BBVectorize: Add initial stores to the write set when tracking uses

When computing the use set of a store, we need to add the store to the write
set prior to iterating over later instructions. Otherwise, if there is a later
aliasing load of that store, that load will not be tagged as a use, and bad
things will happen.

trackUsesOfI still adds later dependent stores of an instruction to that
instruction's write set, but it never sees the original instruction, and so
when tracking uses of a store, the store must be added to the write set by the
caller.

Fixes PR16834.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188329 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp
index cbc1d630d52770f8297df72f567c8c8dd69e28fc..83ee1a459f05135b2ca5ec5a4e496e77c176ee6f 100644 (file)
--- a/lib/Transforms/Vectorize/BBVectorize.cpp
+++ b/lib/Transforms/Vectorize/BBVectorize.cpp
@@ -1182,6 +1182,8 @@ namespace {
       // Look for an instruction with which to pair instruction *I...
       DenseSet<Value *> Users;
       AliasSetTracker WriteSet(*AA);
       // Look for an instruction with which to pair instruction *I...
       DenseSet<Value *> Users;
       AliasSetTracker WriteSet(*AA);

+      if (I->mayWriteToMemory()) WriteSet.add(I);
+

       bool JAfterStart = IAfterStart;
       BasicBlock::iterator J = llvm::next(I);
       for (unsigned ss = 0; J != E && ss <= Config.SearchLimit; ++J, ++ss) {
       bool JAfterStart = IAfterStart;
       BasicBlock::iterator J = llvm::next(I);
       for (unsigned ss = 0; J != E && ss <= Config.SearchLimit; ++J, ++ss) {


@@ -1403,6 +1405,8 @@ namespace {
 
       DenseSet<Value *> Users;
       AliasSetTracker WriteSet(*AA);
 
       DenseSet<Value *> Users;
       AliasSetTracker WriteSet(*AA);

+      if (I->mayWriteToMemory()) WriteSet.add(I);
+

       for (BasicBlock::iterator J = llvm::next(I); J != E; ++J) {
         (void) trackUsesOfI(Users, WriteSet, I, J);
 
       for (BasicBlock::iterator J = llvm::next(I); J != E; ++J) {
         (void) trackUsesOfI(Users, WriteSet, I, J);
 


@@ -2804,6 +2808,8 @@ namespace {
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);

+    if (I->mayWriteToMemory()) WriteSet.add(I);
+

     for (; cast<Instruction>(L) != J; ++L)
       (void) trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs);
 
     for (; cast<Instruction>(L) != J; ++L)
       (void) trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs);
 


@@ -2824,6 +2830,8 @@ namespace {
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);

+    if (I->mayWriteToMemory()) WriteSet.add(I);
+

     for (; cast<Instruction>(L) != J;) {
       if (trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs)) {
         // Move this instruction
     for (; cast<Instruction>(L) != J;) {
       if (trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs)) {
         // Move this instruction


@@ -2853,6 +2861,7 @@ namespace {
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);
 
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);

+    if (I->mayWriteToMemory()) WriteSet.add(I);

 
     // Note: We cannot end the loop when we reach J because J could be moved
     // farther down the use chain by another instruction pairing. Also, J
 
     // Note: We cannot end the loop when we reach J because J could be moved
     // farther down the use chain by another instruction pairing. Also, J

diff --git a/test/Transforms/BBVectorize/X86/wr-aliases.ll b/test/Transforms/BBVectorize/X86/wr-aliases.ll
new file mode 100644 (file)
index 0000000..34b1d4e
--- /dev/null
+++ b/test/Transforms/BBVectorize/X86/wr-aliases.ll
@@ -0,0 +1,144 @@
+; RUN: opt -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -bb-vectorize -S < %s | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%class.QBezier.15 = type { double, double, double, double, double, double, double, double }
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #0
+
+; Function Attrs: uwtable
+declare fastcc void @_ZL12printQBezier7QBezier(%class.QBezier.15* byval nocapture readonly align 8) #1
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #0
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #0
+
+define void @main_arrayctor.cont([10 x %class.QBezier.15]* %beziers, %class.QBezier.15* %agg.tmp.i, %class.QBezier.15* %agg.tmp55.i, %class.QBezier.15* %agg.tmp56.i) {
+newFuncRoot:
+  br label %arrayctor.cont
+
+arrayctor.cont.ret.exitStub:                      ; preds = %arrayctor.cont
+  ret void
+
+; CHECK-LABEL: @main_arrayctor.cont
+; CHECK: <2 x double>
+; CHECK: @_ZL12printQBezier7QBezier
+; CHECK: store double %mul8.i, double* %x3.i, align 16
+; CHECK: load double* %x3.i, align 16
+; CHECK: ret
+
+arrayctor.cont:                                   ; preds = %newFuncRoot
+  %ref.tmp.sroa.0.0.idx = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 0
+  store double 1.000000e+01, double* %ref.tmp.sroa.0.0.idx, align 16
+  %ref.tmp.sroa.2.0.idx1 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 1
+  store double 2.000000e+01, double* %ref.tmp.sroa.2.0.idx1, align 8
+  %ref.tmp.sroa.3.0.idx2 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 2
+  store double 3.000000e+01, double* %ref.tmp.sroa.3.0.idx2, align 16
+  %ref.tmp.sroa.4.0.idx3 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 3
+  store double 4.000000e+01, double* %ref.tmp.sroa.4.0.idx3, align 8
+  %ref.tmp.sroa.5.0.idx4 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 4
+  store double 5.000000e+01, double* %ref.tmp.sroa.5.0.idx4, align 16
+  %ref.tmp.sroa.6.0.idx5 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 5
+  store double 6.000000e+01, double* %ref.tmp.sroa.6.0.idx5, align 8
+  %ref.tmp.sroa.7.0.idx6 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 6
+  store double 7.000000e+01, double* %ref.tmp.sroa.7.0.idx6, align 16
+  %ref.tmp.sroa.8.0.idx7 = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 7
+  store double 8.000000e+01, double* %ref.tmp.sroa.8.0.idx7, align 8
+  %add.ptr = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1
+  %v0 = bitcast %class.QBezier.15* %agg.tmp.i to i8*
+  call void @llvm.lifetime.start(i64 64, i8* %v0)
+  %v1 = bitcast %class.QBezier.15* %agg.tmp55.i to i8*
+  call void @llvm.lifetime.start(i64 64, i8* %v1)
+  %v2 = bitcast %class.QBezier.15* %agg.tmp56.i to i8*
+  call void @llvm.lifetime.start(i64 64, i8* %v2)
+  %v3 = bitcast [10 x %class.QBezier.15]* %beziers to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %v0, i8* %v3, i64 64, i32 8, i1 false)
+  call fastcc void @_ZL12printQBezier7QBezier(%class.QBezier.15* byval align 8 %agg.tmp.i)
+  %x2.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 2
+  %v4 = load double* %x2.i, align 16
+  %x3.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 4
+  %v5 = load double* %x3.i, align 16
+  %add.i = fadd double %v4, %v5
+  %mul.i = fmul double 5.000000e-01, %add.i
+  %x1.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 0
+  %v6 = load double* %x1.i, align 16
+  %add3.i = fadd double %v4, %v6
+  %mul4.i = fmul double 5.000000e-01, %add3.i
+  %x25.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 2
+  store double %mul4.i, double* %x25.i, align 16
+  %v7 = load double* %x3.i, align 16
+  %x4.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 6
+  %v8 = load double* %x4.i, align 16
+  %add7.i = fadd double %v7, %v8
+  %mul8.i = fmul double 5.000000e-01, %add7.i
+  store double %mul8.i, double* %x3.i, align 16
+  %v9 = load double* %x1.i, align 16
+  %x111.i = getelementptr inbounds %class.QBezier.15* %add.ptr, i64 0, i32 0
+  store double %v9, double* %x111.i, align 16
+  %v10 = load double* %x25.i, align 16
+  %add15.i = fadd double %mul.i, %v10
+  %mul16.i = fmul double 5.000000e-01, %add15.i
+  %x317.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 4
+  store double %mul16.i, double* %x317.i, align 16
+  %v11 = load double* %x3.i, align 16
+  %add19.i = fadd double %mul.i, %v11
+  %mul20.i = fmul double 5.000000e-01, %add19.i
+  store double %mul20.i, double* %x2.i, align 16
+  %v12 = load double* %x317.i, align 16
+  %add24.i = fadd double %v12, %mul20.i
+  %mul25.i = fmul double 5.000000e-01, %add24.i
+  store double %mul25.i, double* %x1.i, align 16
+  %x427.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 6
+  store double %mul25.i, double* %x427.i, align 16
+  %y2.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 3
+  %v13 = load double* %y2.i, align 8
+  %y3.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 5
+  %v14 = load double* %y3.i, align 8
+  %add28.i = fadd double %v13, %v14
+  %div.i = fmul double 5.000000e-01, %add28.i
+  %y1.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 1
+  %v15 = load double* %y1.i, align 8
+  %add30.i = fadd double %v13, %v15
+  %mul31.i = fmul double 5.000000e-01, %add30.i
+  %y232.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 3
+  store double %mul31.i, double* %y232.i, align 8
+  %v16 = load double* %y3.i, align 8
+  %y4.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 0, i32 7
+  %v17 = load double* %y4.i, align 8
+  %add34.i = fadd double %v16, %v17
+  %mul35.i = fmul double 5.000000e-01, %add34.i
+  store double %mul35.i, double* %y3.i, align 8
+  %v18 = load double* %y1.i, align 8
+  %y138.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 1
+  store double %v18, double* %y138.i, align 8
+  %v19 = load double* %y232.i, align 8
+  %add42.i = fadd double %div.i, %v19
+  %mul43.i = fmul double 5.000000e-01, %add42.i
+  %y344.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 5
+  store double %mul43.i, double* %y344.i, align 8
+  %v20 = load double* %y3.i, align 8
+  %add46.i = fadd double %div.i, %v20
+  %mul47.i = fmul double 5.000000e-01, %add46.i
+  store double %mul47.i, double* %y2.i, align 8
+  %v21 = load double* %y344.i, align 8
+  %add51.i = fadd double %v21, %mul47.i
+  %mul52.i = fmul double 5.000000e-01, %add51.i
+  store double %mul52.i, double* %y1.i, align 8
+  %y454.i = getelementptr inbounds [10 x %class.QBezier.15]* %beziers, i64 0, i64 1, i32 7
+  store double %mul52.i, double* %y454.i, align 8
+  %v22 = bitcast %class.QBezier.15* %add.ptr to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %v1, i8* %v22, i64 64, i32 8, i1 false)
+  call fastcc void @_ZL12printQBezier7QBezier(%class.QBezier.15* byval align 8 %agg.tmp55.i)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %v2, i8* %v3, i64 64, i32 8, i1 false)
+  call fastcc void @_ZL12printQBezier7QBezier(%class.QBezier.15* byval align 8 %agg.tmp56.i)
+  call void @llvm.lifetime.end(i64 64, i8* %v0)
+  call void @llvm.lifetime.end(i64 64, i8* %v1)
+  call void @llvm.lifetime.end(i64 64, i8* %v2)
+  br label %arrayctor.cont.ret.exitStub
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }