Model MXCSR for all AVX512 instructions
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
My biggest concern here is that the SAE and implicit rounding instructions are all marked as mayRaiseExceptions which is incorrect.
llvm/lib/Target/X86/X86InstrInfo.cpp | ||
---|---|---|
1764–1765 | Use getNumExplicitOperands() - 1 instead. |
Comment Actions
Removed MXCSR and mayRaiseExceptions from SAE opcodes.
llvm/lib/Target/X86/X86InstrInfo.cpp | ||
---|---|---|
1764–1765 | Thanks! |
Comment Actions
I think SAE instructions and embedded rounding instructions still read the DAZ and FTZ bits from MXCSR
This comment was removed by pengfei.
Comment Actions
You are correct, we do need to model DAZ and FTZ. I think I commented "not modeled" just for simplification. I will commit a patch to model them.
Don't we need to suppress MXCSR and exceptions here