[llvm] f0eeb3c - [GISel][CombinerHelper] Combine shuffle_vector scalar to build_vector
Quentin Colombet via llvm-commits
llvm-commits at lists.llvm.org
Wed Oct 30 18:40:22 PDT 2019
Author: Quentin Colombet
Date: 2019-10-30T18:20:37-07:00
New Revision: f0eeb3c7a7136dd043b792ff2d321ad492db203a
URL: https://github.com/llvm/llvm-project/commit/f0eeb3c7a7136dd043b792ff2d321ad492db203a
DIFF: https://github.com/llvm/llvm-project/commit/f0eeb3c7a7136dd043b792ff2d321ad492db203a.diff
LOG: [GISel][CombinerHelper] Combine shuffle_vector scalar to build_vector
Teach the combiner helper how to replace shuffle_vector of scalars
into build_vector.
I am not particularly happy about having to add this combine, but we
currently get those from <1 x iN> from the IR.
Bonus: This fixes an assert in the shuffle_vector combines since before
this patch, we were expecting vector types.
Added:
Modified:
llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-shuffle-vector.mir
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index 854769d283f7..bb7f751de903 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -190,7 +190,7 @@ bool CombinerHelper::matchCombineShuffleVector(MachineInstr &MI,
Register Src1 = MI.getOperand(1).getReg();
LLT SrcType = MRI.getType(Src1);
unsigned DstNumElts = DstType.getNumElements();
- unsigned SrcNumElts = SrcType.getNumElements();
+ unsigned SrcNumElts = SrcType.isVector() ? SrcType.getNumElements() : 1;
// If the resulting vector is smaller than the size of the source
// vectors being concatenated, we won't be able to replace the
@@ -254,7 +254,7 @@ void CombinerHelper::applyCombineShuffleVector(MachineInstr &MI,
Builder.setInsertPt(*MI.getParent(), MI);
Register NewDstReg = MRI.cloneVirtualRegister(DstReg);
- Builder.buildConcatVectors(NewDstReg, Ops);
+ Builder.buildMerge(NewDstReg, Ops);
MI.eraseFromParent();
replaceRegWith(MRI, DstReg, NewDstReg);
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-shuffle-vector.mir b/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-shuffle-vector.mir
index e3a27aa2b16b..8694bff4746c 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-shuffle-vector.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-shuffle-vector.mir
@@ -351,3 +351,66 @@ body: |
%6:_(<4 x p0>) = G_SHUFFLE_VECTOR %4(<2 x p0>), %5(<2 x p0>), shufflemask(0,1,2,3)
RET_ReallyLR implicit %6
...
+
+# Check that shuffle_vector on scalars gets combined into build_vector.
+---
+name: shuffle_vector_on_scalars_to_build_vector_ptr
+tracksRegLiveness: true
+body: |
+ bb.1:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: shuffle_vector_on_scalars_to_build_vector_ptr
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
+ ; CHECK: [[COPY1:%[0-9]+]]:_(p0) = COPY $x1
+ ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x p0>) = G_BUILD_VECTOR [[COPY]](p0), [[COPY1]](p0), [[COPY]](p0), [[COPY1]](p0)
+ ; CHECK: RET_ReallyLR implicit [[BUILD_VECTOR]](<4 x p0>)
+ %0:_(p0) = COPY $x0
+ %1:_(p0) = COPY $x1
+ %6:_(<4 x p0>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0,1,0,1)
+ RET_ReallyLR implicit %6
+...
+
+# Check that shuffle_vector on scalars gets combined into build_vector,
+# even if we swap the order of the operands.
+---
+name: shuffle_vector_on_scalars_to_build_vector_swap_ptr
+tracksRegLiveness: true
+body: |
+ bb.1:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: shuffle_vector_on_scalars_to_build_vector_swap_ptr
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
+ ; CHECK: [[COPY1:%[0-9]+]]:_(p0) = COPY $x1
+ ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x p0>) = G_BUILD_VECTOR [[COPY1]](p0), [[COPY]](p0)
+ ; CHECK: RET_ReallyLR implicit [[BUILD_VECTOR]](<2 x p0>)
+ %0:_(p0) = COPY $x0
+ %1:_(p0) = COPY $x1
+ %6:_(<2 x p0>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(1,0)
+ RET_ReallyLR implicit %6
+...
+
+# Check that we properly use undef values when shuffle_vector
+# on scalars gets lowered to build_vector.
+---
+name: shuffle_vector_on_scalars_to_build_vector_with_undef
+tracksRegLiveness: true
+body: |
+ bb.1:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: shuffle_vector_on_scalars_to_build_vector_with_undef
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+ ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
+ ; CHECK: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
+ ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[COPY]](s64), [[DEF]](s64), [[DEF]](s64), [[COPY1]](s64)
+ ; CHECK: RET_ReallyLR implicit [[BUILD_VECTOR]](<4 x s64>)
+ %0:_(s64) = COPY $x0
+ %1:_(s64) = COPY $x1
+ %6:_(<4 x s64>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0,-1,-1,1)
+ RET_ReallyLR implicit %6
+...
More information about the llvm-commits
mailing list