llvm-project

History

[SLP]Check if masked gather can be emitted as a serie of loads/insert subvector.

Masked gather is very expensive operation and sometimes better to
represent it as a serie of consecutive/strided loads + insertsubvectors
sequences. Patch adds some basic estimation and if loads+insertsubvector
is cheaper, decides to represent it in this way rather than masked
gather.

Reviewers: RKSimon

Reviewed By: RKSimon

Pull Request: https://github.com/llvm/llvm-project/pull/83481

2024-03-01 13:49:29 -05:00

AArch64

[SLP]Improve findReusedOrderedScalars and graph rotation.

2024-02-22 14:32:15 -05:00

AMDGPU

[InstCombine] Replace non-demanded undef vector with poison

2023-12-18 16:12:37 +01:00

ARM

[SLP][NFC]Add insertsubvector test with small source vector, NFC.

2023-10-05 06:03:58 -07:00

NVPTX

[Tests] Add disjoint flag to some tests (NFC)