ADVRAM

Technical Info

When developing ADVRAM, we wanted to avoid the common problems with other MSX hardware. For instance, I never liked to search all slots to find where the Mapper is, and we also didn't want to waste another of the limited MSX I/O ports in this project. The basic idea in this project was to make the hardware simple and easy to program. Based on these guidelines, our solutions to these problems were: ADVRAM Slot The ADVRAM will always be on the page 2 (08000h-0BFFFh) of the slot where the MainROM is. You can get this value by reading the contents of the EXPTBL system variable (0FCC1h). This value will be either 00h (slot 0, in common MSXs) or 80h (slot 0.0, in the Turbo-R). We choose this slot because, except for that Korean machine with V9948, no one places ROM in this area. So this is a free area in almost all MSX, and we don't need to spend an extra slot with ADVRAM. ADVRAM Bank Switching As you can see, ADVRAM uses only 16kb of the MainROM slot. Since the MSX-2 VRAM has 128kb, some kind of bank switching must be done. We choose to use the same I/O port used by the Memory Mapper. This means you just need to make an OUT(0FEh),A to choose which portion of the VRAM you want to map into RAM. This has a side effect of switching the main Mapper of the system, but this also happens every time you add another Mapper to your system. Just think of ADVRAM as another Mapper, but one with the special feature of mirroring the VRAM. ADVRAM Control Register Since ADVRAM behaves like a Memory Mapper, some programs could get confused and try to use it to store data. But this do not happen because the ADVRAM can be switched ON and OFF, and the default state when the MSX start is OFF. To turn the ADVRAM ON, you must use its control register. To store a value in the control register, you must use this sequence: LD A,value IN A,(09Ah) I know it sounds strange, to write a value using an IN instruction, but it works. Let me explain what happens: most programmers use to think that the Z80 can map 256 different I/O ports. This is not true. The Z80 has a 16-bit I/O bus. The 09Ah specify the lower 8-bits of the 16-bit value, and the higher 8-bits are the contents of the accumulator. If you are using the IN A,(C), then the value placed in the higher 8-bits is not the accumulator, but instead the Z80 register B. ADVRAM reads the value of the control register from the higher 8-bits of the I/O address. There's no side effect here, since port 09Ah is a write-only port in a normal MSX system. Of the 8 bits of the control register, only two are actually used: Control Register: TD000000 The T (bit 7) is the Turn-on bit. Set it to "1" and the ADVRAM will be enabled. Set it to "0" and the ADVRAM will be disabled (this means you will not be able to read/write in its area, and bank switching will not work). The D (bit 6) is the Deinterlace bit. People with knowledge of MSX hardware know that, in the SCREENs 7 to 12, the VRAM is interlaced, which means consecutive pixels are stored in different chips of VRAM. Programmers usually never notice that, because the VDP handles it in a way the VRAM looks linear. If you set the D bit to "1", ADVRAM will have the same behaviour, and you can program in higher SCREENs the same way you always programmed. ADVRAM Usage There's no trick here. ADVRAM behaves like a Mapper in every aspect, so you can use LD and even LDIR to write and read from it. You can even execute Z80 code from the VRAM! Another nice feature is that you can execute VDP commands while writing/reading from ADVRAM! You get no corrupted data and the VDP commands will not be slowed down. ADVRAM replaces the dynamic RAM used in the VRAMs with the same static RAM used as cache in PC machines, so the speed of ADVRAM is great even on MSX with high clock ( 7 MHz and even higher! )

Availabilty

Ademir Carchano is producing 30 units of ADVRAM. You can contact him through his page: http://www.carchano.com.br The price is R$ 130.00, which is something like US$ 73.00 The ADVRAM is NOT a cartridge. It is a small circuit board which must be connected inside your MSX. You must have some knowledge of hardware to make that, but these days, every MSX user who don't know about hardware has a friend who does. If you want to try it before buying, get the emulator BrMSX 2.5, which has ADVRAM emulation.

Benchmark

When the concept of ADVRAM was first presented on the MSX international mailing list, people were sceptic about its advantages. Based on this, I made a program to test and compare ADVRAM to other methods of accessing VRAM. Since you may not believe in the incredible results obtained, download Video Test 1.0 and try the program by yourself. This program executes six graphic tests, using three different methods: ADVRAM, writing/reading through port 098h, and VDP commands. The main core of the program is the same, you just need to change a EQU in the beggining of the source to select which video engine you want. The package include the source code, and the three precompiled executables: ADVRAM.COM, OUT98H.COM and VDPCMD.COM. To make the result fair and unbiased, some guidelines were followed when designing the routines: The routines must be as optimized as possible The routines make no checking to test if its arguments are valid The routines must not modify any Z80 register. Every register used must be saved with PUSH and POP VDP commands used must not use VRAM->VRAM transfers. (This means no HMMM, YMMM or LMMM) No look-up tables are permitted No memory variables are permitted Secret Z80 instructions and R800-only opcodes are forbidden Loop unrolling is OK You may want to read the source code to check if I indeed followed these guidelines. If you can optimize further any of the routines, please send me a patch. All tests are performed in 10 seconds, and the program counts how many times the routine has executed. Higher results means higher speed. The tests were performed in three configurations: BrMSX 2.5, which has a very good timing and can be used for comparison. If you don't trust me, just get the emulator source code and check it by yourself. A real MSX2+ CIEL-Turbo, which is a standard MSX2+ with Z80, and user-selected clock of 3.57MHz or 7.14MHz. A real MSX Turbo-R GT in Z80 and R800/DRAM modes. The last thing I must talk before presenting the results is that sometimes the VDP is too slow to receive data in the high-speed I'm sending, causing strange video behaviour, and locking the machine after the test. When this happens, just start the program again and press ESC to skip this test. In the tables below, this case will be marked as too slow. Now let's see the results: Generic SCREEN 8 Point This routine must draw a single pixel in SCREEN 8. The VDP command used was PSET. The inputs are: H = Y coordinate of the pixel (0<Y<192) L = X coordinate of the pixel (0<X<256) C = color of the pixel (0<C<256) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 144181 121467 110494 MSX2+ @ 3.57 none 122327 111276 Turbo-R Z80 none 120204 108512 BrMSX @ 7.14 288721 243236 221262 MSX2+ @ 7.14 none 245090 222947 Turbo-R R800 none 222363 141905 ADVRAM 7.14MHz: 288721 Best MSX2+ result (OUT98H @ 7.14): 245090 Best Turbo-R result (OUT98H @ R800): 222363 The ADVRAM was 17% faster than the MSX2+ The ADVRAM was 29% faster than the Turbo-R Generic SCREEN 5 Point This routine must draw a single pixel in SCREEN 5. The VDP command used was PSET. The inputs are: H = Y coordinate of the pixel (0<Y<192) L = X coordinate of the pixel (0<X<256) C = color of the pixel (0<C<16) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 111012 78644 107807 MSX2+ @ 3.57 none 79175 108534 Turbo-R Z80 none 77701 105830 BrMSX @ 7.14 222301 157483 215881 MSX2+ @ 7.14 none 156975 217697 Turbo-R R800 none 136346 141212 ADVRAM 7.14MHz: 222301 Best MSX2+ result (VDPCMD @ 7.14): 217697 Best Turbo-R result (VDPCMD @ R800): 141212 The ADVRAM was 2% faster than the MSX2+ The ADVRAM was 57% faster than the Turbo-R SCREEN 8 Burst Copy This routine must copy a full width horizontal line from RAM to VRAM in SCREEN 8. The VDP command used was HMMC. The inputs are: HL = address of line in RAM D = Y coordinate of the line (0<Y<192) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 7257 7236 7000 MSX2+ @ 3.57 none 7292 7054 Turbo-R Z80 none 6916 6690 BrMSX @ 7.14 14531 14489 14018 MSX2+ @ 7.14 none 13084 too slow Turbo-R R800 none 4430 4264 ADVRAM 7.14MHz: 14531 Best MSX2+ result (OUT98H @ 7.14): 13084 Best Turbo-R result (OUT98H @ Z80): 6916 The ADVRAM was 11% faster than the MSX2+ The ADVRAM was 110% faster than the Turbo-R SCREEN 5 Aligned 8x8 Block Copy This routine must copy a 8x8 block from RAM to VRAM in SCREEN 5, and the coordinates must be multiples of 8. The VDP command used was HMMC. The inputs are: HL = address of block in RAM D = Y coordinate of the line (0<Y<=184, must be a multiple of 8) E = X coordinate of the line (0<X<=248, must be a multiple of 8) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 34094 25772 32207 MSX2+ @ 3.57 none 25946 32424 Turbo-R Z80 none 25006 31087 BrMSX @ 7.14 68273 51607 64493 MSX2+ @ 7.14 none 50282 too slow Turbo-R R800 none 22705 24737 ADVRAM 7.14MHz: 68273 Best MSX2+ result (OUT98H @ 7.14): 50282 Best Turbo-R result (VDPCMD @ Z80): 31087 The ADVRAM was 35% faster than the MSX2+ The ADVRAM was 119% faster than the Turbo-R SCREEN 8 Aligned 8x8 Block Copy This routine must copy a 8x8 block from RAM to VRAM in SCREEN 8, and the coordinates must be multiples of 8. The VDP command used was HMMC. The inputs are: HL = address of block in RAM D = Y coordinate of the line (0<Y<=184, must be a multiple of 8) E = X coordinate of the line (0<X<=248, must be a multiple of 8) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 22744 18706 21149 MSX2+ @ 3.57 none 18850 21311 Turbo-R Z80 none 18048 20341 BrMSX @ 7.14 45543 37459 42350 MSX2+ @ 7.14 none 35635 too slow Turbo-R R800 none 13902 14666 ADVRAM 7.14MHz: 45543 Best MSX2+ result (OUT98H @ 7.14): 35635 Best Turbo-R result (VDPCMD @ Z80): 20341 The ADVRAM was 27% faster than the MSX2+ The ADVRAM was 123% faster than the Turbo-R SCREEN 8 Generic Transparent 8x8 Block Copy This routine must copy a 8x8 block from RAM to VRAM in SCREEN 8. Values "0" should be treated as transparent. The VDP command used was LMMC. The inputs are: HL = address of block in RAM D = Y coordinate of the line (0<Y<=184) E = X coordinate of the line (0<X<=248) ADVRAM OUT98H VDPCMD BrMSX @ 3.57 11370 8974 20579 MSX2+ @ 3.57 none 9042 too slow Turbo-R Z80 none 8848 too slow BrMSX @ 7.14 22767 17969 41209 MSX2+ @ 7.14 none 18100 too slow Turbo-R R800 none 13852 14639 ADVRAM 7.14MHz: 22767 Best MSX2+ result (OUT98H @ 7.14): 18100 Best Turbo-R result (VDPCMD @ R800): 14639 The ADVRAM was 25% faster than the MSX2+ The ADVRAM was 55% faster than the Turbo-R

Credits

This page was created and maintened by Ricardo Bittencourt Last update: 23/4/2000 The ADVRAM original idea was told to me by Márcio of the Other Side MSX Club, during the ExpoSALT 2000 meeting. I liked the idea and told it to Ademir Carchano, who implemented the first hardware prototype in only four months. All the Turbo-R tests were performed in a machine donated to me by Takamichi Suzukawa. All the tests in the CIEL2+ were performed by Luciano Sturaro. Many thanks to the people who contributed to the standard including Daniel Caetano, Werner Kai, Leonard Oliveira, and the people from the #msxzone IRC channel.