llvm-project
[X86] Allow EVEX compression for VPMOV*2M + KMOV pattern
#175219
Merged
Merged
[X86] Allow EVEX compression for VPMOV*2M + KMOV pattern #175219
RKSimon
merged
16 commits into
llvm:main
from
realqhc:extend-comparebitcastvxi1
[X86] combineBitcastvxi1 - prefer movmsk for i32 truncate
e8b53eb1
Revert "[X86] combineBitcastvxi1 - prefer movmsk for i32 truncate"
851e376c
[X86] Compress VPMOV*2M+KMOV chain to VMOVMSK on AVX-512
104a0481
clang-format
e4d4259b
fixup miscompilation and organize code into separate function
1865989c
add support for vmovmskpd
d40f4a7b
restore variables accidentally changed
b78bc134
fixup! move evex after VPMOV*2M per suggestion
8062e233
fixup! do not allow KMOVBrk as truncation is unsafe
415368ec
Revert "fixup! do not allow KMOVBrk as truncation is unsafe"
e57f5a79
fixup! add support for vpmovmskb
29c083f5
Merge branch 'main' into extend-comparebitcastvxi1
d4a59ea9
fixup! Add check for MaskReg usage in different basic blocks before c…
601cce09
add pr174871.ll
f0bec30f
simplify check code according to hasUseAfterLoop from ModuleSchedule.cpp
ea173855
RKSimon
approved these changes
on 2026-01-12
Merge branch 'main' into extend-comparebitcastvxi1
adb62e61
Login to write a write a comment.
Login via GitHub