Hm, I answer my self on this. Seems ok and expected. I tested the same issue on an intel mac and the log seems similar at least. the reduced 2000x3000 size seems ok at least.
Here is a successful log from my intel mac: https://pastebin.com/BmFD6XMb >From the successful intel machine Now I see some difference in the beginning: nona: using graphics card: NVIDIA Corporation NVIDIA GeForce GT 750M OpenGL Engine destStart=[0, 0] destEnd=[4000, 3000] destSize=[(4000, 3000)] srcSize=[(4000, 3000)] *srcBuffer=0x7f9d97300000 * srcAlphaBuffer=0x0 *destBuffer=0x7f9d99556000* >From the *not* successful M1 machine: nona: using graphics card: NVIDIA Corporation NVIDIA GeForce GT 750M OpenGL Engine destStart=[0, 0] destEnd=[4000, 3000] destSize=[(4000, 3000)] srcSize=[(4000, 3000)] *srcBuffer=0x7fee6ca00000 * srcAlphaBuffer=0x0 *destBuffer=0x7fee78000000* If anyone wants to test fixing this please shout out :). söndag 19 mars 2023 kl. 08:31:22 UTC+1 skrev dudek53: > Hi. Maybe I see something here. In my provided testfile created with nona > I get successfull gpu exports when reducing widht and height destination > size straight into the pro file: > In test.pto I have at the top: > # hugin project file > #hugin_ptoversion 2 > p f0 w4000 h3000 v71.5915 E12.2928 R0 n"TIFF_m c:LZW" > > Change it to: > # hugin project file > #hugin_ptoversion 2 > p f0 *w3700 h2750* v71.5915 E12.2928 R0 n"TIFF_m c:LZW" > > And output is ok. No corrupted half frame. > https://pastebin.com/t6yMfcc8 > > Upon inspection of the output comparing debug log it seems issues are all > happening around *gpu dest chunk* > For instance this place: > > > > > *Source chunks: [(0, 0) to (4000, 3000) = (4000x3000)]Dest chunks: > [(0, 0) to (3704, 2750) = (3704x2750)]Total GPU memory used: 532184000* > > Leads to: > > > > > > > > > > > > > > > *gpu shader program compile time = 0.014gpu shader texture/framebuffer > setup time = 0.001gpu dest chunk=[(0, 0) to (3704, 2750) = (3704x2750)] > coord image render time = 0.049gpu dest chunk=[(0, 0) to (3704, 2750) = > (3704x2750)] source chunk=[(0, 0) to (4000, 3000) = (4000x3000)] src upload > = 0.037gpu dest chunk=[(0, 0) to (3704, 2750) = (3704x2750)] source > chunk=[(0, 0) to (4000, 3000) = (4000x3000)] src render = 0.004gpu dest > chunk=[(0, 0) to (3704, 2750) = (3704x2750)] source chunk=[(0, 0) to (4000, > 3000) = (4000x3000)] interpolation chunk=[(0, 0) to (4, 4) = (4x4)] setup = > 0.009gpu dest chunk=[(0, 0) to (3704, 2750) = (3704x2750)] source > chunk=[(0, 0) to (4000, 3000) = (4000x3000)] interpolation chunk=[(0, 0) to > (4, 4) = (4x4)] render = 0.014gpu dest chunk=[(0, 0) to (3704, 2750) = > (3704x2750)] normalization setup = 0gpu dest chunk=[(0, 0) to (3704, 2750) > = (3704x2750)] normalization render = 0.004gpu dest chunk=[(0, 0) to (3704, > 2750) = (3704x2750)] dest rgb disassembly setup = 0gpu dest chunk=[(0, 0) > to (3704, 2750) = (3704x2750)] dest rgb disassembly render = 0.003gpu dest > chunk=[(0, 0) to (3704, 2750) = (3704x2750)] rgb readback = 0.004gpu dest > chunk=[(0, 0) to (3704, 2750) = (3704x2750)] dest alpha disassembly setup = > 0gpu dest chunk=[(0, 0) to (3704, 2750) = (3704x2750)] dest alpha > disassembly render = 0.003gpu dest chunk=[(0, 0) to (3704, 2750) = > (3704x2750)] alpha readback = 0.002* > > Above is an example of a successful gpu processing. > > > > > > > > Then back to the problematic one: > log file: > https://pastebin.com/BmFD6XMb > > In the problematic test let´s keep following: > In test.pto I have at the top: > # hugin project file > #hugin_ptoversion 2 > p f0 *w4000 h3000* v71.5915 E12.2928 R0 n"TIFF_m c:LZW" > > In the log already we can see a potential conflict: > Testing texture size 16384, result=16384 > maxTextureSize=16384 > Source chunks: > [(0, 0) to (4000, 3000) = (4000x3000)] > Dest chunks: > [(0, 0) to (*2000, 3000*) = (*2000x3000*)] > [(2000, 0) to (4000, 3000) = (2000x3000)] > Total GPU memory used: 348000000 > > *Dest chunk should be 4000, 3000 not as nothing was changed here. T.Modes. > How are your ouput here?* > > *proceeding in the log we see the same issue* > > gpu shader program compile time = 0.014 > gpu shader texture/framebuffer setup time = 0.001 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] coord image render > time = 0.008 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] source chunk=[(0, 0) > to (4000, 3000) = (4000x3000)] src upload = 0.038 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] source chunk=[(0, 0) > to (4000, 3000) = (4000x3000)] src render = 0.003 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] source chunk=[(0, 0) > to (4000, 3000) = (4000x3000)] interpolation chunk=[(0, 0) to (4, 4) = > (4x4)] setup = 0.006 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] source chunk=[(0, 0) > to (4000, 3000) = (4000x3000)] interpolation chunk=[(0, 0) to (4, 4) = > (4x4)] render = 0.009 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] normalization setup > = 0 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] normalization render > = 0.002 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] dest rgb disassembly > setup = 0 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] dest rgb disassembly > render = 0.002 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] rgb readback = 0.004 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] dest alpha > disassembly setup = 0 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] dest alpha > disassembly render = 0.002 > gpu dest chunk=[(0, 0) to (2000, 3000) = (2000x3000)] alpha readback = > 0.001 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] coord image > render time = 0.002 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] source chunk=[(0, > 0) to (4000, 3000) = (4000x3000)] interpolation chunk=[(0, 0) to (4, 4) = > (4x4)] setup = 0.001 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] source chunk=[(0, > 0) to (4000, 3000) = (4000x3000)] interpolation chunk=[(0, 0) to (4, 4) = > (4x4)] render = 0.002 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] normalization > setup = 0 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] normalization > render = 0.001 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] dest rgb > disassembly setup = 0 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] dest rgb > disassembly render = 0.001 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] rgb readback = > 0.013 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] dest alpha > disassembly setup = 0 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] dest alpha > disassembly render = 0.002 > gpu dest chunk=[(2000, 0) to (4000, 3000) = (2000x3000)] alpha readback = > 0.002 > gpu destruct time = 0.005 > > > Any thoughts on this? > > /Daniel > > lördag 18 mars 2023 kl. 17:28:52 UTC+1 skrev dudek53: > >> Hehe, this is beyond the scope of my horizon atm. I tested your >> compilations first when I got here hoping this would fix the issues since >> you compiled on M1. >> Could you maybe test on your worksation if it´s working on your end or >> not with the provided test files? Since it is semiworking here with smaller >> files I am not sure if they remove libraries after Big Sur or if it is >> because of lack of virtual memory or what not on later macbook pros. >> >> I played around with a pto file created in nona and it was able to spit >> out a b&w pgm file and I could also reduce size directly in the pro file >> but other than that I coudl not get any more clues. >> >> By the way. How hard is it to get compiling working on mac m1? I am >> failing miserably over here. makefiles and sources all seem to be using >> older mac architecture. >> >> lördag 18 mars 2023 kl. 17:18:36 UTC+1 skrev [email protected]: >> >>> On my 2015 dual-gpu Mac, I got corrupted images when remapping with the >>> integrated (Intel) GPU forced, but correct images when nona was allowed to >>> use the dedicated (AMD) GPU. >>> >>> From what I remember from the small amount of OpenGL programming I did >>> on that machine, the AMD drivers were looser with extreme values plugged >>> into transcendental functions, whereas the Intel drivers were quicker to >>> return NaN or Infinity. Perhaps that has something to do with it. >>> On Saturday, March 18, 2023 at 3:05:26 AM UTC-4 [email protected] wrote: >>> >>>> Yup. Not opimal at all. Mac philosophy sometimes :(. >>>> At least there´s some direction now what to do if and when that will >>>> happen :). >>>> Big thanks for you help here meanhwile. >>>> /Daniel >>>> >>>> lördag 18 mars 2023 kl. 07:54:16 UTC+1 skrev T. Modes: >>>> >>>>> [email protected] schrieb am Samstag, 18. März 2023 um 03:25:15 UTC+1: >>>>> >>>>> Or port OPenGL to use Metal :). A quick search shows non support for >>>>> OpenGL. I guess this could cause issues. >>>>> >>>>> >>>>> If you (or somebody) rewrite the whole code… >>>>> >>>>> And Metal is Apple only. It will not work on Linux and Windows. >>>>> >>>> -- A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ --- You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/a77df1a5-47a0-4398-802e-ce32acfd9becn%40googlegroups.com.
