[llvm] 3b7f516 - [SLPVectorizer][SVE] Skip scalable-vector instructions before vectorizeSimpleInstructions.

[Huihui Zhang via llvm-commits]

Author: Huihui Zhang
Date: 2020-09-15T13:10:15-07:00
New Revision: 3b7f5166bd11fc6cbf96597d26753e8c3fc0e6ab

URL: https://github.com/llvm/llvm-project/commit/3b7f5166bd11fc6cbf96597d26753e8c3fc0e6ab
DIFF: https://github.com/llvm/llvm-project/commit/3b7f5166bd11fc6cbf96597d26753e8c3fc0e6ab.diff

LOG: [SLPVectorizer][SVE] Skip scalable-vector instructions before vectorizeSimpleInstructions.

For scalable type, the aggregated size is unknown at compile-time.
Skip instructions with scalable type to ensure the list of instructions
for vectorizeSimpleInstructions does not contains any scalable-vector instructions.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D87550

Added: 
    llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index e73113dab6d4..3347419077e3 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -7508,6 +7508,11 @@ bool SLPVectorizerPass::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) {
   SmallVector<Instruction *, 8> PostProcessInstructions;
   SmallDenseSet<Instruction *, 4> KeyNodes;
   for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
+    // Skip instructions with scalable type. The num of elements is unknown at
+    // compile-time for scalable type.
+    if (isa<ScalableVectorType>(it->getType()))
+      continue;
+
     // Skip instructions marked for the deletion.
     if (R.isDeleted(&*it))
       continue;

diff  --git a/llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll
new file mode 100644
index 000000000000..b5cab5d3186a
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll
@@ -0,0 +1,44 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -slp-vectorizer -S 2>%t | FileCheck %s
+; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t
+
+; WARN-NOT: warning
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+define <2 x float> @insertelement-fixed-vector() {
+; CHECK-LABEL: @insertelement-fixed-vector(
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast <2 x float> @llvm.fabs.v2f32(<2 x float> undef)
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x float> [[TMP1]], i32 0
+; CHECK-NEXT:    [[I0:%.*]] = insertelement <2 x float> undef, float [[TMP2]], i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x float> [[TMP1]], i32 1
+; CHECK-NEXT:    [[I1:%.*]] = insertelement <2 x float> [[I0]], float [[TMP3]], i32 1
+; CHECK-NEXT:    ret <2 x float> [[I1]]
+;
+  %f0 = tail call fast float @llvm.fabs.f32(float undef)
+  %f1 = tail call fast float @llvm.fabs.f32(float undef)
+  %i0 = insertelement <2 x float> undef, float %f0, i32 0
+  %i1 = insertelement <2 x float> %i0, float %f1, i32 1
+  ret <2 x float> %i1
+}
+
+; TODO: llvm.fabs could be optimized in vector form. It's legal to extract
+; elements from fixed-length vector and insert into scalable vector.
+define <vscale x 2 x float> @insertelement-scalable-vector() {
+; CHECK-LABEL: @insertelement-scalable-vector(
+; CHECK-NEXT:    [[F0:%.*]] = tail call fast float @llvm.fabs.f32(float undef)
+; CHECK-NEXT:    [[F1:%.*]] = tail call fast float @llvm.fabs.f32(float undef)
+; CHECK-NEXT:    [[I0:%.*]] = insertelement <vscale x 2 x float> undef, float [[F0]], i32 0
+; CHECK-NEXT:    [[I1:%.*]] = insertelement <vscale x 2 x float> [[I0]], float [[F1]], i32 1
+; CHECK-NEXT:    ret <vscale x 2 x float> [[I1]]
+;
+  %f0 = tail call fast float @llvm.fabs.f32(float undef)
+  %f1 = tail call fast float @llvm.fabs.f32(float undef)
+  %i0 = insertelement <vscale x 2 x float> undef, float %f0, i32 0
+  %i1 = insertelement <vscale x 2 x float> %i0, float %f1, i32 1
+  ret <vscale x 2 x float> %i1
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare float @llvm.fabs.f32(float)