Vendor import of llvm trunk r126079:

http://llvm.org/svn/llvm-project/llvm/trunk@126079

19 files changed, 551 insertions, 289 deletions

diff --git a/test/Transforms/ScalarRepl/2003-05-30-InvalidIndices.ll b/test/Transforms/ScalarRepl/2003-05-30-InvalidIndices.ll
deleted file mode 100644
index b147ec9..0000000
--- a/test/Transforms/ScalarRepl/2003-05-30-InvalidIndices.ll
+++ /dev/null
@@ -1,8 +0,0 @@
-; RUN: opt < %s -scalarrepl
-
-define void @main() {
-	%E = alloca { { i32, float, double, i64 }, { i32, float, double, i64 } }	; <{ { i32, float, double, i64 }, { i32, float, double, i64 } }*> [#uses=1]
-	%tmp.151 = getelementptr { { i32, float, double, i64 }, { i32, float, double, i64 } }* %E, i64 0, i32 1, i32 3		; <i64*> [#uses=0]
-	ret void
-}
-
diff --git a/test/Transforms/ScalarRepl/2003-05-30-MultiLevel.ll b/test/Transforms/ScalarRepl/2003-05-30-MultiLevel.ll
deleted file mode 100644
index 89c0b05..0000000
--- a/test/Transforms/ScalarRepl/2003-05-30-MultiLevel.ll
+++ /dev/null
@@ -1,10 +0,0 @@
-; RUN: opt < %s -scalarrepl
-
-define i32 @test() {
-	%X = alloca { [4 x i32] }		; <{ [4 x i32] }*> [#uses=1]
-	%Y = getelementptr { [4 x i32] }* %X, i64 0, i32 0, i64 2		; <i32*> [#uses=2]
-	store i32 4, i32* %Y
-	%Z = load i32* %Y		; <i32> [#uses=1]
-	ret i32 %Z
-}
-
diff --git a/test/Transforms/ScalarRepl/2005-12-14-UnionPromoteCrash.ll b/test/Transforms/ScalarRepl/2005-12-14-UnionPromoteCrash.ll
deleted file mode 100644
index ea23c31..0000000
--- a/test/Transforms/ScalarRepl/2005-12-14-UnionPromoteCrash.ll
+++ /dev/null
@@ -1,28 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-
-target datalayout = "E-p:32:32"
-	%struct.rtx_def = type { [2 x i8], i32, [1 x %union.rtunion_def] }
-	%union.rtunion_def = type { i32 }
-
-define void @find_reloads() {
-entry:
-	%c_addr.i = alloca i8		; <i8*> [#uses=1]
-	switch i32 0, label %return [
-		 i32 36, label %label.7
-		 i32 34, label %label.7
-		 i32 41, label %label.5
-	]
-label.5:		; preds = %entry
-	ret void
-label.7:		; preds = %entry, %entry
-	br i1 false, label %then.4, label %switchexit.0
-then.4:		; preds = %label.7
-	%tmp.0.i = bitcast i8* %c_addr.i to i32*		; <i32*> [#uses=1]
-	store i32 44, i32* %tmp.0.i
-	ret void
-switchexit.0:		; preds = %label.7
-	ret void
-return:		; preds = %entry
-	ret void
-}
-
diff --git a/test/Transforms/ScalarRepl/2006-01-24-IllegalUnionPromoteCrash.ll b/test/Transforms/ScalarRepl/2006-01-24-IllegalUnionPromoteCrash.ll
deleted file mode 100644
index 03c7452..0000000
--- a/test/Transforms/ScalarRepl/2006-01-24-IllegalUnionPromoteCrash.ll
+++ /dev/null
@@ -1,12 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-
-target datalayout = "E-p:32:32"
-
-define i32 @test(i64 %L) {
-	%X = alloca i32		; <i32*> [#uses=2]
-	%Y = bitcast i32* %X to i64*		; <i64*> [#uses=1]
-	store i64 0, i64* %Y
-	%Z = load i32* %X		; <i32> [#uses=1]
-	ret i32 %Z
-}
-
diff --git a/test/Transforms/ScalarRepl/2006-04-20-PromoteCrash.ll b/test/Transforms/ScalarRepl/2006-04-20-PromoteCrash.ll
deleted file mode 100644
index 63840f1..0000000
--- a/test/Transforms/ScalarRepl/2006-04-20-PromoteCrash.ll
+++ /dev/null
@@ -1,18 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-
-define void @output_toc() {
-entry:
-	%buf = alloca [256 x i8], align 16		; <[256 x i8]*> [#uses=1]
-	%name = alloca i8*, align 4		; <i8**> [#uses=1]
-	%real_name = alloca i8*, align 4		; <i8**> [#uses=0]
-	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
-	%buf.upgrd.1 = bitcast [256 x i8]* %buf to i8*		; <i8*> [#uses=1]
-	store i8* %buf.upgrd.1, i8** %name
-	call void @abort( )
-	unreachable
-return:		; No predecessors!
-	ret void
-}
-
-declare void @abort()
-
diff --git a/test/Transforms/ScalarRepl/2006-10-23-PointerUnionCrash.ll b/test/Transforms/ScalarRepl/2006-10-23-PointerUnionCrash.ll
deleted file mode 100644
index dcd7e53..0000000
--- a/test/Transforms/ScalarRepl/2006-10-23-PointerUnionCrash.ll
+++ /dev/null
@@ -1,57 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-; END.
-target datalayout = "e-p:32:32"
-target triple = "i686-apple-darwin8.7.2"
-
-define void @glgProcessColor() {
-entry:
-	%source_ptr = alloca i8*, align 4		; <i8**> [#uses=2]
-	br i1 false, label %bb1357, label %cond_next583
-cond_next583:		; preds = %entry
-	ret void
-bb1357:		; preds = %entry
-	br i1 false, label %bb1365, label %bb27055
-bb1365:		; preds = %bb1357
-	switch i32 0, label %cond_next10377 [
-		 i32 0, label %bb4679
-		 i32 1, label %bb4679
-		 i32 2, label %bb4679
-		 i32 3, label %bb4679
-		 i32 4, label %bb5115
-		 i32 5, label %bb6651
-		 i32 6, label %bb7147
-		 i32 7, label %bb8683
-		 i32 8, label %bb9131
-		 i32 9, label %bb9875
-		 i32 10, label %bb4679
-		 i32 11, label %bb4859
-		 i32 12, label %bb4679
-		 i32 16, label %bb10249
-	]
-bb4679:		; preds = %bb1365, %bb1365, %bb1365, %bb1365, %bb1365, %bb1365
-	ret void
-bb4859:		; preds = %bb1365
-	ret void
-bb5115:		; preds = %bb1365
-	ret void
-bb6651:		; preds = %bb1365
-	ret void
-bb7147:		; preds = %bb1365
-	ret void
-bb8683:		; preds = %bb1365
-	ret void
-bb9131:		; preds = %bb1365
-	ret void
-bb9875:		; preds = %bb1365
-	%source_ptr9884 = bitcast i8** %source_ptr to i8**		; <i8**> [#uses=1]
-	%tmp9885 = load i8** %source_ptr9884		; <i8*> [#uses=0]
-	ret void
-bb10249:		; preds = %bb1365
-	%source_ptr10257 = bitcast i8** %source_ptr to i16**		; <i16**> [#uses=1]
-	%tmp10258 = load i16** %source_ptr10257		; <i16*> [#uses=0]
-	ret void
-cond_next10377:		; preds = %bb1365
-	ret void
-bb27055:		; preds = %bb1357
-	ret void
-}
diff --git a/test/Transforms/ScalarRepl/2006-12-11-SROA-Crash.ll b/test/Transforms/ScalarRepl/2006-12-11-SROA-Crash.ll
deleted file mode 100644
index 2606203..0000000
--- a/test/Transforms/ScalarRepl/2006-12-11-SROA-Crash.ll
+++ /dev/null
@@ -1,20 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-; PR1045
-
-target datalayout = "e-p:32:32"
-target triple = "i686-pc-linux-gnu"
-	%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>" = type { %"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block"* }
-	%"struct.__gnu_cxx::bitmap_allocator<char>" = type { i8 }
-	%"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block" = type { [8 x i8] }
-
-define void @_ZN9__gnu_cxx16bitmap_allocatorIwE27_M_deallocate_single_objectEPw() {
-entry:
-	%this_addr.i = alloca %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*		; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"**> [#uses=3]
-	%tmp = alloca %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>", align 4		; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
-	store %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp, %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i
-	%tmp.i = load %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i		; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
-	%tmp.i.upgrd.1 = bitcast %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp.i to %"struct.__gnu_cxx::bitmap_allocator<char>"*		; <%"struct.__gnu_cxx::bitmap_allocator<char>"*> [#uses=0]
-	%tmp1.i = load %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i		; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
-	%tmp.i.upgrd.2 = getelementptr %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp1.i, i32 0, i32 0		; <%"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block"**> [#uses=0]
-	unreachable
-}
diff --git a/test/Transforms/ScalarRepl/2007-03-19-CanonicalizeMemcpy.ll b/test/Transforms/ScalarRepl/2007-03-19-CanonicalizeMemcpy.ll
deleted file mode 100644
index bd49106..0000000
--- a/test/Transforms/ScalarRepl/2007-03-19-CanonicalizeMemcpy.ll
+++ /dev/null
@@ -1,44 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-output
-
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:64:64-v128:128:128-a0:0:64"
-target triple = "arm-apple-darwin8"
-	%struct.CGPoint = type { float, float }
-	%struct.aal_big_range_t = type { i32, i32 }
-	%struct.aal_callback_t = type { i8* (i8*, i32)*, void (i8*, i8*)* }
-	%struct.aal_edge_pool_t = type { %struct.aal_edge_pool_t*, i32, i32, [0 x %struct.aal_edge_t] }
-	%struct.aal_edge_t = type { %struct.CGPoint, %struct.CGPoint, i32 }
-	%struct.aal_range_t = type { i16, i16 }
-	%struct.aal_span_pool_t = type { %struct.aal_span_pool_t*, [341 x %struct.aal_span_t] }
-	%struct.aal_span_t = type { %struct.aal_span_t*, %struct.aal_big_range_t }
-	%struct.aal_spanarray_t = type { [2 x %struct.aal_range_t] }
-	%struct.aal_spanbucket_t = type { i16, [2 x i8], %struct.anon }
-	%struct.aal_state_t = type { %struct.CGPoint, %struct.CGPoint, %struct.CGPoint, i32, float, float, float, float, %struct.CGPoint, %struct.CGPoint, float, float, float, float, i32, i32, i32, i32, float, float, i8*, i32, i32, %struct.aal_edge_pool_t*, %struct.aal_edge_pool_t*, i8*, %struct.aal_callback_t*, i32, %struct.aal_span_t*, %struct.aal_span_t*, %struct.aal_span_t*, %struct.aal_span_pool_t*, i8, float, i8, i32 }
-	%struct.anon = type { %struct.aal_spanarray_t }
-
-
-declare void @llvm.memcpy.i32(i8*, i8*, i32, i32)
-
-define fastcc void @aal_insert_span() {
-entry:
-	%SB = alloca %struct.aal_spanbucket_t, align 4		; <%struct.aal_spanbucket_t*> [#uses=2]
-	br i1 false, label %cond_true, label %cond_next79
-
-cond_true:		; preds = %entry
-	br i1 false, label %cond_next, label %cond_next114.i
-
-cond_next114.i:		; preds = %cond_true
-	ret void
-
-cond_next:		; preds = %cond_true
-	%SB19 = bitcast %struct.aal_spanbucket_t* %SB to i8*		; <i8*> [#uses=1]
-	call void @llvm.memcpy.i32( i8* %SB19, i8* null, i32 12, i32 0 )
-	br i1 false, label %cond_next34, label %cond_next79
-
-cond_next34:		; preds = %cond_next
-	%i.2.reload22 = load i32* null		; <i32> [#uses=1]
-	%tmp51 = getelementptr %struct.aal_spanbucket_t* %SB, i32 0, i32 2, i32 0, i32 0, i32 %i.2.reload22, i32 1		; <i16*> [#uses=0]
-	ret void
-
-cond_next79:		; preds = %cond_next, %entry
-	ret void
-}
diff --git a/test/Transforms/ScalarRepl/2009-01-09-scalarrepl-empty.ll b/test/Transforms/ScalarRepl/2009-01-09-scalarrepl-empty.ll
deleted file mode 100644
index 62f7d19..0000000
--- a/test/Transforms/ScalarRepl/2009-01-09-scalarrepl-empty.ll
+++ /dev/null
@@ -1,15 +0,0 @@
-; RUN: opt < %s -scalarrepl | llvm-dis
-; PR3304
-
-       %struct.c37304a__vrec = type { i8, %struct.c37304a__vrec___disc___XVN }
-        %struct.c37304a__vrec___disc___XVN = type {
-%struct.c37304a__vrec___disc___XVN___O }
-        %struct.c37304a__vrec___disc___XVN___O = type {  }
-
-define void @_ada_c37304a() {
-entry:
-        %v = alloca %struct.c37304a__vrec
-        %0 = getelementptr %struct.c37304a__vrec* %v, i32 0, i32 0             
-        store i8 8, i8* %0, align 1
-        unreachable
-}
diff --git a/test/Transforms/ScalarRepl/2009-04-21-ZeroLengthMemSet.ll b/test/Transforms/ScalarRepl/2009-04-21-ZeroLengthMemSet.ll
deleted file mode 100644
index c5ebf8e..0000000
--- a/test/Transforms/ScalarRepl/2009-04-21-ZeroLengthMemSet.ll
+++ /dev/null
@@ -1,16 +0,0 @@
-; RUN: opt < %s -scalarrepl | llvm-dis
-; rdar://6808691
-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-f80:128:128"
-target triple = "x86_64-apple-darwin9.0"
-	type <{ i32, i16, i8, i8, i64, i64, i16, [0 x i16] }>		
-
-define i32 @foo() {
-entry:
-	%.compoundliteral = alloca %0		
-	%tmp228 = getelementptr %0* %.compoundliteral, i32 0, i32 7
-	%tmp229 = bitcast [0 x i16]* %tmp228 to i8*		
-	call void @llvm.memset.i64(i8* %tmp229, i8 0, i64 0, i32 2)
-	unreachable
-}
-
-declare void @llvm.memset.i64(i8* nocapture, i8, i64, i32) nounwind
diff --git a/test/Transforms/ScalarRepl/2009-05-08-I1Crash.ll b/test/Transforms/ScalarRepl/2009-05-08-I1Crash.ll
deleted file mode 100644
index aa3487b..0000000
--- a/test/Transforms/ScalarRepl/2009-05-08-I1Crash.ll
+++ /dev/null
@@ -1,12 +0,0 @@
-; RUN: opt < %s -scalarrepl | llvm-dis
-; PR4146
-
- %wrapper = type { i1 }
-
-define void @f() {
-entry:
-        %w = alloca %wrapper, align 8           ; <%wrapper*> [#uses=1]
-        %0 = getelementptr %wrapper* %w, i64 0, i32 0           ; <i1*>
-        store i1 true, i1* %0
-        ret void
-}
diff --git a/test/Transforms/ScalarRepl/2009-06-01-BitcastIntPadding.ll b/test/Transforms/ScalarRepl/2009-06-01-BitcastIntPadding.ll
deleted file mode 100644
index cecbdd4..0000000
--- a/test/Transforms/ScalarRepl/2009-06-01-BitcastIntPadding.ll
+++ /dev/null
@@ -1,17 +0,0 @@
-; RUN: opt < %s -scalarrepl
-; PR4286
-
-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-f80:128:128"
-target triple = "x86_64-undermydesk-freebsd8.0"
-	%struct.singlebool = type <{ i8 }>
-
-define zeroext i8 @doit() nounwind {
-entry:
-	%a = alloca %struct.singlebool, align 1		; <%struct.singlebool*> [#uses=2]
-	%storetmp.i = bitcast %struct.singlebool* %a to i1*		; <i1*> [#uses=1]
-	store i1 true, i1* %storetmp.i
-	%tmp = getelementptr %struct.singlebool* %a, i64 0, i32 0		; <i8*> [#uses=1]
-	%tmp1 = load i8* %tmp		; <i8> [#uses=1]
-	ret i8 %tmp1
-}
-
diff --git a/test/Transforms/ScalarRepl/2009-08-16-VLA.ll b/test/Transforms/ScalarRepl/2009-08-16-VLA.ll
deleted file mode 100644
index d69af11..0000000
--- a/test/Transforms/ScalarRepl/2009-08-16-VLA.ll
+++ /dev/null
@@ -1,23 +0,0 @@
-; RUN: opt < %s -scalarrepl -disable-opt
-
-	%struct.Item = type { [4 x i16], %struct.rule* }
-	%struct.rule = type { [4 x i16], i32, i32, i32, %struct.nonterminal*, %struct.pattern*, i8 }
-	%struct.nonterminal = type { i8*, i32, i32, i32, %struct.plankMap*, %struct.rule* }
-	%struct.plankMap = type { %struct.list*, i32, %struct.stateMap* }
-	%struct.list = type { i8*, %struct.list* }
-	%struct.stateMap = type { i8*, %struct.plank*, i32, i16* }
-	%struct.plank = type { i8*, %struct.list*, i32 }
-	%struct.pattern = type { %struct.nonterminal*, %struct.operator*, [2 x %struct.nonterminal*] }
-	%struct.operator = type { i8*, i8, i32, i32, i32, i32, %struct.table* }
-	%struct.table = type { %struct.operator*, %struct.list*, i16*, [2 x %struct.dimension*], %struct.item_set** }
-	%struct.dimension = type { i16*, %struct.Index_Map, %struct.mapping*, i32, %struct.plankMap* }
-	%struct.Index_Map = type { i32, %struct.item_set** }
-	%struct.item_set = type { i32, i32, %struct.operator*, [2 x %struct.item_set*], %struct.item_set*, i16*, %struct.Item*, %struct.Item* }
-	%struct.mapping = type { %struct.list**, i32, i32, i32, %struct.item_set** }
-
-define void @addHP_2_0() {
-bb4.i:
-	%0 = malloc [0 x %struct.Item]		; <[0 x %struct.Item]*> [#uses=1]
-	%.sub.i.c.i = getelementptr [0 x %struct.Item]* %0, i32 0, i32 0		; <%struct.Item*> [#uses=0]
-	unreachable
-}
diff --git a/test/Transforms/ScalarRepl/basictest.ll b/test/Transforms/ScalarRepl/basictest.ll
index a26b62d..9676873 100644
--- a/test/Transforms/ScalarRepl/basictest.ll
+++ b/test/Transforms/ScalarRepl/basictest.ll
@@ -1,11 +1,30 @@
-; RUN: opt < %s -scalarrepl -mem2reg -S | not grep alloca
+; RUN: opt < %s -scalarrepl -S | FileCheck %s
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
 
-define i32 @test() {
+define i32 @test1() {
 	%X = alloca { i32, float }		; <{ i32, float }*> [#uses=1]
 	%Y = getelementptr { i32, float }* %X, i64 0, i32 0		; <i32*> [#uses=2]
 	store i32 0, i32* %Y
 	%Z = load i32* %Y		; <i32> [#uses=1]
 	ret i32 %Z
+; CHECK: @test1
+; CHECK-NOT: alloca
+; CHECK: ret i32 0
+}
+
+; PR8980
+define i64 @test2(i64 %X) {
+	%A = alloca [8 x i8]
+        %B = bitcast [8 x i8]* %A to i64*
+        
+	store i64 %X, i64* %B
+        br label %L2
+        
+L2:
+	%Z = load i64* %B		; <i32> [#uses=1]
+	ret i64 %Z
+; CHECK: @test2
+; CHECK-NOT: alloca
+; CHECK: ret i64 %X
 }
 
diff --git a/test/Transforms/ScalarRepl/copy-aggregate.ll b/test/Transforms/ScalarRepl/copy-aggregate.ll
index 2992413..997da4b 100644
--- a/test/Transforms/ScalarRepl/copy-aggregate.ll
+++ b/test/Transforms/ScalarRepl/copy-aggregate.ll
@@ -1,9 +1,11 @@
-; RUN: opt < %s -scalarrepl -S | not grep alloca
+; RUN: opt < %s -scalarrepl -S | FileCheck %s
 ; PR3290
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
 
 ;; Store of integer to whole alloca struct.
 define i32 @test1(i64 %V) nounwind {
+; CHECK: test1
+; CHECK-NOT: alloca
 	%X = alloca {{i32, i32}}
 	%Y = bitcast {{i32,i32}}* %X to i64*
 	store i64 %V, i64* %Y
@@ -18,6 +20,8 @@ define i32 @test1(i64 %V) nounwind {
 
 ;; Store of integer to whole struct/array alloca.
 define float @test2(i128 %V) nounwind {
+; CHECK: test2
+; CHECK-NOT: alloca
 	%X = alloca {[4 x float]}
 	%Y = bitcast {[4 x float]}* %X to i128*
 	store i128 %V, i128* %Y
@@ -32,6 +36,8 @@ define float @test2(i128 %V) nounwind {
 
 ;; Load of whole alloca struct as integer
 define i64 @test3(i32 %a, i32 %b) nounwind {
+; CHECK: test3
+; CHECK-NOT: alloca
 	%X = alloca {{i32, i32}}
 
 	%A = getelementptr {{i32,i32}}* %X, i32 0, i32 0, i32 0
@@ -46,6 +52,8 @@ define i64 @test3(i32 %a, i32 %b) nounwind {
 
 ;; load of integer from whole struct/array alloca.
 define i128 @test4(float %a, float %b) nounwind {
+; CHECK: test4
+; CHECK-NOT: alloca
 	%X = alloca {[4 x float]}
 	%A = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 0
 	%B = getelementptr {[4 x float]}* %X, i32 0, i32 0, i32 3
@@ -56,3 +64,45 @@ define i128 @test4(float %a, float %b) nounwind {
 	%V = load i128* %Y
 	ret i128 %V
 }
+
+;; If the elements of a struct or array alloca contain padding, SROA can still
+;; split up the alloca as long as there is no padding between the elements.
+%padded = type { i16, i8 }
+%arr = type [4 x %padded]
+define void @test5(%arr* %p, %arr* %q) {
+entry:
+; CHECK: test5
+; CHECK-NOT: i128
+  %var = alloca %arr, align 4
+  %vari8 = bitcast %arr* %var to i8*
+  %pi8 = bitcast %arr* %p to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %vari8, i8* %pi8, i32 16, i32 4, i1 false)
+  %qi8 = bitcast %arr* %q to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %qi8, i8* %vari8, i32 16, i32 4, i1 false)
+  ret void
+}
+
+;; Check that an array alloca can be split up when it is also accessed with
+;; a load or store as a homogeneous structure with the same element type and
+;; number of elements as the array.
+%homogeneous = type { <8 x i16>, <8 x i16>, <8 x i16> }
+%wrapped_array = type { [3 x <8 x i16>] }
+define void @test6(i8* %p, %wrapped_array* %arr) {
+entry:
+; CHECK: test6
+; CHECK: store <8 x i16>
+; CHECK: store <8 x i16>
+; CHECK: store <8 x i16>
+  %var = alloca %wrapped_array, align 16
+  %res = call %homogeneous @test6callee(i8* %p)
+  %varcast = bitcast %wrapped_array* %var to %homogeneous*
+  store %homogeneous %res, %homogeneous* %varcast
+  %tmp1 = bitcast %wrapped_array* %arr to i8*
+  %tmp2 = bitcast %wrapped_array* %var to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp1, i8* %tmp2, i32 48, i32 16, i1 false)
+  ret void
+}
+
+declare %homogeneous @test6callee(i8* nocapture) nounwind
+
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
diff --git a/test/Transforms/ScalarRepl/crash.ll b/test/Transforms/ScalarRepl/crash.ll
new file mode 100644
index 0000000..7b62f09
--- /dev/null
+++ b/test/Transforms/ScalarRepl/crash.ll
@@ -0,0 +1,260 @@
+; RUN: opt -scalarrepl %s -disable-output
+; RUN: opt -scalarrepl-ssa %s -disable-output
+
+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"
+target triple = "x86_64-apple-darwin10.0.0"
+
+; PR9017
+define void @test1() nounwind readnone ssp {
+entry:
+  %l_72 = alloca i32*, align 8
+  unreachable
+
+for.cond:                                         ; preds = %for.cond
+  %tmp1.i = load i32** %l_72, align 8
+  store i32* %tmp1.i, i32** %l_72, align 8
+  br label %for.cond
+
+if.end:                                           ; No predecessors!
+  ret void
+}
+
+
+define void @test2() {
+  %E = alloca { { i32, float, double, i64 }, { i32, float, double, i64 } }        ; <{ { i32, float, double, i64 }, { i32, float, double, i64 } }*> [#uses=1]
+  %tmp.151 = getelementptr { { i32, float, double, i64 }, { i32, float, double, i64 } }* %E, i64 0, i32 1, i32 3          ; <i64*> [#uses=0]
+  ret void
+}
+
+define i32 @test3() {
+        %X = alloca { [4 x i32] }               ; <{ [4 x i32] }*> [#uses=1]
+        %Y = getelementptr { [4 x i32] }* %X, i64 0, i32 0, i64 2               ; <i32*> [#uses=2]
+        store i32 4, i32* %Y
+        %Z = load i32* %Y               ; <i32> [#uses=1]
+        ret i32 %Z
+}
+
+
+%struct.rtx_def = type { [2 x i8], i32, [1 x %union.rtunion_def] }
+%union.rtunion_def = type { i32 }
+
+define void @test4() {
+entry:
+        %c_addr.i = alloca i8           ; <i8*> [#uses=1]
+        switch i32 0, label %return [
+                 i32 36, label %label.7
+                 i32 34, label %label.7
+                 i32 41, label %label.5
+        ]
+label.5:                ; preds = %entry
+        ret void
+label.7:                ; preds = %entry, %entry
+        br i1 false, label %then.4, label %switchexit.0
+then.4:         ; preds = %label.7
+        %tmp.0.i = bitcast i8* %c_addr.i to i32*                ; <i32*> [#uses=1]
+        store i32 44, i32* %tmp.0.i
+        ret void
+switchexit.0:           ; preds = %label.7
+        ret void
+return:         ; preds = %entry
+        ret void
+}
+
+
+define void @test5() {
+entry:
+        %source_ptr = alloca i8*, align 4               ; <i8**> [#uses=2]
+        br i1 false, label %bb1357, label %cond_next583
+cond_next583:           ; preds = %entry
+        ret void
+bb1357:         ; preds = %entry
+        br i1 false, label %bb1365, label %bb27055
+bb1365:         ; preds = %bb1357
+        switch i32 0, label %cond_next10377 [
+                 i32 0, label %bb4679
+                 i32 1, label %bb4679
+                 i32 2, label %bb4679
+                 i32 3, label %bb4679
+                 i32 4, label %bb5115
+                 i32 5, label %bb6651
+                 i32 6, label %bb7147
+                 i32 7, label %bb8683
+                 i32 8, label %bb9131
+                 i32 9, label %bb9875
+                 i32 10, label %bb4679
+                 i32 11, label %bb4859
+                 i32 12, label %bb4679
+                 i32 16, label %bb10249
+        ]
+bb4679:         ; preds = %bb1365, %bb1365, %bb1365, %bb1365, %bb1365, %bb1365
+        ret void
+bb4859:         ; preds = %bb1365
+        ret void
+bb5115:         ; preds = %bb1365
+        ret void
+bb6651:         ; preds = %bb1365
+        ret void
+bb7147:         ; preds = %bb1365
+        ret void
+bb8683:         ; preds = %bb1365
+        ret void
+bb9131:         ; preds = %bb1365
+        ret void
+bb9875:         ; preds = %bb1365
+        %source_ptr9884 = bitcast i8** %source_ptr to i8**              ; <i8**> [#uses=1]
+        %tmp9885 = load i8** %source_ptr9884            ; <i8*> [#uses=0]
+        ret void
+bb10249:                ; preds = %bb1365
+        %source_ptr10257 = bitcast i8** %source_ptr to i16**            ; <i16**> [#uses=1]
+        %tmp10258 = load i16** %source_ptr10257         ; <i16*> [#uses=0]
+        ret void
+cond_next10377:         ; preds = %bb1365
+        ret void
+bb27055:                ; preds = %bb1357
+        ret void
+}
+
+
+        %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>" = type { %"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block"* }
+        %"struct.__gnu_cxx::bitmap_allocator<char>" = type { i8 }
+        %"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block" = type { [8 x i8] }
+
+; PR1045
+define void @test6() {
+entry:
+        %this_addr.i = alloca %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*                ; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"**> [#uses=3]
+        %tmp = alloca %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>", align 4                ; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
+        store %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp, %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i
+        %tmp.i = load %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i          ; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
+        %tmp.i.upgrd.1 = bitcast %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp.i to %"struct.__gnu_cxx::bitmap_allocator<char>"*              ; <%"struct.__gnu_cxx::bitmap_allocator<char>"*> [#uses=0]
+        %tmp1.i = load %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"** %this_addr.i         ; <%"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"*> [#uses=1]
+        %tmp.i.upgrd.2 = getelementptr %"struct.__gnu_cxx::balloc::_Inclusive_between<__gnu_cxx::bitmap_allocator<char>::_Alloc_block*>"* %tmp1.i, i32 0, i32 0         ; <%"struct.__gnu_cxx::bitmap_allocator<char>::_Alloc_block"**> [#uses=0]
+        unreachable
+}
+
+        %struct.CGPoint = type { float, float }
+        %struct.aal_big_range_t = type { i32, i32 }        %struct.aal_callback_t = type { i8* (i8*, i32)*, void (i8*, i8*)* }        %struct.aal_edge_pool_t = type { %struct.aal_edge_pool_t*, i32, i32, [0 x %struct.aal_edge_t] }        %struct.aal_edge_t = type { %struct.CGPoint, %struct.CGPoint, i32 }
+        %struct.aal_range_t = type { i16, i16 }
+        %struct.aal_span_pool_t = type { %struct.aal_span_pool_t*, [341 x %struct.aal_span_t] }
+        %struct.aal_span_t = type { %struct.aal_span_t*, %struct.aal_big_range_t }
+        %struct.aal_spanarray_t = type { [2 x %struct.aal_range_t] }
+        %struct.aal_spanbucket_t = type { i16, [2 x i8], %struct.anon }
+        %struct.aal_state_t = type { %struct.CGPoint, %struct.CGPoint, %struct.CGPoint, i32, float, float, float, float, %struct.CGPoint, %struct.CGPoint, float, float, float, float, i32, i32, i32, i32, float, float, i8*, i32, i32, %struct.aal_edge_pool_t*, %struct.aal_edge_pool_t*, i8*, %struct.aal_callback_t*, i32, %struct.aal_span_t*, %struct.aal_span_t*, %struct.aal_span_t*, %struct.aal_span_pool_t*, i8, float, i8, i32 }
+        %struct.anon = type { %struct.aal_spanarray_t }
+
+
+declare void @llvm.memcpy.i32(i8*, i8*, i32, i32)
+
+define fastcc void @test7() {
+entry:
+        %SB = alloca %struct.aal_spanbucket_t, align 4          ; <%struct.aal_spanbucket_t*> [#uses=2]
+        br i1 false, label %cond_true, label %cond_next79
+
+cond_true:              ; preds = %entry
+        br i1 false, label %cond_next, label %cond_next114.i
+
+cond_next114.i:         ; preds = %cond_true
+        ret void
+
+cond_next:              ; preds = %cond_true
+        %SB19 = bitcast %struct.aal_spanbucket_t* %SB to i8*            ; <i8*> [#uses=1]
+        call void @llvm.memcpy.i32( i8* %SB19, i8* null, i32 12, i32 0 )
+        br i1 false, label %cond_next34, label %cond_next79
+
+cond_next34:            ; preds = %cond_next
+        %i.2.reload22 = load i32* null          ; <i32> [#uses=1]
+        %tmp51 = getelementptr %struct.aal_spanbucket_t* %SB, i32 0, i32 2, i32 0, i32 0, i32 %i.2.reload22, i32 1      
+        ; <i16*> [#uses=0]
+        ret void
+
+cond_next79:            ; preds = %cond_next, %entry
+        ret void
+}
+
+
+       %struct.c37304a__vrec = type { i8, %struct.c37304a__vrec___disc___XVN }
+        %struct.c37304a__vrec___disc___XVN = type {
+%struct.c37304a__vrec___disc___XVN___O }
+        %struct.c37304a__vrec___disc___XVN___O = type {  }
+
+; PR3304
+define void @test8() {
+entry:
+        %v = alloca %struct.c37304a__vrec
+        %0 = getelementptr %struct.c37304a__vrec* %v, i32 0, i32 0             
+        store i8 8, i8* %0, align 1
+        unreachable
+}
+
+
+
+; rdar://6808691 - ZeroLengthMemSet
+        type <{ i32, i16, i8, i8, i64, i64, i16, [0 x i16] }>           
+
+define i32 @test9() {
+entry:
+        %.compoundliteral = alloca %0           
+        %tmp228 = getelementptr %0* %.compoundliteral, i32 0, i32 7
+        %tmp229 = bitcast [0 x i16]* %tmp228 to i8*             
+        call void @llvm.memset.i64(i8* %tmp229, i8 0, i64 0, i32 2)
+        unreachable
+}
+
+declare void @llvm.memset.i64(i8* nocapture, i8, i64, i32) nounwind
+
+
+; PR4146 - i1 handling
+%wrapper = type { i1 }
+define void @test10() {
+entry:
+        %w = alloca %wrapper, align 8           ; <%wrapper*> [#uses=1]
+        %0 = getelementptr %wrapper* %w, i64 0, i32 0           ; <i1*>
+        store i1 true, i1* %0
+        ret void
+}
+
+
+        %struct.singlebool = type <{ i8 }>
+; PR4286
+define zeroext i8 @test11() nounwind {
+entry:
+        %a = alloca %struct.singlebool, align 1         ; <%struct.singlebool*> [#uses=2]
+        %storetmp.i = bitcast %struct.singlebool* %a to i1*             ; <i1*> [#uses=1]
+        store i1 true, i1* %storetmp.i
+        %tmp = getelementptr %struct.singlebool* %a, i64 0, i32 0               ; <i8*> [#uses=1]
+        %tmp1 = load i8* %tmp           ; <i8> [#uses=1]
+        ret i8 %tmp1
+}
+
+
+       %struct.Item = type { [4 x i16], %struct.rule* }
+        %struct.rule = type { [4 x i16], i32, i32, i32, %struct.nonterminal*, %struct.pattern*, i8 }
+        %struct.nonterminal = type { i8*, i32, i32, i32, %struct.plankMap*, %struct.rule* }
+        %struct.plankMap = type { %struct.list*, i32, %struct.stateMap* }
+        %struct.list = type { i8*, %struct.list* }
+        %struct.stateMap = type { i8*, %struct.plank*, i32, i16* }
+        %struct.plank = type { i8*, %struct.list*, i32 }
+        %struct.pattern = type { %struct.nonterminal*, %struct.operator*, [2 x %struct.nonterminal*] }
+        %struct.operator = type { i8*, i8, i32, i32, i32, i32, %struct.table* }
+        %struct.table = type { %struct.operator*, %struct.list*, i16*, [2 x %struct.dimension*], %struct.item_set** }
+        %struct.dimension = type { i16*, %struct.Index_Map, %struct.mapping*, i32, %struct.plankMap* }
+        %struct.Index_Map = type { i32, %struct.item_set** }
+        %struct.item_set = type { i32, i32, %struct.operator*, [2 x %struct.item_set*], %struct.item_set*, i16*, %struct.Item*, %struct.Item* }
+        %struct.mapping = type { %struct.list**, i32, i32, i32, %struct.item_set** }
+
+; VLAs.
+define void @test12() {
+bb4.i:
+        %0 = malloc [0 x %struct.Item]          ; <[0 x %struct.Item]*> [#uses=1]
+        %.sub.i.c.i = getelementptr [0 x %struct.Item]* %0, i32 0, i32 0                ; <%struct.Item*> [#uses=0]
+        unreachable
+}
+
+; PR8680
+define void @test13() nounwind {
+entry:
+  %memtmp = alloca i32, align 4
+  %0 = bitcast i32* %memtmp to void ()*
+  call void %0() nounwind
+  ret void
+}
diff --git a/test/Transforms/ScalarRepl/memcpy-from-global.ll b/test/Transforms/ScalarRepl/memcpy-from-global.ll
index 8152785..5b25864 100644
--- a/test/Transforms/ScalarRepl/memcpy-from-global.ll
+++ b/test/Transforms/ScalarRepl/memcpy-from-global.ll
@@ -1,12 +1,19 @@
-; RUN: opt < %s -scalarrepl -S | not grep {call.*memcpy}
+; RUN: opt < %s -scalarrepl -S | FileCheck %s
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
 @C.0.1248 = internal constant [128 x float] [ float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00 ], align 32		; <[128 x float]*> [#uses=1]
 
-define float @grad4(i32 %hash, float %x, float %y, float %z, float %w) {
+define float @test1(i32 %hash, float %x, float %y, float %z, float %w) {
 entry:
 	%lookupTable = alloca [128 x float], align 16		; <[128 x float]*> [#uses=5]
 	%lookupTable1 = bitcast [128 x float]* %lookupTable to i8*		; <i8*> [#uses=1]
 	call void @llvm.memcpy.i32( i8* %lookupTable1, i8* bitcast ([128 x float]* @C.0.1248 to i8*), i32 512, i32 16 )
+        
+; CHECK: @test1
+; CHECK-NOT: alloca
+; CHECK-NOT: call{{.*}}@llvm.memcpy
+; CHECK: %lookupTable1 = bitcast [128 x float]* @C.0.1248 to i8*
+; CHECK-NOT: call{{.*}}@llvm.memcpy
+        
 	%tmp3 = shl i32 %hash, 2		; <i32> [#uses=1]
 	%tmp5 = and i32 %tmp3, 124		; <i32> [#uses=4]
 	%tmp753 = getelementptr [128 x float]* %lookupTable, i32 0, i32 %tmp5		; <float*> [#uses=1]
@@ -32,3 +39,58 @@ entry:
 }
 
 declare void @llvm.memcpy.i32(i8*, i8*, i32, i32)
+
+
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+
+%T = type { i8, [123 x i8] }
+
+@G = constant %T {i8 1, [123 x i8] zeroinitializer }
+
+define void @test2() {
+  %A = alloca %T
+  %B = alloca %T
+  %a = bitcast %T* %A to i8*
+  %b = bitcast %T* %B to i8*
+
+; CHECK: @test2
+
+; %A alloca is deleted
+; CHECK-NEXT: %B = alloca %T
+
+; use @G instead of %A
+; CHECK-NEXT: %a = bitcast %T* @G to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* bitcast (%T* @G to i8*), i64 124, i32 4, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %b, i8* %a, i64 124, i32 4, i1 false)
+  call void @bar(i8* %b)
+  ret void
+}
+
+declare void @bar(i8*)
+
+
+;; Should be able to eliminate the alloca.
+define void @test3() {
+  %A = alloca %T
+  %a = bitcast %T* %A to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* bitcast (%T* @G to i8*), i64 124, i32 4, i1 false)
+  call void @bar(i8* %a) readonly
+; CHECK: @test3
+; CHECK-NEXT: %a = bitcast %T* @G to i8*
+; CHECK-NEXT: call void @bar(i8* %a)
+  ret void
+}
+
+define void @test4() {
+  %A = alloca %T
+  %a = bitcast %T* %A to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* bitcast (%T* @G to i8*), i64 124, i32 4, i1 false)
+  call void @baz(i8* byval %a) 
+; CHECK: @test4
+; CHECK-NEXT: %a = bitcast %T* @G to i8*
+; CHECK-NEXT: call void @baz(i8* byval %a)
+  ret void
+}
+
+declare void @baz(i8* byval)
diff --git a/test/Transforms/ScalarRepl/phi-select.ll b/test/Transforms/ScalarRepl/phi-select.ll
new file mode 100644
index 0000000..fa3972d
--- /dev/null
+++ b/test/Transforms/ScalarRepl/phi-select.ll
@@ -0,0 +1,153 @@
+; RUN: opt %s -scalarrepl -S | FileCheck %s
+; Test promotion of allocas that have phis and select users.
+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"
+target triple = "x86_64-apple-darwin10.2"
+
+%struct.X = type { i32 }
+%PairTy = type {i32, i32}
+
+; CHECK: @test1
+; CHECK: %a.0 = alloca i32
+; CHECK: %b.0 = alloca i32
+define i32 @test1(i32 %x) nounwind readnone ssp {
+entry:
+  %a = alloca %struct.X, align 8                  ; <%struct.X*> [#uses=2]
+  %b = alloca %struct.X, align 8                  ; <%struct.X*> [#uses=2]
+  %0 = getelementptr inbounds %struct.X* %a, i64 0, i32 0 ; <i32*> [#uses=1]
+  store i32 1, i32* %0, align 8
+  %1 = getelementptr inbounds %struct.X* %b, i64 0, i32 0 ; <i32*> [#uses=1]
+  store i32 2, i32* %1, align 8
+  %2 = icmp eq i32 %x, 0                          ; <i1> [#uses=1]
+  %p.0 = select i1 %2, %struct.X* %b, %struct.X* %a ; <%struct.X*> [#uses=1]
+  %3 = getelementptr inbounds %struct.X* %p.0, i64 0, i32 0 ; <i32*> [#uses=1]
+  %4 = load i32* %3, align 8                      ; <i32> [#uses=1]
+  ret i32 %4
+}
+
+; CHECK: @test2
+; CHECK: %X.ld = phi i32 [ 1, %entry ], [ 2, %T ]
+; CHECK-NEXT: ret i32 %X.ld
+define i32 @test2(i1 %c) {
+entry:
+  %A = alloca {i32, i32}
+  %B = getelementptr {i32, i32}* %A, i32 0, i32 0
+  store i32 1, i32* %B
+  br i1 %c, label %T, label %F
+T:
+  %C = getelementptr {i32, i32}* %A, i32 0, i32 1
+  store i32 2, i32* %C
+  br label %F
+F:
+  %X = phi i32* [%B, %entry], [%C, %T]
+  %Q = load i32* %X
+  ret i32 %Q
+}
+
+; CHECK: @test3
+; CHECK-NEXT: %Q = select i1 %c, i32 1, i32 2
+; CHECK-NEXT: ret i32 %Q
+; rdar://8904039
+define i32 @test3(i1 %c) {
+  %A = alloca {i32, i32}
+  %B = getelementptr {i32, i32}* %A, i32 0, i32 0
+  store i32 1, i32* %B
+  %C = getelementptr {i32, i32}* %A, i32 0, i32 1
+  store i32 2, i32* %C
+  
+  %X = select i1 %c, i32* %B, i32* %C
+  %Q = load i32* %X
+  ret i32 %Q
+}
+
+;; We can't scalarize this, a use of the select is not an element access.
+define i64 @test4(i1 %c) {
+entry:
+  %A = alloca %PairTy
+  ; CHECK: @test4
+  ; CHECK: %A = alloca %PairTy
+  %B = getelementptr {i32, i32}* %A, i32 0, i32 0
+  store i32 1, i32* %B
+  %C = getelementptr {i32, i32}* %A, i32 0, i32 1
+  store i32 2, i32* %B
+  
+  %X = select i1 %c, i32* %B, i32* %C
+  %Y = bitcast i32* %X to i64*
+  %Q = load i64* %Y
+  ret i64 %Q
+}
+
+
+;;
+;; Tests for promoting allocas used by selects.
+;; rdar://7339113
+;;
+
+define i32 @test5(i32 *%P) nounwind readnone ssp {
+entry:
+  %b = alloca i32, align 8 
+  store i32 2, i32* %b, align 8
+  
+  ;; Select on constant condition should be folded.
+  %p.0 = select i1 false, i32* %b, i32* %P
+  store i32 123, i32* %p.0
+  
+  %r = load i32* %b, align 8
+  ret i32 %r
+  
+; CHECK: @test5
+; CHECK: store i32 123, i32* %P
+; CHECK: ret i32 2
+}
+
+define i32 @test6(i32 %x, i1 %c) nounwind readnone ssp {
+  %a = alloca i32, align 8
+  %b = alloca i32, align 8
+  store i32 1, i32* %a, align 8
+  store i32 2, i32* %b, align 8
+  %p.0 = select i1 %c, i32* %b, i32* %a
+  %r = load i32* %p.0, align 8
+  ret i32 %r
+; CHECK: @test6
+; CHECK-NEXT: %r = select i1 %c, i32 2, i32 1
+; CHECK-NEXT: ret i32 %r
+}
+
+; Verify that the loads happen where the loads are, not where the select is.
+define i32 @test7(i32 %x, i1 %c) nounwind readnone ssp {
+  %a = alloca i32, align 8
+  %b = alloca i32, align 8
+  store i32 1, i32* %a
+  store i32 2, i32* %b
+  %p.0 = select i1 %c, i32* %b, i32* %a
+  
+  store i32 0, i32* %a
+  
+  %r = load i32* %p.0, align 8
+  ret i32 %r
+; CHECK: @test7
+; CHECK-NOT: alloca i32
+; CHECK: %r = select i1 %c, i32 2, i32 0
+; CHECK: ret i32 %r
+}
+
+;; Promote allocs that are PHI'd together by moving the loads.
+define i32 @test8(i32 %x) nounwind readnone ssp {
+; CHECK: @test8
+; CHECK-NOT: load i32
+; CHECK-NOT: store i32
+; CHECK: %p.0.ld = phi i32 [ 2, %entry ], [ 1, %T ]
+; CHECK-NEXT: ret i32 %p.0.ld
+entry:
+  %a = alloca i32, align 8
+  %b = alloca i32, align 8
+  store i32 1, i32* %a, align 8
+  store i32 2, i32* %b, align 8
+  %c = icmp eq i32 %x, 0 
+  br i1 %c, label %T, label %Cont
+T:
+  br label %Cont
+Cont:
+  %p.0 = phi i32* [%b, %entry],[%a, %T]
+  %r = load i32* %p.0, align 8
+  ret i32 %r
+}
diff --git a/test/Transforms/ScalarRepl/vector_promote.ll b/test/Transforms/ScalarRepl/vector_promote.ll
index fe55426..37cb49f 100644
--- a/test/Transforms/ScalarRepl/vector_promote.ll
+++ b/test/Transforms/ScalarRepl/vector_promote.ll
@@ -87,8 +87,6 @@ define i32 @test5(float %X) {  ;; should turn into bitcast.
 }
 
 
-;; should not turn into <1 x i64> - It is a banned MMX datatype.
-;; rdar://8380055
 define i64 @test6(<2 x float> %X) {
 	%X_addr = alloca <2 x float>
         store <2 x float> %X, <2 x float>* %X_addr
@@ -96,7 +94,7 @@ define i64 @test6(<2 x float> %X) {
 	%tmp = load i64* %P
 	ret i64 %tmp
 ; CHECK: @test6
-; CHECK-NEXT: bitcast <2 x float> %X to i64
-; CHECK-NEXT: ret i64
+; CHECK: bitcast <2 x float> %X to <1 x i64>
+; CHECK: ret i64
 }