search

microsoft / DirectXMath

DirectXMath is an all inline SIMD C++ linear algebra library for use in games and graphics apps
https://walbourn.github.io/introducing-directxmath/
MIT License
1.52k stars 234 forks source link

The clang/LLVM returns bad results for specific code optimizations #127

Closed shinjich closed 5 months ago

shinjich commented 3 years ago

The clang/LLVM + /Ox + /fp:fast + XMVectorRound built for SSE/SSE2 generated the bad results. XMVectorRound are used in trigonometric functions such as XMVectorSinCos, etc.

Repro code: https://github.com/shinjich/directxmathtest/tree/master/VerifyRounding The reproduction code compares the results of standard library, original XMVectorSinCos and modified version of XMVectorSinCos.

Results compiled and executed with.. /Ox /fp:fast (Bad results) https://github.com/shinjich/directxmathtest/blob/master/VerifyRounding/_Result_fp_fast.txt /Ox /fp:precise (Good results) https://github.com/shinjich/directxmathtest/blob/master/VerifyRounding/_Result_fp_precise.txt /Ox /fp:fast + SVML (Good results) https://github.com/shinjich/directxmathtest/blob/master/VerifyRounding/_Result_fp_fast_SVML.txt

This issue was not seen with VC.

walbourn commented 5 months ago

The issue here is that #pragma float_control was not in place for clang as it is for MSVC, but more importantly this pragma on clang doesn't work correctly for SSE instrinsics :(

https://github.com/llvm/llvm-project/issues/55713

Therefore the recommendation is going to be to NOT use /fp:fast for clang until this is resolved.