16 files changed, 697 insertions, 0 deletions
diff --git a/test/Transforms/SLPVectorizer/X86/barriercall.ll b/test/Transforms/SLPVectorizer/X86/barriercall.ll
new file mode 100644
index 0000000..04eb8f9
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/barriercall.ll
@@ -0,0 +1,32 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+;CHECK: @foo
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @foo(i32* nocapture %A, i32 %n) {
+entry:
+  %call = tail call i32 (...)* @bar() #2
+  %mul = mul nsw i32 %n, 5
+  %add = add nsw i32 %mul, 9
+  store i32 %add, i32* %A, align 4
+  %mul1 = mul nsw i32 %n, 9
+  %add2 = add nsw i32 %mul1, 9
+  %arrayidx3 = getelementptr inbounds i32* %A, i64 1
+  store i32 %add2, i32* %arrayidx3, align 4
+  %mul4 = shl i32 %n, 3
+  %add5 = add nsw i32 %mul4, 9
+  %arrayidx6 = getelementptr inbounds i32* %A, i64 2
+  store i32 %add5, i32* %arrayidx6, align 4
+  %mul7 = mul nsw i32 %n, 10
+  %add8 = add nsw i32 %mul7, 9
+  %arrayidx9 = getelementptr inbounds i32* %A, i64 3
+  store i32 %add8, i32* %arrayidx9, align 4
+  ret i32 undef
+}
+
+  ; We can still vectorize the stores below.
+
+declare i32 @bar(...)
diff --git a/test/Transforms/SLPVectorizer/X86/cast.ll b/test/Transforms/SLPVectorizer/X86/cast.ll
new file mode 100644
index 0000000..344dbbc
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/cast.ll
@@ -0,0 +1,38 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.9.0"
+
+; int foo(int * restrict A, char * restrict B) {
+;     A[0] = B[0];
+;     A[1] = B[1];
+;     A[2] = B[2];
+;     A[3] = B[3];
+; }
+;CHECK: @foo
+;CHECK: load <4 x i8>
+;CHECK: sext
+;CHECK: store <4 x i32>
+define i32 @foo(i32* noalias nocapture %A, i8* noalias nocapture %B) {
+entry:
+  %0 = load i8* %B, align 1
+  %conv = sext i8 %0 to i32
+  store i32 %conv, i32* %A, align 4
+  %arrayidx2 = getelementptr inbounds i8* %B, i64 1
+  %1 = load i8* %arrayidx2, align 1
+  %conv3 = sext i8 %1 to i32
+  %arrayidx4 = getelementptr inbounds i32* %A, i64 1
+  store i32 %conv3, i32* %arrayidx4, align 4
+  %arrayidx5 = getelementptr inbounds i8* %B, i64 2
+  %2 = load i8* %arrayidx5, align 1
+  %conv6 = sext i8 %2 to i32
+  %arrayidx7 = getelementptr inbounds i32* %A, i64 2
+  store i32 %conv6, i32* %arrayidx7, align 4
+  %arrayidx8 = getelementptr inbounds i8* %B, i64 3
+  %3 = load i8* %arrayidx8, align 1
+  %conv9 = sext i8 %3 to i32
+  %arrayidx10 = getelementptr inbounds i32* %A, i64 3
+  store i32 %conv9, i32* %arrayidx10, align 4
+  ret i32 undef
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/compare-reduce.ll b/test/Transforms/SLPVectorizer/X86/compare-reduce.ll
new file mode 100644
index 0000000..05f8e61
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/compare-reduce.ll
@@ -0,0 +1,53 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.7.0"
+
+@.str = private unnamed_addr constant [6 x i8] c"bingo\00", align 1
+
+;CHECK: @reduce_compare
+;CHECK: load <2 x double>
+;CHECK: fmul <2 x double>
+;CHECK: fmul <2 x double>
+;CHECK: fadd <2 x double>
+;CHECK: extractelement
+;CHECK: extractelement
+;CHECK: ret
+define void @reduce_compare(double* nocapture %A, i32 %n) {
+entry:
+  %conv = sitofp i32 %n to double
+  br label %for.body
+
+for.body:                                         ; preds = %for.inc, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
+  %0 = shl nsw i64 %indvars.iv, 1
+  %arrayidx = getelementptr inbounds double* %A, i64 %0
+  %1 = load double* %arrayidx, align 8
+  %mul1 = fmul double %conv, %1
+  %mul2 = fmul double %mul1, 7.000000e+00
+  %add = fadd double %mul2, 5.000000e+00
+  %2 = or i64 %0, 1
+  %arrayidx6 = getelementptr inbounds double* %A, i64 %2
+  %3 = load double* %arrayidx6, align 8
+  %mul8 = fmul double %conv, %3
+  %mul9 = fmul double %mul8, 4.000000e+00
+  %add10 = fadd double %mul9, 9.000000e+00
+  %cmp11 = fcmp ogt double %add, %add10
+  br i1 %cmp11, label %if.then, label %for.inc
+
+if.then:                                          ; preds = %for.body
+  %call = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([6 x i8]* @.str, i64 0, i64 0))
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body, %if.then
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 100
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.inc
+  ret void
+}
+
+declare i32 @printf(i8* nocapture, ...)
+
diff --git a/test/Transforms/SLPVectorizer/X86/diamond.ll b/test/Transforms/SLPVectorizer/X86/diamond.ll
new file mode 100644
index 0000000..8e85cb6
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/diamond.ll
@@ -0,0 +1,78 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+; int foo(int * restrict B,  int * restrict A, int n, int m) {
+;   B[0] = n * A[0] + m * A[0];
+;   B[1] = n * A[1] + m * A[1];
+;   B[2] = n * A[2] + m * A[2];
+;   B[3] = n * A[3] + m * A[3];
+;   return 0;
+; }
+
+; CHECK: @foo
+; CHECK: load <4 x i32>
+; CHECK: mul <4 x i32>
+; CHECK: store <4 x i32>
+; CHECK: ret
+define i32 @foo(i32* noalias nocapture %B, i32* noalias nocapture %A, i32 %n, i32 %m) #0 {
+entry:
+  %0 = load i32* %A, align 4
+  %mul238 = add i32 %m, %n
+  %add = mul i32 %0, %mul238
+  store i32 %add, i32* %B, align 4
+  %arrayidx4 = getelementptr inbounds i32* %A, i64 1
+  %1 = load i32* %arrayidx4, align 4
+  %add8 = mul i32 %1, %mul238
+  %arrayidx9 = getelementptr inbounds i32* %B, i64 1
+  store i32 %add8, i32* %arrayidx9, align 4
+  %arrayidx10 = getelementptr inbounds i32* %A, i64 2
+  %2 = load i32* %arrayidx10, align 4
+  %add14 = mul i32 %2, %mul238
+  %arrayidx15 = getelementptr inbounds i32* %B, i64 2
+  store i32 %add14, i32* %arrayidx15, align 4
+  %arrayidx16 = getelementptr inbounds i32* %A, i64 3
+  %3 = load i32* %arrayidx16, align 4
+  %add20 = mul i32 %3, %mul238
+  %arrayidx21 = getelementptr inbounds i32* %B, i64 3
+  store i32 %add20, i32* %arrayidx21, align 4
+  ret i32 0
+}
+
+
+; int foo_fail(int * restrict B,  int * restrict A, int n, int m) {
+;   B[0] = n * A[0] + m * A[0];
+;   B[1] = n * A[1] + m * A[1];
+;   B[2] = n * A[2] + m * A[2];
+;   B[3] = n * A[3] + m * A[3];
+;   return A[0];
+; }
+
+; CHECK: @foo_fail
+; CHECK-NOT: load <4 x i32>
+; CHECK: ret
+define i32 @foo_fail(i32* noalias nocapture %B, i32* noalias nocapture %A, i32 %n, i32 %m) {
+entry:
+  %0 = load i32* %A, align 4
+  %mul238 = add i32 %m, %n
+  %add = mul i32 %0, %mul238
+  store i32 %add, i32* %B, align 4
+  %arrayidx4 = getelementptr inbounds i32* %A, i64 1
+  %1 = load i32* %arrayidx4, align 4
+  %add8 = mul i32 %1, %mul238
+  %arrayidx9 = getelementptr inbounds i32* %B, i64 1
+  store i32 %add8, i32* %arrayidx9, align 4
+  %arrayidx10 = getelementptr inbounds i32* %A, i64 2
+  %2 = load i32* %arrayidx10, align 4
+  %add14 = mul i32 %2, %mul238
+  %arrayidx15 = getelementptr inbounds i32* %B, i64 2
+  store i32 %add14, i32* %arrayidx15, align 4
+  %arrayidx16 = getelementptr inbounds i32* %A, i64 3
+  %3 = load i32* %arrayidx16, align 4
+  %add20 = mul i32 %3, %mul238
+  %arrayidx21 = getelementptr inbounds i32* %B, i64 3
+  store i32 %add20, i32* %arrayidx21, align 4
+  ret i32 %0  ;<--------- This value has multiple users and can't be vectorized.
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/flag.ll b/test/Transforms/SLPVectorizer/X86/flag.ll
new file mode 100644
index 0000000..3ca5407
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/flag.ll
@@ -0,0 +1,51 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -slp-threshold=1000 -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+; Check that the command line flag works.
+;CHECK:rollable
+;CHECK-NOT:load <4 x i32>
+;CHECK: ret
+
+define i32 @rollable(i32* noalias nocapture %in, i32* noalias nocapture %out, i64 %n) {
+  %1 = icmp eq i64 %n, 0
+  br i1 %1, label %._crit_edge, label %.lr.ph
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ]
+  %2 = shl i64 %i.019, 2
+  %3 = getelementptr inbounds i32* %in, i64 %2
+  %4 = load i32* %3, align 4
+  %5 = or i64 %2, 1
+  %6 = getelementptr inbounds i32* %in, i64 %5
+  %7 = load i32* %6, align 4
+  %8 = or i64 %2, 2
+  %9 = getelementptr inbounds i32* %in, i64 %8
+  %10 = load i32* %9, align 4
+  %11 = or i64 %2, 3
+  %12 = getelementptr inbounds i32* %in, i64 %11
+  %13 = load i32* %12, align 4
+  %14 = mul i32 %4, 7
+  %15 = add i32 %14, 7
+  %16 = mul i32 %7, 7
+  %17 = add i32 %16, 14
+  %18 = mul i32 %10, 7
+  %19 = add i32 %18, 21
+  %20 = mul i32 %13, 7
+  %21 = add i32 %20, 28
+  %22 = getelementptr inbounds i32* %out, i64 %2
+  store i32 %15, i32* %22, align 4
+  %23 = getelementptr inbounds i32* %out, i64 %5
+  store i32 %17, i32* %23, align 4
+  %24 = getelementptr inbounds i32* %out, i64 %8
+  store i32 %19, i32* %24, align 4
+  %25 = getelementptr inbounds i32* %out, i64 %11
+  store i32 %21, i32* %25, align 4
+  %26 = add i64 %i.019, 1
+  %exitcond = icmp eq i64 %26, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  ret i32 undef
+}
diff --git a/test/Transforms/SLPVectorizer/X86/hoist.ll b/test/Transforms/SLPVectorizer/X86/hoist.ll
new file mode 100644
index 0000000..5074cea
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/hoist.ll
@@ -0,0 +1,59 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
+target triple = "i386-apple-macosx10.9.0"
+
+;int foo(int *A, int n, int k) {
+;  for (int i=0; i < 10000; i+=4) {
+;    A[i]   += n;
+;    A[i+1] += k;
+;    A[i+2] += n;
+;    A[i+3] += k;
+;  }
+;}
+
+; preheader:
+;CHECK: entry
+;CHECK-NEXT: insertelement
+;CHECK-NEXT: insertelement
+;CHECK-NEXT: insertelement
+;CHECK-NEXT: insertelement
+; loop body:
+;CHECK: phi
+;CHECK: load <4 x i32>
+;CHECK: add <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @foo(i32* nocapture %A, i32 %n, i32 %k) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.024 = phi i32 [ 0, %entry ], [ %add10, %for.body ]
+  %arrayidx = getelementptr inbounds i32* %A, i32 %i.024
+  %0 = load i32* %arrayidx, align 4
+  %add = add nsw i32 %0, %n
+  store i32 %add, i32* %arrayidx, align 4
+  %add121 = or i32 %i.024, 1
+  %arrayidx2 = getelementptr inbounds i32* %A, i32 %add121
+  %1 = load i32* %arrayidx2, align 4
+  %add3 = add nsw i32 %1, %k
+  store i32 %add3, i32* %arrayidx2, align 4
+  %add422 = or i32 %i.024, 2
+  %arrayidx5 = getelementptr inbounds i32* %A, i32 %add422
+  %2 = load i32* %arrayidx5, align 4
+  %add6 = add nsw i32 %2, %n
+  store i32 %add6, i32* %arrayidx5, align 4
+  %add723 = or i32 %i.024, 3
+  %arrayidx8 = getelementptr inbounds i32* %A, i32 %add723
+  %3 = load i32* %arrayidx8, align 4
+  %add9 = add nsw i32 %3, %k
+  store i32 %add9, i32* %arrayidx8, align 4
+  %add10 = add nsw i32 %i.024, 4
+  %cmp = icmp slt i32 %add10, 10000
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  ret i32 undef
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/lit.local.cfg b/test/Transforms/SLPVectorizer/X86/lit.local.cfg
new file mode 100644
index 0000000..a8ad0f1
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/lit.local.cfg
@@ -0,0 +1,6 @@
+config.suffixes = ['.ll', '.c', '.cpp']
+
+targets = set(config.root.targets_to_build.split())
+if not 'X86' in targets:
+    config.unsupported = True
+
diff --git a/test/Transforms/SLPVectorizer/X86/loopinvariant.ll b/test/Transforms/SLPVectorizer/X86/loopinvariant.ll
new file mode 100644
index 0000000..4a37fce
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/loopinvariant.ll
@@ -0,0 +1,69 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+;CHECK: @foo
+;CHECK: load <4 x i32>
+;CHECK: add <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: load <4 x i32>
+;CHECK: add <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @foo(i32* nocapture %A, i32 %n) #0 {
+entry:
+  %cmp62 = icmp sgt i32 %n, 0
+  br i1 %cmp62, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %add1 = add nsw i32 %0, %n
+  store i32 %add1, i32* %arrayidx, align 4
+  %1 = or i64 %indvars.iv, 1
+  %arrayidx4 = getelementptr inbounds i32* %A, i64 %1
+  %2 = load i32* %arrayidx4, align 4
+  %add5 = add nsw i32 %2, %n
+  store i32 %add5, i32* %arrayidx4, align 4
+  %3 = or i64 %indvars.iv, 2
+  %arrayidx8 = getelementptr inbounds i32* %A, i64 %3
+  %4 = load i32* %arrayidx8, align 4
+  %add9 = add nsw i32 %4, %n
+  store i32 %add9, i32* %arrayidx8, align 4
+  %5 = or i64 %indvars.iv, 3
+  %arrayidx12 = getelementptr inbounds i32* %A, i64 %5
+  %6 = load i32* %arrayidx12, align 4
+  %add13 = add nsw i32 %6, %n
+  store i32 %add13, i32* %arrayidx12, align 4
+  %7 = or i64 %indvars.iv, 4
+  %arrayidx16 = getelementptr inbounds i32* %A, i64 %7
+  %8 = load i32* %arrayidx16, align 4
+  %add17 = add nsw i32 %8, %n
+  store i32 %add17, i32* %arrayidx16, align 4
+  %9 = or i64 %indvars.iv, 5
+  %arrayidx20 = getelementptr inbounds i32* %A, i64 %9
+  %10 = load i32* %arrayidx20, align 4
+  %add21 = add nsw i32 %10, %n
+  store i32 %add21, i32* %arrayidx20, align 4
+  %11 = or i64 %indvars.iv, 6
+  %arrayidx24 = getelementptr inbounds i32* %A, i64 %11
+  %12 = load i32* %arrayidx24, align 4
+  %add25 = add nsw i32 %12, %n
+  store i32 %add25, i32* %arrayidx24, align 4
+  %13 = or i64 %indvars.iv, 7
+  %arrayidx28 = getelementptr inbounds i32* %A, i64 %13
+  %14 = load i32* %arrayidx28, align 4
+  %add29 = add nsw i32 %14, %n
+  store i32 %add29, i32* %arrayidx28, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 8
+  %15 = trunc i64 %indvars.iv.next to i32
+  %cmp = icmp slt i32 %15, %n
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body, %entry
+  ret i32 undef
+}
+
+attributes #0 = { nounwind ssp 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" "unsafe-fp-math"="false" "use-soft-float"="false" }
diff --git a/test/Transforms/SLPVectorizer/X86/multi_user.ll b/test/Transforms/SLPVectorizer/X86/multi_user.ll
new file mode 100644
index 0000000..aaa6063
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/multi_user.ll
@@ -0,0 +1,47 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.7.0"
+
+;int foo (int *A, int n) {
+;  A[0] += n * 5 + 7;
+;  A[1] += n * 5 + 8;
+;  A[2] += n * 5 + 9;
+;  A[3] += n * 5 + 10;
+;  A[4] += n * 5 + 11;
+;}
+
+;CHECK: @foo
+;CHECK: insertelement <4 x i32>
+;CHECK: load <4 x i32>
+;CHECK: add <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @foo(i32* nocapture %A, i32 %n) {
+  %1 = mul nsw i32 %n, 5
+  %2 = add nsw i32 %1, 7
+  %3 = load i32* %A, align 4
+  %4 = add nsw i32 %2, %3
+  store i32 %4, i32* %A, align 4
+  %5 = add nsw i32 %1, 8
+  %6 = getelementptr inbounds i32* %A, i64 1
+  %7 = load i32* %6, align 4
+  %8 = add nsw i32 %5, %7
+  store i32 %8, i32* %6, align 4
+  %9 = add nsw i32 %1, 9
+  %10 = getelementptr inbounds i32* %A, i64 2
+  %11 = load i32* %10, align 4
+  %12 = add nsw i32 %9, %11
+  store i32 %12, i32* %10, align 4
+  %13 = add nsw i32 %1, 10
+  %14 = getelementptr inbounds i32* %A, i64 3
+  %15 = load i32* %14, align 4
+  %16 = add nsw i32 %13, %15
+  store i32 %16, i32* %14, align 4
+  %17 = add nsw i32 %1, 11
+  %18 = getelementptr inbounds i32* %A, i64 4
+  %19 = load i32* %18, align 4
+  %20 = add nsw i32 %17, %19
+  store i32 %20, i32* %18, align 4
+  ret i32 undef
+}
diff --git a/test/Transforms/SLPVectorizer/X86/reduction.ll b/test/Transforms/SLPVectorizer/X86/reduction.ll
new file mode 100644
index 0000000..70b7c3a
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/reduction.ll
@@ -0,0 +1,47 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
+target triple = "i386-apple-macosx10.8.0"
+
+; int foo(double *A, int n, int m) {
+;   double sum = 0, v1 = 2, v0 = 3;
+;   for (int i=0; i < n; ++i)
+;     sum += 7*A[i*2] + 7*A[i*2+1];
+;   return sum;
+; }
+
+;CHECK: reduce
+;CHECK: load <2 x double>
+;CHECK: fmul <2 x double>
+;CHECK: ret
+define i32 @reduce(double* nocapture %A, i32 %n, i32 %m) {
+entry:
+  %cmp13 = icmp sgt i32 %n, 0
+  br i1 %cmp13, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.015 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  %sum.014 = phi double [ %add6, %for.body ], [ 0.000000e+00, %entry ]
+  %mul = shl nsw i32 %i.015, 1
+  %arrayidx = getelementptr inbounds double* %A, i32 %mul
+  %0 = load double* %arrayidx, align 4
+  %mul1 = fmul double %0, 7.000000e+00
+  %add12 = or i32 %mul, 1
+  %arrayidx3 = getelementptr inbounds double* %A, i32 %add12
+  %1 = load double* %arrayidx3, align 4
+  %mul4 = fmul double %1, 7.000000e+00
+  %add5 = fadd double %mul1, %mul4
+  %add6 = fadd double %sum.014, %add5
+  %inc = add nsw i32 %i.015, 1
+  %exitcond = icmp eq i32 %inc, %n
+  br i1 %exitcond, label %for.cond.for.end_crit_edge, label %for.body
+
+for.cond.for.end_crit_edge:                       ; preds = %for.body
+  %phitmp = fptosi double %add6 to i32
+  br label %for.end
+
+for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
+  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge ], [ 0, %entry ]
+  ret i32 %sum.0.lcssa
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/reduction2.ll b/test/Transforms/SLPVectorizer/X86/reduction2.ll
new file mode 100644
index 0000000..7aa7d7e
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/reduction2.ll
@@ -0,0 +1,32 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
+target triple = "i386-apple-macosx10.8.0"
+
+;CHECK: @foo
+;CHECK: load <2 x double>
+;CHECK: ret
+define double @foo(double* nocapture %D) {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %i.02 = phi i32 [ 0, %0 ], [ %10, %1 ]
+  %sum.01 = phi double [ 0.000000e+00, %0 ], [ %9, %1 ]
+  %2 = shl nsw i32 %i.02, 1
+  %3 = getelementptr inbounds double* %D, i32 %2
+  %4 = load double* %3, align 4
+  %A4 = fmul double %4, %4
+  %5 = or i32 %2, 1
+  %6 = getelementptr inbounds double* %D, i32 %5
+  %7 = load double* %6, align 4
+  %A7 = fmul double %7, %7
+  %8 = fadd double %A4, %A7
+  %9 = fadd double %sum.01, %8
+  %10 = add nsw i32 %i.02, 1
+  %exitcond = icmp eq i32 %10, 100
+  br i1 %exitcond, label %11, label %1
+
+; <label>:11                                      ; preds = %1
+  ret double %9
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/saxpy.ll b/test/Transforms/SLPVectorizer/X86/saxpy.ll
new file mode 100644
index 0000000..b520913
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/saxpy.ll
@@ -0,0 +1,45 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+; SLP vectorization example from http://cs.stanford.edu/people/eschkufz/research/asplos291-schkufza.pdf
+;CHECK: SAXPY
+;CHECK: mul <4 x i32>
+;CHECK: ret
+
+define void @SAXPY(i32* noalias nocapture %x, i32* noalias nocapture %y, i32 %a, i64 %i) {
+  %1 = getelementptr inbounds i32* %x, i64 %i
+  %2 = load i32* %1, align 4
+  %3 = mul nsw i32 %2, %a
+  %4 = getelementptr inbounds i32* %y, i64 %i
+  %5 = load i32* %4, align 4
+  %6 = add nsw i32 %3, %5
+  store i32 %6, i32* %1, align 4
+  %7 = add i64 %i, 1
+  %8 = getelementptr inbounds i32* %x, i64 %7
+  %9 = load i32* %8, align 4
+  %10 = mul nsw i32 %9, %a
+  %11 = getelementptr inbounds i32* %y, i64 %7
+  %12 = load i32* %11, align 4
+  %13 = add nsw i32 %10, %12
+  store i32 %13, i32* %8, align 4
+  %14 = add i64 %i, 2
+  %15 = getelementptr inbounds i32* %x, i64 %14
+  %16 = load i32* %15, align 4
+  %17 = mul nsw i32 %16, %a
+  %18 = getelementptr inbounds i32* %y, i64 %14
+  %19 = load i32* %18, align 4
+  %20 = add nsw i32 %17, %19
+  store i32 %20, i32* %15, align 4
+  %21 = add i64 %i, 3
+  %22 = getelementptr inbounds i32* %x, i64 %21
+  %23 = load i32* %22, align 4
+  %24 = mul nsw i32 %23, %a
+  %25 = getelementptr inbounds i32* %y, i64 %21
+  %26 = load i32* %25, align 4
+  %27 = add nsw i32 %24, %26
+  store i32 %27, i32* %22, align 4
+  ret void
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/simple-loop.ll b/test/Transforms/SLPVectorizer/X86/simple-loop.ll
new file mode 100644
index 0000000..0111b94
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/simple-loop.ll
@@ -0,0 +1,100 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+;CHECK:rollable
+define i32 @rollable(i32* noalias nocapture %in, i32* noalias nocapture %out, i64 %n) {
+  %1 = icmp eq i64 %n, 0
+  br i1 %1, label %._crit_edge, label %.lr.ph
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ]
+  %2 = shl i64 %i.019, 2
+  %3 = getelementptr inbounds i32* %in, i64 %2
+;CHECK:load <4 x i32>
+  %4 = load i32* %3, align 4
+  %5 = or i64 %2, 1
+  %6 = getelementptr inbounds i32* %in, i64 %5
+  %7 = load i32* %6, align 4
+  %8 = or i64 %2, 2
+  %9 = getelementptr inbounds i32* %in, i64 %8
+  %10 = load i32* %9, align 4
+  %11 = or i64 %2, 3
+  %12 = getelementptr inbounds i32* %in, i64 %11
+  %13 = load i32* %12, align 4
+;CHECK:mul <4 x i32>
+  %14 = mul i32 %4, 7
+;CHECK:add <4 x i32>
+  %15 = add i32 %14, 7
+  %16 = mul i32 %7, 7
+  %17 = add i32 %16, 14
+  %18 = mul i32 %10, 7
+  %19 = add i32 %18, 21
+  %20 = mul i32 %13, 7
+  %21 = add i32 %20, 28
+  %22 = getelementptr inbounds i32* %out, i64 %2
+;CHECK:store <4 x i32>
+  store i32 %15, i32* %22, align 4
+  %23 = getelementptr inbounds i32* %out, i64 %5
+  store i32 %17, i32* %23, align 4
+  %24 = getelementptr inbounds i32* %out, i64 %8
+  store i32 %19, i32* %24, align 4
+  %25 = getelementptr inbounds i32* %out, i64 %11
+  store i32 %21, i32* %25, align 4
+  %26 = add i64 %i.019, 1
+  %exitcond = icmp eq i64 %26, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+;CHECK: ret
+  ret i32 undef
+}
+
+;CHECK:unrollable
+;CHECK-NOT: <4 x i32>
+;CHECK: ret
+define i32 @unrollable(i32* %in, i32* %out, i64 %n) nounwind ssp uwtable {
+  %1 = icmp eq i64 %n, 0
+  br i1 %1, label %._crit_edge, label %.lr.ph
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ]
+  %2 = shl i64 %i.019, 2
+  %3 = getelementptr inbounds i32* %in, i64 %2
+  %4 = load i32* %3, align 4
+  %5 = or i64 %2, 1
+  %6 = getelementptr inbounds i32* %in, i64 %5
+  %7 = load i32* %6, align 4
+  %8 = or i64 %2, 2
+  %9 = getelementptr inbounds i32* %in, i64 %8
+  %10 = load i32* %9, align 4
+  %11 = or i64 %2, 3
+  %12 = getelementptr inbounds i32* %in, i64 %11
+  %13 = load i32* %12, align 4
+  %14 = mul i32 %4, 7
+  %15 = add i32 %14, 7
+  %16 = mul i32 %7, 7
+  %17 = add i32 %16, 14
+  %18 = mul i32 %10, 7
+  %19 = add i32 %18, 21
+  %20 = mul i32 %13, 7
+  %21 = add i32 %20, 28
+  %22 = getelementptr inbounds i32* %out, i64 %2
+  store i32 %15, i32* %22, align 4
+  %23 = getelementptr inbounds i32* %out, i64 %5
+  store i32 %17, i32* %23, align 4
+  %barrier = call i32 @goo(i32 0)                      ; <---------------- memory barrier.
+  %24 = getelementptr inbounds i32* %out, i64 %8
+  store i32 %19, i32* %24, align 4
+  %25 = getelementptr inbounds i32* %out, i64 %11
+  store i32 %21, i32* %25, align 4
+  %26 = add i64 %i.019, 1
+  %exitcond = icmp eq i64 %26, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  ret i32 undef
+}
+
+declare i32 @goo(i32)
diff --git a/test/Transforms/SLPVectorizer/X86/simplebb.ll b/test/Transforms/SLPVectorizer/X86/simplebb.ll
new file mode 100644
index 0000000..cd0b99e
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/simplebb.ll
@@ -0,0 +1,25 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+; Simple 3-pair chain with loads and stores
+; CHECK: test1
+; CHECK: store <2 x double>
+; CHECK: ret
+define void @test1(double* %a, double* %b, double* %c) {
+entry:
+  %i0 = load double* %a, align 8
+  %i1 = load double* %b, align 8
+  %mul = fmul double %i0, %i1
+  %arrayidx3 = getelementptr inbounds double* %a, i64 1
+  %i3 = load double* %arrayidx3, align 8
+  %arrayidx4 = getelementptr inbounds double* %b, i64 1
+  %i4 = load double* %arrayidx4, align 8
+  %mul5 = fmul double %i3, %i4
+  store double %mul, double* %c, align 8
+  %arrayidx5 = getelementptr inbounds double* %c, i64 1
+  store double %mul5, double* %arrayidx5, align 8
+  ret void
+}
+
diff --git a/test/Transforms/SLPVectorizer/X86/vector.ll b/test/Transforms/SLPVectorizer/X86/vector.ll
new file mode 100644
index 0000000..02a1897
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/vector.ll
@@ -0,0 +1,14 @@
+; RUN: opt < %s -slp-vectorizer -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-apple-macosx10.8.0"
+
+; Make sure that we are not crashing or changing the code.
+;CHECK: test
+;CHECK: icmp
+;CHECK: ret
+define void @test(<4 x i32> %in, <4 x i32> %in2) {
+  %k = icmp eq <4 x i32> %in, %in2
+  ret void
+}
+
diff --git a/test/Transforms/SLPVectorizer/lit.local.cfg b/test/Transforms/SLPVectorizer/lit.local.cfg
new file mode 100644
index 0000000..19eebc0
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/lit.local.cfg
@@ -0,0 +1 @@
+config.suffixes = ['.ll', '.c', '.cpp']