diff options
| author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2025-03-13 12:58:22 +0000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-03-13 12:58:22 +0000 |
| commit | dff22a0c1173f9d1ec0830b1c039749b6c2241d6 (patch) | |
| tree | 089f6aaad19aae4cf4b9e7d39aae3cecff69eaf2 /flang/lib/Frontend/CompilerInvocation.cpp | |
| parent | a413ef83a16b1069ca953613f1ff495f702970b4 (diff) | |
| download | llvm-dff22a0c1173f9d1ec0830b1c039749b6c2241d6.zip llvm-dff22a0c1173f9d1ec0830b1c039749b6c2241d6.tar.gz llvm-dff22a0c1173f9d1ec0830b1c039749b6c2241d6.tar.bz2 | |
[X86] combineConcatVectorOps - convert X86ISD::BLENDI concatenation to use combineConcatVectorOps recursion (#131121)
Only concatenate X86ISD::BLENDI nodes if at least one operand is beneficial to concatenate
Add AVX1/AVX2 handling to 256-bit BLENDI nodes (accounting for AVX2 v16i16 repeated mask requirements).
Extend existing AVX512BW (which still always concats until I get get rid of the remaining regressions) to handle AVX512F for 32/64-bit scalar types.
Diffstat (limited to 'flang/lib/Frontend/CompilerInvocation.cpp')
0 files changed, 0 insertions, 0 deletions