https://github.com/tex3d commented:
I think we should treat the DXIL op `WavePrefixBitCount` as completely separate from the other `WavePrefix` op that uses `WaveOpKind`. The fact that they share the same OpCodeClass in DXC's DXIL definition was a mistake - there's a [comment about this](https://github.com/microsoft/DirectXShaderCompiler/blob/df50f51f8fcbb19cfefa3d7135256af6c55dd9c3/utils/hct/hctdb.py#L3568-L3570) in hctdb.py. They don't have a compatible signature. That should get rid of the hard-coded `136` in the def and use of `getPrefixOpIntrinsic` in `CGHLSLBuiltins.cpp`. Note that the test generates a bad DXIL op too. https://github.com/llvm/llvm-project/pull/178059 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
