 gnd[6] gnd 207 vdde[10] io supply 3.3v 208 ccfl_off otftqmx ccfl supply control off table 1-5: mb87p2020 pinning sorted by name pin name buffer type description 4 place holder for analog area 18 place holder for analog area 8 a_blue otamx analog blue 5 a_green otamx analog green 14 a_red otamx analog red 11 a_vro otamx dac full scale adjust 29 apll_avdd apll supply 2.5v 30 apll_avss apll gnd 204 ccfl_fet1 otftqmx ccfl fet driver 203 ccfl_fet2 otftqmx ccfl fet driver 205 ccfl_ignit otftqmx ccfl supply control ignition 208 ccfl_off otftqmx ccfl supply control off 13 dac1_vdda itavdx dac supply 2.5v 9 dac1_vdda1 itavdx dac supply 2.5v 12 dac1_vssa itavsx dac ground 10 dac1_vssa1 itavsx dac ground 7 dac2_vdda1 itavdx dac supply 2.5v 6 dac2_vssa1 itavsx dac ground 15 dac3_vdda1 itavdx dac supply 2.5v 16 dac3_vssa1 itavsx dac ground 166 dis_ckey b3nnlmx display colour key 135 dis_d[0] b3nnlmx display data table 1-4: mb87p2020 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 246 148 dis_d[10] b3nnlmx display data 149 dis_d[11] b3nnlmx display data 150 dis_d[12] b3nnlmx display data 151 dis_d[13] b3nnlmx display data 152 dis_d[14] b3nnlmx display data 153 dis_d[15] b3nnlmx display data 156 dis_d[16] b3nnlmx display data 157 dis_d[17] b3nnlmx display data 158 dis_d[18] b3nnlmx display data 159 dis_d[19] b3nnlmx display data 136 dis_d[1] b3nnlmx display data 160 dis_d[20] b3nnlmx display data 161 dis_d[21] b3nnlmx display data 162 dis_d[22] b3nnlmx display data 163 dis_d[23] b3nnlmx display data 137 dis_d[2] b3nnlmx display data 139 dis_d[3] b3nnlmx display data 140 dis_d[4] b3nnlmx display data 141 dis_d[5] b3nnlmx display data 143 dis_d[6] b3nnlmx display data 144 dis_d[7] b3nnlmx display data 145 dis_d[8] b3nnlmx display data 146 dis_d[9] b3nnlmx display data 169 dis_hsync b3nnlmx display programmable sync 167 dis_pixclk b3nnnmx display pixel clock (programmable in/out) 170 dis_vref b3nnlmx display programmable sync 168 dis_vsync b3nnlmx display programmable sync 3 gnd gnd 26 gnd gnd 38 gnd gnd gnd 50 gnd gnd gnd 60 gnd gnd gnd 72 gnd gnd gnd table 1-5: mb87p2020 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 247 85 gnd gnd gnd 96 gnd gnd gnd 107 gnd gnd gnd 119 gnd gnd gnd 130 gnd gnd gnd 142 gnd gnd gnd 154 gnd gnd gnd 164 gnd gnd gnd 176 gnd gnd gnd 189 gnd gnd gnd 200 gnd gnd gnd 66 gnd[1] gnd 80 gnd[2] gnd 91 gnd[3] gnd 103 gnd[4] gnd 113 gnd[5] gnd 206 gnd[6] gnd 1 mode[0] bfnnqlx mode pin 2 mode[1] bfnnqlx mode pin 179 mode[2] bfnnqlx mode pin 180 mode[3] bfnnqlx mode pin 25 osc_in yi002aex xtal input 23 osc_out yb002aax xtal output 31 rclk itfhx reserved clock 28 rdy_trien b3nnlmx control rdy pin behaviour 22 resetx itfuhx reset (pull up) 128 sdram_pbi ipbix sdram test mode 132 sdram_tbst itbstx sdram test mode 133 sdram_ttst ittstx sdram test mode 110 sdram_vcc[0] sdram supply 2.5v 115 sdram_vcc[1] sdram supply 2.5v 118 sdram_vcc[2] sdram supply 2.5v 125 sdram_vcc[3] sdram supply 2.5v table 1-5: mb87p2020 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 248 199 spb_bus bfnnqhx spb interface 202 spb_tst bfnnqhx spb test 20 test itchx fujitsu test pin 57 ulb_a[0] bfnnqlx ulb interface address 46 ulb_a[10] bfnnqlx ulb interface address 45 ulb_a[11] bfnnqlx ulb interface address 42 ulb_a[12] bfnnqlx ulb interface address 41 ulb_a[13] bfnnqlx ulb interface address 40 ulb_a[14] bfnnqlx ulb interface address 39 ulb_a[15] bfnnqlx ulb interface address 37 ulb_a[16] bfnnqlx ulb interface address 36 ulb_a[17] bfnnqlx ulb interface address 35 ulb_a[18] bfnnqlx ulb interface address 33 ulb_a[19] bfnnqlx ulb interface address 56 ulb_a[1] bfnnqlx ulb interface address 32 ulb_a[20] bfnnqlx ulb interface address 55 ulb_a[2] bfnnqlx ulb interface address 54 ulb_a[3] bfnnqlx ulb interface address 53 ulb_a[4] bfnnqlx ulb interface address 52 ulb_a[5] bfnnqlx ulb interface address 51 ulb_a[6] bfnnqlx ulb interface address 49 ulb_a[7] bfnnqlx ulb interface address 48 ulb_a[8] bfnnqlx ulb interface address 47 ulb_a[9] bfnnqlx ulb interface address 44 ulb_clk itfhx ulb interface clock 58 ulb_cs bfnnqlx ulb interface chip select 112 ulb_d[0] bfnnqmx ulb interface data 95 ulb_d[10] bfnnqmx ulb interface data 94 ulb_d[11] bfnnqmx ulb interface data 93 ulb_d[12] bfnnqmx ulb interface data 90 ulb_d[13] bfnnqmx ulb interface data 89 ulb_d[14] bfnnqmx ulb interface data 88 ulb_d[15] bfnnqmx ulb interface data table 1-5: mb87p2020 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 249 87 ulb_d[16] bfnnqmx ulb interface data 84 ulb_d[17] bfnnqmx ulb interface data 83 ulb_d[18] bfnnqmx ulb interface data 82 ulb_d[19] bfnnqmx ulb interface data 111 ulb_d[1] bfnnqmx ulb interface data 81 ulb_d[20] bfnnqmx ulb interface data 78 ulb_d[21] bfnnqmx ulb interface data 77 ulb_d[22] bfnnqmx ulb interface data 76 ulb_d[23] bfnnqmx ulb interface data 75 ulb_d[24] bfnnqmx ulb interface data 71 ulb_d[25] bfnnqmx ulb interface data 70 ulb_d[26] bfnnqmx ulb interface data 69 ulb_d[27] bfnnqmx ulb interface data 68 ulb_d[28] bfnnqmx ulb interface data 65 ulb_d[29] bfnnqmx ulb interface data 108 ulb_d[2] bfnnqmx ulb interface data 64 ulb_d[30] bfnnqmx ulb interface data 63 ulb_d[31] bfnnqmx ulb interface data 106 ulb_d[3] bfnnqmx ulb interface data 105 ulb_d[4] bfnnqmx ulb interface data 104 ulb_d[5] bfnnqmx ulb interface data 101 ulb_d[6] bfnnqmx ulb interface data 100 ulb_d[7] bfnnqmx ulb interface data 99 ulb_d[8] bfnnqmx ulb interface data 98 ulb_d[9] bfnnqmx ulb interface data 62 ulb_dack bfnnqlx ulb interface dma acknowledge 120 ulb_dreq otftqmx ulb interface dma request 117 ulb_dstp bfnnqmx ulb interface dma stop 121 ulb_intrq otftqmx ulb interface interrupt request 59 ulb_rdx bfnnqlx ulb interface read 116 ulb_rdy otftqmx ulb interface ready 122 ulb_wrx[0] itfhx ulb interface write enable (d[31:24]) 124 ulb_wrx[1] itfhx ulb interface write enable (d[23:16]) table 1-5: mb87p2020 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 250 126 ulb_wrx[2] itfhx ulb interface write enable (d[15:8]) 127 ulb_wrx[3] itfhx ulb interface write enable (d[7:0]) 43 vdde vdde# io supply 3.3v 61 vdde vdde# io supply 3.3v 97 vdde vdde# io supply 3.3v 114 vdde vdde# io supply 3.3v 147 vdde vdde# io supply 3.3v 165 vdde vdde# io supply 3.3v 201 vdde vdde# io supply 3.3v 24 vdde[0] io supply 3.3v 207 vdde[10] io supply 3.3v 67 vdde[1] io supply 3.3v 74 vdde[2] io supply 3.3v 92 vdde[3] io supply 3.3v 102 vdde[4] io supply 3.3v 109 vdde[5] io supply 3.3v 129 vdde[7] io supply 3.3v 155 vdde[8] io supply 3.3v 178 vdde[9] io supply 3.3v 19 vddi vddi# core supply 2.5 v 27 vddi vddi# core supply 2.5 v 34 vddi vddi# core supply 2.5 v 73 vddi vddi# core supply 2.5 v 79 vddi vddi# core supply 2.5 v 86 vddi vddi# core supply 2.5 v 123 vddi vddi# core supply 2.5 v 131 vddi vddi# core supply 2.5 v 138 vddi vddi# core supply 2.5 v 177 vddi vddi# core supply 2.5 v 183 vddi vddi# core supply 2.5 v 190 vddi vddi# core supply 2.5 v 134 vpd vpdx fujitsu tester pin 17 vref itamx dac test pin vref table 1-5: mb87p2020 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 251 1.2.2 buffer types table 1-6 shows all used buffers for mb87p2020. 197 vsc_alpha bfnnqlx video scaler alpha 21 vsc_clkv itfhx video scaler clock 171 vsc_d[0] bfnnqlx video scaler data input 187 vsc_d[10] bfnnqlx video scaler data input 188 vsc_d[11] bfnnqlx video scaler data input 191 vsc_d[12] bfnnqlx video scaler data input 192 vsc_d[13] bfnnqlx video scaler data input 193 vsc_d[14] bfnnqlx video scaler data input 194 vsc_d[15] bfnnqlx video scaler data input 172 vsc_d[1] bfnnqlx video scaler data input 173 vsc_d[2] bfnnqlx video scaler data input 174 vsc_d[3] bfnnqlx video scaler data input 175 vsc_d[4] bfnnqlx video scaler data input 181 vsc_d[5] bfnnqlx video scaler data input 182 vsc_d[6] bfnnqlx video scaler data input 184 vsc_d[7] bfnnqlx video scaler data input 185 vsc_d[8] bfnnqlx video scaler data input 186 vsc_d[9] bfnnqlx video scaler data input 198 vsc_ident bfnnqlx video scaler field identi?cation 196 vsc_vact bfnnqlx video scaler vact 195 vsc_vref bfnnqlx video scaler vertical reference table 1-6: buffer types for mb87p2020 buffer type description b3nnlmx bidirectional true buffer (3.3v cmos, iol=4ma,low noise type) b3nnnmx bidirectional true buffer (3.3v cmos, iol=4ma) bfnnqhx bidirectional true buffer (5v tolerant, iol=8ma, high speed type) bfnnqlx bidirectional true buffer (5v tolerant, iol=2ma, high speed type) bfnnqmx 5v tolerant, bidirectional true buffer 3.3v cmos, iol/ioh=4ma ipbix input true buffer for dram test (2.5v cmos with 25k pull-up) (sdram test only) table 1-5: mb87p2020 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 252 1.3 pinning for MB87J2120 1.3.1 pinning table 1-7 shows the pinning for MB87J2120 (lavender) sorted by pin number while table 1-8 is sorted by names. both tables refer to a standard fujitsu 256-pin bga package (bga-256p-m01). itamx analog input buffer itavdx analog power supply itavsx analog gnd itbstx input true buffer for dram test (2.5v cmos with 25k pull-down) (sdram test only) itchx input true buffer (2.5v cmos) itfhx 5v tolerant 3.3v cmos input itfuhx 5v tolerant 3.3v cmos input, 25 k pull-up ittstx input true buffer for dram test control (2.5v cmos with 25k pull-down) otamx analog output otftqmx 5 v tolerant 3.3v tri-state output, iol/ioh=4ma vpdx 3.3v cmos input, disable input for pull up/down resistors, connect to gnd yb002aax oscillator output yi002aex oscillator input table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description 1 gnd gnd gnd 2 dac3_vssa1_2 itavsx dac ground 3 a_blue otamx analog blue 4 spacer, n.c. 5 a_green otamx analog green 6 dac1_vdda itavdx dac supply 2.5v 7 vpd vpdx fujitsu tester pin 8 dis_d[1] yb3nnlmx display data 9 dis_d[6] yb3nnlmx display data 10 dis_d[8] yb3nnlmx display data table 1-6: buffer types for mb87p2020 buffer type description pinning for lavender and jasmine pinning and buffer types page 253 11 vdde[0] ovdd3x io supply 3.3v 12 dis_d[13] yb3nnlmx display data 13 gnd[0] ovssx gnd 14 dis_d[14] yb3nnlmx display data 15 dis_d[19] yb3nnlmx display data 16 gnd[1] ovssx gnd 17 dis_d[22] yb3nnlmx display data 18 dis_vsync yb3nnlmx display programmable sync 19 dis_d[20] yb3nnlmx display data 20 gnd gnd gnd 21 ulb_d[14] bfnnqmx ulb interface data 22 ulb_d[8] bfnnqmx ulb interface data 23 ulb_d[4] bfnnqmx ulb interface data 24 ulb_d[5] bfnnqmx ulb interface data 25 ulb_d[11] bfnnqmx ulb interface data 26 vdde[9] ovdd3x io supply 3.3v 27 ulb_d[20] bfnnqmx ulb interface data 28 ulb_wrx[2] itfhx ulb interface byte 2 write enable 29 ulb_wrx[1] itfhx ulb interface byte 1 write enable 30 ulb_d[26] bfnnqmx ulb interface data 31 ulb_d[30] bfnnqmx ulb interface data 32 ulb_d[25] bfnnqmx ulb interface data 33 ulb_d[31] bfnnqmx ulb interface data 34 ulb_clk itfhx ulb interface clock 35 ulb_a[7] itfhx ulb interface address 36 ulb_wrx[3] itfhx ulb interface byte 3 write enable 37 ulb_a[10] itfhx ulb interface address 38 ulb_a[2] itfhx ulb interface address 39 gnd gnd gnd 40 ulb_intrq otftqmx ulb interface interrupt output 41 gnd[3] ovssx gnd 42 ulb_a[17] itfhx ulb interface address 43 ulb_a[18] itfhx ulb interface address table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 254 44 ulb_dack itfhx ulb interface dma acknowledge 45 sdc_d[3] yb3nnlmx sdram data 46 sdc_d[7] yb3nnlmx sdram data 47 sdc_d[11] yb3nnlmx sdram data 48 vdde[4] ovdd3x io supply 3.3v 49 gnd[10] ovssx gnd 50 sdc_d[15] yb3nnlmx sdram data 51 sdc_cke yb3dnlmx sdram clock enable 52 sdc_d[16] yb3nnlmx sdram data 53 sdc_d[21] yb3nnlmx sdram data 54 gnd[5] ovssx gnd 55 sdc_a[0] yb3nnlmx sdram address 56 sdc_d[28] yb3nnlmx sdram data 57 sdc_d[22] yb3nnlmx sdram data 58 gnd gnd gnd 59 sdc_cas yb3nnlmx sdram cas 60 sdc_a[7] yb3nnlmx sdram address 61 sdc_a[3] yb3nnlmx sdram address 62 sdc_a[4] yb3nnlmx sdram address 63 sdc_a[10] yb3nnlmx sdram address 64 sbp_bus bfunqhx spb interface 65 mode[3] itfuhx gdc mode pin (pull up) 66 ccfl_fet2 otftqmx ccfl fet driver 67 ccfl_ignit otftqmx ccfl supply control ignition 68 gnd[11] ovssx gnd 69 mode[0] itfuhx gdc mode pin (pull up) 70 mode[1] itfuhx gdc mode pin (pull up) 71 avdd[5] apll apll supply 2.5v 72 vsc_d[7] itfhx video scaler data input 73 vsc_alpha itfhx video scaler alpha 74 vsc_clkv itfhx video scaler clock 75 vsc_d[14] itfhx video scaler data input 76 vsc_d[6] itfhx video scaler data input table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 255 77 vsc_vref itfhx video scaler vertical refernce 78 a_red otamx analog red 79 dac2_vssa1_2 itavsx dac ground 80 vsc_d[8] itfhx video scaler data input 81 a_vro otavx dac full scale adjust 82 vref itamx dac testpin vref 83 vsc_d[1] itfhx video scaler data input 84 dis_d[4] yb3nnlmx display data 85 dis_d[5] yb3nnlmx display data 86 dis_d[7] yb3nnlmx display data 87 dis_d[15] yb3nnlmx display data 88 dis_d[10] yb3nnlmx display data 89 dis_d[16] yb3nnlmx display data 90 dis_d[23] yb3nnlmx display data 91 dis_d[2] yb3nnlmx display data 92 dis_ck yb3nnlmx display color key 93 ulb_wrx[0] itfhx ulb interface byte 0 write enable 94 ulb_d[1] bfnnqmx ulb interface data 95 ulb_d[12] bfnnqmx ulb interface data 96 ulb_d[6] bfnnqmx ulb interface data 97 ulb_d[3] bfnnqmx ulb interface data 98 vdde[2] ovdd3x io supply 3.3v 99 ulb_d[15] bfnnqmx ulb interface data 100 ulb_d[18] bfnnqmx ulb interface data 101 ulb_d[22] bfnnqmx ulb interface data 102 ulb_d[23] bfnnqmx ulb interface data 103 ulb_d[24] bfnnqmx ulb interface data 104 ulb_cs itfhx ulb interface chip select 105 ulb_d[27] bfnnqmx ulb interface data 106 ulb_a[0] itfhx ulb interface address 107 ulb_a[5] itfhx ulb interface address 108 ulb_a[13] itfhx ulb interface address 109 ulb_a[12] itfhx ulb interface address table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 256 110 ulb_a[4] itfhx ulb interface address 111 vdde[10] ovdd3x io supply 3.3v 112 ulb_deop bfnnqmx ulb interface dma stop output (dstop) 113 ulb_a[19] itfhx ulb interface address 114 ulb_a[16] itfhx ulb interface address 115 ulb_rdy otftqmx ulb interface ready 116 sdc_d[2] yb3nnlmx sdram data 117 sdc_d[6] yb3nnlmx sdram data 118 sdc_d[9] yb3nnlmx sdram data 119 sdc_d[10] yb3nnlmx sdram data 120 gnd[4] ovssx gnd 121 sdc_d[17] yb3nnlmx sdram data 122 sdc_d[12] yb3nnlmx sdram data 123 sdc_d[18] yb3nnlmx sdram data 124 sdc_d[25] yb3nnlmx sdram data 125 sdc_d[31] yb3nnlmx sdram data 126 sdc_d[30] yb3nnlmx sdram data 127 sdc_d[24] yb3nnlmx sdram data 128 sdc_d[20] yb3nnlmx sdram data 129 sdc_a[11] yb3nnlmx sdram address 130 sdc_a[5] yb3nnlmx sdram address 131 sdc_a[2] yb3nnlmx sdram address 132 vdde[6] ovdd3x io supply 3.3v 133 sdc_ras yb3nnlmx sdram ras 134 mode[2] itfuhx gdc mode pin (pull up) 135 vdde[8] ovdd3x io supply 3.3v 136 ccfl_fet1 otftqmx ccfl fet driver 137 ccfl_off otftqmx ccfl supply control off 138 avss[5] apll apll gnd 139 osc_in yi002aex xtal input 140 vsc_ident itfhx video scaler field identi?cation 141 vsc_d[3] itfhx video scaler data input table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 257 142 vsc_d[11] itfhx video scaler data input 143 vsc_d[12] itfhx video scaler data input 144 vsc_d[4] itfhx video scaler data input 145 spacer, n.c. 146 dac1_vssa itavsx dac ground 147 vsc_d[0] itfhx video scaler data input 148 dac2_vdda1 itavdx dac supply 2.5v 149 dac3_vdda1 itavdx dac supply 2.5v 150 dis_pixclk b3nnnmx display pixel clock (programmable in/out) 151 dis_d[0] yb3nnlmx display data 152 dis_d[3] yb3nnlmx display data 153 dis_d[11] yb3nnlmx display data 154 dis_d[17] yb3nnlmx display data 155 dis_d[12] yb3nnlmx display data 156 dis_d[21] yb3nnlmx display data 157 dis_vref yb3nnlmx display programmable sync 158 dis_d[18] yb3nnlmx display data 159 ulb_d[0] bfnnqmx ulb interface data 160 ulb_d[16] bfnnqmx ulb interface data 161 ulb_d[10] bfnnqmx ulb interface data 162 ulb_d[2] bfnnqmx ulb interface data 163 ulb_d[7] bfnnqmx ulb interface data 164 ulb_d[13] bfnnqmx ulb interface data 165 ulb_d[17] bfnnqmx ulb interface data 166 ulb_d[19] bfnnqmx ulb interface data 167 ulb_d[21] bfnnqmx ulb interface data 168 ulb_d[28] bfnnqmx ulb interface data 169 ulb_a[1] itfhx ulb interface address 170 ulb_d[29] bfnnqmx ulb interface data 171 ulb_a[3] itfhx ulb interface address 172 ulb_a[11] itfhx ulb interface address 173 ulb_a[14] itfhx ulb interface address table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 258 174 ulb_a[6] itfhx ulb interface address 175 ulb_a[20] itfhx ulb interface address 176 ulb_dreq otftqmx ulb interface dma request 177 ulb_a[15] itfhx ulb interface address 178 sdc_d[0] yb3nnlmx sdram data 179 sdc_d[1] yb3nnlmx sdram data 180 sdc_d[4] yb3nnlmx sdram data 181 sdc_d[5] yb3nnlmx sdram data 182 sdc_d[8] yb3nnlmx sdram data 183 sdc_d[13] yb3nnlmx sdram data 184 sdc_d[19] yb3nnlmx sdram data 185 sdc_d[14] yb3nnlmx sdram data 186 sdc_d[23] yb3nnlmx sdram data 187 sdc_d[29] yb3nnlmx sdram data 188 sdc_we yb3nnlmx sdram write enable 189 sdc_d[26] yb3nnlmx sdram data 190 sdc_dmq yb3nnlmx sdram dqm 191 sdc_a[9] yb3nnlmx sdram address 192 sdc_a[1] yb3nnlmx sdram address 193 sdc_a[6] yb3nnlmx sdram address 194 sdc_a[12] yb3nnlmx sdram address 195 test yb3dnlmx fujitsu testpin 196 rstx itfuhx gdc reset (pull up) 197 gnd[8] ovssx gnd 198 vdde[7] ovdd3x io supply 3.3v 199 vsc_d[9] itfhx video scaler data input 200 osc_out yb002aax xtal output 201 vsc_d[5] itfhx video scaler data input 202 vsc_d[13] itfhx video scaler data input 203 vsc_d[10] itfhx video scaler data input 204 vsc_d[2] itfhx video scaler data input 205 gnd gnd gnd 206 dac1_vssa1_2 itavsx dac supply 2.5v table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 259 207 vddi vddi# core supply 2.5 v 208 dac1_vdda1 itavdx dac supply 2.5v 209 gnd gnd gnd 210 vdde vdde#5 io supply 3.3v 211 vddi vddi# core supply 2.5 v 212 dis_d[9] yb3nnlmx display data 213 vdde vdde# io supply 3.3v 214 gnd gnd gnd 215 dis_hsync yb3nnlmx display programmable sync 216 vdde vdde# io supply 3.3v 217 vdde[1] ovdd3x io supply 3.3v 218 gnd gnd gnd 219 gnd[2] ovssx gnd 220 vddi vddi# core supply 2.5 v 221 ulb_d[9] bfnnqmx ulb interface data 222 gnd gnd gnd 223 vddi vddi# core supply 2.5 v 224 vdde vdde# io supply 3.3v 225 gnd[9] ovssx gnd 226 vddi vddi# core supply 2.5 v 227 gnd gnd gnd 228 ulb_a[9] itfhx ulb interface address 229 vdde vdde# io supply 3.3v 230 ulb_a[8] itfhx ulb interface address 231 gnd gnd gnd 232 ulb_rdx itfhx ulb interface read 233 vddi vddi# core supply 2.5 v 234 vdde[3] ovdd3x io supply 3.3v 235 gnd gnd gnd 236 vdde vdde# io supply 3.3v 237 vddi vddi# core supply 2.5 v 238 sdc_clk b3nnnmx sdram clock 239 vdde vdde# io supply 3.3v table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 260 240 gnd gnd gnd 241 sdc_d[27] yb3nnlmx sdram data 242 vdde vdde# io supply 3.3v 243 vdde[5] ovdd3x io supply 3.3v 244 gnd gnd gnd 245 gnd[6] ovssx gnd 246 vddi vddi# core supply 2.5 v 247 sdc_a[8] yb3nnlmx sdram address 248 gnd gnd gnd 249 vddi vddi# core supply 2.5 v 250 vdde vdde# io supply 3.3v 251 gnd[7] ovssx gnd 252 vddi vddi# core supply 2.5 v 253 gnd gnd gnd 254 vsc_d[15] itfhx video scaler data input 255 vdde vdde# io supply 3.3v 256 vsc_vact itfhx video scaler vact table 1-8: MB87J2120 pinning sorted by name pin name buffer type description 4 spacer, n.c. 145 spacer, n.c. 3 a_blue otamx analog blue 5 a_green otamx analog green 78 a_red otamx analog red 81 a_vro otavx dac full scale adjust 71 avdd[5] apll apll supply 2.5v 138 avss[5] apll apll gnd 136 ccfl_fet1 otftqmx ccfl fet driver 66 ccfl_fet2 otftqmx ccfl fet driver 67 ccfl_ignit otftqmx ccfl supply control ignition 137 ccfl_off otftqmx ccfl supply control off table 1-7: MB87J2120 pinning sorted by pin number pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 261 6 dac1_vdda itavdx dac supply 2.5v 208 dac1_vdda1 itavdx dac supply 2.5v 146 dac1_vssa itavsx dac ground 206 dac1_vssa1_2 itavsx dac supply 2.5v 148 dac2_vdda1 itavdx dac supply 2.5v 79 dac2_vssa1_2 itavsx dac ground 149 dac3_vdda1 itavdx dac supply 2.5v 2 dac3_vssa1_2 itavsx dac ground 92 dis_ck yb3nnlmx display color key 151 dis_d[0] yb3nnlmx display data 88 dis_d[10] yb3nnlmx display data 153 dis_d[11] yb3nnlmx display data 155 dis_d[12] yb3nnlmx display data 12 dis_d[13] yb3nnlmx display data 14 dis_d[14] yb3nnlmx display data 87 dis_d[15] yb3nnlmx display data 89 dis_d[16] yb3nnlmx display data 154 dis_d[17] yb3nnlmx display data 158 dis_d[18] yb3nnlmx display data 15 dis_d[19] yb3nnlmx display data 8 dis_d[1] yb3nnlmx display data 19 dis_d[20] yb3nnlmx display data 156 dis_d[21] yb3nnlmx display data 17 dis_d[22] yb3nnlmx display data 90 dis_d[23] yb3nnlmx display data 91 dis_d[2] yb3nnlmx display data 152 dis_d[3] yb3nnlmx display data 84 dis_d[4] yb3nnlmx display data 85 dis_d[5] yb3nnlmx display data 9 dis_d[6] yb3nnlmx display data 86 dis_d[7] yb3nnlmx display data 10 dis_d[8] yb3nnlmx display data 212 dis_d[9] yb3nnlmx display data table 1-8: MB87J2120 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 262 215 dis_hsync yb3nnlmx display programmable sync 150 dis_pixclk b3nnnmx display pixel clock (programmable in/out) 157 dis_vref yb3nnlmx display programmable sync 18 dis_vsync yb3nnlmx display programmable sync 1 gnd gnd gnd 20 gnd gnd gnd 39 gnd gnd gnd 58 gnd gnd gnd 205 gnd gnd gnd 209 gnd gnd gnd 214 gnd gnd gnd 218 gnd gnd gnd 222 gnd gnd gnd 227 gnd gnd gnd 231 gnd gnd gnd 235 gnd gnd gnd 240 gnd gnd gnd 244 gnd gnd gnd 248 gnd gnd gnd 253 gnd gnd gnd 13 gnd[0] ovssx gnd 49 gnd[10] ovssx gnd 68 gnd[11] ovssx gnd 16 gnd[1] ovssx gnd 219 gnd[2] ovssx gnd 41 gnd[3] ovssx gnd 120 gnd[4] ovssx gnd 54 gnd[5] ovssx gnd 245 gnd[6] ovssx gnd 251 gnd[7] ovssx gnd 197 gnd[8] ovssx gnd 225 gnd[9] ovssx gnd table 1-8: MB87J2120 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 263 69 mode[0] itfuhx gdc mode pin (pull up) 70 mode[1] itfuhx gdc mode pin (pull up) 134 mode[2] itfuhx gdc mode pin (pull up) 65 mode[3] itfuhx gdc mode pin (pull up) 139 osc_in yi002aex xtal input 200 osc_out yb002aax xtal output 196 rstx itfuhx gdc reset (pull up) 64 sbp_bus bfunqhx spb interface 55 sdc_a[0] yb3nnlmx sdram address 63 sdc_a[10] yb3nnlmx sdram address 129 sdc_a[11] yb3nnlmx sdram address 194 sdc_a[12] yb3nnlmx sdram address 192 sdc_a[1] yb3nnlmx sdram address 131 sdc_a[2] yb3nnlmx sdram address 61 sdc_a[3] yb3nnlmx sdram address 62 sdc_a[4] yb3nnlmx sdram address 130 sdc_a[5] yb3nnlmx sdram address 193 sdc_a[6] yb3nnlmx sdram address 60 sdc_a[7] yb3nnlmx sdram address 247 sdc_a[8] yb3nnlmx sdram address 191 sdc_a[9] yb3nnlmx sdram address 59 sdc_cas yb3nnlmx sdram cas 51 sdc_cke yb3dnlmx sdram clock enable 238 sdc_clk b3nnnmx sdram clock 178 sdc_d[0] yb3nnlmx sdram data 119 sdc_d[10] yb3nnlmx sdram data 47 sdc_d[11] yb3nnlmx sdram data 122 sdc_d[12] yb3nnlmx sdram data 183 sdc_d[13] yb3nnlmx sdram data 185 sdc_d[14] yb3nnlmx sdram data 50 sdc_d[15] yb3nnlmx sdram data 52 sdc_d[16] yb3nnlmx sdram data 121 sdc_d[17] yb3nnlmx sdram data table 1-8: MB87J2120 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 264 123 sdc_d[18] yb3nnlmx sdram data 184 sdc_d[19] yb3nnlmx sdram data 179 sdc_d[1] yb3nnlmx sdram data 128 sdc_d[20] yb3nnlmx sdram data 53 sdc_d[21] yb3nnlmx sdram data 57 sdc_d[22] yb3nnlmx sdram data 186 sdc_d[23] yb3nnlmx sdram data 127 sdc_d[24] yb3nnlmx sdram data 124 sdc_d[25] yb3nnlmx sdram data 189 sdc_d[26] yb3nnlmx sdram data 241 sdc_d[27] yb3nnlmx sdram data 56 sdc_d[28] yb3nnlmx sdram data 187 sdc_d[29] yb3nnlmx sdram data 116 sdc_d[2] yb3nnlmx sdram data 126 sdc_d[30] yb3nnlmx sdram data 125 sdc_d[31] yb3nnlmx sdram data 45 sdc_d[3] yb3nnlmx sdram data 180 sdc_d[4] yb3nnlmx sdram data 181 sdc_d[5] yb3nnlmx sdram data 117 sdc_d[6] yb3nnlmx sdram data 46 sdc_d[7] yb3nnlmx sdram data 182 sdc_d[8] yb3nnlmx sdram data 118 sdc_d[9] yb3nnlmx sdram data 190 sdc_dmq yb3nnlmx sdram dqm 133 sdc_ras yb3nnlmx sdram ras 188 sdc_we yb3nnlmx sdram write enable 195 test yb3dnlmx fujitsu testpin 106 ulb_a[0] itfhx ulb interface address 37 ulb_a[10] itfhx ulb interface address 172 ulb_a[11] itfhx ulb interface address 109 ulb_a[12] itfhx ulb interface address 108 ulb_a[13] itfhx ulb interface address 173 ulb_a[14] itfhx ulb interface address table 1-8: MB87J2120 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 265 177 ulb_a[15] itfhx ulb interface address 114 ulb_a[16] itfhx ulb interface address 42 ulb_a[17] itfhx ulb interface address 43 ulb_a[18] itfhx ulb interface address 113 ulb_a[19] itfhx ulb interface address 169 ulb_a[1] itfhx ulb interface address 175 ulb_a[20] itfhx ulb interface address 38 ulb_a[2] itfhx ulb interface address 171 ulb_a[3] itfhx ulb interface address 110 ulb_a[4] itfhx ulb interface address 107 ulb_a[5] itfhx ulb interface address 174 ulb_a[6] itfhx ulb interface address 35 ulb_a[7] itfhx ulb interface address 230 ulb_a[8] itfhx ulb interface address 228 ulb_a[9] itfhx ulb interface address 34 ulb_clk itfhx ulb interface clock 104 ulb_cs itfhx ulb interface chip select 159 ulb_d[0] bfnnqmx ulb interface data 161 ulb_d[10] bfnnqmx ulb interface data 25 ulb_d[11] bfnnqmx ulb interface data 95 ulb_d[12] bfnnqmx ulb interface data 164 ulb_d[13] bfnnqmx ulb interface data 21 ulb_d[14] bfnnqmx ulb interface data 99 ulb_d[15] bfnnqmx ulb interface data 160 ulb_d[16] bfnnqmx ulb interface data 165 ulb_d[17] bfnnqmx ulb interface data 100 ulb_d[18] bfnnqmx ulb interface data 166 ulb_d[19] bfnnqmx ulb interface data 94 ulb_d[1] bfnnqmx ulb interface data 27 ulb_d[20] bfnnqmx ulb interface data 167 ulb_d[21] bfnnqmx ulb interface data 101 ulb_d[22] bfnnqmx ulb interface data 102 ulb_d[23] bfnnqmx ulb interface data table 1-8: MB87J2120 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 266 103 ulb_d[24] bfnnqmx ulb interface data 32 ulb_d[25] bfnnqmx ulb interface data 30 ulb_d[26] bfnnqmx ulb interface data 105 ulb_d[27] bfnnqmx ulb interface data 168 ulb_d[28] bfnnqmx ulb interface data 170 ulb_d[29] bfnnqmx ulb interface data 162 ulb_d[2] bfnnqmx ulb interface data 31 ulb_d[30] bfnnqmx ulb interface data 33 ulb_d[31] bfnnqmx ulb interface data 97 ulb_d[3] bfnnqmx ulb interface data 23 ulb_d[4] bfnnqmx ulb interface data 24 ulb_d[5] bfnnqmx ulb interface data 96 ulb_d[6] bfnnqmx ulb interface data 163 ulb_d[7] bfnnqmx ulb interface data 22 ulb_d[8] bfnnqmx ulb interface data 221 ulb_d[9] bfnnqmx ulb interface data 44 ulb_dack itfhx ulb interface dma acknowledge 112 ulb_deop bfnnqmx ulb interface dma stop output (dstop) 176 ulb_dreq otftqmx ulb interface dma request 40 ulb_intrq otftqmx ulb interface interrupt output 232 ulb_rdx itfhx ulb interface read 115 ulb_rdy otftqmx ulb interface ready 93 ulb_wrx[0] itfhx ulb interface byte 0 write enable 29 ulb_wrx[1] itfhx ulb interface byte 1 write enable 28 ulb_wrx[2] itfhx ulb interface byte 2 write enable 36 ulb_wrx[3] itfhx ulb interface byte 3 write enable 210 vdde vdde#5 io supply 3.3v 213 vdde vdde# io supply 3.3v 216 vdde vdde# io supply 3.3v 224 vdde vdde# io supply 3.3v 229 vdde vdde# io supply 3.3v 236 vdde vdde# io supply 3.3v table 1-8: MB87J2120 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 267 239 vdde vdde# io supply 3.3v 242 vdde vdde# io supply 3.3v 250 vdde vdde# io supply 3.3v 255 vdde vdde# io supply 3.3v 11 vdde[0] ovdd3x io supply 3.3v 111 vdde[10] ovdd3x io supply 3.3v 217 vdde[1] ovdd3x io supply 3.3v 98 vdde[2] ovdd3x io supply 3.3v 234 vdde[3] ovdd3x io supply 3.3v 48 vdde[4] ovdd3x io supply 3.3v 243 vdde[5] ovdd3x io supply 3.3v 132 vdde[6] ovdd3x io supply 3.3v 198 vdde[7] ovdd3x io supply 3.3v 135 vdde[8] ovdd3x io supply 3.3v 26 vdde[9] ovdd3x io supply 3.3v 207 vddi vddi# core supply 2.5 v 211 vddi vddi# core supply 2.5 v 220 vddi vddi# core supply 2.5 v 223 vddi vddi# core supply 2.5 v 226 vddi vddi# core supply 2.5 v 233 vddi vddi# core supply 2.5 v 237 vddi vddi# core supply 2.5 v 246 vddi vddi# core supply 2.5 v 249 vddi vddi# core supply 2.5 v 252 vddi vddi# core supply 2.5 v 7 vpd vpdx fujitsu tester pin 82 vref itamx dac testpin vref 73 vsc_alpha itfhx video scaler alpha 74 vsc_clkv itfhx video scaler clock 147 vsc_d[0] itfhx video scaler data input 203 vsc_d[10] itfhx video scaler data input 142 vsc_d[11] itfhx video scaler data input 143 vsc_d[12] itfhx video scaler data input table 1-8: MB87J2120 pinning sorted by name pin name buffer type description MB87J2120, mb87p2020-a hardware manual page 268 1.3.2 buffer types table 1-9 shows all used buffers for MB87J2120. 202 vsc_d[13] itfhx video scaler data input 75 vsc_d[14] itfhx video scaler data input 254 vsc_d[15] itfhx video scaler data input 83 vsc_d[1] itfhx video scaler data input 204 vsc_d[2] itfhx video scaler data input 141 vsc_d[3] itfhx video scaler data input 144 vsc_d[4] itfhx video scaler data input 201 vsc_d[5] itfhx video scaler data input 76 vsc_d[6] itfhx video scaler data input 72 vsc_d[7] itfhx video scaler data input 80 vsc_d[8] itfhx video scaler data input 199 vsc_d[9] itfhx video scaler data input 140 vsc_ident itfhx video scaler field identi?cation 256 vsc_vact itfhx video scaler vact 77 vsc_vref itfhx video scaler vertical refernce table 1-9: buffer types for MB87J2120 buffer type description b3nnlmx bidirectional true buffer (3.3v cmos, iol=4ma,low noise type) b3nnnmx bidirectional true buffer (3.3v cmos, iol=4ma) bfnnqmx 5v tolerant, bidirectional true buffer 3.3v cmos, iol/ioh=4ma bfunqhx bidirectional true buffer (5v tolerant, 25k pull-up, iol=8ma, high speed type) itamx analog input buffer itavdx analog power supply itavsx analog gnd itfhx 5v tolerant 3.3v cmos input itfuhx 5v tolerant 3.3v cmos input, 25 k pull-up otamx analog output otavx analog output table 1-8: MB87J2120 pinning sorted by name pin name buffer type description pinning for lavender and jasmine pinning and buffer types page 269 otftqmx 5 v tolerant 3.3v tri-state output, iol/ioh=4ma ovdd3x power supply for 3.3v vdd ovssx power supply for vss vpdx 3.3v cmos input, disable input for pull up/down resistors, connect to gnd yb002aax oscillator output yb3dnlmx bidirectional true buffer (3.3v cmos, 25k pull-down, iol=4ma,low noise type) yb3nnlmx bidirectional true buffer (3.3v cmos, iol=4ma,low noise type) yi002aex oscillator pin input table 1-9: buffer types for MB87J2120 buffer type description MB87J2120, mb87p2020-a hardware manual page 270 2 electrical speci?ation 2.1 maximum ratings the maximum ratings are the limit values that must never be exceeded even for an instant. as long as the device is used within the maximum ratings specified range, it will never be damaged. the cx71 series of cmos asics has five types of output buffers for driving current values, each of which has a different maximum output current rating. 2.1.1 power-on sequence lavender and jasmine are dual power supply devices. for power on/off sequence, there is no specific restriction, but the following sequences are recommended: power-on:vddi (internal, 2.5v)-> vdde (external, 3.3v)-> signal power-off:signal-> vdde (external, 3.3v)-> vddi (internal, 2.5v) it is restricted that vdde only is supplied continuously for more than 1 minute while vddi/dram supply is off. if the time exceeds 1 minute, it may affect the reliability of the internal transistors. when vdde is changed from off to on, the internal state of the circuit may not be maintained due to the noise by power supply. therefore, the circuit should be initialized after power is on. table 2-1: maximum ratings parameter symbol requirements unit supply voltage vddi vdde sdram_vcc apll_avdd dac_vdda -0.5 to +3.0 -0.5 to +4.0 vddi vddi -0.5 to +3.0 v input voltage vi -0.5 to vdd + 0.5 (<= 4.0v) a -0.5 to vdde + 4.0 (<= 6.0v) b a.for 3.3v interface b.for 5.0v tolerant v output voltage vo -0.5 to vdd +0.5 (<= 4.0v) a -0.5 to vdde + 4.0 b electrical specififcation electrical specification page 271 2.1.2 external signal levels external signal levels must not be higher than power supply voltage by 0.5v or more (3.3v inputs). if a signal with 0.5v or more than vdde is given to an input buffer the current will flow internally to supply, which can give a permanent damage to the lsi. in addition, when power supply becomes on or off, signal levels must not be higher than the power supply voltage by 0.5v or more. this means that signals must not be applied before power on / after power off. if an external signal (5v) is input at a 5v tolerant input before the device in question is powered-on, it will give the lsi a permanent damage. 2.1.3 apll power supply level apll (analog pll) power supply level must not be higher than power supply voltage vddi. please take care of apll power supply not to be over vddi level at power on/off sequence. 2.1.4 dac supply dac supply is isolated from other 2.5v supply. 2.1.5 sdram supply sdram supply must be as same level as vddi (jasmine only). MB87J2120, mb87p2020-a hardware manual page 272 2.2 recommended operating conditions the recommended operating conditions are the recommended values for assuring normal logic operation. as long as the device is used within the recommended operating conditions, the electrical characteristics described below are assured. 2.3 dc characteristics the dc characteristics assure the worst values of the static characteristics of input/output buffers within the range specified at the recommended operating conditions. table 2-2: operating conditions parameter symbol requirements unit min typ max supply voltage vdde 3.0 3.3 3.6 v vddi 2.3 2.5 2.7 dac_vdda 2.3 2.5 2.7 high-level input voltage 3.3v vih 2.0 - vdde+ 0.3 v 5v tolerant 2.0 - 5.5 low-level input voltage 3.3v vil -0.3 - 0.8 v 5v tolerant -0.3 - 0.8 junction temperature tj -40 - 125 ?c ambient temperature ta -40 - 85 ?c table 2-3: dc characteristics parameter symbol test conditions requirements unit min typ max supply cur- rent ab idds asic master type t7 - - 0.2 ma high-level out- put voltage voh ioh= -100ua vdde-0.2 - vdde v low-level out- put voltage vol iol= 100ua 0 - 0.2 v high-level out- put current ioh l type voh=vdde- 0.4v -2 - - ma m type voh=vdde- 0.4v -4 - - ma h type voh=vdde- 0.4v -8 - - ma electrical specififcation electrical specification page 273 following table shows current/power consumption for jasmine under special operating conditions. core clock, which has most influence is varied over specified range. please note, if other parameters varied that given values can be exceeded. measurement conditions: oscillator 12.0 mhz video clock 13.5 mhz, pixel clock (display) 6.0 mhz, ulb_clk 16 mhz vddi = 2.7v, vdde = 3.6v low-level out- put current iol l type vol=0.4v 2.0 - - ma m type vol=0.4v 4.0 - - ma h type vol=0.4v 8.0 - - ma input leakage current per pin b il - - +/-5 ua input pull-up/ pull-down resistor c rp 10 25 70 kohm output short- circuit current d ios l type - - +/-40 ma m type - - +/-60 ma h type - - +/-120 ma a.vih = vdd and vil = vss, memory is in stand-by mode, analog cells (apll, dacs, dac- vref) are at power down mode, tj = 25?c b.input pins have to be static. if an input buffer with pull-up/pull-down resistor is used, the input leakage current may exceed the above value c.either a buffer without a resistor or with a pull-up/pull-down resistor can be selected from the input and bidirectional buffers. d.maximum supply current at the short circuit of output and vdd or vss. for 1 second per pin. table 2-4: maximum core current consumption for jasmine frequency [mhz] apll divider/mul- tiplier setup a a.values interpreted with n+1 core/analog supply [ma] dram supply [ma] power consump- tion [mw] 16 5 / 7 84.1 9.2 (3.8) b b.values for dram supply in parenthesis are measured while running an usual application. 360 (346) 20 5 / 9 104.4 11.5 (4.8) 421 (403) 36 6 / 20 180.7 20.7 (9.1) 652 (621) 48 5 / 23 232.4 27.6 (10.6) 811 (765) 64 2 / 15 294.0 36.8 (12.6) 1001 (936) table 2-3: dc characteristics parameter symbol test conditions requirements unit min typ max MB87J2120, mb87p2020-a hardware manual page 274 i/o current assumed 30 ma, this varies in given environments/applications. part of i/o power con- sumtion was 30ma * 3.6v = 108mw (?xed within this mesurement environment). 2.4 mounting / soldering mounting and soldering is explained in fujitsu package data book, chapter 2. ac characteristics page 275 2.5 ac characteristics 2.5.1 measurement conditions for the determination of gdc timing values three independent conditions are used: 4. external capacities : 20pf and 50pf . internal capacitance of the pin itself is included in this measurement condition (20/50pf already including 5...7pf pin capacitance) and has to be subtracted from external capacitive load. 5. operating conditions : maximum conditions (min. voltage/max. temperature/slowest process) and minimum conditions (max. voltage/min. temperature/fastest process). the allowed values for voltage and temperature are described in the operating condition chapter. 6. timing path : longest and shortest (internal) path to/from a given output/input pin. for every timing type for input and output signals always the worst case for the environment circuits is listed in this specification. table 2-5 shows the used conditions for all signal types in this specification. input setup and hold timings are minimum requirements from the application point of view, thus specified in the ?min? column. however the input parameters are evaluated at maximum operating conditions and longest timing path (see table 2-5). for tri-state outputs the timing is evaluated for the 0 -> z and 1 -> z transitions without any load capacitance dependency. measurement points are defined if current through the output pin becomes below 5% of its nominal value. the voltage level at the output pin is not of interest when switching into high-z state. the timing of voltage level in high-z state depends only on the external rc combination and is not related to this ac specification. 2.5.2 de?itions figure 2-1 shows the definition of setup and hold relations for inputs (din1...3) and hold and delay relations for outputs (dout1...2) regarding the interface clock pin. din2, din3 and dout2 are special cases when negative setup or hold timings are given. table 2-5 shows the kind of timing values defined in this specification, its symbols and the evaluation con- ditions used to determine a specific value according to chapter 2.5.1. d2 d1 d2 d1 t t s h din1 h ? t s din3 ? t h din2 t t t od oh s dout1 t ? t oh t od dout2 clk figure 2-1: input and output timing relations MB87J2120, mb87p2020-a hardware manual page 276 for a given pin the symbol name is expanded by the pin name (e.g. t od ac characteristics page 277 rclk (routed to clkk, core clock without apll) rsv clock cycle time t cyr 15.625 ns - rsv clock high pulse width t phr 7 ns - rsv clock low pulse width t plr 7 ns - vsc_clkv (video interface clock) video clock cycle time t cyv 18.50 ns - video clock high pulse width t phv 9.25 ns - video clock low pulse width t plv 9.25 ns - dis_pixclk (routed to clkd as external display clock) display clock cycle time t cyd 18.50 ns - display clock high pulse width t phd 9.25 ns - display clock low pulse width t pld 9.25 ns - table 2-6: timing speci?cation clock parameter symbol min max MB87J2120, mb87p2020-a hardware manual page 278 2.5.4 mcu user logic bus interface table 2-7: ulb input signal timing speci?cation parameter symbol min [ps] max [ps] jas lav jas lav chip select (ulb_cs) setup time (falling) t scsf 1450 410 - chip select (ulb_cs) hold time (falling) t hcsf -130 2380 - chip select (ulb_cs) hold time (rising) a t hcsr * 1800 3120 - ulb_cs ulb_clk ulb_rdx ulb_a ulb_d ulb_wrx[n] t swrx t hdi swrx t hwrx t t hwrx ha t sa t t sdi di t hcsf t scsf t hcsr * t hwrx t hcsf figure 2-3: ulb write access (followed by another write) ulb_cs ulb_clk ulb_rdx ulb_a ulb_wrx[n] srdx tt t hcsf t scsf t hcsr hcsr t hcsf tt scsf t srdx acc t hrdx t exr t hrdx * * sa t ulb_d ulb_rdy t ohrdy t ohrdy t odd t do t odrdy odrdy t t rising edge first rdx or cs ha t t ohdrdx ohdrdx oddrdx figure 2-4: ulb read access ac characteristics page 279 read (ulb_rdx) setup time t srdx 980 5020 - read (ulb_rdx) hold time t hrdx 1890 2620 - write (ulb_wrx) setup time t swrx 5820 7890 - write (ulb_wrx) hold time t hwrx 1700 3850 - address (ulb_a) setup time b t sa 2600 2410 - address (ulb_a) hold time t ha 1910 3610 - input data (ulb_d) setup time t sdi 5200 1070 - input data (ulb_d) hold time t hdi 2130 3750 - access time t acc 1 t ulb_clk variable c external reaction time on ulb_rdy t exr 0 t ulb_clk - a. more restrictive speci?cation to rising edge of clk_ulb is only needed if spb will be used. b. address setup timing related to falling edge. c. access time varies with whole number of ulb_clk periods for different register addresses. required tim- ing is controlled with ulb_rdy handshake. table 2-8: ulb timing speci?cation, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. ready (ulb_rdy) output delay time t odrdy 13960 13070 15690 15110 min. ready (ulb_rdy) output hold time t ohrdy 4550 4560 4660 5190 max. output data (ulb_d) delay time (regard. clk) t odd 17980 21160 20060 23230 min. output data (ulb_d) hold time (regard. clk) t ohd 2640 3980 2640 4120 max. output data (ulb_d) delay time (regard. rdx) t oddrdx 12010 16490 14120 18560 min. output data (ulb_d) hold time (regard. rdx) t ohdrdx 2450 3560 3120 3750 table 2-7: ulb input signal timing speci?cation parameter symbol min [ps] max [ps] jas lav jas lav MB87J2120, mb87p2020-a hardware manual page 280 2.5.5 interrupt for a functional description of gdc interrupt controller and supported settings see user logic bus (ulb) controller description. this specification describes only the physical timing of interrupt request signal. table 2-9: intrq timing speci?cation parameter symbol min [ps] max [ps] jas lav jas lav interrupt (ulb_intrq) pulse width t pwi 2t / intreq a a. setup value in periods of ulb_clk, for edge sensitive mode only. in level sensitive mode minimum 2 cycles. - table 2-10: intrq timing speci?cation, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. interrupt (ulb_intrq) output delay time t odi 12360 12870 14400 14900 min. interrupt (ulb_intrq) output hold time t ohi 4020 4420 4360 4640 ulb_clk ulb_intrq l t t ohi ohi odi t odi t pwi t h/z figure 2-5: interrupt output timing ac characteristics page 281 2.5.6 dma control ports for a functional description of gdc dma controller and supported settings see user logic bus (ulb) con- troller description. this specification describes only the physical timing of dma control signals. table 2-11: dma timing speci?cation parameter symbol min [ps] max [ps] jas lav jas lav dma acknowledge (ulb_dack) setup time (rising) t sdack -80 50 - dma acknowledge (ulb_dack) hold time (rising) a a. jasmine has ulb_dack timing requirement for rising clock edge only. t hdack 1600 - - dma acknowledge (ulb_dack) hold time (falling) b b. lavender has ulb_dack timing requirement to both edge types. dack 1-> 0 transition only allowed between falling and rising clock edge. t hdack - 3130 - dma request (ulb_dreq) reaction time t rdreq 1t / 3t c c. ulb_dreq reaction time on ulb_dack. 1 cycle in block/step/burst mode. minimum 3 cycles in demand mode. - table 2-12: dma timing speci?cation, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. dma request (ulb_dreq) output delay time t oddreq 14140 15480 16170 17530 min. dma request (ulb_dreq) output hold time t ohdreq 4200 4460 4600 4690 ulb_dreq ulb_cs ulb_clk ulb_a ulb_rdx/wrx ulb_dack t t t ha sa t tt t ohdreq t oddreq t ohdreq oddreq rdreq sdack t t hdack (jas) t hdack (lav) t hcsf t scsf t hcsf hwrx/rdx hwrx/rdx t swrx/rdx swrx/rdx t figure 2-6: dma block/burst access MB87J2120, mb87p2020-a hardware manual page 282 2.5.7 display interface for display settings, supported colour formats, functional timing and configuration possibilities of display interface see graphic processing unit (gpu) description. this specification describes only the physical tim- ing of display signals. pixel clock output pin (dis_pixclk) has a clock invert option.thus timings are valid for both clock edges. the timing values given in table 2-14 and 2-15 are related to display clock output pin (dis_pixclk). the values are only valid if the capacitive load is the same for clock and data pins. because the driving clock edge for display clock (dis_pixclk) is programmable (see gpu description for more details) all possible edge combinations have to be taken into account for timing calculation. table 2-13: pixel clock output timing parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. pixel clock (dis_pixclk) output delay (r/f) a a. related to internal pixel_clk t odpixclk 15740 11240 17900 13410 min. pixel clock (dis_pixclk) output hold t ohpixclk 5240 3610 6090 4460 table 2-14: digital display interface, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. display data (dis_d) output delay time (r/f) t oddisd 3510 4540 4020 4930 min. display data (dis_d) output hold time (r/f) t ohdisd -690 -820 -430 -600 max. color key (dis_ckey/dis_ck) out- put delay (r/f) t odckey 1960 2880 2450 3320 min. color key (dis_ckey/dis_ck) out- put hold (r/f) t ohckey -800 200 -530 300 max. dis_vsync output delay (r/f) t odvsync 2800 3770 3220 4200 min. dis_vsync output hold (r/f) t ohvsync -4770 -1610 -6930 -3780 max. dis_hsync output delay (r/f) t odhsync 2420 3720 2840 4140 min. dis_hsync output hold (r/f) t ohhsync -5050 -1720 -7210 -3890 max. dis_vref output delay (r/f) t odvref 2560 3710 2980 4140 min. dis_vref output hold (r/f) t ohvref -4920 -1720 -7080 -3890 ac characteristics page 283 2.5.8 video input for a functional timing description of video input signals and for a description of supported video modes see video interface controller (vic) description. this specification describes only the physical timing of video input signals. 2.5.9 ccfl fet driver for pulse shape and functional timing of ccfl fet driver see ccfl description. this information is for capacitance influence on operating conditions and skew estimation only. table 2-15: analog display interface, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. disp. analog output delay time (r) a t oddisa 14200 14430 13350 13580 min. display analog output hold time t ohdisa 8410 9080 6250 6910 a. related to dis_pixclk output pin. the capacitive in?uence on the pixel clock output is greater than the in?uence on the analog outputs. due to subtracted clock delay the relative delay values are smaller in the 50pf column. table 2-16: video input timing parameter symbol min [ps] max [ps] jas lav jas lav video data (vsc_d) setup time t svid -360 -1660 - video data (vsc_d) hold time t hvid 2360 2800 - vsc_vref setup time t svivref -1940 -1580 - vsc_vref hold time t hvivref 2660 2450 - vsc_vact setup t svivact -2100 -1610 - vsc_vact hold t hvivact 2820 2510 - vsc_alpha setup t svialpha -2100 -1590 - vsc_alpha hold t hvialpha 2840 2540 - vsc_ident setup t sviidenti -2100 -1620 - vsc_ident hold t hviidenti 2850 2540 - table 2-17: ccfl driver outputs, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. ccfl_fet1 output delay time a t odccfet1 8460 5310 10490 7340 min. ccfl_fet1 output hold time t ohccfet1 3040 1830 3670 2460 max. ccfl_fet2 output delay time t odccfet2 8370 5270 10420 7290 MB87J2120, mb87p2020-a hardware manual page 284 2.5.10 serial peripheral bus for pulse shape and functional timing see spb documentation. spb timing regarding ulb_clk is not of interest. it is given for completeness only. 2.5.11 special and mode pins mode[3:0], rdy_trien, vpd, test are static configuration and test pins. resetx is asynchronous, thus no timing relation to any clk can be specified. maximum low pulse width suppressed by spike filter is 5.5 ns. only jasmine has a spike filter implemented. 2.5.12 sdram ports (lavender) see sdc documentation for description of external sdram interface timing configuration. different from measurement conditions above, sdram interface timing is evaluated from load capacitance connection of c l = 10pf for min conditions and c l = 30pf for max conditions. sdram clock output delay regarding internal core clock is shown in table 2-20. other sdram interface signals are specified regarding the sdc_clk clock output pin. same capacitive load is assumed for clock and address/command/data pins. if different load capacitance is connected to the sdc_clk pin than to the min. ccfl_fet2 output hold time t ohccfet2 3020 1830 3640 2450 max. ccfl_off output delay time t odccoff 8130 4730 10160 6760 min. ccfl_off output hold time t ohccoff 2920 1730 3540 2350 max. ccfl_ignit output delay time t odccignit 8200 4740 10250 6770 min. ccfl_ignit output hold time t ohccignit 2940 1710 3570 2330 a. for jasmine related to internal master_clk, rising edge; for lavender only the path delay was used. table 2-18: spb pin timing parameter symbol min [ps] max [ps] jas lav jas lav spb_bus input setup time (falling) t sspb -1530 -3780 - spb_bus input hold time t hspb 3100 4870 - table 2-19: spb pin timing, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav max. spb_bus output delay time t odspb 11470 14890 12500 14890 min. spb_bus output hold time t ohspb 2990 4670 3010 5110 table 2-17: ccfl driver outputs, output characteristics parameter symbol @20pf [ps] @50pf [ps] jas lav jas lav ac characteristics page 285 other sdram interface pins timing is not guaranteed to follow this specification. thus all connections to the sdram should have same wire length. due to configurable sdram interface timing, setup value of sdif register should be considered. the con- figuration register has two bits for each signal group: sdif_tao for address, control (ras/cas/we), cke and dqm output delays sdif_tdo for data output delay sdif_tdi for data input sampling delay and sdif_toe for controlling tri-state enable of the data output buffers. table 2-21 specifies the possible delay shifts, commonly valid for address, command, cke, dqm and data ports. sdram port timing specifications in table 2-22 are valid for a configured delay value 00 b . the user can add the given delay shifts to the appropriate minimum and maximum values for different con- figurations. note that additional delay shifts have to be subtracted for input setup timings and delay shifts have to be added for output delay, output hold and input hold timings. for data input this could be used to shift the data sampling point later with regard to the internal clock. for calculating shifted output parameters the user should take minimum delay shifts for min delay/hold and maximum delay shifts for max delay/hold. however input characteristics are always valid under max operating conditions and have to be shifted by delay shift value given in max column. input characteristics are given in the min column because they are minimum requirements. the timing values given in table 2-22 and table 2-23 are related to lavender sdram clock pin (sdc_clk). because the clock output path depends on external capacitance also the input setup and hold times are ca- pacitance dependend which is normally not the case (see also chapter 2.5.2). table 2-20: sdram clock output timing parameter symbol @10pf [ps] @30pf [ps] max. sdram clock (sdc_clk) output delay time t odsdcclk 7100 8470 min. sdram clock (sdc_clk) output hold time t ohsdcclk 2850 3480 table 2-21: con?gurable delay shifts by sdif sdif setup min delay shift [ps] max delay shift [ps] 00 b 00 01 b 650 1920 10 b 1240 3620 11 b 1720 5030 MB87J2120, mb87p2020-a hardware manual page 286 table 2-22: sdram port timing for lavender parameter symbol @10pf[ps] @30pf[ps] sdram data (sdc_d) input setup time (max cond.) a a. related to sdram clock pin (sdc_clk). t ssdcd 2520 3890 sdram data (sdc_d) input hold time (max cond.) t hsdcd -850 -1480 table 2-23: sdram port timing for lavender, output characteristics parameter symbol @10pf [ps] @30pf [ps] max. sdram address (sdc_a) output delay time t odsdca 3870 4250 min. sdram address (sdc_a) output hold time t ohsdca 1280 1410 max. sdram command output delay time (sdc_ras/sdc_cas/sdc_we) t odsdccmd 3820 4180 min. sdram command output hold time (sdc_ras/sdc_cas/sdc_we) t ohsdccmd 1270 1400 max. sdram clock enable (sdc_cke) output delay time t odsdccke 3520 3930 min. sdram clock enable (sdc_cke) output hold time t ohsdccke 1150 1290 max. sdram data enable/mask (sdc_dmq) out- put delay time t odsdcdqm 3800 4180 min. sdram data enable/mask (sdc_dmq) out- put hold time t ohsdcdqm 1280 1410 max. sdram data (sdc_d) output delay time t odsdcd 6160 6590 min. sdram data (sdc_d) output hold time t ohsdcd 1240 660 page 287 part d - appendix MB87J2120, mb87p2020-a hardware manual page 288 page 289 d-1 jasmine command and regis- ter description MB87J2120, mb87p2020-a hardware manual page 290 register description register description page 291 1 register description table 1-2 shows all registers and bit groups of mb87p2020-a (jasmine) and MB87J2120 (lavender) with a short explanation. a more detailed description can be found in the component specific manual parts. in table 1-2 valid registers or bitgroups for every device are marked in additional columns. some registers or bitgroups are listed twice, separated for every device, since they are different for lavender and jasmine. additionaly some special registers exist which are marked in table 1-2 with a special code explained in table 1-1. table 1-1: marking for special registers mark description r read only register or bitgroup w write only register or bitgroup f hardware ?ag (value can be manipulated by hardware) m register write access is locked if mtimon_on=1 (gpu master switch) c these registers have an internal shadow register which is synchronized to command execution. if a command is running the shadow register is locked, written values are stored and take effect only when next command is started. for reading always the user writeable register is returned. d like ?c? but reading is con?gurable with readinit_rcr table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address ulb (user logic bus controller) cmd 0x0000 command register 31:8 par o o c command parameter 0 7:0 code o o c command code (for writing check flnom_cwen ) 0xff (noop) ififo 0x0004 31:0 - o o w input fifo only word access is allowed. - ofifo 0x0008 31:0 - o o r output fifo only word access is allowed. - flnom 0x000c 31:0 - o o f flag register (normal write access) a 0x20400000 flrst 0x0010 31:0 - o o f flag register (reset write access) a 1: reset ?ag at this position 0x20400000 MB87J2120, mb87p2020-a hardware manual page 292 flset 0x0014 31:0 - o o f flag register (set write access) a 1: set ?ag at this position 0x20400000 intnom 0x0018 31:0 - o o interrupt mask register (nor- mal write access) a 1: use ?ag for interrupt 0 intrst 0x001c 31:0 - o o interrupt mask register (reset write access) a 1: reset mask at this position 0 intset 0x0020 31:0 - o o interrupt mask register (set write access) a 1: set mask at this position 0 intlvl 0x0024 31:0 oo interrupt level/edge settings a 1: positive edge of ?ag trig- gers interrupt 0: high level of ?ag triggers interrupt 0 intreq 0x0028 interrupt request length 5:0 intclk o interrupt request length in ulb clocks (clkm) 0x10 rdyto 0x002c rdy timeout control register 7:0 rdyto o rdy timeout length in ulb clocks (clkm) 0xff 8 rdtoen o rdy timeout enable 1: enable rdy timeout 1 rdyaddr 0x0030 rdy timeout address register (read only) 20:0 addr o r address where rdy timeout occurred 0 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 293 ifctrl 0x0034 mcu interface control regis- ter 9:8 smode o sample mode for bus control signals 00: 3 point mode 01: 2 of 3 point mode 10: 2 point mode 11: 1 point mode 0 5 drinv o 1: invert ulb_dreq 0 4 dsinv o 1: invert ulb_dstp 0 3 intinv o 1: invert ulb_intrq 0 2 drtri o 1: open drain for ulb_dreq 0 1 dstri o 1: open drain for ulb_dstp 0 0 inttri o 1: open drain for ulb_intrq 0 wndof0 0x0040 20:0 off o o mcu offset for sdram window 0 0x10?000 wndsz0 0x0044 20:0 size o o size of sdram window 0 0x10?000 wndof1 0x0048 20:0 off o o mcu offset for sdram window 1 0x10?000 wndsz1 0x004c 20:0 size size of sdram window 1 0x10?000 wndsd0 0x0050 19:0 off o sdram offset for sdram window 0 0 22:0 o wndsd1 0x0054 19:0 off o sdram offset for sdram window 1 0 22:0 o sdflag 0x0058 0 en o o ?1?: enable sdram space ?0?: any access to sdram space is ignored 0 iful 0x0080 input fifo limits 22:16 ul o input fifo upper limit for ?ag- or interrupt controlled ?ow control ifh=1 if ifload >= iful_ul 0x0c 23:16 o 6:0 ll o input fifo lower limit for ?ag- or interrupt controlled ?ow control ifl=1 if ifload <= iful_ll 0x03 7:0 o table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 294 oful 0x0084 output fifo limits 22:16 ul o output fifo upper limit for ?ag- or interrupt controlled ?ow control ofh=1 if ofload >= oful_ul 0x3c 23:16 o 6:0 ll output fifo lower limit for ?ag- or interrupt controlled ?ow control ofl=1 if ofload <= iful_ll 0x0f ifdma 0x0088 input fifo limits for dma transfer 22:16 ul o upper limit for dma access to input fifo (not used) 0x0a 23:16 o 6:0 ll o lower limit for dma access to input fifo 0x3c 7:0 o ofdma 0x008c output fifo limits for dma transfer 22:16 ul o upper limit for dma access from output fifo 0x0a 23:16 o 6:0 ll o lower limit for dma access from output fifo (not used) 0x3c 7:0 o table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 295 dmaflag 0x0090 dma ?ag register 12:8 dstp o o duration of dstp signal. this value can be set in order to ensure a save mcu- dmac reset. normally the default value should work. 7 4 tri o 1: open drain for ulb_dreq , ulb_dstp , ulb_intrq 0 3 inv o 1: invert ulb_dreq , ulb_dstp , ulb_intrq 0 2 mode o o ?1?: dma demand mode ?0?: dma block/step- or burst mode 0 1en o o ?1?: enable dma 0 0io o o ?1?: use dma for input fifo ?0?: use dma for output fifo 1 flagres 0x0094 flag behaviour register a 31:0 - o o 1: set ?ag to dynamic behav- iour b 0: set ?ag to static behaviour 0x20400000 ulbdeb 0x0098 fifo debug register (read only) 23:16 iflc o r input fifo load for current command attention : this value changes with core clock; cor- rect sampling by mcu can?t be ensured. 0x00 22:16 o 15:8 of o r output fifo load attention : this value changes with core clock; cor- rect sampling by mcu can?t be ensured. 0x00 14:8 o 7:0 if o r input fifo load independent from current command attention : this value changes with core clock; cor- rect sampling by mcu can?t be ensured. 0x00 6:0 o table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 296 cmddeb 0x009c command debug register (read only) 31:8 par o r parameter for currently exe- cuted command 0x000000 7:0 cmd o r code for currently executed command 0xff sdc (sdram controller) sdseqram [32] (jas- mine) sdseqram [64] (laven- der) 0x0100- 0x017c (jas- mine) 0x0100- 0x01fc (laven- der) sdram sequencer ram (32 words) 12:7 addr o microcode sequencer address argument undef 11:7 o 6:4 inst o o microcode instruction: run, ret, call, loop, srw, rrw, pde, pdx - coded 0 to 7) undef 3 ras o o container command for sdram (ras) undef 2 cas o o container command for sdram (cas) undef 1we o o container command for sdram (we) undef sdmode 0x0200 sdram mode register (external sdram mrs reg- ister; lavender only), bit [12:10, 8:7] reserved, set to ?0? 9 brst o write enable 0: burst 1: single 0 6:4 cl o cas latency (2/3) 0x3 3 ilb o 1: interleave burst 0: sequential burst 0 2:0 blen o burst length (0=1, 1=2, 2=4, 3=8, 7=full) 0x3 sdinit 0x0204 15:0 ip o o sysclocks of sdram power up initialization period (200 us) 0x4e20 sdrfsh 0x0208 15:0 rp o o sysclocks of sdram row refresh period (16 us) 0x0640 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 297 sdwait 0x020c sdram timings (refer to sdram manual) 20 opt o bank interleave optimization 1 19:16 trp o o ras precharge time - 1 0x2 15:12 trrd o ras to ras bank active delay time - 1 1 11:8 tras o o ras active time - 1 0x5 7:4 trcd o o ras to cas delay time - 1 0x2 3:0 trw o o read to write recovery time recommended values: lavender: 10 (8) c jasmine: 7 (5) c 0x3 value for- bidden! use recom- mended val- ues! sdif 0x0210 sdram port interface timing (scalable clock delay) - is ignored by jasmine 7:6 tao o address output delay 0 5:4 tdo o data output delay 0 3:2 tdi o data sampling delay 0 1:0 toe o tristate control delay 0 sdcflag 0x0214 sdc control register 1 dqmen o 1: enable dqm partial write optimization 0 0 busy o o set busy ?ag during micro- program upload 0 gpu (graphic processing unit) gpu-ldr (layer description record) pha[16] 0x1000 - 0x103c physical layer address in sdram 19:0 ofs o address offset (ra, ca, byte) bits[9:0] ?xed to zero undef 22:0 o address offset (ra, ba, ca, byte) bits[11:0] ?xed to zero dsz[16] 0x1040 - 0x107c layer domain size 29:16 o o o o dimension of layer size undef table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 298 dp1[16] 0x1080 - 0x10bc first pixel in domain (mem- ory offset) 29:16 o o o offset for o dimension undef 13:0 y o o offset for y dimension undef wsz[16] 0x10c0 - 0x10fc display window size for layer 29:16 o o o size in o dimension undef 13:0 y o o size in y dimension undef wof[16] 0x1100 - 0x113c layer offset for display 29:16 o o o offset for o dimension undef 13:0 y o o offset for y dimension undef ltc[16] 0x1140 - 0x117c transparent colour for layer 23:0 col o o colour code depending on layer colour depth (lsb aligned) undef lbc[16] 0x1180 - 0x11bc blink colour for layer 23:0 col o o colour code depending on layer colour depth (lsb aligned) undef lac[16] 0x11c0 - 0x11fc blink alternative colour for layer 23:0 col o o colour code depending on layer colour depth (lsb aligned) undef lbr[16] 0x1200 - 0x123c blink rate for layer 15:8 off o o number of frames for alter- nate colour undef 7:0 on o o number of frames for blink colour undef cspc[16] 0x1240 - 0x127c colour space code and ?ag register 9 lde o o line doubling enable undef 8te o o transparency enable undef 3:0 csc o o colour space code undef gpu-mdr (merging description record) table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 299 cformat 0x1300 31:16 litc o o layer to intermediate trans- fer codes (one bit for each layer) 0x0000 7 gamen o 1: gamma table enable 0 6 ipolen o 1: interpolation for yuv422 colour code enable 0 5 gforce o 1: force gamma conversation for rgb>= 16bpp 0 3:0 csc o o intermediate colour space code 0x0 backcol 0x1304 background colour register 24 en o o 1: background colour enable 1 23:0 col o o background colour depend- ing on colour depth for inter- mediate colour space 0x000000 mbc 0x1308 blink control register 15:0 en o o blink enable (one bit for each layer) 0x0000 31:16 cbs o o r current blink state (read only) 0x0000 zorder 0x130c z-order register 19:16 en o o enable (one bit for each plane) 0x0 15:12 tm3 o o topmost layer number 0x0 11:8 tm2 oo 0x0 7:4 tm1 oo 0x0 3:0 tm0 o o bottom layer number 0x0 clutof [16] 0x1340 - 0x137c 16 clut offsets (1 per layer) 7:0 ofs o clut offset for layer undef 8:0 o drm[14] 0x1380- 0x13b4 duty ratio modulator pseudo levels, 14 words 5:0 pgl o o pseudo level (pgl/64 frames) 0x00 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 300 prebias 0x1400 pre matrix bias values 23:16 y o y value 0xf0 15:8 cb o cb value 0x80 7:0 cr o cr value 0x80 coeffr 0x1404 matrix coef?cients to red channel 23:16 yxr o y to red 0x80 15:8 cbxr o cb to red 0x00 7:0 crxr o cr to red 0xb3 coeffg 0x1408 matrix coef?cients to green channel 23:16 yxg o y to green 0x80 15:8 cbxg o cb to green 0x2c 7:0 crxg o cr to green 0x5b coeffb 0x140c matrix coef?cients to blue channel 23:16 yxb o y to blue 0x80 15:8 cbxb o cb to blue 0xe2 7:0 crxb o cr to blue 0x00 clut [512] (jas- mine) clut [256] (laven- der) 0x2000 - 0x27fc (jas- mine) 0x2000 - 0x23fc (laven- der) colour look-up table 23:0 col o o colour for intermediate col- our space undef gamma [256] 0x2800 - 0x2bfc gamma table 23:16 r o red mapping undef 15:8 g o green mapping undef 7:0 b o blue mapping undef gpu-dir (display interface record) phsize 0x3000 display physical size 29:16 o o o m width 0x0000 13:0 y o o m height 0x0000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 301 phfrm 0x3004 physical format register 27 pol o o m xfref polarity 0=odd ?eld ref is low 0 26 vsyae o o m xvsync edge (0=low/high edge, 1=high/low edge) 0 25 hsyae o o m xhsync edge (0=low/high edge, 1=high/low edge) 0 24 ies o o m 0: internal sync 1: external sync 0 18 tdm o m 1: enable twin display mode 0 17:16 sm o o m scan mode 0 12 fte o o m 1: field toggle enable 0 11:8 bsc o o m bitstream format code 0x0 4 rbsw o o m swap r/b channel for rgb111 0 3:0 csc o o m physical colour space code 0x0 phscan 0x3008 29:16 sclk o o m physical size in scan clocks (pixel*bpp/bitsperscan- clock) 0 dualscof 0x300c 13:0 ofs o o m dual scan y offset 0 mtimodd [2] 0x3010 - 0x3014 master timing, odd ?eld, first word start, next word stop. 30:16 o o o m o dimension (2?s comple- ment) 0x0000 14:0 y o o m y dimension (2?s comple- ment) 0x0000 mtmeven [2] 0x3018 - 0x301c master timing, even ?eld, first word start, next word stop. (o values from mti- modd [0,1]) 14:0 y o o m y dimension (2?s comple- ment) 0x0000 mtimon 0x3020 0 on o o master timing switch (0=off,1=on) 0 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 302 timdiag 0x3024 diagnostic timing position output (read only) 30:16 o o o r current o position (2?s com- plement) 0x0000 15 field o o r current ?eld 0 14:0 y o o r current y position (2?s com- plement) 0x0000 spg [6,4] 0x3030 - 0x308c these registers contain 6 sync pulse generators (spg0-spg5) with each 4 registers. 0 spgpson [6] 0: 0x3030, 1: 0x3040, 2: 0x3050, 3: 0x3060, 4: 0x3070, 5: 0x3080 position to switch on 30:16 o o o o position (2?s complement) 0x0000 15 field o o field ?ag 0: odd; 1: even 0 14:0 y o o y position (2?s complement) 0x0000 spgmkon 0: 0x3034, 1: 0x3044, 2: 0x3054, 3: 0x3064, 4: 0x3074, 5: 0x3084 don?t care vector for ?posi- tion on? match. (1=do not include this bit into position matching) 30:16 o o o o mask 0x0000 15 field o o field mask 0 14:0 y o o y mask 0x0000 spgpsof 0: 0x3038, 1: 0x3048, 2: 0x3058, 3: 0x3068, 4: 0x3078, 5: 0x3088 position to switch off 30:16 o o o o position (2?s complement) 0x0000 15 field o o field ?ag (0=odd, 1=even ?eld) 0 14:0 y o o y position (2?s complement) 0x0000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 303 spgmkof 0: 0x303c, 1: 0x304c, 2: 0x305c, 3: 0x306c, 4: 0x307c, 5: 0x308c don?t care vector for ?posi- tion off? match (1=do not include this bit into position matching) 30:16 o o o o mask 0x0000 15 field o o field mask 0 14:0 y o o y mask 0x0000 ssqcycle 0x30fc 5:0 sc o actual length of sequencer cycle ( sc = num -1) 0x0 4:0 o ssqcnts [64] 0x3100 - 0x31fc sync sequencer ram con- tents (64 words) 31 out o o output value for scan position undef 30:16 seqx o o o scan position (2?s comple- ment) undef 15 field o o field ?ag (0: odd;1: even) undef 14:0 seqy o o y scan position (2?s comple- ment) undef smx [8,2] 0x3200 - 0x323c array de?nition for sync mix- ers (smx0-smx7 with each 2 registers) 0 smxsigs 0: 0x3200, 1: 0x3208, 2: 0x3210, 3: 0x3218, 4: 0x3220, 5: 0x3228, 6: 0x3230, 7: 0x3238 sync mixer 14:12 s4 o o signal select signal to select: 0:const.zero 1:sequencer out 2...7: output of spg0...spg5 0 11:9 s3 oo 0 8:6 s2 oo 0 5:3 s1 oo 0 2:0 s0 oo 0 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 304 smxfct 0: 0x3204, 1: 0x320c, 2: 0x3214, 3: 0x321c, 4: 0x3224, 5: 0x322c, 6: 0x3234, 7: 0x323c function table 31:0 ft o o output value = function_table[a] a= s4 *2 4 + s3 *2 3 + s2 *2 2 + s1 *2 1 + s0 *2 0 sn : value (binary) of signal selected with smxsigs_sn 0x00000000 sswitch 0x3240 output signal delay (sync switch) register 7:0 cd o o m sync switch, (0=no; 1=0.5 display clock (clkd) cycles delay) 0x00 pixclkgt 0x3248 pixel clock gate register; gate is output of sm7 3hc o o m clock divider (0=1:1;1=1:2) 0 2cp o o m clock polarity (0=true; 1=inverted) 0 1 gon o o m clock gate enable (1=on/ 0=off) 0 0gt o o m gate type (0=and; 1=or) 0 cklow 0x3250 colour key lower limits (according to physical colour space) pin dis_ckey is activated when all pixel channels lie within their limits (including limits). 24 op o o key out polarity (0=active high, 1=active low) 0 23:16 llr o o red channel 0x00 15:8 llg o o green channel 0x00 7:0 llb o o blue/monochrome channel 0x00 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 305 ckup 0x3254 colour key upper limits (according to physical colour space) pin dis_ckey is activated when all pixel channels lie within their limits (including limits). 23:16 ulr o o red channel 0x00 15:8 ulg o o green channel 0x00 7:0 ulb o o blue/monochrome channel 0x00 aclamp 0x3258 jasmine: clamping values for analog outputs (dacs) lavender: clamping values for analog outputs (dacs) and digital output 23:16 aclr o red channel 0x00 15:8 aclg o green channel 0x00 7:0 aclb o blue channel 0x00 23:0 dao o clamping value for analog and digital output 0x000000 dclamp 0x325c 23:0 dcl o blanking clamping value for digital outputs 0x000000 mainen 0x3260 main display output enable ?ags 29 refoe o 1: (internal) reference volt- age disable 0: (internal) reference volt- age enable for lavender this reference is connected to dacoe . 0 28 dacoe o o 1: dac output ( a_blue , a_green , a_red ) enable 0 27 ckoe oo 1: dis_ckey output enable 0 26 vroe oo 1: dis_vref output enable 0 25 vsoe oo 1: dis_vsync output ena- ble 0 24 hsoe oo 1: dis_hsync output ena- ble 0 23:0 doe o o 1: dis_d[23:0] output enable 0x000000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 306 gpu-sdcr (sdc record) sdcp 0x3270 sdram controller request priorities 15:8 ifl o o r input fifo load (read-only) 0x00 6:4 hp o o high priority 0x7 2:0 lp o o low priority 0x3 vic (video interface controller) vicstart 0x4000 input layer start coordinates (memory offset for video) 29:16 o o o o offset 0x0000 13:0 y o o y offset 0x0000 vicalpha 0x4004 replace colour when alpha pin ( vsc_alpha ) is active colour depth depends on layer 23:0 col o o alpha colour 0x000000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 307 vicctrl 0x4008 input control word 7 bswap o swap external channels a and b to internal channels ia and ib. 0: a->ia, b->ib 1: a->ib, b->ia 0 6 alen o o 1:enable alpha; 0:disable alpha 0 5 port o o port mode 1: double port (port ia and ib) 0: single port (port ia only) 1 4 clock o o video clock ( vsc_clkv ) mode for data sampling 1: double clock mode (both edges) 0: single clock mode (rising edge only) 0 3:0 mode o o colour mode for video input: 0x4: rgb555 0x5: rgb565 0x6: rgb888 0x7: yuv422 0x8: yuv444 0xe: yuv555 0xf: yuv655 0x6 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 308 vicfctrl 0x400c field control word 22 oddfst o o field order within a frame: 1:odd ?eld is top ?eld 0:even ?eld is top ?eld 0 21 frame o o video mode 1:frame mode (interleave ?elds in one layer) 0:field mode (store one ?eld in one layer) 0 20 skip o o field skip enable for selected ?elds 1: skip every 2nd ?led of each type 0:use every ?eld 0 18 vicen o o 1: general vic enable 0 17 evenen o o 1: enable even ?elds 0 16 odden o o 1: enable odd ?elds 0 11:8 trd o o 3rd layer number 0x7 7:4 sec o o 2nd layer number 0x0 3:0 fst o o 1st layer number 0x6 vicpctrl 0x4010 polarity settings for video control pins 3 alpha oo vsc_alpha: 1: low active 0: high active 0 2 field oo vsc_ident : 1: odd ?eld low active 0: odd ?eld high active 0 1 vact oo vsc_vact : 1: low active, 0: high active 0 0 vref oo vsc_vref : 1: low active, 0: high active 0 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 309 vicfsync 0x4014 control register for video i/o synchronization 16 rel o o 1: vic is faster than gpu 0: gpu is faster than vic 0 15 sync o o 1: three layer mode 0: two layer mode 0 14:0 swl o o switch level for layer switch in two layer sync mode 0 viccsync 0x4018 control word for clock syn- chronization (lavender only) 1:0 mode o sync mode 00: core clock > video clock 01: core clock > 2*video clock 10: core clock = video clock 11: reserved 00 sdram 0x401c sdram request priority con- trol register 2:0 lp o o low priority 0x2 6:4 hp o o high priority 0x6 vicbsta 0x4020 o o r vic status register for test purpose (read only) vicrlay 0x4024 video layer debug register for test purpose (read only) 19:16 aovl o r current output layer (to gpu) undef 11:8 livl o r last input layer (vic) undef 3:0 aivl o r current input layer (vic) undef table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 310 vicvisyn 0x4028 video path control register 25:24 del o negative data delay with respect to vsc_vact signal 0 20:16 shuff o bus shuf?er for data ports ia, ib, ia_delayed and ia_negedge 0 8 start o selector for vic_sync ?ag a source: 0: video write start 1: video read start 0 1:0 sel o selector for video input pro- tocol: 00: vpx 01: ccir-trc 10: ccir external sync 11: reserved 00 viclimen 0x402c 0 limena o 1: enable video limitation unit 0 viclimh 0x4030 horizontal limitation settings 26:16 hen o horizontal window length including offset 0x000 10:0 hoff o horizontal window offset 0x000 viclimv 0x4034 vertical limitation settings 26:16 ven o vertical window length including offset 0x000 10:0 voff o vertical window offset 0x000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 311 extpctrl 0x4038 polarity control register for ccir mode with external video synchronization. 3 alpha o alpha (pin vsc_alpha ) 0: high active 1: low active 0 2 parity o parity (pin vsc_ident ) 0: polarity unchanged 1: invert polarity 0 1 href o horizontal reference signal (pin vsc_vact ) 0: high active 1: low active 0 0 vref o vertical reference signal (pin vsc_vref ) 0: high active 1: low active 0 pp (pixel processor) bgcol 0x4100 background pixel colour 24 en o o d 1: enable background colour 0 23:0 col o o d background colour data 0x000000 fgcol 0x4104 23:0 o o d foreground pixel colour 0x000000 ignorcol 0x4108 ignored pixel colour (putbm, putcp) 24 en o o d 1: enable ignore colour 0 23:0 col o o d ignore colour data 0x000000 linecol 0x410c 23:0 col o o d line colour (dwline) 0x000000 pixcol 0x4110 23:0 col o o d colour for pixel with ?xed colour (putpxfc) 0x000000 plcol 0x4114 23:0 col o o d polygon colour (dwpoly) 0x000000 rectccol 0x4118 23:0 col o o d rectangle colour (dwrect) 0x000000 xymax 0x411c pixel stop address for pixel processor bitmap commands (putbm, putcp, puttxtbm, puttxtcp) 29:16 xmax o o d o dimension for stop point 0x0000 13:0 ymax o o d y dimension for stop point 0x0000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 312 xymin 0x4120 start address for pixel proces- sor bitmap commands (putbm, putcp, puttxtbm, puttxtcp) 29:16 xmin o o d o dimension for start point 0x0000 13:0 ymin o o d y dimension for start point 0x0000 ppcmd 0x4124 con?guration for pixel proc- essor commands 28 ulay o d 1: use target layer for draw- ing commands 0 27:24 lay o o d target layer for commands putbm, putcp, puttxtbm,puttxtcp 0x0 8 dir o o d direction for sequential com- mands putbm, putcp, puttxtbm, puttxtcp 0: horizontal, 1:vertical 0 1:0 mir o o d mirror for sequential com- mands putbm, putcp, puttxtbm, puttxtcp 00: no mirror, 01: o-mirror, 10: y-mirror,11:xy-mirror 0x0 sdcprio 0x4128 2:0 prio o o d priority for sdc interface 0x0 reqcnt 0x412c 7:0 mfb ood mfb +1: minimum fifo- block size before activating a sdram request (0<= mfb <64) 0x00 readinit 0x4130 0 rcr o o read control registers for pixel processor (address range: 0x4100-0x4130) 0: read back pp internal regis- ter 1: read back pp user writea- ble register 0 dipa (direct and indirect physical memery access) dipactrl 0x4200 dipa control register 18:16 pdpa o o dpa priority for sdc access 0x0000 2:0 pipa o o ipa priority for sdc access 0x0000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 313 dipaif 0x4204 ipa input fifo control regis- ter 23:16 max o input fifo max. block size 0x0002 22:16 o 7:0 min o input fifo min. block size 0x0002 6:0 o dipaof 0x4208 ipa output fifo control reg- ister 23:16 max o output fifo max. block size 0x0001 22:16 o 7:0 min o output fifo min. block size 0x0001 6:0 o ccfl (cold cathode fluorescence light driver) ccfl1 0x4400 ccfl control register 28 ssel o synchronization select: 0: use sncs ?ag 1: sync to output of gpu sync mixer 5 0 27 en o o 1: ccfl enable 0 26 prot o o 1: protect old settings during con?guration 0 25 sncs o o synchronization select: 0: internal (vsync from gpu), 1: software 0 24 sync o o 1: synchronization trigger by software 0 7:0 scl o o timebase scale factor (derived from system clock (clkk)) 0x00 ccfl2 0x4404 ccfl duration register (unit: 4 timebase clocks) 31:16 fls o o flash duration 0x0000 15:8 pse o o pause duration 0x00 7:0 ignt o o ignition duration 0x00 aaf (anti aliasing filter) table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 314 aatr 0x4500 thresholds for red channel 15:8 th2 oo threshold2 0x00 7:0 th1 oo threshold1 0xff aatg 0x4504 thresholds for green channel 15:8 th2 oo threshold2 0x00 7:0 th1 oo threshold1 0xff aatb 0x4508 thresholds for blue channel 15:8 th2 oo threshold2 0x00 7:0 th1 oo threshold1 0xff aaoe 0x450c aaf options 1 bx4 o 1: aaf block size 4x4 0: aaf block size 2x2 0 0 wb0 oo 1: enable write back optimi- zation 0 cu (clock unit) table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 315 clkcr 0xfc00 clock con?guration register 31:30 dcs o o direct clock source 00: crystal (pins osc_in and osc_out ), 01: pixel clock (pin dis_pxclk ) 10: mcu clock (pin ulb_clk ) 11: reserved clock (pin rclk ) 00 29:24 scp o o system clock (clkk) pres- caler 0 23:22 pcs o o pll clock source 00: crystal (pins osc_in and osc_out ), 01: pixel clock (pin dis_pxclk ) 10: mcu clock (pin ulb_clk ) 11: reserved clock (pin rclk ) 00 21:16 pfd o o pll feedback divider 0x00 15 scsl o o system clock (clkk) select 0: direct clock source 1: pll clock source 0 14 pcsl o o pixel clock (clkd) select 0: direct clock source 1: pll clock source 0 13 ipc o o pixel clock (clkd) invert 0: not inverted 1: inverted 0 12 pcod o o pixel clock output ( dis_pxclk ) disable 0: internal pixel clock ( clkd output) 1: external pixel clock ( dis_pxclk input) 1 11 dbg o 1: core clock (clkk) debug mode (output at pin spb_tst ) 0 10:0 pcp o o pixel clock prescaler value 0x000 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 316 clkpdr 0xfc04 clock power down register 31:24 id o o r chip id (read only) 00: MB87J2120 (lavender) 01: mb87p2020(-a) (jas- mine) 01 23:16 sid o o r chip sub-id (read only) 00: MB87J2120 (lavender) and mb87p2020 (jasmine) 01: mb87p2020-a (jasmine redesign) 01 15 mrst o w master hardware reset write 1: start reset write 0: stop reset read: returns always ?0? 0 o f master hardware reset write 1: start reset (stops automatically) write 0: no effect read: returns current value 14 lck o r pll lock (read only) 1: pll has locked to input frequency undef 13 - o reserved; set to 0 0 12 vii o 1: invert video clock (clkv) 0 o r pll lock (read only) 1: pll has locked to input frequency undef 11 run o o pll enable 1: pll on 0: pll off 0 10 gpu o o 1: enable gpu clocks 0 9 ulb o o 1: enable ulb clocks 0 8 spb o o 1: enable spb clock 0 7 ccfl o o 1: enable ccfl clock 0 6 sdc o o 1: enable sdc clock 0 5 vis o o 1: enable vic clocks 0 4 dipa o o 1: enable dipa clock 0 3 aaf o o 1: enable aaf clock 0 2 mcp o o 1: enable pp-mcp clock 0 table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address register description register description page 317 1 mau o o 1: enable pp-mau clock 0 0pe o o 1: enable pp-pe clock 0 a. see chapter 2 for a description of bit groups for ?ags. b. dynamic behaviour means that a hardware ?ag reset is possible. c. setting in ?()? is an optimized setting if no aaf is used. table 1-2: register address space for jasmine and lavender register bits group name lavender jasmine special description default value name address MB87J2120, mb87p2020-a hardware manual page 318 flag description flag description page 319 2 flag description table 2-1 contains all flags for mb87p2020-a (jasmine) and MB87J2120 (lavender). in order to avoid data inconsistencies during bit masking within flag register a mask (and/or gating) process is implemented in hardware for flag register. to distinguish between flag set-, reset- and direct write access different addresses are used 1 : flnom (0x000c):normal write operation flrst (0x0010):reset operation (1: reset flag on specified position; 0: don?t touch) flset (0x0014):set operation (1: set flag on specified position; 0: don?t touch) all of these three addresses write physically to one register with three different methods. for reading all three addresses return the value of flag register. every flag can have a different reset behaviour. with help of register flagres the application can choose whether the hardware is allowed to reset the desired flag ( dynamic behaviour, flagres_x=1 ) or not ( stat- ic behaviour, flagres_x=0 ). with dynamic behaviour the flag follows the driving hardware signal while with static behaviour the application is responsible for resetting the flag in order to catch next event. in table 2-1 the default reset behaviour at system start up is given in last column. note that some flags are only available for jasmine.. 1. additionally all access types (word, halfword and byte) are possible for each of these addresses. table 2-1: flags for MB87J2120 (lavender) and mb87p2020-a (jasmine) name bit lavender jasmine description default behaviour ( flagres ) vicsyn 31 x a frame or ?eld has been written to or read from sdram which event is signalled by this ?ag depends on vic set- tings: vicfctrl_frame determines the storage type within sdram (?eld or frame). for details see vic description and register list. vicvisyn_start determines whether a write or read start should trigger the ?ag static (0) erdy 30 x rdy timeout error has occurred see ulb description and register description for rdyto and rdyaddr static (0) rdpa 29 x x 1: dpa write access is enabled. this ?ag has to be polled before each dpa (write-) access to ensure a save sdram access a . otherwise data loss may occur . dynamic (1) fdpa 28 x x 1: dpa has ?nished sdram access and is ready for next one. static (0) ripa 27 x x 1: ipa is ready for command execution static (0) rmcp 26 x x 1: mcp is ready for command execution static (0) MB87J2120, mb87p2020-a hardware manual page 320 rmau 25 x x 1: mau is ready for command execution static (0) rpe 24 x x 1: pe is ready for command execution static (0) ebpp 23 x x 1: colour depth for source- and target layer is different during memcp command in this error case mcp reads data from input fifo but performs no further actions. static (0) cwen 22 x x 1: command register can be written. this ?ag has to be polled before a command can be writ- ten a . otherwise data loss and synchronization loss between jasmine/lavender and mcu may occur . dynamic (1) eint0 21 x x 1: lavender/jasmine external interrupt occurred this interrupt is currently assigned to spb device which is implemented on chip but outside the lavender/jasmine core. static (0) stout 20 x x after reset: flag is set (?1?) when sdram initialisation time is over else: flag is set (?1?) when sdc forces sdram refresh. this indicates a high sdram bus load. static (0) gsync 19 x x this ?ag is directly connected to gpu sync mixer 6 b . the sync mixer default settings generate no sync signal. static (0) bwvio 18 x x 1: gpu bandwidth violation occurred which means that the gpu didn?t receive requested data from sdram. this ?ag indicates a high sdram bus load. static (0) eov 17 x x 1: a command pipeline over?ow has occurred. a command was sent to jasmine while cwen ?ag was ?0?. static (0) ecode 16 x x 1: wrong error code was written to command register this command code is internally treated as noop com- mand. static (0) edata 15 x 1: execution device (pp or ipa) tried to read from empty input fifo this may indicate a wrong behaviour of execution device. static (0) bdpa 12 x x 1: dpa is performing an sdram access. static (0) bipa 11 x x 1: ipa is executing a command static (0) bmcp 10 x x 1: mcp is executing a command static (0) bmau 9xx 1: mau is executing a command static (0) table 2-1: flags for MB87J2120 (lavender) and mb87p2020-a (jasmine) name bit lavender jasmine description default behaviour ( flagres ) flag description flag description page 321 bpe 8xx 1: pe is executing a command static (0) ofl 7xx 1: output fifo load is equal or lower than programmable output fifo lower limit ( oful_ll ). static (0) ofh 6xx 1: output fifo load is equal or higher than programma- ble output fifo upper limit ( oful_ul ). static (0) ofe 5xx 1: output fifo is empty. static (0) off 4xx 1: output fifo is full. static (0) ifl 3xx 1: input fifo load is equal or lower than programmable input fifo lower limit ( iful_ll ). static (0) ifh 2xx 1: input fifo load is equal or higher than programmable input fifo upper limit ( iful_ul ). static (0) ife 1xx 1: input fifo is empty. static (0) iff 0xx 1: input fifo is full. static (0) a. alternative this ?ag can cause an interrupt and writing can be done inside interrupt service routine (isr). b. see gpu description for details about sync mixer settings. table 2-1: flags for MB87J2120 (lavender) and mb87p2020-a (jasmine) name bit lavender jasmine description default behaviour ( flagres ) MB87J2120, mb87p2020-a hardware manual page 322 command description command description page 323 3 command description jasmine command register width is 32 bit. it is divided into command code and parameters: partial writing of command register is supported. write to ?code?, byte 3 triggers command execution. not all commands need parameters. in this case parameter section is ignored. 3.1 command list all commands are listed with mnemonic, command code and command parameters (if necessary), com- mand function, registers evaluated by by the command, source and target for command data (input and out- put data if required). registers shown in table 3-1 are double buffered. write access during running command is possible without affecting current operation. values become valid after next command is started. table 3-1: command list mnemonic code device function registers swreset 00h ulb, ipa, pe, mau, mcp, aaf reset ulb command controller, ipa and drawing devices of pp, initialize fifos (pp-fifos, ififo, ofifo). only devices participated on command procession are initialized. a - putbm 01h pe store an uncompressed bitmap in video ram (?nite command). source: ififo with n*[bitmap colour data] n=(xmax-xmin+1)*(ymax- ymin+1)*bpp/32 target: video ram area which is de?ned by {xmax,xmin}, {ymax,ymin}, ppcm_lay and ppcmd_mir. pixel with ignorcol_col are not written if ignorcol_en = ?1?. xymax xymin ignorcol_en ignorcol_col ppcmd_lay ppcmd_en ppcmd_dir ppcmd_mir ififo 0 7 code 31 parameters MB87J2120, mb87p2020-a hardware manual page 324 putcp 02h pe store an compressed bitmap (tga run- length coded) in video ram (?nite com- mand). source: ififo with n*[bitmap rle data] n depends on compression factor. if number of decompressed pixels is not suf?- cient (xmax-xmin+1)*(ymax-ymin+1), pe will not become ready and waits for data. target: video ram area which is de?ned by {xmax, xmin}, {ymax, ymin}, ppcm_lay and ppcmd_mir. pixel with ignorcol_col are not written if ignorcol_en = ?1?. xymax xymin ignorcol_en ignorcol_col ppcmd_lay ppcmd_en ppcmd_dir ppcmd_mir ififo dwline 03h pe draw lines. the calculated pixel data are stored into video ram (in?nite com- mand). colour is de?ned by linecol. source: ulb input ?fo with start and end points n*([xs,ys], [le,xe,ye]), n = number of lines. layer of end point ignored if ppcmd_ulay is set. then ppcmd_lay is used instead of. target: line of pixels between start and end point in selected layer in video ram. notes: linecol or ppcmd changes apply after new command only. layer information in pixel address can change with each line. only relevant bits of linecol register are used, depending on colour depht. linecol ppcmd_lay ppcmd_ulay ppcmd_en ififo table 3-1: command list mnemonic code device function registers command description command description page 325 dwrect 04h pe draw rectangles. the calculated pixel data are stored into video ram (in?nite com- mand). colour is de?ned by rectcol. source: ulb input ?fo with start and end point n*([xs,ys], [le,xe,ye]), n = number of rectangles. layer of endpoint ignored if ppcmd_ulay is set. then ppcmd_lay is used instead of. target: rectangular area of pixels between start and end point in selected layer in video ram. notes: rectcol or ppcmd changes apply after new command only. layer information in pixel address can change with each line. only relevant bits of rectcol register are used, depending on colour depht. rectcol ppcmd_lay ppcmd_ulay ppcmd_en ififo puttxtbm 05h pe store uncompressed pixel data with ?xed foreground or background colour into video ram (?nite command). source: ulb input ?fo with n*[colour enable data] n = (xmax-xmin+1)*(ymax-ymin+1)/32 target: rectangular area of pixels between start and end point in selected layer in video ram. background pixels are written only if bgcol_en = ?1?. note: only the relevant bits of bgcol and fgcol register are used (depends on the colour depth). xymax xymin fgcol bgcol_en bgcol_col ppcmd_lay ppcmd_en ppcmd_dir ppcmd_mir ififo table 3-1: command list mnemonic code device function registers MB87J2120, mb87p2020-a hardware manual page 326 puttxtcp 06h pe store compressed pixel data with ?xed foreground or background colour into video ram (?nite command). source: ulb input ?fo with n*[coded colour ena- ble data], n depends on compression factor target: rectangular area of pixels between start and end point in selected layer in video ram. background pixels are written only if bgcol_en = ?1?. note: only the relevant bits of bgcol and fgcol register are used (depends on the colour depth). xymax xymin fgcol bgcol_en bgcol_col ppcmd_lay ppcmd_en ppcmd_dir ppcmd_mir ififo putpixel 07h mau transfer single pixel data into video ram (in?nite command). source: ififo with address/data pairs. n*([l,x,y], [single colour data]), n = number of pixels target: addressed pixel in video ram. note: ififo: lsb aligned pixel data, however physical data in video ram is msb aligned ppcmd_en ififo putpxwd 08h mau transfer packetized pixel in data words (pixel bursts) into video ram (in?nite command). source: ififo with n*([l,x,y],[data word]) n = number of packet words target: successive pixels in video ram on addressed position. note: the number of pixels in the 32-bit word depends on colour depth/layer. in opposi- tion to putpixel command ififo data is directly in video ram format. especially for 24 bit colour depht ififo data is msb aligned in this case. ppcmd_en ififo table 3-1: command list mnemonic code device function registers command description command description page 327 putpxfc 09h mau set ?xed coloured pixels in video ram at appropriate adresses (in?nite command). colour is de?ned by pixcol register. source: ififo with pixel adresses n*([l,x,y]), n = number of pixels target: addressed pixels in video ram. note: only the relevant bits of pixcol register are used (depends on the colour depth). pixcol ppcmd_en ififo getpixel 0ah mau read pixel data from video ram (in?nite command). source: ififo with pixel addresses, n*([l,x,y]), n = number of pixels. video ram with pixel data from pixel address. target: ofifo, n*([single colour data]). ififo ofifo xchpixel 0bh mau transfer single pixel data into video ram and read the old value from video ram (in?nite command). source: ififo with pixel address/data pairs, n*([l,x,y],[single colour data]), n = number of pixels. old pixel data from video ram. target: video ram for new pixel from ififo. ofifo for old pixel from video ram. ififo ofifo table 3-1: command list mnemonic code device function registers MB87J2120, mb87p2020-a hardware manual page 328 memcp 0ch mcp copy rectangular region from video ram to video ram (in?nite command). source: video ram area de?ned by start point [ls,xs,ys] and end point [xe,ye]. ififo control parameter: n*([ls,xs,ys], [xe,ye], [lt,xt,yt]), n = number of rectangular areas to copy target: videoram area with start point [lt,xt,yt] of same size as source area. note: layer information of source end point is ignored in pixel address. there is no end point of target area to specify. source and target layer should have same colour depth. if mismatch detected, the error ?ag flnom_ebpp is activated. ififo putpa 0dh ipa store data in video ram physically with address auto-increment (in?nite com- mand). source: ififo with ([physical address], n*[physi- cal data]). target: video ram on physical address and suc- ceeding. physical address means there is no direct point relation. ififo getpa 0eh ipa load data from video ram with address auto-increment, stop after cmd_par words. source: ififo with [physical address] video ram address starting at physical address and succeeding words. number of words given by cmd_par. target: physical data in ofifo. cmd_par ififo ofifo table 3-1: command list mnemonic code device function registers command description command description page 329 legend, symbols from command list table: physical byte address note: depending on video memory size not all addresses are used. data word single colour data (lsb aligned) bpp...24, 16, 8, 4, 2, 1 bit colour data bpp...24, 16, 8, 4, 2, 1 bit n...count of pixel per word bpp n dwpoly 0fh pe draw polygon lines. calculated pixel are stored in video ram (in?nite command). source: ififo with start and next points of a poly- gon ([xs,ys], n*[ln,xn,yn]). n = number of polygon lines. each ?next? point is the start point of the next polygon line. target: polygon pixels in video ram. note: drawing of more than one polygon on dif- ferent layers with one dwpoly command is possible by changing layer of next point. the colour of the polygon can?t change in same command cycle. only the relevant bits of plcol register are used (depends on colour depth). plcol ppcmd_lay ppcmd_ulay ppcmd_en ififo noop ffh ulb no operation. dummy cycle for command controller. - a.sdc, gpu, vic, spb, ccfl did not react on swreset. table 3-1: command list mnemonic code device function registers 31 0 31 byte 1 byte 0 byte 2 byte 3 0 0 bpp-1 0 31 31-(bpp-1) ... p 0 bpp-1 p (n-1) MB87J2120, mb87p2020-a hardware manual page 330 24 1 * 16 2 84 48 216 132 * in case of 24 bpp colour data is packet over word boundaries. there are no lower bits left empty and ?rst pixel is msb aligned on the 32-bit word grid. colour enable data (...) source or target fifo data [...]n deliver data in ?[...]? n times [...]+ deliver data in ?[...]? at least one time {l,x,y} pixel coordinates, point de?nition if l is not stated i.e. in {x,y} form, information in bits 31, 30, 15, 14 kept as don?t care. 3.2 command and i/o control command queue consits of two registers, an internal actual processed command and an additional buffer for the next planned command. before writing to command register the flnom_cwen (command write enable) flag polling is required to ensure that the new command can be written. all execution devices pro- vide ready or busy information via flnom flag register. normally this dedicated device information is not needed to evaluate. command controller takes care on operation. also the ififo and ofifo has its status flags in flnom. these are usable for i/o flow control by polling or interrupt. additional dma support is possible to transfer data from/to the fifos. 0 31 ... p 0 p 31 31 0 13 15 29 layer (to be ignored if layer register exists) 0 1 2 3 x coordinate y coordinate page 331 d-2 hints and restrictions for laven- der and jasmine MB87J2120, mb87p2020-a hardware manual page 332 hints and restrictions special hints page 333 1 special hints 1.1 ipa resistance against wrong settings invalid settings for minimum transfer block sizes lead to ipa deadlock from which it can only recover by reset. at minimum a value of 1 has to be given for the output fifo (block size of zero makes no sense) and a minimum value of 2 for the input fifo should be setup. input fifo needs at least one address/data pair for ipa. table 1-1 gives an overview and a classification about the described problem. 1.2 ulb_dreq pin timing to host mcu ulb_dreq reaction time is critical if only one wait state for user logic bus (ulb) cycle is set up. writing per dma ?demand mode? can lead to additional transferred data after ulb_dreq tied low. jasmine can handle this additional data by an two words deep overflow buffer. thus no data will be lost. for lavender the count of waitstates can be set equal or greater than two. writing to input fifo in dma block/step/burst mode and reading from output fifo with dma (both block/step/burst and demand mode) transfers correct amount of data in any case for jasmine and lavender. a detailed description about this topic can be found in hardware manual (chapter b-2.1.7) and in a dma application note. table 1-2 gives an overview and a classification about the described problem. table 1-1: overview for ipa resistance subject description description smaller block sizes does not make sense since a block should contain at least one word. input fifo contains address and at least one data word. output fifo contains at least one data word. classi?cation hint effects without workaround the system may hang with wrong settings. do not use forbidden settings. escape only with hw-reset. solution/workaround no workaround required. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ipa.1 and sdc.3 (limits can be set in ?mkctrl?.) table 1-2: overview for ulb_dreq timing subject description description reaction time for ulb_dreq pin for write accesses is critical if one wait- state is set up. classi?cation hint MB87J2120, mb87p2020-a hardware manual page 334 1.3 clkpdr master reset clkpdr, bit[15] is write only. read access returns always ?0?. writing ?1? initiates global master reset, writing a ?0? releases master reset. table 1-3 gives an overview and a classification about the described problem. 1.4 mau (memory access unit) commands for lavender drawing commands (putpixel, putpxwd, putpxfc, getpixel, xchpixel, dwline, dwrect and dwpoly) the target layer has to be always included into data stream (pixel addresses). for jasmine only commands putpixel, putpxwd, putpxfc, getpixel and xchpixel can?t use the layer reg- ister ppcmd_lay while the other drawing commands can use it. mau has its own dedicated layer man- agement and could not have benefit from the enhancement. 1 effects without workaround one data word more than expected can be delivered by mcu. solution/workaround for mb87p2020 (jasmine) a hardware solution is already implemented with a two word over?ow buffer. for MB87J2120 (lavender) the count of waitstates during dma transfer should be set equal or larger than two. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ulb.5 table 1-3: overview for master reset subject description description the reset status can not read back. classi?cation hint effects without workaround clkpdr_mrst is write only. reading always return ?0?. solution/workaround if the reset status has to be memorized it has to be done externally. normally this sequence should reset jasmine/lavender safely: g0clkpdr_mrst =1; g0clkpdr_mrst =0; concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: ctrl.1 (i/o march) 1. the target layer register ppcmd_lay can be used for pixel engine commands if ppcmd_ulay is set to ?1?. table 1-2: overview for ulb_dreq timing subject description hints and restrictions special hints page 335 a good workaround is to configure layer 0 with the same parameters as the target layer 1 and make all draw- ings to layer 0. the advantage is that no layer number has to be merged into pixel addresses. after drawing access to the layer data is possible via target layer configuration. also gpu can be set up to target layer. table 1-4 gives an overview and a classification about the described problem. 1.5 pixel processor (pp) double buffering the pixel processor contains a double buffering mechanism for its registers which is synchronized to com- mand execution. data are always written to configuration register which is not used for command execution. instead of con- figuration register directly an internal register is used for data storing during command execution. the time to write internal register is determined by hardware. for reading from pp registers an application can choose whether to read from configuration register ( readinit_rcr=1 ) or to read from internal register ( readinit_rcr=0 ). for bitwise modification of pixel processor configuration registers the double buffering mechanism has to be considered. if parts of the registers should be changed with read-modify-write operations there is the risk to read from internal double buffered register and write the result to the external register back. data incon- sistency could be the result. it is recommended to initialize readinit_rcr=1 to have read access to the external registers too (same register is accessed for reading and writing). note that the reset value is readinit_rcr=0 and has not the recommended value. table 1-5 gives an overview and a classification about the described problem. 1. in this case layer 0 is a temporary layer only; it should not be used for normal drawings. table 1-4: overview for mau commands subject description description for jasmine target layer register ( ppcmd_lay ) is ignored for mau com- mands even if ppcmd_ulay is set to ?1?. classi?cation hint effects without workaround commands putpixel, putpxwd, putpxfc, getpixel and xchpixel uses only the target layer delivered in input fifo. see command list for a detailed command syntax description. solution/workaround use layer 0 as a temporary drawing layer (see description for further details). concerned devices MB87J2120 (lavender) partly ?xed for mb87p2020 (jasmine) partly ?xed for mb87p2020-a (jasmine redesign) testcase emdc: sdc.log2phy table 1-5: overview for pp double buffering subject description description application note for readint register and its consequences. classi?cation hint MB87J2120, mb87p2020-a hardware manual page 336 1.6 robustness of ulb_rdy signal disturbances at external bus interface (ulb interface) can cause the ulb_rdy pin to stay in active state (logic ?0?) during a bus read access. this stops a mb91360 series mcu completely so that no further com- mands will be executed (blocking of external bus). only a mcu watchdog can reset the system. to avoid this problem the pcb layout should be done very carefully especially for the signals ulb_cs, ulb_rdx and ulb_wrx[3:0]. the following hints may help: place mcu and lavender as close as possible together on pcb. use separate ground plane on pcb in order to avoid interferences with other signals. place sdram and lavender as close as possible together in order to avoid interferences from sdram interface to ulb interface the interface circuit has been redesigned for jasmine so that this problem should not occur with this device. additionally for jasmine an ulb_rdy timeout has been added so that a ?hanging? ulb_rdy can not block the hole system. see hardware manual for further details. table 1-6 gives an overview and a classification about the described problem. effects without workaround readinit_rcr=0 : read internal register (reset value) readinit_rcr=1 : read con?guration register be careful with read-modify-write accesses; use readinit_rcr=1 in this case. solution/workaround no workaround required. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: sdc.log2phy table 1-6: overview for ulb_rdy robustness subject description description ulb_rdy signal may hang as a result of signal distortions on pcb. classi?cation hint effects without workaround a hanging ulb_rdy pin blocks the command execution of a mb91360 series mcu. solution/workaround keep special attention to pcb layout. this problem should not occur with jasmine. additionally a timeout for ulb_rdy has been added. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase every testcase. table 1-5: overview for pp double buffering subject description hints and restrictions special hints page 337 1.7 robustness of command pipeline against software errors as described in detail in hardware manual a command should only be written to lavender/jasmine when the flag flnom_cwen is ?1?. if this rule is followed the command execution of lavender works correctly but if a command is written although the flnom_cwen flag is ?0? the command execution of lavender may hang (command pipeline overflow). only a hardware reset can solve this situation. in case of a command pipeline overflow the command dataflow via fifos is not automatically corrected but the application can detect a command pipeline overflow with help of the flag flnom_eov and should perform corrective actions. please note that not only wrong written software can cause this problem but also signal distortions on ulb data bus (ulb_d). normally an application polls the flnom_cwen flag so it may be read many times inside a loop. if at one of these read accesses this flag changes its value from ?0? to ?1? due to pcb disturbances the software stops polling and sends a new command to the graphic controller. lavender has not finished executing pre- vious command and a command pipeline overflow occurs. to avoid this communication problem between mcu and lavender signal distortions on ulb_d should be reduced by improving pcb layout as already described in chapter 1.6. please note that additionally to the signals described in chapter 1.6 the ulb data bus (ulb_d) is also a critical signal for pcb layout. the command controller for jasmine has been redesigned and a command pipeline overflow can not cause the command execution to hang. but the application has still to take care about the command dataflow syn- chronization as described above. table 1-7 gives an overview and a classification about the described problem. 1.8 dma resistance against wrong settings in dma demand mode the dma controller inside lavender/jasmine counts the amount of data words to be transferred. see hardware manual chapter b-2.1.7 (ulb description) for more details about dma. if the amount of data to be transferred in dma demand mode is equal to ?0? the lavender dma controller may hang. the data amount of ?0? can be calculated 1 if: ifdma_ll is set to max. input fifo size (128 for lavender) for dma to input fifo table 1-7: overview for command pipeline robustness subject description description if a command is written to lavender while the ?ag flnom_cwen is ?0? the command pipeline may hang. this can be watched with flnom_eov ?ag. classi?cation hint effects without workaround the command controller does not work properly after sending a new com- mand while flnom_cwen = ?0?. only a hardware reset can solve this situation. for jasmine no hang-up can occur but the command<-> data synchroniza- tion can be disturbed. solution/workaround write only commands to lavender when flnom_cwen is ?1?. keep special attention to pcb layout for ulb data bus (ulb_d). concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase every testcase with command execution. MB87J2120, mb87p2020-a hardware manual page 338 ofdma_ul is set to ?0? for dma from output fifo. in order to avoid this problem the described settings are forbidden for lavender. this problem has been solved for jasmine. table 1-8 gives an overview and a classification about the described problem. 1. for calculation of data amount not the programmed limits are used but the current fifo load. therefore the data amount is not necessarily ?0? if the described settings are applied. table 1-8: overview for dma resistance against wrong settings subject description description the dma controller for lavender may hang if transfer size in dma demand mode is equal to ?0?. classi?cation hint effects without workaround no dma transfer is started even if the start condition is met a . a. start condition for input fifo is: fifo load <= ifdma_ll; for output fifo: fifo load >= ofdma_ul. solution/workaround avoid critical settings ifdma_ll = 128 and ofdma_ul = 0. for jasmine this problem has been ?xed. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ulb.5 and ulb.13 hints and restrictions restrictions page 339 2 restrictions 2.1 esd characteristics for i/o buffers table 2-1 shows the esd characteristics for all i/o buffers for lavender and jasmine. for a complete pin- ning table see hardware manual. (a) listed values are determined by using the human body model(c=100pf, r=1.5k ohm) . the test pro- cedure is described in mil-std-883. (b) the worst values among the four conditions (vdd positive, vdd negative, vss positive, and vss neg- ative) are shown. (c) the pass/fail criteria of 1ua leakage current was used for all the i/os except for the i/os described with ?leak mode?. (d) in the case of ?leak mode?, the leakage current will increase but the i/os will continue to be functional. (e) the i/os marked with ?destruction mode?, however, may be destroyed if the higher voltage than shown in the table is applied. once they are destroyed, they may become non-functional. table 2-1: esd performance for buffer types buffer type description esd performance of mb87p2020 b3nnlmx bidirectional true buffer (3.3v cmos, iol=4ma,low noise type) +/-500v, destruction mode b3nnnmx bidirectional true buffer (3.3v cmos, iol=4ma) +/-500v, destruction mode bfnnqhx bidirectional true buffer (5v tolerant, iol=8ma, high speed type) +/-2000v, leak mode bfnnqlx bidirectional true buffer (5v tolerant, iol=2ma, high speed type) +/-2000v, leak mode bfnnqmx 5v tolerant, bidirectional true buffer 3.3v cmos, iol/ ioh=4ma +/-2000v, leak mode ipbix input true buffer for dram test (2.5v cmos with 25k pull-up) (sdram test only) +/-2000v itamx analog input buffer +/-2000v itavdx analog power supply n.a. itavsx analog gnd n.a. itbstx input true buffer for dram test (2.5v cmos with 25k pull-down) (sdram test only) +/-2000v itchx input true buffer (2.5v cmos) +/-2000v MB87J2120, mb87p2020-a hardware manual page 340 table 2-2 gives an overview and a classification about the esd characteristics. itfhx 5v tolerant 3.3v cmos input +/-2000v, leak mode itfuhx 5v tolerant 3.3v cmos input, 25 k pull-up +/-2000v, leak mode ittstx input true buffer for dram test control (2.5v cmos with 25k pull-down) +/-2000v otamx analog output +/-2000v otavx analog output buffer +/-2000v otftqmx 5 v tolerant 3.3v tri-state out- put, iol/ioh=4ma +/-2000v, leak mode vpdx 3.3v cmos input, disable input for pull up/down resistors, con- nect to gnd +/-2000v yb002aax oscillator output +/-500v, destruction mode yb3dnlmx bidirectional true buffer (3.3v cmos, 25k pull-down, iol=4ma,low noise type) +/- 500v, destruction mode yb3nnlmx bidirectional true buffer (3.3v cmos, iol=4ma,low noise type) +/- 500v, destruction mode yi002aex oscillator pin input +/-2000v table 2-2: overview for esd characteristics subject description description some i/o buffer show a weak esd performance (see table 2-1). classi?cation hw limitation effects without workaround affected buffers may be destroyed or may have a higher leakage current if a voltage higher than speci?ed in table 2-1 is applied to the buffer. solution/workaround no workaround possible. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ulb.5 and ulb.13 table 2-1: esd performance for buffer types buffer type description esd performance of mb87p2020 hints and restrictions restrictions page 341 2.2 command fsm during emulation for jasmine development a hang-up of the command fsm has been observed while the commands getpa and noop have been executed within a loop. this hang-up has never been observed for lavender. based on this emulation result the command fsm for jasmine has been redesigned in order to avoid this problem in the future. at the present time it can not securely excluded that the timing gap problem is not present on lavender. therefore further investigation and verification is necessary to check the lavender behaviour with respect to this problem. table 2-3 gives an overview and a classification about the described problem. 2.3 gpu mastertiming synchronization for lavender the internal gpu mastertiming signal (master switch for gpu) is not synchronized to display clock domain (output pixel clock) properly. if the pixel clock runs asynchronous to core clock (e.g. external pixel clock) and the flag mtimon_on is switched on display output may hang afterwards. in order to avoid this problem no asynchronous display clock should be used. asynchronous display clock can either be an external display clock or an internal clock derived from another clock than core clock. with other words core and display clock should have the same clock source. see chapter b-1 (clock unit (cu)) description in hardware manual for further details about lavender?s clock concept. for jasmine this problem is solved. table 2-4 gives an overview and a classification about the described problem. table 2-3: overview for timing gap problem subject description description a command fsm hang up has been observed for an intermediate version a of jasmine. a. the term ?intermediate? means a rtl version between lavender and jasmine. classi?cation hw limitation effects without workaround the command execution is stopped at next command. only a hardware reset can terminate this state. solution/workaround for jasmine this problem is ?xed. for lavender it is not clear whether this problem occurs. therefore further investigation is necessary. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ulb.1 table 2-4: overview for gpu mastertiming subject description description if an asynchronous display clock (compared to core clock) is used, turning on/off the gpu with help of mtimon_on can cause the display output to hang. MB87J2120, mb87p2020-a hardware manual page 342 2.4 read limitation for 16 bit data interface to mcu the user logic bus interface of MB87J2120 to mb91xxxx mcus has read limitations in 16bit data mode (pin mode[2]=0). read access in 32-bit data mode (pin mode[2]=1) is fully functional as also write ac- cesses in all modes and access types. writing to registers works with all modes properly. this read limitation affects only half word (16 bit) and byte (8 bit) read access to lavender internal regis- ters (address range 0x0000-0xfbff) in register space and direct sdram read access. it does not affect a word (32 bit) read access. in this case the mcu splits the word into two half words (16 bit) which are sub- mitted sequentially. for this transmission the order has to be msb first (big endian format). table 2-5 gives an overview on all possible modes and read access types. an access to address offset 2 or 3 may lead to delivery of wrong values, depending on address of previous read access.figure 2-1 shows these offsets relative to a given address (addr0, addr1) in byte addressed mcu memory space. the right part of the figure shows the mapping from this memory to internal 32 bit registers. there is a possible work-around because reading from an address with offset 0 or 1 to an aligned word ad- dress (divisible by 4) is fully operable. after the access to an address with offset 0 or 1, read access from offset 2 or 3 is possible and delivers correct data. if not otherwise guaranteed, a dummy read access on offset 0 or 1 should be included before reading from offset 2 or 3 in order to prepare the correct internal read se- quence and deliver correct values also for offsets 2 and 3. 1 alternative to the described procedure read ac- cesses can be limited to 32-bit accesses only (see table 2-5). classi?cation hw limitation effects without workaround a wrong output picture is visible. only a hardware reset can solve this problem. solution/workaround no asynchronous display clock should be used for lavender. core and dis- play clock should have the same clock source a . for jasmine this problem is solved. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: gpu-p10 a. as common input clock the pins osc_in, ulb_clk, dis_pixclk or rclk/mode[3] can be used. table 2-5: overview for read accesses in different modes read access mode word access (32bit) halfword access (16bit) byte access (8bit) 32bit mode read (mode[2]=1) supported supported supported 16bit mode read (mode[2]=0) supported a a. since mcu delivers data in big endian order (msb ?rst). not supported for offset 2 not supported for offset 2 and 3 1. make sure that this sequence is not interrupted by other read accesses (for instance by an interrupt request with ?ag read access). table 2-4: overview for gpu mastertiming subject description hints and restrictions restrictions page 343 in c-api the described workaround is already implemented. within api read accesses from address offset 2 and 3 are limited to registers flnom_iff and flnom_ofe . these accesses are implemented as 32-bit read accesses. table 2-6 gives an overview and a classification about the described problem. table 2-6: overview for read limitation in 16bit data mode subject description description read access from addr0+2 or addr0+3 does not work reliable. classi?cation hw limitation effects without workaround halfword and byte read access from addr0+2 or addr0+3 may deliver wrong values in 16bit data mode ( mode[2]=0 ). solution/workaround read data from addr0+0 or addr0+1 before reading from addr0+2 or addr0+3. or limit read accesses to 32bit read accesses only. all c-api functions which have to read back values implement this worka- round already. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: ctrl.1 (i/o march) and ipa.1 in 16bit data mode. +2 +0 +1 +3 +0 +1 byte 0 byte 3 31 0 byte 2 byte 1 0 7 ....... addr0 addr1 figure 2-1: address offsets in byte addressed memory and its mapping to jasmine/lavender in- ternal registers MB87J2120, mb87p2020-a hardware manual page 344 2.5 sdc sequencer readback for lavender the sdc sequencer is not readable in 16bit data mode (pin mode[2]=0). normally the se- quencer data will only be written once at initialization time. a readback of sequencer data is usually not necessary. this problem has been fixed for jasmine. table 2-7 gives an overview and a classification about the described problem. 2.6 direct sdram access with 16bit and 8bit data mode lavender and jasmine can perform a memory mapped sdram access (see hardware manual sections for ulb and dipa for further details). in the case of 16bit (halfword) or 8bit (byte) read access from sdram wrong read data may be delivered for some addresses for lavender. direct sdram write access works for all data sizes (32,16 and 8bit). to avoid this problem 32bit (word) access should always be used for sdram read access. there is no re- striction for read access from register space 1 . for jasmine this problem is solved and every data size can be used for reading from sdram. table 2-8 gives an overview and a classification about the described problem. table 2-7: overview for sdc sequencer readback subject description description the sdc sequencer can not be read back in 16bit data mode. classi?cation hw limitation effects without workaround lavender delivers wrong data when the sdc sequencer is read in 16bit data mode. solution/workaround none. this problem has been ?xed for jasmine. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: ctrl.1 (i/o march) 1. the gdc address space is divided into register and sdram space. see chapter b-2.1.4 in hardware manual for further details. table 2-8: overview for direct sdram access with 16bit and 8bit data mode subject description description read access from sdram space with 16bit (halfword) and 8bit (byte) data size may deliver wrong values for some addresses. classi?cation hw limitation effects without workaround the mcu reads wrong data from an address mapped to lavender sdram space when reading a 16bit (halfword) or 8bit (byte) value. reading a 32bit (word) value from sdram space works as well as reading from register space. hints and restrictions restrictions page 345 2.7 input fifo read in 16bit mode one hidden feature of lavender and jasmine is the reading from input fifo 1 . normally reading from input fifo makes no sense and is therefore not officially supported but it may be the reason for a mcu blocking which can occur when reading from this address is performed by accident. in 16bit data mode (pin mode[2]=0) reading from input fifo can cause the ulb_rdy pin to stay in ac- tive state (logic ?0?). as already described in chapter 1.6 this can block the command execution of mcu. to avoid this problem reading from input fifo should not be used in 16bit data mode. this problem occurs on both gdc devices but for jasmine the ulb_rdy timeout can be set so that no system blocking can occur. table 2-9 gives an overview and a classification about the described problem. 2.8 ulb_dstp pin function ulb_dstp output of lavender/jasmine must not be used because it can not be guaranteed that ulb_dstp signal is always generated at dma interruption by sw-reset or falling edge of dmaflag_en 2 . it is rec- solution/workaround always read 32bit (word) values from lavender sdram space and mask the data afterwards if necessary. for jasmine this problem is solved. concerned devices MB87J2120 (lavender) ?xed for mb87p2020 (jasmine) ?xed for mb87p2020-a (jasmine redesign) testcase emdc: sdc.3 and dpa.1 1. the read value is the current fifo output value which will be delivered to pixel processor (pp) with next fifo read access. note that the pp reads with core clock data from input fifo while mcu reading can only be performed with ulb clock. because ulb clock is always slower (or at most equal) than core clock not all fifo output data can be sampled. table 2-9: overview for input fifo read access in 16bit data mode subject description description reading from input fifo in 16bit data mode causes ulb_rdy = 0. classi?cation hw limitation effects without workaround the hanging ulb_rdy signal blocks the command execution of a mb91360 series mcu. solution/workaround do not read from input fifo (address 0x0004) in 16bit data mode. for jasmine the ulb_rdy timeout can be activated. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: ctrl.1 (i/o march) 2. see ulb speci?cation in hardware manual for further details. table 2-8: overview for direct sdram access with 16bit and 8bit data mode subject description MB87J2120, mb87p2020-a hardware manual page 346 ommended to disable/interrupt mcu-dmac directly. see table 2-10 for a reset sequence example. the reset of mcu-dmac is already included in c-api function ulb_dma_hdg . in case of direct mcu reset dmaflag_dstp should be set to ?0? in order to avoid unintentional interrup- tion. with this setting no ulb_dstp impulse is generated by the graphic controller. the ulb_dstp pin must not be used; it will not be supported any longer. table 2-10 gives an overview and a classification about the described problem. 2.9 software reset for command execution for jasmine the swreset command does not reset parts of command controller correctly. as a consequence the following commands are not executed properly. they may be finished too early. lavender does not have this problem due to a different command controller structure. as workaround a noop can be executed after swreset. this is already included into c-api. table 2-11 gives an overview and a classification about the described problem. table 2-10: overview for dstp function subject description description ulb_dstp signal is sometimes not generated. classi?cation hw limitation effects without workaround mcu dma-controller is not reset correctly. additional data may be sent after swreset or dma turn off. solution/workaround ulb_dstp signal must not be used. instead turn off mcu dma-control- ler before sending swreset to jasmine/lavender or disable dma via dmaflag_en . mcu-dmac reset sequence: 1.) dmaca[30] = 0 2.) dmaca[31] = 0 gdc-api function ulb_dma_hdg takes care of this issue. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: ulb.5 and ulb.13 table 2-11: overview for software reset (incomplete reset) subject description description parts of command controller will not be reset with swreset command. classi?cation hw limitation effects without workaround after swreset next command may not be executed properly. solution/workaround send a noop command after swreset command. the noop command is already included in api function ?gdc_cmd_swrs?. concerned devices not present for MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) hints and restrictions restrictions page 347 in case of software reset and anti aliasing (aaf) enabled, data for logical address [layer 0, pixel {0,0}] can be destroyed. the swreset command is executed by aaf despite a pending sdram access. therefore layer and pixel address is reset which causes a wrong address for sdram access. for this restriction two possible workarounds exist which are described in table 2-12. table 2-12 gives an overview and a classification about the described problem. 2.10 aaf settings double buffering the command synchronization for aaf settings (register: ppcmd) works not properly. it can happen that the currently executed command uses the settings from previous command. in order to avoid this problem a noop command should be inserted before aaf settings are manipulated. a sequence for a drawing command with aaf (turn on before command execution and turn off afterwards) looks as follows: gdc_cmd_nop(); g0ppcmd_en = 1; // aaf enable g0aaoe = ..; // aaf init gdc_cmd_dw...; // drawing command gdc_cmd_nop(); g0ppcmd_en = 0; // aaf disable testcase emdc: ulb.1, ulb.2 and sdc.6 table 2-12: overview for software reset (aaf) subject description description if aaf is enabled swreset command causes an aaf reset without waiting for sdc. classi?cation hw limitation effects without workaround if aaf is enabled data for [layer 0, pixel {0,0}] can be destroyed. solution/workaround workaround 1: do not display [layer 0, pixel {0,0}] (either do not use entire layer or do not use row 0 and column 0 for layer 0) workaround 2: save physical word at start address of layer 0. this word contains pixel {0,0}. physical access does not need command control. optional: disable mcu interrupts and wait for gpu frame sync in order to avoid visible effects on display. send command swreset (including a noop according to table 2-11) to jasmine/lavender restore physical word at start address of layer 0. pixel {0,0} will be re- stored in this way. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase code review table 2-11: overview for software reset (incomplete reset) subject description MB87J2120, mb87p2020-a hardware manual page 348 please note that the c-api functions already include the necessary flag polling of flnom_cwen. the c-api function ?pxp_aaf? implements the described workaround already and should be used to turn on or off aaf within an application. a special situation occurs when a command, which uses the aaf, is interrupted by a software reset (c-api function ?gdc_cmd_swrs?). in this case even a noop command can not ensure reinitialization of aaf. it is necessary to execute a dummy drawing command after changing the aaf settings in order to restore double buffered values inside aaf. note that aaf is enabled during dummy command and the options of the command executed before swreset are activated. after the dummy command the new settings are ap- plied to aaf. table 2-13 gives an overview and a classification about the described problem. 2.11 pixel engine (pe) commands there are four pe commands which load rectangular shapes of bitmaps or textures to video memory. puttxtbm, puttxtcp, putbm and putcp commands have the option to disable the write process of back- ground data to video memory. for puttxtbm and puttxtcp relevant configuration is bgcol_en = 0. for putbm and putcp it is ignorcol_en = 1. in both cases situation can occur where the data stream to video memory is finished before procession of the command is terminated. this is the case if not transferred background or ignore-coloured pixels were generated at the end. due to the finished data transfer the pp output fifos are empty already. if the command has finished the ready condition could be propagated to ulb command controller immediately, also if ulb has not stopped the pe device. resulting from this too early sent ?command ready? the ulb command controller fails. only in the special situations with suppressed background pixels an additional {gdc_cmd_nop(); gdc_cmd_swrs();} command sequence is required as work around after using the described commands with its appropriate parameters where the error may occur (example for puttxtbm): gdc_cmd_txbm (xmin,xmax,ymin,ymax,fgcol,bgcol,bgen,0,0,lay); gdc_fifo_inp(p_pat, font->offset, 0); if (bgen == 0) { gdc_cmd_nop(); /* finish puttxtbm without data loss */ gdc_cmd_swrs(); /* puttxtbm with bgcol_en=0 work around */ table 2-13: overview for aafen double buffering subject description description the aaf double buffer for aafen signal works not properly. classi?cation hw limitation effects without workaround the aaf settings from previous command are still valid for currently exe- cuted command. solution/workaround workaround if no swreset is used: execute a noop command before aaf settings will be changed if the drawing command with aaf is interrupted by a swreset a special workaround is needed: execute a noop command after swreset (see also chapter 2.9) execute dummy draw command concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase code review hints and restrictions restrictions page 349 } the inserted nop command is for de-coupling puttxtbm from software reset. all data were processed finally and closed. to bring the ulb command processor the software reset function is called afterwards. gdc_cmd_swrs() itself executes the swres and nop commands to bring all command processing de- vices to an initial state (see also section 2.9). an insertion of gdc_cmd_nop() before each command register write (introduced in c-api former ver- sions) is not required. only at dedicated points in the software at situations described above the work around is required. other commands and with background enabled the error can not occur. application examples encountered only marginal performance loss at transparent text output over background drawings. for bit- map transfer the additional two commands are not important. table 2-14 gives an overview and a classification about the described problem. because of the splitting between command and data write functions within api the noop - swreset com- mand sequence can not be included selectively after affected commands. this can only be done in target application or in a higher level api. the c-api can?t solve this problem by inserting a gdc_cmd_nop () and gdc_cmd_swrs() before each command register write activity without performance loss. it is rec- ommended to solve this problem only at dedicated points in the higher level software if special background conditions are relevant only. 2.12 pixel read back commands (getpixel, xchpixel) internal data flow blocks if reqcnt+1 data words does not fit into output fifo. therefore the full output fifo size is not useable in some cases. if an application wants to collect data in output fifo in order to transfer a complete block at once without checking fifo load for every data word for instance via dma it is possible that not all data appear in output fifo and the initialized limit is not reached. even if the next command was sent after getpixel or xch- pixel which is normally suitable to flush input fifo data flow blocking is not escaped. note that this behaviour does not occur if output fifo is read with flag polling for every data word be- cause the amount of words in output fifo falls below the limit fifosize-reqcnt-1 at a certain time. getpa is not affected because it uses other registers than reqcnt for block size calculation. table 2-14: pixel engine commands subject description description pixel engine signals end of command execution without stop signal from command controller. this causes possible malfunction of command con- troller for the next command processed. classi?cation hw limitation effects without workaround commands puttxtbm , puttxtcp , putbm and putcp with background disable ( bgcol_en=0 ) or ignore colour enable ( ignorcol_en=1 ) can cause command control malfunction. this may lead to an incorrect execu- tion of following commands. solution/workaround send noop -swreset command sequence after puttxt[bm|cp] if con?g- ured with bgcol_en=0 or after put[bm|cp] with ignorcol_en=1. this description is valid for lavender and jasmine. concerned devices MB87J2120 (lavender) mb87p2020 (jasmine) mb87p2020-a (jasmine redesign) testcase emdc: sdc.6 MB87J2120, mb87p2020-a hardware manual page 350 table 2-15 gives an overview and a classification about the described problem. a possibility to utilize the full output fifo size is to ensure that always reqcnt+1 words can be placed in output fifo. this limits the maximal package size (number of words to transfer for one output fifo fill) for a given reqcnt . the maximal package size can be calculated according to (1). (27) the function ? trunc ? in (27) means that only the natural part of this fraction should be taken for calcula- tion. the parameter ? fifosize ? is the size of output fifo. note that ? pkg_size ? is the maximal pack- age size, sizes smaller than the calculated size can be used. figure 2-2 shows an example on how to calculate the correct package size based on a given reqcnt and to read back a block of data. in order to keep the example simple flag polling for flnom_ofh is used but it is also possible to generate an interrupt with this flag or to set up ofdma_ul with the package size for dma transfer. // ---------------------------------------- // calculate optimal package size for // a given request-count // ---------------------------------------- reqcnt = g0reqcnt; // set fifo size if (g0clkpdr_id==0) { // lavender of_size = 128; } else { // jasmine of_size = 64; } pkg_size = ((byte)(of_size/(reqcnt+1)))*(reqcnt+1); // ---------------------------------------- // write command to display controller and // set command registers for getpixel // (no registers required) // ---------------------------------------- gdc_cmd_gtpx(); // ---------------------------------------- // write data to input fifo // ---------------------------------------- for (jj=0;jj hints and restrictions restrictions page 351 gdc_fifo_inp((dword*)buildifdata(x,y,layer),1,0); } // ---------------------------------------- // send noop command to force fifo flush // ---------------------------------------- gdc_cmd_nop(); // ---------------------------------------- // wait for data in of // ---------------------------------------- g0oful_ul = pkg_size; while (g0flnom_ofh==0); // read data from output fifo for (jj=0;jj page 353 d-3 abbreviations MB87J2120, mb87p2020-a hardware manual page 354 abbreviations abbreviations page 355 abbreviations aaf anti aliasing filter apll analog phase locked loop bpp bits per pixel (colour depth) bsf bit stream formatter cbp control bus port ccfl cold cathode fluorescent lamp ccu color conversion unit clut color look-up table crt cathode ray tube ctrl control unit cu clock unit dac digital to analog converter dfu data fetching unit dipa direct and indirect physical memory access dma direct memory access dmac dma controller dpa direct physical memory access drm duty ratio modulator/modulation emc electromagnetic compatibility fet field effect transistor fsm finite state machine gdc graphic display controller (lavender and/or jasmine in this manual) gpu graphic processing unit hw hardware ipa indirect physical memory access isr interrupt service routine lcd liquid crystal display lsa line segment accumulator mau memory access unit mcp memory copy unit mcu micro controller unit pcb printed circuit board MB87J2120, mb87p2020-a hardware manual page 356 pe pixel engine pp pixel processor rgb red / green / blue color format rle run length encoding sdc sdram controller spb serial peripheral bus spg sync pulse generator sw software tft thin film transistor tga tga is a trademark of truevision, inc. ulb user logic bus (controller) vic video interface controller yuv luminance / chrominance color format |
Price & Availability of MB87J2120
 |
|
|
|
|
All Rights Reserved © IC-ON-LINE 2003 - 2022 |
| [Add Bookmark] [Contact Us] [Link exchange] [Privacy policy] |
|
Mirror Sites : [www.datasheet.hk]
[www.maxim4u.com] [www.ic-on-line.cn]
[www.ic-on-line.com] [www.ic-on-line.net]
[www.alldatasheet.com.cn]
[www.gdcy.com]
[www.gdcy.net] |