BSD Commit Log Search

DAG: Avoid breaking legal vector_shuffle with multiple uses (#123712)

Previously this combine would undo AMDGPU's new custom legalization of
wide vector shuffles into 2 element pieces. The comment also
states that this combine is only done before legalization,
but the case with a build_vector source was unconditional.

We probably don't want to do this if the multiple uses are full
scalarization of the vector, but this seems to work well enough.
Scalarizing extracts should have folded out pre-legalize.

Merge branch 'main' into users/kparzysz/spr/m04-sema-checks

Merge branch 'main' into users/kparzysz/spr/m03-meta-simple

Remove DAGCombiner copy

DAG: Move scalarizeExtractedVectorLoad to TargetLowering

SimplifyDemandedVectorElts should be able to use this on loads

Rebase

Created using spr 1.3.5

DAG: Avoid breaking legal vector_shuffle with multiple uses

Previously this combine would undo AMDGPU's new custom legalization of
wide vector shuffles into 2 element pieces. The comment also
states that this combine is only done before legalization,
but the case with a build_vector source was unconditional.

We probably don't want to do this if the multiple uses are full
scalarization of the vector, but this seems to work well enough.
Scalarizing extracts should have folded out pre-legalize.

DAG: Avoid breaking legal vector_shuffle with multiple uses

Previously this combine would undo AMDGPU's new custom legalization of
wide vector shuffles into 2 element pieces. The comment also
states that this combine is only done before legalization,
but the case with a build_vector source was unconditional.

We probably don't want to do this if the multiple uses are full
scalarization of the vector, but this seems to work well enough.
Scalarizing extracts should have folded out pre-legalize.

Merge branch 'main' into linearize_bitcast

Rebase

Created using spr 1.3.5

Merge branch 'users/meinersbur/flang_runtime_shared' into users/meinersbur/flang_runtime_remove-FLANG_INCLUDE_RUNTIME

Merge branch 'users/meinersbur/flang_runtime' into users/meinersbur/flang_runtime_shared

Merge branch 'users/meinersbur/flang_runtime_move-files' into users/meinersbur/flang_runtime

Merge branch 'users/meinersbur/flang_runtime_flang_rt' into users/meinersbur/flang_runtime_move-files

Merge branch 'users/meinersbur/flang_runtime_FLANG_INCLUDE_RUNTIME' into users/meinersbur/flang_runtime_flang_rt

Merge branch 'users/meinersbur/flang_runtime_FortranDecimal' into users/meinersbur/flang_runtime_FLANG_INCLUDE_RUNTIME

Merge branch 'users/meinersbur/flang_runtime_Testing' into users/meinersbur/flang_runtime_FortranDecimal

Merge branch 'users/meinersbur/flang_runtime_FortranSupport' into users/meinersbur/flang_runtime_Testing

Merge branch 'main' into users/meinersbur/flang_runtime_FortranSupport

Merge branch 'main' into users/boomanaiden154/windows-premerge

Rebase

Created using spr 1.3.5

Merge branch 'main' into users/zhaoqi5/jitlink-support-add-sub

use llvm::StringMap

Created using spr 1.3.5-bogner

merge with main branch

rebase

Created using spr 1.3.4

.

Created using spr 1.3.5-bogner

DAG: Avoid breaking legal vector_shuffle with multiple uses

Previously this combine would undo AMDGPU's new custom legalization of
wide vector shuffles into 2 element pieces. The comment also
states that this combine is only done before legalization,
but the case with a build_vector source was unconditional.

We probably don't want to do this if the multiple uses are full
scalarization of the vector, but this seems to work well enough.
Scalarizing extracts should have folded out pre-legalize.

[SDAG] Use BatchAAResults for querying alias analysis (AA) results (#123934)

Once we get to SelectionDAG the IR should not be changing anymore, so we
can use BatchAAResults rather than AAResults to cache AA queries.

This should be a NFC change for targets that enable AA during codegen
(such as AArch64), but also give a nice compile-time improvement in some
cases. See:
https://github.com/llvm/llvm-project/pull/123787#issuecomment-2606797041

Note: This follows Nikita's suggestion on #123787.

Address comments

Created using spr 1.3.5

Rebase, rework

Created using spr 1.3.5