diff options
| author | Sanjay Patel <spatel@rotateright.com> | 2018-09-30 13:50:42 +0000 |
|---|---|---|
| committer | Sanjay Patel <spatel@rotateright.com> | 2018-09-30 13:50:42 +0000 |
| commit | 26c119a9c2f1d6866fe5996ef5a039b4fc3749ca (patch) | |
| tree | 8caae0b7b55785ce1fd1fd5922fe0c360fb68734 /llvm/test | |
| parent | 818cfc40ff464d426be37a552c3db8e895c94321 (diff) | |
| download | bcm5719-llvm-26c119a9c2f1d6866fe5996ef5a039b4fc3749ca.tar.gz bcm5719-llvm-26c119a9c2f1d6866fe5996ef5a039b4fc3749ca.zip | |
[InstCombine] allow lengthening of insertelement to eliminate shuffles
As noted in post-commit comments for D52548, the limitation on
increasing vector length can be applied by opcode.
As a first step, this patch only allows insertelement to be
widened because that has no logical downsides for IR and has
little risk of pessimizing codegen.
This may cause PR39132 to go into hiding during a full compile,
but that bug is not fixed.
llvm-svn: 343406
Diffstat (limited to 'llvm/test')
| -rw-r--r-- | llvm/test/Transforms/InstCombine/vec_shuffle.ll | 9 |
1 files changed, 4 insertions, 5 deletions
diff --git a/llvm/test/Transforms/InstCombine/vec_shuffle.ll b/llvm/test/Transforms/InstCombine/vec_shuffle.ll index b3f2ca54be9..7a5ff3628d2 100644 --- a/llvm/test/Transforms/InstCombine/vec_shuffle.ll +++ b/llvm/test/Transforms/InstCombine/vec_shuffle.ll @@ -218,14 +218,13 @@ define <3 x i32> @div_wider(i32 %y, i32 %z) { ret <3 x i32> %ext } -; TODO: Increasing length of insertelements (no math ops) is a good canonicalization. +; Increasing length of insertelements (no math ops) is a good canonicalization. define <3 x i8> @fold_inselts_with_widening_shuffle(i8 %x, i8 %y) { ; CHECK-LABEL: @fold_inselts_with_widening_shuffle( -; CHECK-NEXT: [[INS0:%.*]] = insertelement <2 x i8> undef, i8 [[X:%.*]], i32 0 -; CHECK-NEXT: [[INS1:%.*]] = insertelement <2 x i8> [[INS0]], i8 [[Y:%.*]], i32 1 -; CHECK-NEXT: [[WIDEN:%.*]] = shufflevector <2 x i8> [[INS1]], <2 x i8> undef, <3 x i32> <i32 0, i32 1, i32 undef> -; CHECK-NEXT: ret <3 x i8> [[WIDEN]] +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x i8> undef, i8 [[X:%.*]], i32 0 +; CHECK-NEXT: [[TMP2:%.*]] = insertelement <3 x i8> [[TMP1]], i8 [[Y:%.*]], i32 1 +; CHECK-NEXT: ret <3 x i8> [[TMP2]] ; %ins0 = insertelement <2 x i8> undef, i8 %x, i32 0 %ins1 = insertelement <2 x i8> %ins0, i8 %y, i32 1 |
