[test] Precommit test about hoisting invariant loads from globals

And run update_test_checks.py on hoisting.ll.

diff --git a/llvm/test/Transforms/LICM/hoisting.ll b/llvm/test/Transforms/LICM/hoisting.ll
index 88c5b04..c77b1d6 100644 (file)
--- a/llvm/test/Transforms/LICM/hoisting.ll
+++ b/llvm/test/Transforms/LICM/hoisting.ll
@@ -1,5 +1,5 @@
-; RUN: opt < %s -aa-pipeline=basic-aa -passes='require<opt-remark-emit>,loop-mssa(licm)' -S | FileCheck %s
-; RUN: opt < %s -licm -verify-memoryssa -S | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -aa-pipeline=basic-aa -passes='loop-mssa(licm)' -S | FileCheck %s
 
 @X = global i32 0              ; <i32*> [#uses=1]
 
@@ -7,28 +7,42 @@ declare void @foo()
 
 declare i32 @llvm.bitreverse.i32(i32)
 
-; This testcase tests for a problem where LICM hoists 
+; This testcase tests for a problem where LICM hoists
 ; potentially trapping instructions when they are not guaranteed to execute.
 define i32 @test1(i1 %c) {
 ; CHECK-LABEL: @test1(
-       %A = load i32, i32* @X          ; <i32> [#uses=2]
-       br label %Loop
+; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @X, align 4
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       Loop:
+; CHECK-NEXT:    call void @foo()
+; CHECK-NEXT:    br i1 [[C:%.*]], label [[LOOPTAIL:%.*]], label [[IFUNEQUAL:%.*]]
+; CHECK:       IfUnEqual:
+; CHECK-NEXT:    [[B1:%.*]] = sdiv i32 4, [[A]]
+; CHECK-NEXT:    br label [[LOOPTAIL]]
+; CHECK:       LoopTail:
+; CHECK-NEXT:    [[B:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[B1]], [[IFUNEQUAL]] ]
+; CHECK-NEXT:    br i1 [[C]], label [[LOOP]], label [[OUT:%.*]]
+; CHECK:       Out:
+; CHECK-NEXT:    [[B_LCSSA:%.*]] = phi i32 [ [[B]], [[LOOPTAIL]] ]
+; CHECK-NEXT:    [[C:%.*]] = sub i32 [[A]], [[B_LCSSA]]
+; CHECK-NEXT:    ret i32 [[C]]
+;
+  %A = load i32, i32* @X               ; <i32> [#uses=2]
+  br label %Loop
 Loop:          ; preds = %LoopTail, %0
-       call void @foo( )
-       br i1 %c, label %LoopTail, label %IfUnEqual
-        
+  call void @foo( )
+  br i1 %c, label %LoopTail, label %IfUnEqual
+
 IfUnEqual:             ; preds = %Loop
-; CHECK: IfUnEqual:
-; CHECK-NEXT: sdiv i32 4, %A
-       %B1 = sdiv i32 4, %A            ; <i32> [#uses=1]
-       br label %LoopTail
-        
+  %B1 = sdiv i32 4, %A         ; <i32> [#uses=1]
+  br label %LoopTail
+
 LoopTail:              ; preds = %IfUnEqual, %Loop
-       %B = phi i32 [ 0, %Loop ], [ %B1, %IfUnEqual ]          ; <i32> [#uses=1]
-       br i1 %c, label %Loop, label %Out
+  %B = phi i32 [ 0, %Loop ], [ %B1, %IfUnEqual ]               ; <i32> [#uses=1]
+  br i1 %c, label %Loop, label %Out
 Out:           ; preds = %LoopTail
-       %C = sub i32 %A, %B             ; <i32> [#uses=1]
-       ret i32 %C
+  %C = sub i32 %A, %B          ; <i32> [#uses=1]
+  ret i32 %C
 }
 
 
@@ -38,8 +52,19 @@ declare void @foo2(i32) nounwind
 ;; It is ok and desirable to hoist this potentially trapping instruction.
 define i32 @test2(i1 %c) {
 ; CHECK-LABEL: @test2(
-; CHECK-NEXT: load i32, i32* @X
-; CHECK-NEXT: %B = sdiv i32 4, %A
+; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @X, align 4
+; CHECK-NEXT:    [[B:%.*]] = sdiv i32 4, [[A]]
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       Loop:
+; CHECK-NEXT:    br label [[LOOP2:%.*]]
+; CHECK:       loop2:
+; CHECK-NEXT:    call void @foo2(i32 [[B]])
+; CHECK-NEXT:    br i1 [[C:%.*]], label [[LOOP]], label [[OUT:%.*]]
+; CHECK:       Out:
+; CHECK-NEXT:    [[B_LCSSA:%.*]] = phi i32 [ [[B]], [[LOOP2]] ]
+; CHECK-NEXT:    [[C:%.*]] = sub i32 [[A]], [[B_LCSSA]]
+; CHECK-NEXT:    ret i32 [[C]]
+;
   %A = load i32, i32* @X
   br label %Loop
 
@@ -60,24 +85,45 @@ Out:
 
 ; This loop invariant instruction should be constant folded, not hoisted.
 define i32 @test3(i1 %c) {
-; CHECK-LABEL: define i32 @test3(
-; CHECK: call void @foo2(i32 6)
-       %A = load i32, i32* @X          ; <i32> [#uses=2]
-       br label %Loop
+; CHECK-LABEL: @test3(
+; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @X, align 4
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       Loop:
+; CHECK-NEXT:    call void @foo2(i32 6)
+; CHECK-NEXT:    br i1 [[C:%.*]], label [[LOOP]], label [[OUT:%.*]]
+; CHECK:       Out:
+; CHECK-NEXT:    [[B_LCSSA:%.*]] = phi i32 [ 6, [[LOOP]] ]
+; CHECK-NEXT:    [[C:%.*]] = sub i32 [[A]], [[B_LCSSA]]
+; CHECK-NEXT:    ret i32 [[C]]
+;
+  %A = load i32, i32* @X               ; <i32> [#uses=2]
+  br label %Loop
 Loop:
-       %B = add i32 4, 2               ; <i32> [#uses=2]
-       call void @foo2( i32 %B )
-       br i1 %c, label %Loop, label %Out
+  %B = add i32 4, 2            ; <i32> [#uses=2]
+  call void @foo2( i32 %B )
+  br i1 %c, label %Loop, label %Out
 Out:           ; preds = %Loop
-       %C = sub i32 %A, %B             ; <i32> [#uses=1]
-       ret i32 %C
+  %C = sub i32 %A, %B          ; <i32> [#uses=1]
+  ret i32 %C
 }
 
-; CHECK-LABEL: @test4(
-; CHECK: call
-; CHECK: sdiv
-; CHECK: ret
 define i32 @test4(i32 %x, i32 %y) nounwind uwtable ssp {
+; CHECK-LABEL: @test4(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[I_02:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[N_01:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    call void @foo_may_call_exit(i32 0)
+; CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT:    [[ADD]] = add nsw i32 [[N_01]], [[DIV]]
+; CHECK-NEXT:    [[INC]] = add nsw i32 [[I_02]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[INC]], 10000
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
+; CHECK:       for.end:
+; CHECK-NEXT:    [[N_0_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_BODY]] ]
+; CHECK-NEXT:    ret i32 [[N_0_LCSSA]]
+;
 entry:
   br label %for.body
 
@@ -99,13 +145,23 @@ for.end:                                          ; preds = %for.body
 declare void @foo_may_call_exit(i32)
 
 ; PR14854
-; CHECK-LABEL: @test5(
-; CHECK: extractvalue
-; CHECK: br label %tailrecurse
-; CHECK: tailrecurse:
-; CHECK: ifend:
-; CHECK: insertvalue
 define { i32*, i32 } @test5(i32 %i, { i32*, i32 } %e) {
+; CHECK-LABEL: @test5(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[OUT:%.*]] = extractvalue { i32*, i32 } [[E:%.*]], 1
+; CHECK-NEXT:    br label [[TAILRECURSE:%.*]]
+; CHECK:       tailrecurse:
+; CHECK-NEXT:    [[I_TR:%.*]] = phi i32 [ [[I:%.*]], [[ENTRY:%.*]] ], [ [[CMP2:%.*]], [[THEN:%.*]] ]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[OUT]], [[I_TR]]
+; CHECK-NEXT:    br i1 [[CMP1]], label [[THEN]], label [[IFEND:%.*]]
+; CHECK:       then:
+; CHECK-NEXT:    call void @foo()
+; CHECK-NEXT:    [[CMP2]] = add i32 [[I_TR]], 1
+; CHECK-NEXT:    br label [[TAILRECURSE]]
+; CHECK:       ifend:
+; CHECK-NEXT:    [[D_LE:%.*]] = insertvalue { i32*, i32 } [[E]], i32* null, 0
+; CHECK-NEXT:    ret { i32*, i32 } [[D_LE]]
+;
 entry:
   br label %tailrecurse
 
@@ -125,10 +181,21 @@ ifend:                                            ; preds = %tailrecurse
   ret { i32*, i32 } %d
 }
 
-; CHECK: define void @test6(float %f)
-; CHECK: fneg
-; CHECK: br label %for.body
 define void @test6(float %f) #2 {
+; CHECK-LABEL: @test6(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[NEG:%.*]] = fneg float [[F:%.*]]
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    call void @foo_may_call_exit(i32 0)
+; CHECK-NEXT:    call void @use(float [[NEG]])
+; CHECK-NEXT:    [[INC]] = add nsw i32 [[I]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[INC]], 10000
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
+; CHECK:       for.end:
+; CHECK-NEXT:    ret void
+;
 entry:
   br label %for.body
 
@@ -147,10 +214,25 @@ for.end:                                          ; preds = %for.body
 
 declare void @use(float)
 
-; CHECK: define i32 @hoist_bitreverse(i32 %0)
-; CHECK: bitreverse
-; CHECK: br label %header
 define i32 @hoist_bitreverse(i32 %0)  {
+; CHECK-LABEL: @hoist_bitreverse(
+; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.bitreverse.i32(i32 [[TMP0:%.*]])
+; CHECK-NEXT:    br label [[HEADER:%.*]]
+; CHECK:       header:
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[TMP1:%.*]] ], [ [[TMP5:%.*]], [[LATCH:%.*]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = phi i32 [ 0, [[TMP1]] ], [ [[TMP6:%.*]], [[LATCH]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp slt i32 [[TMP3]], 1024
+; CHECK-NEXT:    br i1 [[TMP4]], label [[BODY:%.*]], label [[RETURN:%.*]]
+; CHECK:       body:
+; CHECK-NEXT:    [[TMP5]] = add i32 [[SUM]], [[TMP2]]
+; CHECK-NEXT:    br label [[LATCH]]
+; CHECK:       latch:
+; CHECK-NEXT:    [[TMP6]] = add nsw i32 [[TMP3]], 1
+; CHECK-NEXT:    br label [[HEADER]]
+; CHECK:       return:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[HEADER]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
   br label %header
 
 header:
@@ -175,9 +257,14 @@ return:
 ; Can neither sink nor hoist
 define i32 @test_volatile(i1 %c) {
 ; CHECK-LABEL: @test_volatile(
-; CHECK-LABEL: Loop:
-; CHECK: load volatile i32, i32* @X
-; CHECK-LABEL: Out:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       Loop:
+; CHECK-NEXT:    [[A:%.*]] = load volatile i32, i32* @X, align 4
+; CHECK-NEXT:    br i1 [[C:%.*]], label [[LOOP]], label [[OUT:%.*]]
+; CHECK:       Out:
+; CHECK-NEXT:    [[A_LCSSA:%.*]] = phi i32 [ [[A]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[A_LCSSA]]
+;
   br label %Loop
 
 Loop:
@@ -195,12 +282,31 @@ declare void @escaping.invariant.start({}*) nounwind
 ; invariant.start dominates the load, and in this scope, the
 ; load is invariant. So, we can hoist the `addrld` load out of the loop.
 define i32 @test_fence(i8* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence
-; CHECK-LABEL: entry
-; CHECK: invariant.start
-; CHECK: %addrld = load atomic i32, i32* %addr.i unordered, align 8
-; CHECK: br label %loop
-entry: 
+; CHECK-LABEL: @test_fence(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, i8* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[ADDR_I:%.*]] = bitcast i8* [[GEP]] to i32*
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 4, i8* [[GEP]])
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
+entry:
   %gep = getelementptr inbounds i8, i8* %addr, i64 8
   %addr.i = bitcast i8* %gep to i32 *
   store atomic i32 5, i32 * %addr.i unordered, align 8
@@ -230,11 +336,31 @@ loopexit:
 ; Same as test above, but the load is no longer invariant (presence of
 ; invariant.end). We cannot hoist the addrld out of loop.
 define i32 @test_fence1(i8* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence1
-; CHECK-LABEL: entry
-; CHECK: invariant.start
-; CHECK-NEXT: invariant.end
-; CHECK-NEXT: br label %loop
+; CHECK-LABEL: @test_fence1(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, i8* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[ADDR_I:%.*]] = bitcast i8* [[GEP]] to i32*
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 4, i8* [[GEP]])
+; CHECK-NEXT:    call void @llvm.invariant.end.p0i8({}* [[INVST]], i64 4, i8* [[GEP]])
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
 entry:
   %gep = getelementptr inbounds i8, i8* %addr, i64 8
   %addr.i = bitcast i8* %gep to i32 *
@@ -264,10 +390,31 @@ loopexit:
 ; same as test above, but instead of invariant.end, we have the result of
 ; invariant.start escaping through a call. We cannot hoist the load.
 define i32 @test_fence2(i8* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence2
-; CHECK-LABEL: entry
-; CHECK-NOT: load
-; CHECK: br label %loop
+; CHECK-LABEL: @test_fence2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, i8* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[ADDR_I:%.*]] = bitcast i8* [[GEP]] to i32*
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 4, i8* [[GEP]])
+; CHECK-NEXT:    call void @escaping.invariant.start({}* [[INVST]])
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
 entry:
   %gep = getelementptr inbounds i8, i8* %addr, i64 8
   %addr.i = bitcast i8* %gep to i32 *
@@ -299,12 +446,31 @@ loopexit:
 ; Consider the loadoperand addr.i bitcasted before being passed to
 ; invariant.start
 define i32 @test_fence3(i32* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence3
-; CHECK-LABEL: entry
-; CHECK: invariant.start
-; CHECK-NOT: %addrld = load atomic i32, i32* %addr.i unordered, align 8
-; CHECK: br label %loop
-entry: 
+; CHECK-LABEL: @test_fence3(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[ADDR_I:%.*]] = getelementptr inbounds i32, i32* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[GEP:%.*]] = bitcast i32* [[ADDR_I]] to i8*
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 4, i8* [[GEP]])
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
+entry:
   %addr.i = getelementptr inbounds i32, i32* %addr, i64 8
   %gep = bitcast i32* %addr.i to i8 *
   store atomic i32 5, i32 * %addr.i unordered, align 8
@@ -331,11 +497,31 @@ loopexit:
 
 ; We should not hoist the addrld out of the loop.
 define i32 @test_fence4(i32* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence4
-; CHECK-LABEL: entry
-; CHECK-NOT: %addrld = load atomic i32, i32* %addr.i unordered, align 8
-; CHECK: br label %loop
-entry: 
+; CHECK-LABEL: @test_fence4(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[ADDR_I:%.*]] = getelementptr inbounds i32, i32* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[GEP:%.*]] = bitcast i32* [[ADDR_I]] to i8*
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 4, i8* [[GEP]])
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
+entry:
   %addr.i = getelementptr inbounds i32, i32* %addr, i64 8
   %gep = bitcast i32* %addr.i to i8 *
   br label %loop
@@ -363,11 +549,30 @@ loopexit:
 ; We can't hoist the invariant load out of the loop because
 ; the marker is given a variable size (-1).
 define i32 @test_fence5(i8* %addr, i32 %n, i8* %volatile) {
-; CHECK-LABEL: @test_fence5
-; CHECK-LABEL: entry
-; CHECK: invariant.start
-; CHECK-NOT: %addrld = load atomic i32, i32* %addr.i unordered, align 8
-; CHECK: br label %loop
+; CHECK-LABEL: @test_fence5(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, i8* [[ADDR:%.*]], i64 8
+; CHECK-NEXT:    [[ADDR_I:%.*]] = bitcast i8* [[GEP]] to i32*
+; CHECK-NEXT:    store atomic i32 5, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    fence release
+; CHECK-NEXT:    [[INVST:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 -1, i8* [[GEP]])
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[INDVAR:%.*]] = phi i32 [ [[INDVAR_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[SUM_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT:    [[VOLLOAD:%.*]] = load atomic i8, i8* [[VOLATILE:%.*]] unordered, align 8
+; CHECK-NEXT:    fence acquire
+; CHECK-NEXT:    [[VOLCHK:%.*]] = icmp eq i8 [[VOLLOAD]], 0
+; CHECK-NEXT:    [[ADDRLD:%.*]] = load atomic i32, i32* [[ADDR_I]] unordered, align 8
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[VOLCHK]], i32 0, i32 [[ADDRLD]]
+; CHECK-NEXT:    [[SUM_NEXT]] = add i32 [[SEL]], [[SUM]]
+; CHECK-NEXT:    [[INDVAR_NEXT]] = add i32 [[INDVAR]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[INDVAR_NEXT]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
 entry:
   %gep = getelementptr inbounds i8, i8* %addr, i64 8
   %addr.i = bitcast i8* %gep to i32 *
@@ -392,3 +597,32 @@ loop:
 loopexit:
   ret i32 %sum
 }
+
+declare void @g(i1)
+
+@a = external global i8
+
+; FIXME: Don't hoist invariant loads of globals.
+define void @test_fence6() {
+; CHECK-LABEL: @test_fence6(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* @a)
+; CHECK-NEXT:    [[TMP0:%.*]] = load i8, i8* @a, align 1
+; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[TMP0]], 0
+; CHECK-NEXT:    [[T:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:    br label [[F:%.*]]
+; CHECK:       f:
+; CHECK-NEXT:    tail call void @g(i1 [[T]])
+; CHECK-NEXT:    br label [[F]]
+;
+entry:
+  %i = call {}* @llvm.invariant.start.p0i8(i64 1, i8* @a)
+  br label %f
+
+f:
+  %0 = load i8, i8* @a
+  %1 = and i8 %0, 0
+  %t = icmp eq i8 %1, 0
+  tail call void @g(i1 %t)
+  br label %f
+}