}
}
-// For a global variable with one store, if the store dominates any loads,
-// those loads will always load the stored value (as opposed to the
-// initializer), even in the presence of recursion.
-static bool forwardStoredOnceStore(
- GlobalVariable *GV, const StoreInst *StoredOnceStore,
- function_ref<DominatorTree &(Function &)> LookupDomTree) {
- const Value *StoredOnceValue = StoredOnceStore->getValueOperand();
- SmallVector<LoadInst *> Loads;
- const Function *F = StoredOnceStore->getFunction();
- for (User *U : GV->users()) {
- if (auto *LI = dyn_cast<LoadInst>(U)) {
- if (LI->getFunction() == F &&
- LI->getType() == StoredOnceValue->getType() && LI->isSimple())
- Loads.push_back(LI);
- }
- }
- // Only compute DT if we have any loads to examine.
- bool MadeChange = false;
- if (!Loads.empty()) {
- auto &DT = LookupDomTree(*const_cast<Function *>(F));
- for (auto *LI : Loads) {
- if (DT.dominates(StoredOnceStore, LI)) {
- LI->replaceAllUsesWith(const_cast<Value *>(StoredOnceValue));
- LI->eraseFromParent();
- MadeChange = true;
- }
- }
- }
- return MadeChange;
-}
-
/// Analyze the specified global variable and optimize
/// it if possible. If we make a change, return true.
static bool
if (optimizeOnceStoredGlobal(GV, StoredOnceValue, DL, GetTLI))
return true;
- // Try to forward the store to any loads.
- if (forwardStoredOnceStore(GV, GS.StoredOnceStore, LookupDomTree))
- return true;
-
// Otherwise, if the global was not a boolean, we can shrink it to be a
// boolean. Skip this optimization for AS that doesn't allow an initializer.
if (SOVConstant && GS.Ordering == AtomicOrdering::NotAtomic &&
; CHECK-NEXT: [[MALLOCCALL:%.*]] = tail call i8* @malloc(i64 4)
; CHECK-NEXT: [[P:%.*]] = bitcast i8* [[MALLOCCALL]] to i32*
; CHECK-NEXT: store i32* [[P]], i32** @G, align 8
-; CHECK-NEXT: store i32 0, i32* [[P]], align 4
+; CHECK-NEXT: [[GV:%.*]] = load i32*, i32** @G, align 8
+; CHECK-NEXT: store i32 0, i32* [[GV]], align 4
; CHECK-NEXT: ret void
;
%malloccall = tail call i8* @malloc(i64 4)
; CHECK-NEXT: [[MALLOCCALL:%.*]] = tail call i8* @malloc(i64 400)
; CHECK-NEXT: [[P:%.*]] = bitcast i8* [[MALLOCCALL]] to i32*
; CHECK-NEXT: store i32* [[P]], i32** @G, align 8
-; CHECK-NEXT: [[GVE:%.*]] = getelementptr i32, i32* [[P]], i32 40
+; CHECK-NEXT: [[GV:%.*]] = load i32*, i32** @G, align 8
+; CHECK-NEXT: [[GVE:%.*]] = getelementptr i32, i32* [[GV]], i32 40
; CHECK-NEXT: store i32 20, i32* [[GVE]], align 4
; CHECK-NEXT: ret void
;
; CHECK-NEXT: [[MALLOCCALL:%.*]] = tail call i8* @malloc(i64 400) #[[ATTR0:[0-9]+]]
; CHECK-NEXT: [[P:%.*]] = bitcast i8* [[MALLOCCALL]] to i32*
; CHECK-NEXT: store i32* [[P]], i32** @G, align 8
-; CHECK-NEXT: [[GVE:%.*]] = getelementptr i32, i32* [[P]], i32 40
+; CHECK-NEXT: [[GV:%.*]] = load i32*, i32** @G, align 8
+; CHECK-NEXT: [[GVE:%.*]] = getelementptr i32, i32* [[GV]], i32 40
; CHECK-NEXT: store i32 20, i32* [[GVE]], align 4
; CHECK-NEXT: ret void
;
@foo = internal global i32 0, align 4
-define void @store() {
+define dso_local i32 @bar() {
entry:
store i32 5, i32* @foo, align 4
- ret void
-}
-
-define i32 @bar() {
-entry:
%0 = load i32, i32* @foo, align 4
ret i32 %0
}
;CHECK: @bar
;CHECK-NEXT: entry:
+;CHECK-NEXT: store i1 true, i1* @foo, align 1, !dbg ![[DbgLocStore:[0-9]+]]
;CHECK-NEXT: %.b = load i1, i1* @foo, align 1, !dbg ![[DbgLocLoadSel:[0-9]+]]
;CHECK-NEXT: %0 = select i1 %.b, i32 5, i32 0, !dbg ![[DbgLocLoadSel]]
;CHECK-NEXT: call void @llvm.dbg.value({{.*}}), !dbg ![[DbgLocLoadSel]]
;CHECK-NEXT: ret i32 %0, !dbg ![[DbgLocRet:[0-9]+]]
-;CHECK: ![[DbgLocLoadSel]] = !DILocation(line: 3,
-;CHECK: ![[DbgLocRet]] = !DILocation(line: 4,
+;CHECK: ![[DbgLocStore]] = !DILocation(line: 1,
+;CHECK: ![[DbgLocLoadSel]] = !DILocation(line: 2,
+;CHECK: ![[DbgLocRet]] = !DILocation(line: 3,
; Negative test for AS(3). Skip shrink global to bool optimization.
; CHECK: @lvar = internal unnamed_addr addrspace(3) global i32 undef
-define void @test_global_var(i1 %i) {
+define void @test_global_var() {
; CHECK-LABEL: @test_global_var(
; CHECK: store volatile i32 10, i32* undef, align 4
;
entry:
- br i1 %i, label %bb1, label %exit
-bb1:
store i32 10, i32* @gvar
br label %exit
exit:
ret void
}
-define void @test_lds_var(i1 %i) {
+define void @test_lds_var() {
; CHECK-LABEL: @test_lds_var(
; CHECK: store i32 10, i32 addrspace(3)* @lvar, align 4
; CHECK: [[LD:%.*]] = load i32, i32 addrspace(3)* @lvar, align 4
; CHECK: store volatile i32 [[LD]], i32* undef, align 4
;
entry:
- br i1 %i, label %bb1, label %exit
-bb1:
store i32 10, i32 addrspace(3)* @lvar
br label %exit
exit:
define i1 @dom_const() {
; CHECK-LABEL: @dom_const(
+; CHECK-NEXT: store i1 true, ptr @g1, align 1
; CHECK-NEXT: call void @b()
-; CHECK-NEXT: ret i1 true
+; CHECK-NEXT: [[R:%.*]] = load i1, ptr @g1, align 1
+; CHECK-NEXT: ret i1 [[R]]
;
store i1 true, ptr @g1
call void @b()
define i32 @dom_arg(i32 %a) {
; CHECK-LABEL: @dom_arg(
+; CHECK-NEXT: store i32 [[A:%.*]], ptr @g2, align 4
; CHECK-NEXT: call void @b()
-; CHECK-NEXT: ret i32 [[A:%.*]]
+; CHECK-NEXT: [[R:%.*]] = load i32, ptr @g2, align 4
+; CHECK-NEXT: ret i32 [[R]]
;
store i32 %a, ptr @g2
call void @b()
; CHECK-LABEL: @dom_multiple_function_loads(
; CHECK-NEXT: store i1 true, ptr @g5, align 1
; CHECK-NEXT: call void @b()
-; CHECK-NEXT: ret i1 true
+; CHECK-NEXT: [[R:%.*]] = load i1, ptr @g5, align 1
+; CHECK-NEXT: ret i1 [[R]]
;
store i1 true, ptr @g5
call void @b()
define i32 @main() {
; CHECK-LABEL: @main(
; CHECK-NEXT: entry:
+; CHECK-NEXT: store i1 true, i1* @a, align 4
; CHECK-NEXT: [[TMP0:%.*]] = load i32*, i32** @e, align 8
; CHECK-NEXT: store i32 0, i32* [[TMP0]], align 4
; CHECK-NEXT: store i32* null, i32** @e, align 8