diff options
| author | Sanjay Patel <spatel@rotateright.com> | 2019-11-24 10:06:26 -0500 |
|---|---|---|
| committer | Sanjay Patel <spatel@rotateright.com> | 2019-11-24 10:06:26 -0500 |
| commit | f575f12c646544a3200852cf72212045fdf2e0b4 (patch) | |
| tree | 3fa3ccc309cfe279115e940d33e844994d62116e /llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp | |
| parent | f04a3e981d3175d7f3d0f5008b842823034f47ed (diff) | |
| download | bcm5719-llvm-f575f12c646544a3200852cf72212045fdf2e0b4.tar.gz bcm5719-llvm-f575f12c646544a3200852cf72212045fdf2e0b4.zip | |
[InstCombine] remove identity shuffle simplification for mask with undefs
And simultaneously enhance SimplifyDemandedVectorElts() to rcognize that
pattern. That preserves some of the old optimizations in IR.
Given a shuffle that includes undef elements in an otherwise identity mask like:
define <4 x float> @shuffle(<4 x float> %arg) {
%shuf = shufflevector <4 x float> %arg, <4 x float> undef, <4 x i32> <i32 undef, i32 1, i32 2, i32 3>
ret <4 x float> %shuf
}
We were simplifying that to the input operand.
But as discussed in PR43958:
https://bugs.llvm.org/show_bug.cgi?id=43958
...that means that per-vector-element poison that would be stopped by the shuffle can now
leak to the result.
Also note that we still have (and there are tests for) the same transform with no undef
elements in the mask (a fully-defined identity mask). I don't think there's any
controversy about that case - it's a valid transform under any interpretation of
shufflevector/undef/poison.
Looking at a few of the diffs into codegen, I don't see any difference in final asm. So
depending on your perspective, that's good (no real loss of optimization power) or bad
(poison exists in the DAG, so we only partially fixed the bug).
Differential Revision: https://reviews.llvm.org/D70246
Diffstat (limited to 'llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp')
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp | 24 |
1 files changed, 24 insertions, 0 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp index c772c351cb8..ef8604fc3f2 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp @@ -1274,6 +1274,30 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts, APInt RHSUndefElts(OpWidth, 0); simplifyAndSetOp(I, 1, RightDemanded, RHSUndefElts); + // If this shuffle does not change the vector length and the elements + // demanded by this shuffle are an identity mask, then this shuffle is + // unnecessary. + // + // We are assuming canonical form for the mask, so the source vector is + // operand 0 and operand 1 is not used. + // + // Note that if an element is demanded and this shuffle mask is undefined + // for that element, then the shuffle is not considered an identity + // operation. The shuffle prevents poison from the operand vector from + // leaking to the result by replacing poison with an undefined value. + if (VWidth == OpWidth) { + bool IsIdentityShuffle = true; + for (unsigned i = 0; i < VWidth; i++) { + unsigned MaskVal = Shuffle->getMaskValue(i); + if (DemandedElts[i] && i != MaskVal) { + IsIdentityShuffle = false; + break; + } + } + if (IsIdentityShuffle) + return Shuffle->getOperand(0); + } + bool NewUndefElts = false; unsigned LHSIdx = -1u, LHSValIdx = -1u; unsigned RHSIdx = -1u, RHSValIdx = -1u; |

