Bug while DMA from msu to vram.

[gizaha]

Test1: I have vram graphics in msu file, address $34d3. I dma them to vram but sends only $b2d length (34d3+b2d=4000). Test2: I move the data to msu address $4000 and dma is working ok.

Other times (ingame), i set address to $2000-$2003 wait for data busy, dma data and (sometimes) it reads from wrong (advance) address.

Both problems at vblank. I tried f-blank and close hdma, no difference. No MSU music is playing to lighten the load. US sd2snes 1.10.3, pal snes 2/1/3 SD card has avg 1.275ms and max 1.696ms

[mrehkopf]

Hi,

1. what happens after copying $b2d bytes? SNES DMA should definitely continue but what data is returned by MSU1?
2. can you estimate how much the address is advanced after the busy loop? Could you provide example code?

[gizaha]

1) Dma stops. 2) Because one picture is a thousand words, check it yourself to see the buffer overflow. Use left right and keys. Jump from address 200 to 4000 (keys X-A) to see what happens. If you press A again, its correct. If you have access to uart for read printf you will gain time.

On Tue, Jun 2, 2020 at 1:57 AM mrehkopf notifications@github.com wrote:

Hi,

1. what happens after copying $b2d bytes?
2. can you estimate how much the address is advanced after the busy loop? Could you provide example code?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mrehkopf/sd2snes/issues/137#issuecomment-637168697, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKWMIZAOSQNMFBDXGLOWZGDRUQW7JANCNFSM4NIQR6IA .

;sd2snes test rom

hirom ;vectors org $ffc0 db "sd2snes test rom" org $ffd5 db $31 ;3.58mhz hirom map mode 21 db 0 ;without coprocessor db $9 ;32mbit (9=4mbit, c=32) db 0 ;no ram db $14 ;region=other db $33 ;fixed value

org $ffea
    dw nmi_code

org $fffc
    dw reset

org $8000 reset: ;set cpu native mode sei lda #$01 sta $4200 stz $420c lda #$8f sta $2100 clc xce ;setup stack rep #$10 ldx #$01ff txs ;setup bank phk plb ;setup direct page rep #$20 lda #$0000 tcd

;setup screen mode sep #$30 lda #$01 sta $2105 ;screen mode stz $2106 ;mosaic lda #$f8/2 sta $2107 lda #$f0/2 sta $2108 sta $2109

stz $210b       ;bg1/2 char address
stz $210c       ;bg3/4 char address 
stz $210d   
stz $210d       ;bg1 hor scroll
stz $210e
stz $210e       ;bg1 vertical scroll

lda #$80
sta $2115       ;video port control (2118/9 inc)

stz $2121       ;cgram address
stz $2122       ;first color in palette=black
stz $2122

stz $2123       ;bg12 window disable
stz $2124       ;bg34 window disable
stz $2125       ;oam window disable

stz $212a       ;mask
stz $212b       ;mask
lda #$01
sta $212c       ;bg1 mainscreen
sta $212d       ;subscreen
stz $212e       ;mainscreen window disable
stz $212f       ;subscreen window disable

stz $2130       ;color addition
lda #$01
sta $2131       ;color math
lda #$e0
sta $2132       ;bg color
stz $2133       ;disable pseudohires

;set linear tilemap rep #$20 lda #$f800/2 sta $2116 lda #$2000

• sta $2118 inc cmp #$2400 bne - ;enable nmi ldx #$80 stx $4200 ;reset ram positions
  stz $10 stz $12 stz $14 stz $16 stz $18 ;loop forever
• ;nop ;bra -

-- lda $18 beq --

;wait line 0

• ldx $4212 bmi - ;read keypresses rep #$20 stz $14 lda $4218 sta $10 ;current keypress cmp $12 ;previous keypress beq + sta $12 sta $14 ;one frame keypress

• stz $18 bra --

nmi_code: sep #$20 lda $4210 ;clear nmi flag lda #$8f sta $2100 ;close beam lda #$81 sta $4200 ;80=enable nmi, 1=joypad read

inc $18

.check_keypresses ;that latched in previous frame nmi rep #$20 lda $14 bne + jmp .exit

;RIGHT

• cmp #$100
  bne + lda $16 cmp.w #(.keys_end-.keys)-2/2 beq ++ inc $16 bra ++

• ;LEFT cmp #$200
  bne + lda $16 beq ++ dec $16 bra ++

  ;L

• cmp #$0020 bne + tdc bra +++ ;R

• cmp #$0010 bne + lda #$4200 bra +++ ;Y

• cmp #$4000 bne + lda #$100 bra +++ ;X

• cmp #$0040 bne + lda #$200 bra +++ ;A

• cmp #$0080 bne + lda #$4000 bra +++ ;B

• cmp #$8000 bne .exit lda #$3400 bra +++

;set msu address
++ lda $16 asl tax lda .keys,x +++ sta $2000 stz $2002 ;dma tiles stz $2116 lda #$1809 ;read from msu file sta $4300 lda #$2001 sta $4302 ldx #$00 stx $4304 lda #$3000 sta $4305

• ldx $2000 bmi - ldx #$01 stx $420b

.exit ldx #$0f stx $2100 ;open beam rti

.keys dw 0, $800, $1000, $1800, $2000, $2800, $3000, $3800, $4000, $4800 .keys_end

[gizaha]

I forgot: 1) When the bug occurs, dma stops. If i read 4305 is zero. Also it's NOT at "data busy". So i couldn't detect when the bug occurs to make a workaround. I cannot predict how advance is, i see advance frames from next videos but couldn't calculate the address.

On Tue, Jun 2, 2020 at 3:01 AM George Zachariadis gizaha1@gmail.com wrote:

1) Dma stops. 2) Because one picture is a thousand words, check it yourself to see the buffer overflow. Use left right and keys. Jump from address 200 to 4000 (keys X-A) to see what happens. If you press A again, its correct. If you have access to uart for read printf you will gain time.

On Tue, Jun 2, 2020 at 1:57 AM mrehkopf notifications@github.com wrote:

Hi,

1. what happens after copying $b2d bytes?
2. can you estimate how much the address is advanced after the busy loop? Could you provide example code?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mrehkopf/sd2snes/issues/137#issuecomment-637168697, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKWMIZAOSQNMFBDXGLOWZGDRUQW7JANCNFSM4NIQR6IA .

[gizaha]

It's not DMA problem but sd2snes msu buffer. Instead of DMA, i upload byte by byte reading 2001 and writing to 2118-2119. I have the same result. This took 3 frames, so it's not read speed error. msu1.c has bug at creating it's data buffer.

[mrehkopf]

Indeed there was a bug where the read offset status flag wasn't updated correctly on a buffer hit optimization case. 47bd526f should fix this.

[gizaha]

Any update on this? How can i use the 47bd526 changes? Shall i compile the firmware myself? I work at sfa2 msu1 and this bug is holding me back again. Also please mark this as bug, and an important one, as you read corrupted data.

[mrehkopf]

Fixed in v1.11.0b1. On a semi-related note, some performance improvements may still be required for data seeking in general. Unfortunately MSU1 doesn't provide a way to tell the hardware how much data the program needs in advance so it always buffers the entire 16384 bytes which makes it uneconomical to try and load from multiple offsets per VBlank. I have some ideas for optimization but want to sort out bugs of the current beta release first.