Got it switched over to 16bpp. Still looks the same though because using same graphics, just switched to 16bpp...
Next is to make some better graphics.
Also, need to figure out how to dim a 16bpp color.
Think need to RGBEXP on it, then do something magic, then do RGBSQZ
With 16 bit wide PSRAM, don't need to break the buffer copying into two parts.
Doesn't actually work when broken in two. Have to get back to 4-bit PSRAM and see if that was actually working...
Sadly, seems that 4bit PSRAM is not fast enough for display buffer to be in PSRAM. 16bit works, but would really like to support 4bit.
Think will change direction and have both working and display buffers be in HUB.
Hopefully, there is still enough HUB space left to add sprites and such later...
4 bit PSRAM will have a raw bandwidth of about 75MBps at a P2 clock of 300MHz. Some of that is lost to overheads - I think I can manage to get 8bpp VGA and SVGA video out of it but it then doesn't leave a lot spare for writes. 16 bit colour doubles that requirement and if you are writing an entire frame at the same time I doubt it can work for VGA. 320x200 is smaller so may still be possible though. You'd have to do the numbers and probe the CS lines of a PSRAM to see how much free bandwidth is left over when attempting it. It was surprising to me how much a single nibble of PSRAM could do in the end after tuning the burst sizes to maximize read bandwidth possible.
This is based on the "fast block move" section of the P2 Silicon Doc.
There is something mysterious going on though... First tried doing 320 longs at a time start from the beginning of LUT.
This failed miserably, then remembered moved some code to LUT, so that's a problem...
Then, changed the FIT at the end of LUT code to this "fit $2FF".
Tried doing 160 long reads from $300, but that also fails.
Seems like something is using the end of LUT, but have no idea what...
This is fine though. Maximum frame rate increased from 90 fps to 100 fps with P2 clock at 300 MHz.
Well above the 60 Hz needed.
Using Midjourney to generate new graphics. It's going to be a process to get it to do something useful.
Think it's getting there though.
Having but buffers in HUB takes up most of HUB RAM. But, there's still 90 kB left.
Want to add variety of floor and wall tiles next.
Wall and floor tiles are 8 kB each. So, maybe don't need to dynamically load those to hub right away...
Added a static status bar as an image in PSRAM that gets copied to buffer every frame.
In future, this can be written on top of to show useful things...
Max. frame rate dropped from 90 to 85 fps, still good.
Seeing some strangeness, think due to graphics cogs being nearly full.
The "fit $1F0" doesn't save it from going off the rails, but a smaller fit does.
Also, seems to depend on whether in debug mode or not...
Shouldn't asm debug statements change the fit count?
Was hoping to try this out but am having issues getting this code to work. Are you using PNut or flexspin @Rayman? I'm having trouble with your latest SpinMain1c version above and the startup of the FAT engine - it's failing right away before even mounting the partition yet all it does is assign pins. I modified to use the normal P2-EVAL pins 58-61 and am running a recent flexspin 7.5.0 that I built about a month ago. Not sure what is happening here, seems like it is written to work with aborts but I don't see any in the called code at init time.
...snip...
'Start uSD driver
debug("Starting fat engine")
if \FS.FATEngineStart(uSD_MISO, uSD_CLK, uSD_MOSI, uSD_CS, 0)
debug("Missing SD Card 1?")
repeat
if \FS.mountPartition(0)
debug("Can't mount FS")
debug("Missing SD Card?")
repeat
'if \FS.openfile(string("bgr4x.bmp"),"R")
if \FS.openfile(string("back1b.bmp"),"R")
debug("Missing file?")
repeat
Then when I forcibly bypass the first FATEngineStart failure by not processing any abort, it gets further but stops below. Can't tell what is going wrong. I tried a couple of SD cards and the requested file is definitely on the card in the top level folder.
Cog0 SD clock divider set to sysclock/750 (0.4 MHz)
Cog0 v2 Card Idle
Cog0 SD clock divider set to sysclock/8 (37.5 MHz)
Cog0 OCR register $80FF_8000
Cog0 SDSC Card
Cog0 Can't mount FS
Cog0 Missing SD Card?
EDIT: I was able to bypass the bmp file loading stuff and at least see your 3d graphics engine working with a USB keyboard. So I'm happier now. Be awesome to see this at true 640x480 instead of pixel doubled but I guess the performance won't be fast enough...
Looking at the source code for FATEngineStart() it always returns a non-zero value. Positive values are "cog+1" it is running on, and -1 for error.
Yeah the code didn't make sense running the abort. The FATEngineStart seemed to call sdspi.getCog() method which returned 1 (hard coded) but this somehow triggered the abort code to be executed.
As for mountPartition(), it seems to be a confused return code according to its own docs.
'' // Returns true if the partition was improperly unmounted the last time it was mounted and false if not.
'' //
'' // If an error occurs this method will abort and return a pointer to a string describing that error.
@rogloh said:
Yeah the code didn't make sense running the abort. The FATEngineStart seemed to call sdspi.getCog() method which returned 1 (hard coded) but this somehow triggered the abort code to be executed.
It's just the IF true condition right there in Rayman's code. All results are true so it's always going to abort. I suspect other examples don't check anything. This might be the first time anyone has tried to check return codes from FATEngine.
@rogloh said:
Yeah the code didn't make sense running the abort. The FATEngineStart seemed to call sdspi.getCog() method which returned 1 (hard coded) but this somehow triggered the abort code to be executed.
It's just the IF true condition right there in Rayman's code. All results are true so it's always going to abort. I suspect other examples don't check anything. This might be the first time anyone has tried to check return codes from FATEngine.
Yeah I figured that sequence was bad so I skipped past the abort by converting to look for a negative error code instead, but the rest of the code still failed to mount and read the file. I guess it's my choice of SD card or something. I've sort of expected SDs to just fail to work on the P2 these days after being burned so often in the past before. It's hit or miss. I have the P2-EVAL dip switches set to BOD=off, Flash=off, P59 up=on, P59 down=off.
This is where it fails now (it pauses for a second after the clock divider output).
Cog0 INIT $0000_0000 $0000_0000 load
Cog0 INIT $0000_0404 $0000_0000 load
Cog0 Starting fat engine
Cog0 SD clock divider set to sysclock/750 (0.4 MHz)
Cog0 Can't mount FS
Cog0 Missing SD Card?
Mount code has this - so it looks like the ocr register cannot be read. This was the 32GB Sandisk. EDIT: actually it must be failing before this as it only prints the first divider message, not the second.
Comments
Got QVGA in 16bpp from PSRAM going. That's the first step...
Got it switched over to 16bpp. Still looks the same though because using same graphics, just switched to 16bpp...
Next is to make some better graphics.
Also, need to figure out how to dim a 16bpp color.
Think need to RGBEXP on it, then do something magic, then do RGBSQZ
With 16 bit wide PSRAM, don't need to break the buffer copying into two parts.
Doesn't actually work when broken in two. Have to get back to 4-bit PSRAM and see if that was actually working...
Sadly, seems that 4bit PSRAM is not fast enough for display buffer to be in PSRAM. 16bit works, but would really like to support 4bit.
Think will change direction and have both working and display buffers be in HUB.
Hopefully, there is still enough HUB space left to add sprites and such later...
4 bit PSRAM will have a raw bandwidth of about 75MBps at a P2 clock of 300MHz. Some of that is lost to overheads - I think I can manage to get 8bpp VGA and SVGA video out of it but it then doesn't leave a lot spare for writes. 16 bit colour doubles that requirement and if you are writing an entire frame at the same time I doubt it can work for VGA. 320x200 is smaller so may still be possible though. You'd have to do the numbers and probe the CS lines of a PSRAM to see how much free bandwidth is left over when attempting it. It was surprising to me how much a single nibble of PSRAM could do in the end after tuning the burst sizes to maximize read bandwidth possible.
Switched display buffer back to hub.
Updated the code that copies the offscreen buffer to the display buffer from this:
mov my,#120 myloop4 mov mx,#2*320/4 rdfast #320*2,ptra mxloop4 rflong ml wrlong ml,ptrb++ djnz mx,#mxloop4 add ptra,##320*2 djnz my,#myloop4to this:
mov my,#120 myloop4 setq2 #40-1 rdlong LutBuf,ptra++ setq2 #40-1 wrlong LutBuf,ptrb++ setq2 #40-1 rdlong LutBuf,ptra++ setq2 #40-1 wrlong LutBuf,ptrb++ setq2 #40-1 rdlong LutBuf,ptra++ setq2 #40-1 wrlong LutBuf,ptrb++ setq2 #40-1 rdlong LutBuf,ptra++ setq2 #40-1 wrlong LutBuf,ptrb++ djnz my,#myloop4This is based on the "fast block move" section of the P2 Silicon Doc.
There is something mysterious going on though... First tried doing 320 longs at a time start from the beginning of LUT.
This failed miserably, then remembered moved some code to LUT, so that's a problem...
Then, changed the FIT at the end of LUT code to this "fit $2FF".
Tried doing 160 long reads from $300, but that also fails.
Seems like something is using the end of LUT, but have no idea what...
This is fine though. Maximum frame rate increased from 90 fps to 100 fps with P2 clock at 300 MHz.
Well above the 60 Hz needed.
Using Midjourney to generate new graphics. It's going to be a process to get it to do something useful.
Think it's getting there though.
Having but buffers in HUB takes up most of HUB RAM. But, there's still 90 kB left.
Want to add variety of floor and wall tiles next.
Wall and floor tiles are 8 kB each. So, maybe don't need to dynamically load those to hub right away...
Feeling like simpler tiles are better...
Also, remembered that want to darken far away tiles...
Implemented darkening of further away walls and floors.
Used the simplest method, but may look into the other ways from here:
https://forums.parallax.com/discussion/174921/
This works, but max. frame rate drops from 100 to 90... Still good.
rdword f2,ptra DarkenPixel mov dista,dist mul dista,#50 shr dista,#8 fle dista,#255 mov level,#255 sub level,dista movbyts level, #%%0000 ' make multiplier rgbexp f2 mulpix f2, level ' scale the color rgbsqz f2Found some floor tiles to play with here:
https://opengameart.org/content/200-tile-floor-textures
Added a static status bar as an image in PSRAM that gets copied to buffer every frame.
In future, this can be written on top of to show useful things...
Max. frame rate dropped from 90 to 85 fps, still good.
Seeing some strangeness, think due to graphics cogs being nearly full.
The "fit $1F0" doesn't save it from going off the rails, but a smaller fit does.
Also, seems to depend on whether in debug mode or not...
Shouldn't asm debug statements change the fit count?
Was hoping to try this out but am having issues getting this code to work. Are you using PNut or flexspin @Rayman? I'm having trouble with your latest SpinMain1c version above and the startup of the FAT engine - it's failing right away before even mounting the partition yet all it does is assign pins. I modified to use the normal P2-EVAL pins 58-61 and am running a recent flexspin 7.5.0 that I built about a month ago. Not sure what is happening here, seems like it is written to work with aborts but I don't see any in the called code at init time.
...snip... 'Start uSD driver debug("Starting fat engine") if \FS.FATEngineStart(uSD_MISO, uSD_CLK, uSD_MOSI, uSD_CS, 0) debug("Missing SD Card 1?") repeat if \FS.mountPartition(0) debug("Can't mount FS") debug("Missing SD Card?") repeat 'if \FS.openfile(string("bgr4x.bmp"),"R") if \FS.openfile(string("back1b.bmp"),"R") debug("Missing file?") repeatPUB start_explicit(iDO, iCLK, iDI, iCS) : rc debug("start explicit") do_pin := iDO clk_pin := iCLK di_pin := iDI cs_pin := iCSloadp2 -t -b 2000000 raycast7a_SpinMain1c.binary
( Entering terminal mode. Press Ctrl-] or Ctrl-Z to exit. )
Cog0 INIT $0000_0000 $0000_0000 load
Cog0 INIT $0000_0404 $0000_0000 load
Cog0 Starting fat engine
Cog0 start explicit
Cog0 Missing SD Card 1?
Then when I forcibly bypass the first FATEngineStart failure by not processing any abort, it gets further but stops below. Can't tell what is going wrong. I tried a couple of SD cards and the requested file is definitely on the card in the top level folder.
Cog0 SD clock divider set to sysclock/750 (0.4 MHz)
Cog0 v2 Card Idle
Cog0 SD clock divider set to sysclock/8 (37.5 MHz)
Cog0 OCR register $80FF_8000
Cog0 SDSC Card
Cog0 Can't mount FS
Cog0 Missing SD Card?
EDIT: I was able to bypass the bmp file loading stuff and at least see your 3d graphics engine working with a USB keyboard. So I'm happier now. Be awesome to see this at true 640x480 instead of pixel doubled but I guess the performance won't be fast enough...
hmm, I see it's reporting as a SDSC card. Is it 2GB?
Looking at the source code for FATEngineStart() it always returns a non-zero value. Positive values are "cog+1" it is running on, and -1 for error.
One was yeah, and this one typically works with the P2. The other is a 32GB Sandisk Ultra (a Raspi disk but has a FAT32 boot partition).
Yeah the code didn't make sense running the abort. The FATEngineStart seemed to call sdspi.getCog() method which returned 1 (hard coded) but this somehow triggered the abort code to be executed.
PUB getCog() : rc return 1 ' no cogs here!As for
mountPartition(), it seems to be a confused return code according to its own docs.It's just the IF true condition right there in Rayman's code. All results are true so it's always going to abort. I suspect other examples don't check anything. This might be the first time anyone has tried to check return codes from FATEngine.
Yeah I figured that sequence was bad so I skipped past the abort by converting to look for a negative error code instead, but the rest of the code still failed to mount and read the file. I guess it's my choice of SD card or something. I've sort of expected SDs to just fail to work on the P2 these days after being burned so often in the past before. It's hit or miss. I have the P2-EVAL dip switches set to BOD=off, Flash=off, P59 up=on, P59 down=off.
This is where it fails now (it pauses for a second after the clock divider output).
Cog0 INIT $0000_0000 $0000_0000 load
Cog0 INIT $0000_0404 $0000_0000 load
Cog0 Starting fat engine
Cog0 SD clock divider set to sysclock/750 (0.4 MHz)
Cog0 Can't mount FS
Cog0 Missing SD Card?
Mount code has this - so it looks like the ocr register cannot be read. This was the 32GB Sandisk. EDIT: actually it must be failing before this as it only prints the first divider message, not the second.
PUB mountCard() : rc | clk, di, cs, ocr clk := clk_pin di := di_pin cs := cs_pin ocr := (clkfreq + 400_000) / 800_000 ' 400 kHz during init xwait := ocr - 6 rc := clkfreq +/ (ocr * 200_000) debug("SD clock divider set to sysclock/",udec_(ocr*2)," (",udec_(rc/10),".",udec_(rc//10)," MHz)") pinf(cs) waitus(10) pinh(cs) ' Deselect the card pinh(di) pinh(clk) repeat 600 read() cmd(0, 0) cmd(8, $1aa) read32() if cmd(0, 0) <> 1 ' R1 response deselect() return -40 ' abort, card not responding if cmd(8, $1aa) == 1 ' R7 response (R1 + IFC) if read32() <> $1aa deselect() return -40 ' abort, card error debug("v2 Card Idle") else debug("v1 Card Idle") cmd(16, 512) ' R1 response, set block size repeat 200 cmd(55, 0) ' R1 response rc := cmd(41, $4020_0000) ' R1 response if rc <> 1 quit waitms(1) if rc deselect() return -40 ' could not initialize card xwait := 0 ' full speed rc := clkfreq +/ (8 * 100_000) debug("SD clock divider set to sysclock/8 (",udec_(rc/10),".",udec_(rc//10)," MHz)") cmd(58, 0) ' R3 response (R1 + OCR) ocr := read32() deselect() debug("OCR register ",uhex_(ocr)) if ocr>>30 <> 3 ' check CCS bit sdhc := 0 debug("SDSC Card") ' byte addressed card type else sdhc := 1 debug("SDHC/SDXC Card") ' block addressed card type return sdhc +1 ' return card type 2=SDHC,1=SDThe other card (2GB) gets further but still fails.
Cog0 INIT $0000_0000 $0000_0000 load
Cog0 INIT $0000_0404 $0000_0000 load
Cog0 Starting fat engine
Cog0 SD clock divider set to sysclock/750 (0.4 MHz)
Cog0 v2 Card Idle
Cog0 SD clock divider set to sysclock/8 (37.5 MHz)
Cog0 OCR register $80FF_8000
Cog0 SDSC Card
Cog0 Can't mount FS
Cog0 Missing SD Card?
Right, not going "Idle". So not even the initial CMD0/CMD8 working.
I assume all is fine with those cards using my C based testers?
Haven't tried yet, though I know the smaller 2GB card works out okay with Wuerfel_21's emulator stuff - that's why I sort of use it.
One thing I'm finding out about is that mountPartition seems to be returning -1 due to this strange looking code:
readWriteFATBlock(0, "R") debug("fstype=", UDEC_LONG(fileSystemType), UHEX_LONG(readFATEntry(1))) result := ((readFATEntry(1) >> (14 + (12 & fileSystemType))) <> $3) debug("got a result here", UDEC_LONG(result))Cog0 v2 Card Idle
Cog0 SD clock divider set to sysclock/8 (37.5 MHz)
Cog0 OCR register $80FF_8000
Cog0 SDSC Card
Cog0 fstype=fileSystemType = 0, readFATEntry(1) = $0000_3FFF
Cog0 got a result hereresult = 4_294_967_295
Cog0 Can't mount FS i = -1
Cog0 Missing SD Card?