[PATCH] D87550: [SLPVectorizer][SVE] Avoid blind cast to FixedVectorType in getAggregateSize.

[Huihui Zhang via Phabricator via llvm-commits]

huihuiz created this revision.
huihuiz added reviewers: efriedma, sdesmalen, spatel, paulwalker-arm, david-arm.
huihuiz added a project: LLVM.
Herald added subscribers: psnobl, hiraditya, tschuett.
huihuiz requested review of this revision.

For SLPVectorizerPass::vectorizeInsertElementInst, the 'insertelement' instruction
could be a scalable type. The aggregated size for scalable type is unknown at
compile-time. Therefore, return 'None' for scalable type.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D87550

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll


Index: llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll
===================================================================
--- /dev/null
+++ 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)
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -7076,8 +7076,14 @@
 } // end anonymous namespace
 
 static Optional<unsigned> getAggregateSize(Instruction *InsertInst) {
-  if (auto *IE = dyn_cast<InsertElementInst>(InsertInst))
-    return cast<FixedVectorType>(IE->getType())->getNumElements();
+  if (auto *IE = dyn_cast<InsertElementInst>(InsertInst)) {
+    auto *FVTy = dyn_cast<FixedVectorType>(IE->getType());
+    // For scalable type, num of elements is unknown at compile-time.
+    if (!FVTy)
+      return None;
+    else
+      return FVTy->getNumElements();
+  }
 
   unsigned AggregateSize = 1;
   auto *IV = cast<InsertValueInst>(InsertInst);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D87550.291351.patch
Type: text/x-patch
Size: 3270 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200911/b5698f62/attachment.bin>