667fff7c6ae29ac3843316984845e4d4e6f2b8eb - llvm-project/llvm

commit: 667fff7c6ae29ac3843316984845e4d4e6f2b8eb
[log]
author: Gergo Stomfai <55883018+stomfaig@users.noreply.github.com>
Fri Jan 09 10:39:30 2026 +0000
committer: Copybara-Service <copybara-worker@google.com>
Fri Jan 09 02:43:02 2026 -0800
tree: 3d5223f9138c4cb74bd1bbced672bde2195d1a57
parent: 0358a223d28ae604132dc944c9a66a855a223c8a [diff]

[X86] Add rewrite pattern for SSE41/AVX1 roundss/sd + blendps/pd (#172056)

Due to a previous PR (https://github.com/llvm/llvm-project/pull/171227),
operations like `_mm_ceil_sd` compile to suboptimal assembly:
```asm
roundsd xmm1, xmm1, 10
blendpd xmm0, xmm1, 1
```
This PR introduces a rewrite pattern to mitigate this, and fuse the corresponding operations.

GitOrigin-RevId: a484de1e06efc4c80accfa4d8be97575133cb26b

lib/Target/X86/X86InstrAVX512.td[diff]
lib/Target/X86/X86InstrSSE.td[diff]
test/CodeGen/X86/avx512fp16-arith.ll[diff]
test/CodeGen/X86/vec_floor.ll[diff]

4 files changed

tree: 3d5223f9138c4cb74bd1bbced672bde2195d1a57