[ARM/AArch64] Fix cost model for interleaved accesses

Summary:
Fix the cost of interleaved accesses for ARM/AArch64.
We were calling getTypeAllocSize and using it to check
the number of bits, when we should have called
getTypeAllocSizeInBits instead.

This would pottentially cause the vectorizer to
generate loads/stores and shuffles which cannot
be matched with an interleaved access instruction.

No performance changes are expected for now since
matching/generating interleaved accesses is still
disabled by default.

Reviewers: rengolin

Subscribers: aemerson, llvm-commits, rengolin

Differential Revision: http://reviews.llvm.org/D11524

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243270 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index e085cca35f1c1b8fab7ba2ffe0afd64fa9596ef8..1ffc6d960e94ab28a92be5220969d0d77d52ad99 100644 (file)
--- a/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -416,7 +416,7 @@ unsigned AArch64TTIImpl::getInterleavedMemoryOpCost(
   if (Factor <= TLI->getMaxSupportedInterleaveFactor()) {
     unsigned NumElts = VecTy->getVectorNumElements();
     Type *SubVecTy = VectorType::get(VecTy->getScalarType(), NumElts / Factor);
-    unsigned SubVecSize = DL.getTypeAllocSize(SubVecTy);
+    unsigned SubVecSize = DL.getTypeAllocSizeInBits(SubVecTy);
 
     // ldN/stN only support legal vector types of size 64 or 128 in bits.
     if (NumElts % Factor == 0 && (SubVecSize == 64 || SubVecSize == 128))

diff --git a/lib/Target/ARM/ARMTargetTransformInfo.cpp b/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 2f194cf7ae06ab0cd787579dcdbd5d1493a90062..e3c64ee8a59ccabb310ef8cea959a20d3df6edd3 100644 (file)
--- a/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -493,7 +493,7 @@ unsigned ARMTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
   if (Factor <= TLI->getMaxSupportedInterleaveFactor() && !EltIs64Bits) {
     unsigned NumElts = VecTy->getVectorNumElements();
     Type *SubVecTy = VectorType::get(VecTy->getScalarType(), NumElts / Factor);
-    unsigned SubVecSize = DL.getTypeAllocSize(SubVecTy);
+    unsigned SubVecSize = DL.getTypeAllocSizeInBits(SubVecTy);
 
     // vldN/vstN only support legal vector types of size 64 or 128 in bits.
     if (NumElts % Factor == 0 && (SubVecSize == 64 || SubVecSize == 128))

diff --git a/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll b/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
new file mode 100644 (file)
index 0000000..218dc64
--- /dev/null
+++ b/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
@@ -0,0 +1,39 @@
+; RUN: opt -S -debug-only=loop-vectorize -loop-vectorize -instcombine -enable-interleaved-mem-accesses=true  < %s |& FileCheck %s
+
+
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64--linux-gnueabi"
+
+@AB = common global [1024 x i8] zeroinitializer, align 4
+@CD = common global [1024 x i8] zeroinitializer, align 4
+
+define void @test_byte_interleaved_cost(i8 %C, i8 %D) {
+entry:
+  br label %for.body
+
+; 8xi8 and 16xi8 are valid i8 vector types, so the cost of the interleaved
+; access group is 2.
+
+; CHECK: LV: Found an estimated cost of 2 for VF 8 For instruction:   %tmp = load i8, i8* %arrayidx0, align 4
+; CHECK: LV: Found an estimated cost of 2 for VF 16 For instruction:   %tmp = load i8, i8* %arrayidx0, align 4
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx0 = getelementptr inbounds [1024 x i8], [1024 x i8]* @AB, i64 0, i64 %indvars.iv
+  %tmp = load i8, i8* %arrayidx0, align 4
+  %tmp1 = or i64 %indvars.iv, 1
+  %arrayidx1 = getelementptr inbounds [1024 x i8], [1024 x i8]* @AB, i64 0, i64 %tmp1
+  %tmp2 = load i8, i8* %arrayidx1, align 4
+  %add = add nsw i8 %tmp, %C
+  %mul = mul nsw i8 %tmp2, %D
+  %arrayidx2 = getelementptr inbounds [1024 x i8], [1024 x i8]* @CD, i64 0, i64 %indvars.iv
+  store i8 %add, i8* %arrayidx2, align 4
+  %arrayidx3 = getelementptr inbounds [1024 x i8], [1024 x i8]* @CD, i64 0, i64 %tmp1
+  store i8 %mul, i8* %arrayidx3, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
+  %cmp = icmp slt i64 %indvars.iv.next, 1024
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  ret void
+}

diff --git a/test/Transforms/LoopVectorize/ARM/interleaved_cost.ll b/test/Transforms/LoopVectorize/ARM/interleaved_cost.ll
new file mode 100644 (file)
index 0000000..9453894
--- /dev/null
+++ b/test/Transforms/LoopVectorize/ARM/interleaved_cost.ll
@@ -0,0 +1,39 @@
+; RUN: opt -S -debug-only=loop-vectorize -loop-vectorize -instcombine -enable-interleaved-mem-accesses=true  < %s |& FileCheck %s
+
+
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "armv8--linux-gnueabihf"
+
+@AB = common global [1024 x i8] zeroinitializer, align 4
+@CD = common global [1024 x i8] zeroinitializer, align 4
+
+define void @test_byte_interleaved_cost(i8 %C, i8 %D) {
+entry:
+  br label %for.body
+
+; 8xi8 and 16xi8 are valid i8 vector types, so the cost of the interleaved
+; access group is 2.
+
+; CHECK: LV: Found an estimated cost of 2 for VF 8 For instruction:   %tmp = load i8, i8* %arrayidx0, align 4
+; CHECK: LV: Found an estimated cost of 2 for VF 16 For instruction:   %tmp = load i8, i8* %arrayidx0, align 4
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx0 = getelementptr inbounds [1024 x i8], [1024 x i8]* @AB, i64 0, i64 %indvars.iv
+  %tmp = load i8, i8* %arrayidx0, align 4
+  %tmp1 = or i64 %indvars.iv, 1
+  %arrayidx1 = getelementptr inbounds [1024 x i8], [1024 x i8]* @AB, i64 0, i64 %tmp1
+  %tmp2 = load i8, i8* %arrayidx1, align 4
+  %add = add nsw i8 %tmp, %C
+  %mul = mul nsw i8 %tmp2, %D
+  %arrayidx2 = getelementptr inbounds [1024 x i8], [1024 x i8]* @CD, i64 0, i64 %indvars.iv
+  store i8 %add, i8* %arrayidx2, align 4
+  %arrayidx3 = getelementptr inbounds [1024 x i8], [1024 x i8]* @CD, i64 0, i64 %tmp1
+  store i8 %mul, i8* %arrayidx3, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
+  %cmp = icmp slt i64 %indvars.iv.next, 1024
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  ret void
+}