9 files changed, 384 insertions, 7 deletions
diff --git a/test/Transforms/ObjCARC/apelim.ll b/test/Transforms/ObjCARC/apelim.ll
new file mode 100644
index 0000000..8c7b5b1
--- /dev/null
+++ b/test/Transforms/ObjCARC/apelim.ll
@@ -0,0 +1,53 @@
+; RUN: opt -S -objc-arc-apelim < %s | FileCheck %s
+; rdar://10227311
+
+@llvm.global_ctors = appending global [2 x { i32, void ()* }] [{ i32, void ()* } { i32 65535, void ()* @_GLOBAL__I_x }, { i32, void ()* } { i32 65535, void ()* @_GLOBAL__I_y }]
+
+@x = global i32 0
+
+declare i32 @bar() nounwind
+
+define i32 @foo() nounwind {
+entry:
+  ret i32 5
+}
+
+define internal void @__cxx_global_var_init() {
+entry:
+  %call = call i32 @foo()
+  store i32 %call, i32* @x, align 4
+  ret void
+}
+
+define internal void @__dxx_global_var_init() {
+entry:
+  %call = call i32 @bar()
+  store i32 %call, i32* @x, align 4
+  ret void
+}
+
+; CHECK: define internal void @_GLOBAL__I_x()
+; CHECK-NOT: @objc
+; CHECK: }
+define internal void @_GLOBAL__I_x() {
+entry:
+  %0 = call i8* @objc_autoreleasePoolPush() nounwind
+  call void @__cxx_global_var_init()
+  call void @objc_autoreleasePoolPop(i8* %0) nounwind
+  ret void
+}
+
+; CHECK: define internal void @_GLOBAL__I_y()
+; CHECK: %0 = call i8* @objc_autoreleasePoolPush() nounwind
+; CHECK: call void @objc_autoreleasePoolPop(i8* %0) nounwind
+; CHECK: }
+define internal void @_GLOBAL__I_y() {
+entry:
+  %0 = call i8* @objc_autoreleasePoolPush() nounwind
+  call void @__dxx_global_var_init()
+  call void @objc_autoreleasePoolPop(i8* %0) nounwind
+  ret void
+}
+
+declare i8* @objc_autoreleasePoolPush()
+declare void @objc_autoreleasePoolPop(i8*)
diff --git a/test/Transforms/ObjCARC/basic.ll b/test/Transforms/ObjCARC/basic.ll
index 44c2602..552f4e0 100644
--- a/test/Transforms/ObjCARC/basic.ll
+++ b/test/Transforms/ObjCARC/basic.ll
@@ -786,7 +786,7 @@ C:
 @__block_holder_tmp_1 = external constant %block1
 define void @test23() {
 entry:
-  %0 = call i8* @objc_retainBlock(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
+  %0 = call i8* @objc_retainBlock(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind, !clang.arc.copy_on_escape !0
   call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
   call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
   call void @objc_release(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
@@ -801,13 +801,28 @@ entry:
 ; CHECK: }
 define void @test23b(i8* %p) {
 entry:
-  %0 = call i8* @objc_retainBlock(i8* %p) nounwind
+  %0 = call i8* @objc_retainBlock(i8* %p) nounwind, !clang.arc.copy_on_escape !0
   call void @callee()
   call void @use_pointer(i8* %p)
   call void @objc_release(i8* %p) nounwind
   ret void
 }
 
+; Don't optimize objc_retainBlock, because there's no copy_on_escape metadata.
+
+; CHECK: define void @test23c(
+; CHECK: @objc_retainBlock
+; CHECK: @objc_release
+; CHECK: }
+define void @test23c() {
+entry:
+  %0 = call i8* @objc_retainBlock(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
+  call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
+  call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
+  call void @objc_release(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
+  ret void
+}
+
 ; Any call can decrement a retain count.
 
 ; CHECK: define void @test24(
diff --git a/test/Transforms/ObjCARC/contract-storestrong-ivar.ll b/test/Transforms/ObjCARC/contract-storestrong-ivar.ll
index 4ad78e7..4a9b314 100644
--- a/test/Transforms/ObjCARC/contract-storestrong-ivar.ll
+++ b/test/Transforms/ObjCARC/contract-storestrong-ivar.ll
@@ -1,6 +1,6 @@
 ; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
 
-; CHECK: call void @objc_storeStrong(i8**
+; CHECK: tail call void @objc_storeStrong(i8**
 
 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-darwin11.0.0"
diff --git a/test/Transforms/ObjCARC/contract-storestrong.ll b/test/Transforms/ObjCARC/contract-storestrong.ll
index fda2ff4..4ff0596 100644
--- a/test/Transforms/ObjCARC/contract-storestrong.ll
+++ b/test/Transforms/ObjCARC/contract-storestrong.ll
@@ -9,7 +9,7 @@ declare void @objc_release(i8*)
 
 ; CHECK: define void @test0(
 ; CHECK: entry:
-; CHECK-NEXT: call void @objc_storeStrong(i8** @x, i8* %p) nounwind
+; CHECK-NEXT: tail call void @objc_storeStrong(i8** @x, i8* %p) nounwind
 ; CHECK-NEXT: ret void
 define void @test0(i8* %p) {
 entry:
diff --git a/test/Transforms/ObjCARC/dg.exp b/test/Transforms/ObjCARC/dg.exp
deleted file mode 100644
index f200589..0000000
--- a/test/Transforms/ObjCARC/dg.exp
+++ /dev/null
@@ -1,3 +0,0 @@
-load_lib llvm.exp
-
-RunLLVMTests [lsort [glob -nocomplain $srcdir/$subdir/*.{ll,c,cpp}]]
diff --git a/test/Transforms/ObjCARC/lit.local.cfg b/test/Transforms/ObjCARC/lit.local.cfg
new file mode 100644
index 0000000..19eebc0
--- /dev/null
+++ b/test/Transforms/ObjCARC/lit.local.cfg
@@ -0,0 +1 @@
+config.suffixes = ['.ll', '.c', '.cpp']
diff --git a/test/Transforms/ObjCARC/no-objc-arc-exceptions.ll b/test/Transforms/ObjCARC/no-objc-arc-exceptions.ll
new file mode 100644
index 0000000..9728f6e
--- /dev/null
+++ b/test/Transforms/ObjCARC/no-objc-arc-exceptions.ll
@@ -0,0 +1,122 @@
+; RUN: opt -S -objc-arc < %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"
+%struct.__block_byref_x = type { i8*, %struct.__block_byref_x*, i32, i32, i32 }
+%struct.__block_descriptor = type { i64, i64 }
+@_NSConcreteStackBlock = external global i8*
+@__block_descriptor_tmp = external hidden constant { i64, i64, i8*, i8*, i8*, i8* }
+
+; The optimizer should make use of the !clang.arc.no_objc_arc_exceptions
+; metadata and eliminate the retainBlock+release pair here.
+; rdar://10803830.
+
+; CHECK: define void @test0(
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test0() {
+entry:
+  %x = alloca %struct.__block_byref_x, align 8
+  %block = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>, align 8
+  %byref.isa = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 0
+  store i8* null, i8** %byref.isa, align 8
+  %byref.forwarding = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 1
+  store %struct.__block_byref_x* %x, %struct.__block_byref_x** %byref.forwarding, align 8
+  %byref.flags = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 2
+  store i32 0, i32* %byref.flags, align 8
+  %byref.size = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 3
+  store i32 32, i32* %byref.size, align 4
+  %block.isa = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 0
+  store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** %block.isa, align 8
+  %block.flags = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 1
+  store i32 1107296256, i32* %block.flags, align 8
+  %block.reserved = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 2
+  store i32 0, i32* %block.reserved, align 4
+  %block.invoke = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 3
+  store i8* bitcast (void (i8*)* @__foo_block_invoke_0 to i8*), i8** %block.invoke, align 8
+  %block.descriptor = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 4
+  store %struct.__block_descriptor* bitcast ({ i64, i64, i8*, i8*, i8*, i8* }* @__block_descriptor_tmp to %struct.__block_descriptor*), %struct.__block_descriptor** %block.descriptor, align 8
+  %block.captured = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 5
+  %t1 = bitcast %struct.__block_byref_x* %x to i8*
+  store i8* %t1, i8** %block.captured, align 8
+  %t2 = bitcast <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block to i8*
+  %t3 = call i8* @objc_retainBlock(i8* %t2) nounwind, !clang.arc.copy_on_escape !4
+  %t4 = getelementptr inbounds i8* %t3, i64 16
+  %t5 = bitcast i8* %t4 to i8**
+  %t6 = load i8** %t5, align 8
+  %t7 = bitcast i8* %t6 to void (i8*)*
+  invoke void %t7(i8* %t3)
+          to label %invoke.cont unwind label %lpad, !clang.arc.no_objc_arc_exceptions !4
+
+invoke.cont:                                      ; preds = %entry
+  call void @objc_release(i8* %t3) nounwind, !clang.imprecise_release !4
+  call void @_Block_object_dispose(i8* %t1, i32 8)
+  ret void
+
+lpad:                                             ; preds = %entry
+  %t8 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__objc_personality_v0 to i8*)
+          cleanup
+  call void @_Block_object_dispose(i8* %t1, i32 8)
+  resume { i8*, i32 } %t8
+}
+
+; There is no !clang.arc.no_objc_arc_exceptions
+; metadata here, so the optimizer shouldn't eliminate anything.
+
+; CHECK: define void @test0_no_metadata(
+; CHECK: call i8* @objc_retainBlock(
+; CHECK: invoke
+; CHECK: call void @objc_release(
+; CHECK: }
+define void @test0_no_metadata() {
+entry:
+  %x = alloca %struct.__block_byref_x, align 8
+  %block = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>, align 8
+  %byref.isa = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 0
+  store i8* null, i8** %byref.isa, align 8
+  %byref.forwarding = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 1
+  store %struct.__block_byref_x* %x, %struct.__block_byref_x** %byref.forwarding, align 8
+  %byref.flags = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 2
+  store i32 0, i32* %byref.flags, align 8
+  %byref.size = getelementptr inbounds %struct.__block_byref_x* %x, i64 0, i32 3
+  store i32 32, i32* %byref.size, align 4
+  %block.isa = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 0
+  store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** %block.isa, align 8
+  %block.flags = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 1
+  store i32 1107296256, i32* %block.flags, align 8
+  %block.reserved = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 2
+  store i32 0, i32* %block.reserved, align 4
+  %block.invoke = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 3
+  store i8* bitcast (void (i8*)* @__foo_block_invoke_0 to i8*), i8** %block.invoke, align 8
+  %block.descriptor = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 4
+  store %struct.__block_descriptor* bitcast ({ i64, i64, i8*, i8*, i8*, i8* }* @__block_descriptor_tmp to %struct.__block_descriptor*), %struct.__block_descriptor** %block.descriptor, align 8
+  %block.captured = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block, i64 0, i32 5
+  %t1 = bitcast %struct.__block_byref_x* %x to i8*
+  store i8* %t1, i8** %block.captured, align 8
+  %t2 = bitcast <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>* %block to i8*
+  %t3 = call i8* @objc_retainBlock(i8* %t2) nounwind, !clang.arc.copy_on_escape !4
+  %t4 = getelementptr inbounds i8* %t3, i64 16
+  %t5 = bitcast i8* %t4 to i8**
+  %t6 = load i8** %t5, align 8
+  %t7 = bitcast i8* %t6 to void (i8*)*
+  invoke void %t7(i8* %t3)
+          to label %invoke.cont unwind label %lpad
+
+invoke.cont:                                      ; preds = %entry
+  call void @objc_release(i8* %t3) nounwind, !clang.imprecise_release !4
+  call void @_Block_object_dispose(i8* %t1, i32 8)
+  ret void
+
+lpad:                                             ; preds = %entry
+  %t8 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__objc_personality_v0 to i8*)
+          cleanup
+  call void @_Block_object_dispose(i8* %t1, i32 8)
+  resume { i8*, i32 } %t8
+}
+
+declare i8* @objc_retainBlock(i8*)
+declare void @objc_release(i8*)
+declare void @_Block_object_dispose(i8*, i32)
+declare i32 @__objc_personality_v0(...)
+declare void @__foo_block_invoke_0(i8* nocapture) uwtable ssp
+
+!4 = metadata !{}
diff --git a/test/Transforms/ObjCARC/retain-block-load.ll b/test/Transforms/ObjCARC/retain-block-load.ll
new file mode 100644
index 0000000..a5170e3
--- /dev/null
+++ b/test/Transforms/ObjCARC/retain-block-load.ll
@@ -0,0 +1,51 @@
+; RUN: opt -objc-arc -S < %s | FileCheck %s
+
+; rdar://10803830
+; The optimizer should be able to prove that the block does not
+; "escape", so the retainBlock+release pair can be eliminated.
+
+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"
+
+%struct.__block_descriptor = type { i64, i64 }
+
+@_NSConcreteStackBlock = external global i8*
+@__block_descriptor_tmp = external global { i64, i64, i8*, i8* }
+
+; CHECK: define void @test() {
+; CHECK-NOT: @objc
+; CHECK: declare i8* @objc_retainBlock(i8*)
+; CHECK: declare void @objc_release(i8*)
+
+define void @test() {
+entry:
+  %block = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, align 8
+  %block.isa = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 0
+  store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** %block.isa, align 8
+  %block.flags = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 1
+  store i32 1073741824, i32* %block.flags, align 8
+  %block.reserved = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 2
+  store i32 0, i32* %block.reserved, align 4
+  %block.invoke = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 3
+  store i8* bitcast (i32 (i8*)* @__test_block_invoke_0 to i8*), i8** %block.invoke, align 8
+  %block.descriptor = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 4
+  store %struct.__block_descriptor* bitcast ({ i64, i64, i8*, i8* }* @__block_descriptor_tmp to %struct.__block_descriptor*), %struct.__block_descriptor** %block.descriptor, align 8
+  %block.captured = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block, i64 0, i32 5
+  store i32 4, i32* %block.captured, align 8
+  %tmp = bitcast <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* %block to i8*
+  %tmp1 = call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  %tmp2 = getelementptr inbounds i8* %tmp1, i64 16
+  %tmp3 = bitcast i8* %tmp2 to i8**
+  %tmp4 = load i8** %tmp3, align 8
+  %tmp5 = bitcast i8* %tmp4 to i32 (i8*)*
+  %call = call i32 %tmp5(i8* %tmp1)
+  call void @objc_release(i8* %tmp1) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+declare i32 @__test_block_invoke_0(i8* nocapture %.block_descriptor) nounwind readonly
+
+declare i8* @objc_retainBlock(i8*)
+
+declare void @objc_release(i8*)
+
+!0 = metadata !{}
diff --git a/test/Transforms/ObjCARC/retain-block.ll b/test/Transforms/ObjCARC/retain-block.ll
new file mode 100644
index 0000000..b3b62d3
--- /dev/null
+++ b/test/Transforms/ObjCARC/retain-block.ll
@@ -0,0 +1,138 @@
+; RUN: opt -objc-arc -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+!0 = metadata !{}
+
+declare i8* @objc_retain(i8*)
+declare void @callee(i8)
+declare void @use_pointer(i8*)
+declare void @objc_release(i8*)
+declare i8* @objc_retainBlock(i8*)
+declare i8* @objc_autorelease(i8*)
+
+; Basic retainBlock+release elimination.
+
+; CHECK: define void @test0(i8* %tmp) {
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test0(i8* %tmp) {
+entry:
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Same as test0, but there's no copy_on_escape metadata, so there's no
+; optimization possible.
+
+; CHECK: define void @test0_no_metadata(i8* %tmp) {
+; CHECK: %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind
+; CHECK: tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+; CHECK: }
+define void @test0_no_metadata(i8* %tmp) {
+entry:
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Same as test0, but the pointer escapes, so there's no
+; optimization possible.
+
+; CHECK: define void @test0_escape(i8* %tmp, i8** %z) {
+; CHECK: %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+; CHECK: tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+; CHECK: }
+define void @test0_escape(i8* %tmp, i8** %z) {
+entry:
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  store i8* %tmp2, i8** %z
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Same as test0_escape, but there's no intervening call.
+
+; CHECK: define void @test0_just_escape(i8* %tmp, i8** %z) {
+; CHECK: %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+; CHECK: tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+; CHECK: }
+define void @test0_just_escape(i8* %tmp, i8** %z) {
+entry:
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  store i8* %tmp2, i8** %z
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Basic nested retainBlock+release elimination.
+
+; CHECK: define void @test1(i8* %tmp) {
+; CHECK-NOT: @objc
+; CHECK: tail call i8* @objc_retain(i8* %tmp) nounwind
+; CHECK-NOT: @objc
+; CHECK: tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test1(i8* %tmp) {
+entry:
+  %tmp1 = tail call i8* @objc_retain(i8* %tmp) nounwind
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Same as test1, but there's no copy_on_escape metadata, so there's no
+; retainBlock+release optimization possible. But we can still eliminate
+; the outer retain+release.
+
+; CHECK: define void @test1_no_metadata(i8* %tmp) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: tail call i8* @objc_retainBlock(i8* %tmp) nounwind
+; CHECK-NEXT: @use_pointer(i8* %tmp2)
+; CHECK-NEXT: @use_pointer(i8* %tmp2)
+; CHECK-NEXT: tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test1_no_metadata(i8* %tmp) {
+entry:
+  %tmp1 = tail call i8* @objc_retain(i8* %tmp) nounwind
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; Same as test1, but the pointer escapes, so there's no
+; retainBlock+release optimization possible. But we can still eliminate
+; the outer retain+release
+
+; CHECK: define void @test1_escape(i8* %tmp, i8** %z) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+; CHECK-NEXT: store i8* %tmp2, i8** %z
+; CHECK-NEXT: @use_pointer(i8* %tmp2)
+; CHECK-NEXT: @use_pointer(i8* %tmp2)
+; CHECK-NEXT: tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test1_escape(i8* %tmp, i8** %z) {
+entry:
+  %tmp1 = tail call i8* @objc_retain(i8* %tmp) nounwind
+  %tmp2 = tail call i8* @objc_retainBlock(i8* %tmp) nounwind, !clang.arc.copy_on_escape !0
+  store i8* %tmp2, i8** %z
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @use_pointer(i8* %tmp2)
+  tail call void @objc_release(i8* %tmp2) nounwind, !clang.imprecise_release !0
+  tail call void @objc_release(i8* %tmp) nounwind, !clang.imprecise_release !0
+  ret void
+}