johnfea

This would be probably more work in your case than re-writing dylib deps but I built OpenImageIO deps manually with cmake option: -DBUILD_SHARED_LIBS=OFF. Then OpenImageIO dylib has no external dependencies...

I can reproduce the same or similar issue. This is with MSVC 2022 either with clang compiler supplied with it or with intel one api compiler. liboslexec/wide/wide_opcolor.cpp: ``` __OSL_MASKED_OP2(blackbody, Wv,...

It was a few different issues. Unnecessary dllimport function attribute was causing the issue I showed above and many similar issues. Then, functions definitions in headers followed by unused code...

I tried this again with improved ending for test.cpp that passed asan: ``` delete oslmat; delete ss; ``` And for what ever reason now the sample failed the assert in...

The problem occurs only when there's this type of usage of closure color: ``` BSDF = layer(generalized_schlick_bsdf(),BSDF) ``` So a workaround is to always do this instead: ``` BSDF =...

Based on a quick benchmark, batched SSE2 was 20% to 40% faster in total render speed with low complexity material when a single material filled the screen. 20% was with...

b4_SSE2 ran at 60% the speed of b8_AVX2 mode with single 3D fBm when measured in total render time, but for this measurement all parts of rendering were switched and...

I've updated the pull request, now "closure-string" test fails in batched mode only and for some reason also for non-batched icc/icx. This is the only test that fails for b8_AVX2...

I removed closure-string and testminimal from this PR and added those into a separate PR #1831. Now for b4_SSE2 only "'Invalid bitcast' to i8" should fail in tests and b8_AVX2...

All tests for this PR pass now. I'm not sure why AVX512 needs cases for width 8 in src/liboslexec/llvm_util.cpp and why AVX needs cases for width 16. I tried removing...