3 files changed, 204 insertions, 16 deletions

diff --git a/test/Analysis/BasicAA/2008-12-09-GEP-IndicesAlias.ll b/test/Analysis/BasicAA/2008-12-09-GEP-IndicesAlias.ll
deleted file mode 100644
index aaf9061953e7..000000000000
--- a/test/Analysis/BasicAA/2008-12-09-GEP-IndicesAlias.ll
+++ /dev/null
@@ -1,16 +0,0 @@
-; RUN: opt < %s -aa-eval -print-all-alias-modref-info -disable-output |& grep {MustAlias:.*%R,.*%r}
-; Make sure that basicaa thinks R and r are must aliases.
-
-define i32 @test(i8 * %P) {
-entry:
-	%Q = bitcast i8* %P to {i32, i32}*
-	%R = getelementptr {i32, i32}* %Q, i32 0, i32 1
-	%S = load i32* %R
-
-	%q = bitcast i8* %P to {i32, i32}*
-	%r = getelementptr {i32, i32}* %q, i32 0, i32 1
-	%s = load i32* %r
-
-	%t = sub i32 %S, %s
-	ret i32 %t
-}
diff --git a/test/Analysis/BasicAA/gep-alias.ll b/test/Analysis/BasicAA/gep-alias.ll
new file mode 100644
index 000000000000..1ed031224713
--- /dev/null
+++ b/test/Analysis/BasicAA/gep-alias.ll
@@ -0,0 +1,171 @@
+; RUN: opt < %s -gvn -instcombine -S |& 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"
+
+; Make sure that basicaa thinks R and r are must aliases.
+define i32 @test1(i8 * %P) {
+entry:
+	%Q = bitcast i8* %P to {i32, i32}*
+	%R = getelementptr {i32, i32}* %Q, i32 0, i32 1
+	%S = load i32* %R
+
+	%q = bitcast i8* %P to {i32, i32}*
+	%r = getelementptr {i32, i32}* %q, i32 0, i32 1
+	%s = load i32* %r
+
+	%t = sub i32 %S, %s
+	ret i32 %t
+; CHECK: @test1
+; CHECK: ret i32 0
+}
+
+define i32 @test2(i8 * %P) {
+entry:
+	%Q = bitcast i8* %P to {i32, i32, i32}*
+	%R = getelementptr {i32, i32, i32}* %Q, i32 0, i32 1
+	%S = load i32* %R
+
+	%r = getelementptr {i32, i32, i32}* %Q, i32 0, i32 2
+  store i32 42, i32* %r
+
+	%s = load i32* %R
+
+	%t = sub i32 %S, %s
+	ret i32 %t
+; CHECK: @test2
+; CHECK: ret i32 0
+}
+
+
+; This was a miscompilation.
+define i32 @test3({float, {i32, i32, i32}}* %P) {
+entry:
+  %P2 = getelementptr {float, {i32, i32, i32}}* %P, i32 0, i32 1
+	%R = getelementptr {i32, i32, i32}* %P2, i32 0, i32 1
+	%S = load i32* %R
+
+	%r = getelementptr {i32, i32, i32}* %P2, i32 0, i32 2
+  store i32 42, i32* %r
+
+	%s = load i32* %R
+
+	%t = sub i32 %S, %s
+	ret i32 %t
+; CHECK: @test3
+; CHECK: ret i32 0
+}
+
+
+;; This is reduced from the SmallPtrSet constructor.
+%SmallPtrSetImpl = type { i8**, i32, i32, i32, [1 x i8*] }
+%SmallPtrSet64 = type { %SmallPtrSetImpl, [64 x i8*] }
+
+define i32 @test4(%SmallPtrSet64* %P) {
+entry:
+  %tmp2 = getelementptr inbounds %SmallPtrSet64* %P, i64 0, i32 0, i32 1
+  store i32 64, i32* %tmp2, align 8
+  %tmp3 = getelementptr inbounds %SmallPtrSet64* %P, i64 0, i32 0, i32 4, i64 64
+  store i8* null, i8** %tmp3, align 8
+  %tmp4 = load i32* %tmp2, align 8
+	ret i32 %tmp4
+; CHECK: @test4
+; CHECK: ret i32 64
+}
+
+; P[i] != p[i+1]
+define i32 @test5(i32* %p, i64 %i) {
+  %pi = getelementptr i32* %p, i64 %i
+  %i.next = add i64 %i, 1
+  %pi.next = getelementptr i32* %p, i64 %i.next
+  %x = load i32* %pi
+  store i32 42, i32* %pi.next
+  %y = load i32* %pi
+  %z = sub i32 %x, %y
+  ret i32 %z
+; CHECK: @test5
+; CHECK: ret i32 0
+}
+
+; P[i] != p[(i*4)|1]
+define i32 @test6(i32* %p, i64 %i1) {
+  %i = shl i64 %i1, 2
+  %pi = getelementptr i32* %p, i64 %i
+  %i.next = or i64 %i, 1
+  %pi.next = getelementptr i32* %p, i64 %i.next
+  %x = load i32* %pi
+  store i32 42, i32* %pi.next
+  %y = load i32* %pi
+  %z = sub i32 %x, %y
+  ret i32 %z
+; CHECK: @test6
+; CHECK: ret i32 0
+}
+
+; P[1] != P[i*4]
+define i32 @test7(i32* %p, i64 %i) {
+  %pi = getelementptr i32* %p, i64 1
+  %i.next = shl i64 %i, 2
+  %pi.next = getelementptr i32* %p, i64 %i.next
+  %x = load i32* %pi
+  store i32 42, i32* %pi.next
+  %y = load i32* %pi
+  %z = sub i32 %x, %y
+  ret i32 %z
+; CHECK: @test7
+; CHECK: ret i32 0
+}
+
+; P[zext(i)] != p[zext(i+1)]
+; PR1143
+define i32 @test8(i32* %p, i32 %i) {
+  %i1 = zext i32 %i to i64
+  %pi = getelementptr i32* %p, i64 %i1
+  %i.next = add i32 %i, 1
+  %i.next2 = zext i32 %i.next to i64
+  %pi.next = getelementptr i32* %p, i64 %i.next2
+  %x = load i32* %pi
+  store i32 42, i32* %pi.next
+  %y = load i32* %pi
+  %z = sub i32 %x, %y
+  ret i32 %z
+; CHECK: @test8
+; CHECK: ret i32 0
+}
+
+define i8 @test9([4 x i8] *%P, i32 %i, i32 %j) {
+  %i2 = shl i32 %i, 2
+  %i3 = add i32 %i2, 1
+  ; P2 = P + 1 + 4*i
+  %P2 = getelementptr [4 x i8] *%P, i32 0, i32 %i3
+
+  %j2 = shl i32 %j, 2
+  
+  ; P4 = P + 4*j
+  %P4 = getelementptr [4 x i8]* %P, i32 0, i32 %j2
+
+  %x = load i8* %P2
+  store i8 42, i8* %P4
+  %y = load i8* %P2
+  %z = sub i8 %x, %y
+  ret i8 %z
+; CHECK: @test9
+; CHECK: ret i8 0
+}
+
+define i8 @test10([4 x i8] *%P, i32 %i) {
+  %i2 = shl i32 %i, 2
+  %i3 = add i32 %i2, 4
+  ; P2 = P + 4 + 4*i
+  %P2 = getelementptr [4 x i8] *%P, i32 0, i32 %i3
+  
+  ; P4 = P + 4*i
+  %P4 = getelementptr [4 x i8]* %P, i32 0, i32 %i2
+
+  %x = load i8* %P2
+  store i8 42, i8* %P4
+  %y = load i8* %P2
+  %z = sub i8 %x, %y
+  ret i8 %z
+; CHECK: @test10
+; CHECK: ret i8 0
+}
diff --git a/test/Analysis/BasicAA/modref.ll b/test/Analysis/BasicAA/modref.ll
index 02db861c609f..3f642cff195c 100644
--- a/test/Analysis/BasicAA/modref.ll
+++ b/test/Analysis/BasicAA/modref.ll
@@ -4,6 +4,7 @@ target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:1
 declare void @llvm.memset.i32(i8*, i8, i32, i32)
 declare void @llvm.memset.i8(i8*, i8, i8, i32)
 declare void @llvm.memcpy.i8(i8*, i8*, i8, i32)
+declare void @llvm.memcpy.i32(i8*, i8*, i32, i32)
 declare void @llvm.lifetime.end(i64, i8* nocapture)
 
 declare void @external(i32*) 
@@ -90,3 +91,35 @@ define void @test3a(i8* %P, i8 %X) {
   ret void
 ; CHECK: ret void
 }
+
+@G1 = external global i32
+@G2 = external global [4000 x i32]
+
+define i32 @test4(i8* %P) {
+  %tmp = load i32* @G1
+  call void @llvm.memset.i32(i8* bitcast ([4000 x i32]* @G2 to i8*), i8 0, i32 4000, i32 1)
+  %tmp2 = load i32* @G1
+  %sub = sub i32 %tmp2, %tmp
+  ret i32 %sub
+; CHECK: @test4
+; CHECK: load i32* @G
+; CHECK: memset.i32
+; CHECK-NOT: load
+; CHECK: sub i32 %tmp, %tmp
+}
+
+; Verify that basicaa is handling variable length memcpy, knowing it doesn't
+; write to G1.
+define i32 @test5(i8* %P, i32 %Len) {
+  %tmp = load i32* @G1
+  call void @llvm.memcpy.i32(i8* bitcast ([4000 x i32]* @G2 to i8*), i8* bitcast (i32* @G1 to i8*), i32 %Len, i32 1)
+  %tmp2 = load i32* @G1
+  %sub = sub i32 %tmp2, %tmp
+  ret i32 %sub
+; CHECK: @test5
+; CHECK: load i32* @G
+; CHECK: memcpy.i32
+; CHECK-NOT: load
+; CHECK: sub i32 %tmp, %tmp
+}
+