[GlobalOpt] Perform store->dominated load forwarding for stored once globals

Compile time tracker:
https://llvm-compile-time-tracker.com/compare.php?from=1e556f459b44dd0ca4073e932f66ecb6f40fe31a&to=6d7bed4e1e72c6a8592748626091274209740a40&stat=instructions

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D128128

diff --git a/llvm/lib/Transforms/IPO/GlobalOpt.cpp b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
index a2b4716..6ea2242 100644 (file)
--- a/llvm/lib/Transforms/IPO/GlobalOpt.cpp
+++ b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
@@ -1438,6 +1438,37 @@ static void makeAllConstantUsesInstructions(Constant *C) {
   }
 }
 
+// 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
@@ -1597,6 +1628,10 @@ processInternalGlobal(GlobalVariable *GV, const GlobalStatus &GS,
     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 &&

diff --git a/llvm/test/Transforms/GlobalOpt/malloc-promote-1-no-null-opt.ll b/llvm/test/Transforms/GlobalOpt/malloc-promote-1-no-null-opt.ll
index 4cbe837..ccda0b8 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/malloc-promote-1-no-null-opt.ll
+++ b/llvm/test/Transforms/GlobalOpt/malloc-promote-1-no-null-opt.ll
@@ -12,8 +12,7 @@ define void @init() #0 {
 ; 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:    [[GV:%.*]] = load i32*, i32** @G, align 8
-; CHECK-NEXT:    store i32 0, i32* [[GV]], align 4
+; CHECK-NEXT:    store i32 0, i32* [[P]], align 4
 ; CHECK-NEXT:    ret void
 ;
   %malloccall = tail call i8* @malloc(i64 4)

diff --git a/llvm/test/Transforms/GlobalOpt/malloc-promote-2-no-null-opt.ll b/llvm/test/Transforms/GlobalOpt/malloc-promote-2-no-null-opt.ll
index 0b31aaf..9c17c83 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/malloc-promote-2-no-null-opt.ll
+++ b/llvm/test/Transforms/GlobalOpt/malloc-promote-2-no-null-opt.ll
@@ -12,8 +12,7 @@ define void @t() #0 {
 ; 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:    [[GV:%.*]] = load i32*, i32** @G, align 8
-; CHECK-NEXT:    [[GVE:%.*]] = getelementptr i32, i32* [[GV]], i32 40
+; CHECK-NEXT:    [[GVE:%.*]] = getelementptr i32, i32* [[P]], i32 40
 ; CHECK-NEXT:    store i32 20, i32* [[GVE]], align 4
 ; CHECK-NEXT:    ret void
 ;

diff --git a/llvm/test/Transforms/GlobalOpt/malloc-promote-3.ll b/llvm/test/Transforms/GlobalOpt/malloc-promote-3.ll
index fe7b475..2a99402 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/malloc-promote-3.ll
+++ b/llvm/test/Transforms/GlobalOpt/malloc-promote-3.ll
@@ -12,8 +12,7 @@ define void @t() {
 ; 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:    [[GV:%.*]] = load i32*, i32** @G, align 8
-; CHECK-NEXT:    [[GVE:%.*]] = getelementptr i32, i32* [[GV]], i32 40
+; CHECK-NEXT:    [[GVE:%.*]] = getelementptr i32, i32* [[P]], i32 40
 ; CHECK-NEXT:    store i32 20, i32* [[GVE]], align 4
 ; CHECK-NEXT:    ret void
 ;

diff --git a/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool-check-debug.ll b/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool-check-debug.ll
index da39636..1c91741 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool-check-debug.ll
+++ b/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool-check-debug.ll
@@ -2,21 +2,24 @@
 
 @foo = internal global i32 0, align 4
 
-define dso_local i32 @bar() {
+define void @store() {
 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: ![[DbgLocStore]] = !DILocation(line: 1,
-;CHECK: ![[DbgLocLoadSel]] = !DILocation(line: 2,
-;CHECK: ![[DbgLocRet]] = !DILocation(line: 3,
+;CHECK: ![[DbgLocLoadSel]] = !DILocation(line: 3,
+;CHECK: ![[DbgLocRet]] = !DILocation(line: 4,

diff --git a/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool.ll b/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool.ll
index 74c3546..bdd515b 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool.ll
+++ b/llvm/test/Transforms/GlobalOpt/shrink-global-to-bool.ll
@@ -10,11 +10,13 @@
 ; 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() {
+define void @test_global_var(i1 %i) {
 ; 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:
@@ -23,13 +25,15 @@ exit:
   ret void
 }
 
-define void @test_lds_var() {
+define void @test_lds_var(i1 %i) {
 ; 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:

diff --git a/llvm/test/Transforms/GlobalOpt/stored-once-forward-value.ll b/llvm/test/Transforms/GlobalOpt/stored-once-forward-value.ll
index b043a03..33a3891 100644 (file)
--- a/llvm/test/Transforms/GlobalOpt/stored-once-forward-value.ll
+++ b/llvm/test/Transforms/GlobalOpt/stored-once-forward-value.ll
@@ -12,10 +12,8 @@ declare void @b()
 
 define i1 @dom_const() {
 ; CHECK-LABEL: @dom_const(
-; CHECK-NEXT:    store i1 true, ptr @g1, align 1
 ; CHECK-NEXT:    call void @b()
-; CHECK-NEXT:    [[R:%.*]] = load i1, ptr @g1, align 1
-; CHECK-NEXT:    ret i1 [[R]]
+; CHECK-NEXT:    ret i1 true
 ;
   store i1 true, ptr @g1
   call void @b()
@@ -25,10 +23,8 @@ define i1 @dom_const() {
 
 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:    [[R:%.*]] = load i32, ptr @g2, align 4
-; CHECK-NEXT:    ret i32 [[R]]
+; CHECK-NEXT:    ret i32 [[A:%.*]]
 ;
   store i32 %a, ptr @g2
   call void @b()
@@ -74,8 +70,7 @@ define i1 @dom_multiple_function_loads() {
 ; CHECK-LABEL: @dom_multiple_function_loads(
 ; CHECK-NEXT:    store i1 true, ptr @g5, align 1
 ; CHECK-NEXT:    call void @b()
-; CHECK-NEXT:    [[R:%.*]] = load i1, ptr @g5, align 1
-; CHECK-NEXT:    ret i1 [[R]]
+; CHECK-NEXT:    ret i1 true
 ;
   store i1 true, ptr @g5
   call void @b()

diff --git a/llvm/test/Transforms/PhaseOrdering/recompute-globalsaa.ll b/llvm/test/Transforms/PhaseOrdering/recompute-globalsaa.ll
index 3dbddbd..f4290c6 100644 (file)
--- a/llvm/test/Transforms/PhaseOrdering/recompute-globalsaa.ll
+++ b/llvm/test/Transforms/PhaseOrdering/recompute-globalsaa.ll
@@ -9,7 +9,6 @@
 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