[llvm] r340447 - [Hexagon] Enable interleaving in loop vectorizer

Krzysztof Parzyszek via llvm-commits llvm-commits at lists.llvm.org
Wed Aug 22 13:15:04 PDT 2018


Author: kparzysz
Date: Wed Aug 22 13:15:04 2018
New Revision: 340447

URL: http://llvm.org/viewvc/llvm-project?rev=340447&view=rev
Log:
[Hexagon] Enable interleaving in loop vectorizer

Added:
    llvm/trunk/test/CodeGen/Hexagon/autohvx/interleave.ll
Modified:
    llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp
    llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.h

Modified: llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp?rev=340447&r1=340446&r2=340447&view=diff
==============================================================================
--- llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp (original)
+++ llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp Wed Aug 22 13:15:04 2018
@@ -207,8 +207,10 @@ unsigned HexagonTTIImpl::getGatherScatte
 unsigned HexagonTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode,
       Type *VecTy, unsigned Factor, ArrayRef<unsigned> Indices,
       unsigned Alignment, unsigned AddressSpace) {
-  return BaseT::getInterleavedMemoryOpCost(Opcode, VecTy, Factor, Indices,
-                                           Alignment, AddressSpace);
+  if (Indices.size() != Factor)
+    return BaseT::getInterleavedMemoryOpCost(Opcode, VecTy, Factor, Indices,
+                                             Alignment, AddressSpace);
+  return getMemoryOpCost(Opcode, VecTy, Alignment, AddressSpace, nullptr);
 }
 
 unsigned HexagonTTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,

Modified: llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.h?rev=340447&r1=340446&r2=340447&view=diff
==============================================================================
--- llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.h (original)
+++ llvm/trunk/lib/Target/Hexagon/HexagonTargetTransformInfo.h Wed Aug 22 13:15:04 2018
@@ -98,6 +98,9 @@ public:
   bool prefersVectorizedAddressing() {
     return false;
   }
+  bool enableInterleavedAccessVectorization() {
+    return true;
+  }
 
   unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract);
   unsigned getOperandsScalarizationOverhead(ArrayRef<const Value*> Args,

Added: llvm/trunk/test/CodeGen/Hexagon/autohvx/interleave.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Hexagon/autohvx/interleave.ll?rev=340447&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/Hexagon/autohvx/interleave.ll (added)
+++ llvm/trunk/test/CodeGen/Hexagon/autohvx/interleave.ll Wed Aug 22 13:15:04 2018
@@ -0,0 +1,52 @@
+; RUN: opt -march=hexagon -hexagon-autohvx -loop-vectorize -S < %s | FileCheck %s
+; Check that the loop has been interleaved.
+; CHECK: store <64 x i32> %interleaved.vec
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+target triple = "hexagon"
+
+define void @f0(i32* noalias nocapture %a0, i32* noalias nocapture readonly %a1, i32 %a2) #0 {
+b0:
+  %v0 = icmp eq i32 %a2, 0
+  br i1 %v0, label %b3, label %b1
+
+b1:                                               ; preds = %b0
+  br label %b4
+
+b2:                                               ; preds = %b4
+  br label %b3
+
+b3:                                               ; preds = %b2, %b0
+  ret void
+
+b4:                                               ; preds = %b4, %b1
+  %v1 = phi i32 [ %v13, %b4 ], [ 0, %b1 ]
+  %v2 = getelementptr inbounds i32, i32* %a1, i32 %v1
+  %v3 = load i32, i32* %v2, align 4, !tbaa !1
+  %v4 = getelementptr inbounds i32, i32* %a0, i32 %v1
+  %v5 = load i32, i32* %v4, align 4, !tbaa !1
+  %v6 = add nsw i32 %v5, %v3
+  store i32 %v6, i32* %v4, align 4, !tbaa !1
+  %v7 = or i32 %v1, 1
+  %v8 = getelementptr inbounds i32, i32* %a1, i32 %v7
+  %v9 = load i32, i32* %v8, align 4, !tbaa !1
+  %v10 = getelementptr inbounds i32, i32* %a0, i32 %v7
+  %v11 = load i32, i32* %v10, align 4, !tbaa !1
+  %v12 = add nsw i32 %v11, %v9
+  store i32 %v12, i32* %v10, align 4, !tbaa !1
+  %v13 = add nuw nsw i32 %v1, 2
+  %v14 = icmp eq i32 %v13, %a2
+  br i1 %v14, label %b2, label %b4, !llvm.loop !5
+}
+
+attributes #0 = { norecurse nounwind "target-cpu"="hexagonv60" "target-features"="+hvx-length128b,+hvxv60" }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!2, !2, i64 0}
+!2 = !{!"int", !3, i64 0}
+!3 = !{!"omnipotent char", !4, i64 0}
+!4 = !{!"Simple C/C++ TBAA"}
+!5 = distinct !{!5, !6}
+!6 = !{!"llvm.loop.unroll.disable"}




More information about the llvm-commits mailing list