 Notice. New forum software under development. It's going to miss a few functions and look a bit ugly for a while, but I'm working on it full time now as the old forum was too unstable. Couple days, all good. If you notice any issues, please contact me. |
Forum Index : Microcontroller and PC projects : PicoMite V6.00.02 release candidates - all versions
  Page 26 of 37   |
|||||
| Author | Message | ||||
| Bleep Guru  Joined: 09/01/2022 Location: United KingdomPosts: 570 |
Yes, I found that, as I mentioned, when I changed between using individual pixel blanking, which was faster on a 2040 VGA and LCD, to a frame buffer, which is faster on a 2350 (don't understand why though, memory bandwidth?), and clearing the whole screen with CLS, this freed up several large arrays ',y,z,f(16,65),l(16,65) 'Dim Integer e(65) Initially I just left them in place, until I noticed that by removing them it made a significant difference to the speed of the program? PS, I'm currently failing to get my 2040 VGA going again after updating, the USB is not working for some reason, almost certainly a problem with my hardware. Edited 2025-04-14 20:41 by Bleep |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
OK, got my 2040 VGA USB running on RC12. Just found another interesting feature. These observations apply to the 2040 VGA, so far, will now try 2350 HDMI. If I run Bubble at 378MHz, I get the following. Bubble running with and without file opening. 416mS and 607mS Un comment the two places in the code, which have. ,y,z,f(16,65),l(16,65) Dim Integer e(65) They are not used, but get allocated. 416mS and 669mS Note that the speed with the extra memory allocated slows a lot, but with the file open it does not change? Very curious. Edited 2025-04-14 21:19 by Bleep |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
V6.00.02RC14 is available on https://geoffg.net/Downloads/picomite/PicoMite_Beta.zip Fixes bug in play pause with i2s audio Enables mouse cursor in higher res VGA and HDMI editing Mouse cursor now updates line and character after left press New temporary option OPTION FAST TRIG ON/OFF Setting this to ON in a program tells the firmware to use single precision calculations for the various trig functions. This can significantly speed up processing for programs that do not require full resolution trig calcs and use a lot of trig I don't think I can do anything about the performance variations discussed above. I'm now pretty sure they are to do with memory allocation of code and variables. The RP chips have a paged memory structure where things like DMA can read from one page without affecting performance of system activity on another page. However, they don't have enough memory to allow me to partition the application code and data into different pages so all I can suggest is that for performance critical applications you play with how you order dimensioning data and things like framebuffers to achieve the optimum performance in your specific case. |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
I get exactly the same effect on the 2350 HDMI running at 372MHz. Bubble running with and without file open. 235mS and 285mS Un comment large arrays. 235mS and 330mS ? Crossed posts, I'll try your new reduced precision Trig function. Edited 2025-04-14 21:29 by Bleep |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
Using latest RC14. Using the new 'OPTION FAST TRIG ON' I only see a couple of mS speed up on a 2350 HDMI, presumably because of the partial hardware acceleration, is that what you'd expect?. On the 2040 VGA, I'm seeing about a 6 to 9% speed up, when using Fast Trig. Exact figures at 378MHz. Standard Bubble. 622mS - Bubble with open file. 429mS Now with fast trig. Standard Bubble. 588mS - Bubble with open file. 394mS A useful amout of speed up, Thanks. :-) |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
One last data point, incase it triggers a brain wave. Standard Bubble, with file open, & heap and without. & heap Exactly as original, 225mS 165888 and 276mS 166400 Add 'Dim Integer e(1)' 252mS 165632 and 276mS 166144 No change by increasing dim size. Make 'Dim Integer e(31)' 252mS 165632 and 276mS 166144 Make 'Dim Integer e(32)' 278mS 165376 and 277mS 165888 So the speed drops in steps as the heap gets larger by increasing the dim size when a file is open, but the speed does not change with no file open. Note that the speed when the heap jumps to 165376, with an open file, gives the exact same speed as with no file open, ie no speed up. (when e is an array of 33). This also happens to be the point when the heap is 165888, without an open file, which is exactly the same size as the original, with no dim defined, but with a file open and the fastest speed?? But whatever I try I can't get even close to the 225mS by simply playing around with variable definitions or Frame buffers. Is there anything I can define that shifts or alligns the memory layout in different ways, maybe defining a C function or something, so I can experiment a bit more? Edited 2025-04-15 01:30 by Bleep |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4825 |
Kevin, The moment of creating the framebuffer will move all arrays in ram memory. This was essential for the logic analyzer, where I needed to create a ring buffer before i created the framebuffer. Volhput PicomiteVGA PETSCII ROBOTS |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
Not in HDMI or RP2350 VGA as the framebuffer memory is pre-allocated and the same symptoms occur in both  |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
Eureka !!! The issue is to do with how long it takes to allocate temporary memory to parse the commands. This is fastest if the program finds free memory right at the top of the heap and doesn't need to search down it. By fluke, the OPEN command allocates 512 bytes of memory which it then frees leaving a hole at the top of memory. We can simulate this by allocating 512 bytes of memory which we then free before entering the tight loop. See the attached. Now I need to discuss with Geoff, Tom and Gerry to see if we can do this in some sort of systemic way to improve performance of MMbasic ' Bubble Universe '2 Nested loops each 52 =2704 points calculated VGA 'Pico 2040Anal @ 378MHz normal 316mS Framebuff 397mS 'Pico 2350Anal @ 378MHz normal 185mS Framebuff 185mS '2 Nested loops each 66 =4356 points calculated VGA 'Pico 2040Anal @ 378MHz normal 553mS Framebuff 683mS 'Pico 2040Anal @ 378MHz Framebuff/CLS 607ms 'Pico 2350Anal @ 378MHz normal 341mS Framebuff 341mS 'Pico 2350HDMI @ 372MHz normal 333mS Framebuf 333mS 'Pico 2350HDMI @ 372MHz normal 278mS Framebuf 279mS CLS,remove arrays MODE 2 Font 8 Option fast trig on Dim SSS$,qqq$ 'allocate some memory at the top of the heap FRAMEBUFFER create FRAMEBUFFER write f 'Play mp3 "B:/hmp3" 'Play wav "b:/16" Pause 1000 'Play pause Dim Float u(65),w(65),p,q,t,v=0,x=0,b Dim Integer c(65),d(65),n(65),m(32,65),nn,dd,xc,yc,xs,ys Dim Integer a,g,i,j,cc',y,z,f(16,65),l(16,65) 'Dim Integer e(65):Math Set 2,e() Const r=(2*Pi)/235,k=255,s=50 CLS RGB(black) t=Rnd*10 't=1 nn=Peek(varaddr n()) dd=Peek(varaddr d()) cc=Peek(varaddr c()) 'calculate centre and scale factor xc=MM.HRES\2:yc=MM.VRES\2 xs=MM.HRES/4.2:ys=MM.VRES/4.0 'Oval 'xs=MM.HRes/6:ys=MM.VRes/4 'Circular Erase sss$,qqq$ 'free the memory to create space for command parsing For a=0 To 65 For g=0 To 65 If a<18 And g<18 Then n(g)=RGB(0,255,0) Else n(g)=RGB(a*3.93,g*3.93,128*(a+g<65)) EndIf Next 'g Memory pack nn, Peek(varaddr m(0,a)),66,32 'pack pixel colours Next 'a Do CLS Inc t,0.035:g=0:Print Timer:Timer =0 For i=60To 255Step 3 'z=Peek(varaddr f(0,g)):Memory unpack z,cc,66,16 'y=Peek(varaddr l(0,g)):Memory unpack y,dd,66,16 'Pixel c(),d(),0'Box c(),d(),e(),e(),0,,0 b=r*i+t For j=0To 65:u(j)=Sin(i+v)+Sin(x):v=Cos(i+v)+Cos(x):x=u(j)+b:w(j)=v:Next Math Scale u(),xs,c():Math Scale w(),ys,d():Math add c(),xc,c():Math Add d(),yc,d() Memory unpack Peek(varaddr m(0,g)),nn,66,32 'unpack pixel colours Pixel c(),d(),n()'Box c(),d(),e(),e(),0,,n() 'Memory pack cc,z,66,16:Memory pack dd,y,66,16 Inc g Next 'i FRAMEBUFFER copy f,n Loop |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4825 |
@Peter, If you allocate a fixed amount of memory on the top of the heap, before user can allocate heap with LOCAL, DIM and FRAMEBUFFER, can you please check if user space can start at a 16k boundary for the logic analyzer ring buffer ? -Or- eat 1k from the LOCAL variables RAM space. Thanks, Volhout B.t.w: good work that you found it! Solving this will indeed make the performance more consistent. Edited 2025-04-15 03:50 by Volhout PicomiteVGA PETSCII ROBOTS |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
Fantastic, let's hope it makes a difference elsewhere, not just the bubble. :-) Is 512 bytes enough for any sized program, or only enough for something small like Bubble? Or does the free ram on the top of the heap somehow need to be dynamically sized? Edited 2025-04-15 06:59 by Bleep |
||||
| JanVolk Senior Member  Joined: 28/01/2023 Location: NetherlandsPosts: 217 |
Peter, > option list PicoMiteHDMI MMBasic USB RP2350A Edition V6.00.02RC14 OPTION SERIAL CONSOLE COM1,GP0,GP1 OPTION SYSTEM I2C GP20,GP21 OPTION FLASH SIZE 4194304 OPTION COLOURCODE ON OPTION KEYBOARD US OPTION CPUSPEED (KHz) 372000 OPTION RESOLUTION 1280x720 OPTION DISPLAY 30, 80 OPTION HDMI PINS 1, 3, 7, 5 OPTION SDCARD GP22, GP6, GP7, GP4 OPTION AUDIO GP26,GP27', ON PWM CHANNEL 5 OPTION RTC AUTO ENABLE OPTION MODBUFF ENABLE 192 OPTION PLATFORM OLIMEX USB OPTION DEFAULT FONT 3, 1 Various tests done with a separate HDMI monitor and a somewhat older TV with HDMI. With this version the mouse cursor works perfectly with the right mouse button and the Ln: and Cur: are adjusted neatly and with Pg Up and Pg Dn and Backspace it can now be used perfectly. What is noticeable is that if you want to move the mouse outside the last line, a sync problem occurs in the block of the cursor and disappears again when you move up. It would also be useful to mention in the manual for each option resolution how many lines and how many characters the different modes in the editor can handle. Here are some results with Option resolution xxx with HDMI and USB so a separate computer with SD, HDMI, MOUSE, Keyboard, AUDIO, RTC etc. Compare it with a Commodore64 and Floppy drive and this composition. The HDMI monitor. xxx Ln: Char: 640 38 80 720 31 90 800 48 100 848 38 106 1024 30 64 1280 28 80 The HDMI TV. xxx Ln: Char: 848 38 106 1024 30 64 1280 does give a very nice image but unfortunately the outer characters fall outside the image on all sides of the screen. The other 640, 720, 800 do work but are very small on this screen. Conclusion with the HDMI monitor with 1280 a beautiful image and once you have worked with it for a while you do not want to go back to the terminal. With HDMI TV 1024 gives a good clear image spread over the entire screen. So try all resolutions for the best result. Thanks Peter. Greetings, Jan. |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
Harm and anyone else interested Please could you try the attached with your bank of programs. Tested OK with Petscii and LA_24_2 + my usuals. Fixes the issue in bubble and give a performance improvement in all programs tested varying from very little to almost 50% All "fixes" in bubble should be removed to prove the change PicoMiteVGARP2040.zip |
||||
| Bleep Guru Joined: 09/01/2022 Location: United KingdomPosts: 570 |
Hi Peter, I've tried with your sugestion of declaring a couple of strings and then erasing them and that definately works. Unfortunately all my VGA Mites that are easily updatable are USB so I can't currently test your new improved firmware. :-( Footnote added 2025-04-15 19:45 by Bleep or they are 2350. |
||||
| bfwolf Regular Member  Joined: 03/01/2025 Location: GermanyPosts: 68 |
Hi Peter. Are you ready to reveal the secret of which strategy you used to provide "fast RAM" to the command parser?  |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4825 |
@Peter, The version looks stable to me. Tried the variety of games and programs, all run. Just started a game on the basic chess game. If that still runs tomorrow, testing should be complete. I can confirm some performance increase on 2040 VGA PS2 (varies per program, but averages around 8-10%) This is great !! Volhout Edited 2025-04-15 21:44 by Volhout PicomiteVGA PETSCII ROBOTS |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
I've just re-engineered the solution a different (better/faster?) way. Please test to destruction Here is the HDMIUSB version for test. PicoMiteHDMIUSBV6.00.02RC14.zip and the RP2040 VGA version PicoMiteRP2040VGAV6.00.02RC14.zip |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4825 |
The VGA version may be USB version ? It does not connect on USB to my teraterm. Volhout PicomiteVGA PETSCII ROBOTS |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
Please try again - works for me loaded from the zip. If not try clear-flash but shouldn't be needed. PicoMiteRP2040VGAV6.00.02RC14.zip Edited 2025-04-15 22:09 by matherp |
||||
| matherp Guru Joined: 11/12/2012 Location: United KingdomPosts: 10015 |
Please ignore for the moment - found one small quirk |
||||
| Page 26 of 37 |
|||||
 Print this page |
 |
The Back Shed's forum code is written, and hosted, in Australia. | © JAQ Software 2025 |
