[llvm] [AArch64] C1-Premium SME definitions (PR #212482)
Asher Dobrescu via llvm-commits
llvm-commits at lists.llvm.org
Fri Aug 14 04:45:22 PDT 2026
https://github.com/Asher8118 commented:
Thank you for your changes. I have looked over the SVE floating-point and BF16instructions. I think a lot have been missed from the C1 SME2 SWOG.
Other than what I pointed out in the diff:
- Floating point reciprocal F16, F32, F64 step. All of those forms are showing incorrect information when in streaming mode. They should all show L4 T2 pipeline VXMUL based on the C1 SME2 SWOG.
- Floating point reciprocal step `FRECPX` is howing incorrect information when in streaming mode. It should show L4 T2 pipeline VXMUL based on the C1 SME2 SWOG.
- Convert, F32 to BF16, both `BFCVT` and `BFCVTNT` are showing as L3 T2 and using pipeline V when it should be L4 T2 and using pipeline VXALU when in streaming mode.
Additionally, a lot of the instructions that appear in the C1 SME2 SWOG do not show up in the tests. Please add them in the relevant places, ie: `C1Premium-neon-instructions.s`, `C1Premium-sve-instructions.s`, etc.
I am surprised there is not a common input file for instructions that are in streaming mode. The C1 SME2 SWOG is the same for all C1 cores, so streaming mode instructions should have a lot of common ground.
https://github.com/llvm/llvm-project/pull/212482
More information about the llvm-commits
mailing list