Yes, this is one of known problems, which sneaked to 1.9.1. It was reported in issue #1273 and was recently fixed ( https://github.com/ispc/ispc/commit/3009c6fa608bbc01ba50a178ee8cc33e24e38823). My estimation is that it was not showing up too frequently, though apparently it was frequent enough to get noticed and reported.
Thanks for letting us know that it works now! On Wed, Jun 14, 2017 at 10:40 PM, psmilek <[email protected]> wrote: > I have further info about nice behavior of new version ! I ported simple > diffusive water simulation over heightfield terrain > to ISPC and here are results: > > ISPC 1.9.1 > This version displays performance warnings (quite unexpected in this case) > about using scatter / gather instructions > when compiling kernel and it results in very low performance: > > Nexus::TestWaterSimPerformance: water sim took 51.172697 second(s) (1000 > iters, 4096x4096) > > ISPC 1.9.2rc1 > No scatter / gather performance warnings and much better performance: > > Nexus::TestWaterSimPerformance: water sim took 13.282045 second(s) (1000 > iters, 4096x4096) > > > On Monday, June 5, 2017 at 7:01:38 PM UTC+2, Dmitry Babokin wrote: >> >> Nice! Thanks for trying! >> >> On Mon, Jun 5, 2017 at 2:06 AM, psmilek <[email protected]> wrote: >> >>> Hi Dimitry, >>> I tested new release candidate on my ISPC code (various noise >>> generators) and I confirm, that in my case it works fine. >>> There is also nice speedup of generated code as a bonus ! :) >>> >>> Following is result of my benchmark, which consists of filling texture >>> of size 512x512 (float32 format) 100 times with various noises. >>> 'Mp' in following log means fillrate in mega-pixels per second. >>> >>> >>> ISPC 1.9.1 >>> >>> Value2D SIMD4: (512x512), 12 octaves, 17.118988 Mp, 1.531305 second(s) >>> Value2D SIMD8: (512x512), 12 octaves, 37.554026 Mp, 0.698045 second(s) >>> Perlin2D SIMD4: (512x512), 12 octaves, 8.259425 Mp, 3.173877 second(s) >>> Perlin2D SIMD8: (512x512), 12 octaves, 16.031377 Mp, 1.635193 second(s) >>> Cellular2D SIMD4: (512x512), 12 octaves, 6.949015 Mp, 3.772391 second(s) >>> Cellular2D SIMD8: (512x512), 12 octaves, 14.123636 Mp, 1.856066 second(s) >>> >>> >>> ISPC 1.9.2rc1 >>> >>> Value2D SIMD4: (512x512), 12 octaves, 18.599472 Mp, 1.409416 second(s) >>> Value2D SIMD8: (512x512), 12 octaves, 40.285944 Mp, 0.650708 second(s) >>> Perlin2D SIMD4: (512x512), 12 octaves, 8.349595 Mp, 3.139601 second(s) >>> Perlin2D SIMD8: (512x512), 12 octaves, 16.741686 Mp, 1.565816 second(s) >>> Cellular2D SIMD4: (512x512), 12 octaves, 6.973369 Mp, 3.759216 second(s) >>> Cellular2D SIMD8: (512x512), 12 octaves, 13.938580 Mp, 1.880708 second(s) >>> >>> >>> Thanks for your effort in improving ISCP compiler ! >>> >>> >>> >>> On Wednesday, May 17, 2017 at 9:33:58 AM UTC+2, Dmitry Babokin wrote: >>>> >>>> Hello, >>>> >>>> We are going to release ispc 1.9.2 soon and have prepared a release >>>> candidate for those of you who prefer to use pre-built binaries. Please >>>> give it a try and let us know if you see any problems with your code. >>>> >>>> Windows (VS2015): https://drive.google.com/open?id=0Bxh4sVF04yhxYnRR >>>> czhlNzZOblU >>>> Linux: https://drive.google.com/open?id=0Bxh4sVF04yhxdzhwbzlzY1prU1U >>>> MacOS: https://drive.google.com/open?id=0Bxh4sVF04yhxUDlja3BkNGZyTzQ >>>> >>>> Main changes: >>>> - debug support on Windows fixed >>>> - SVML support for AVX512 targets >>>> - several stability fixes >>>> - based on LLVM 3.9 (we are still avoiding LLVM 4.0 due to know >>>> stability and performance problems) >>>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "Intel SPMD Program Compiler Developers" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> -- > You received this message because you are subscribed to the Google Groups > "Intel SPMD Program Compiler Developers" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Intel SPMD Program Compiler Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
