[CodeGen] Initialize large arrays by copying from a global

Currently, clang compiles explicit initializers for array
elements into series of store instructions. For large arrays of
built-in types this results in bloated output code and
significant amount of time spent on the instruction selection
phase. This patch fixes the issue by initializing such arrays
with global constants that store the binary image of the
initializer.

Differential Revision: https://reviews.llvm.org/D43181

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@325478 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGExprAgg.cpp b/lib/CodeGen/CGExprAgg.cpp
index d2b29e06ae3ade963bc338ad0ac8c4591202961a..34ce0d3398fbd7793eedb3a19f216621ec462311 100644 (file)
--- a/lib/CodeGen/CGExprAgg.cpp
+++ b/lib/CodeGen/CGExprAgg.cpp
@@ -14,6 +14,7 @@
 #include "CodeGenFunction.h"
 #include "CGObjCRuntime.h"
 #include "CodeGenModule.h"
+#include "ConstantEmitter.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclTemplate.h"
@@ -85,7 +86,7 @@ public:
   void EmitMoveFromReturnSlot(const Expr *E, RValue Src);
 
   void EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
-                     QualType elementType, InitListExpr *E);
+                     QualType ArrayQTy, InitListExpr *E);
 
   AggValueSlot::NeedsGCBarriers_t needsGC(QualType T) {
     if (CGF.getLangOpts().getGC() && TypeRequiresGCollection(T))
@@ -394,12 +395,15 @@ static bool isTrivialFiller(Expr *E) {
 
 /// \brief Emit initialization of an array from an initializer list.
 void AggExprEmitter::EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
-                                   QualType elementType, InitListExpr *E) {
+                                   QualType ArrayQTy, InitListExpr *E) {
   uint64_t NumInitElements = E->getNumInits();
 
   uint64_t NumArrayElements = AType->getNumElements();
   assert(NumInitElements <= NumArrayElements);
 
+  QualType elementType =
+      CGF.getContext().getAsArrayType(ArrayQTy)->getElementType();
+
   // DestPtr is an array*.  Construct an elementType* by drilling
   // down a level.
   llvm::Value *zero = llvm::ConstantInt::get(CGF.SizeTy, 0);
@@ -411,6 +415,29 @@ void AggExprEmitter::EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
   CharUnits elementAlign =
     DestPtr.getAlignment().alignmentOfArrayElement(elementSize);
 
+  // Consider initializing the array by copying from a global. For this to be
+  // more efficient than per-element initialization, the size of the elements
+  // with explicit initializers should be large enough.
+  if (NumInitElements * elementSize.getQuantity() > 16 &&
+      elementType.isTriviallyCopyableType(CGF.getContext())) {
+    CodeGen::CodeGenModule &CGM = CGF.CGM;
+    ConstantEmitter Emitter(CGM);
+    LangAS AS = ArrayQTy.getAddressSpace();
+    if (llvm::Constant *C = Emitter.tryEmitForInitializer(E, AS, ArrayQTy)) {
+      auto GV = new llvm::GlobalVariable(
+          CGM.getModule(), C->getType(),
+          CGM.isTypeConstant(ArrayQTy, /* ExcludeCtorDtor= */ true),
+          llvm::GlobalValue::PrivateLinkage, C, "constinit",
+          /* InsertBefore= */ nullptr, llvm::GlobalVariable::NotThreadLocal,
+          CGM.getContext().getTargetAddressSpace(AS));
+      Emitter.finalize(GV);
+      CharUnits Align = CGM.getContext().getTypeAlignInChars(ArrayQTy);
+      GV->setAlignment(Align.getQuantity());
+      EmitFinalDestCopy(ArrayQTy, CGF.MakeAddrLValue(GV, ArrayQTy, Align));
+      return;
+    }
+  }
+
   // Exception safety requires us to destroy all the
   // already-constructed members if an initializer throws.
   // For that, we'll need an EH cleanup.
@@ -1158,11 +1185,8 @@ void AggExprEmitter::VisitInitListExpr(InitListExpr *E) {
 
   // Handle initialization of an array.
   if (E->getType()->isArrayType()) {
-    QualType elementType =
-        CGF.getContext().getAsArrayType(E->getType())->getElementType();
-
     auto AType = cast<llvm::ArrayType>(Dest.getAddress().getElementType());
-    EmitArrayInit(Dest.getAddress(), AType, elementType, E);
+    EmitArrayInit(Dest.getAddress(), AType, E->getType(), E);
     return;
   }
 

diff --git a/test/CodeGen/init.c b/test/CodeGen/init.c
index 5d086723cc0e1d0a9ab145d7d3f8e96c32ae24cf..9924662cddfeb504699fddb7c7efdb0c4fc55503 100644 (file)
--- a/test/CodeGen/init.c
+++ b/test/CodeGen/init.c
@@ -8,8 +8,9 @@ unsigned v2[2][3] = {[0 ... 1][0 ... 1] = 2222, 3333};
 // CHECK-DAG: %struct.M = type { [2 x %struct.I] }
 // CHECK-DAG: %struct.I = type { [3 x i32] }
 
-// CHECK: [1 x %struct.M] [%struct.M { [2 x %struct.I] [%struct.I { [3 x i32] [i32 4, i32 4, i32 0] }, %struct.I { [3 x i32] [i32 4, i32 4, i32 5] }] }],
-// CHECK: [2 x [3 x i32]] {{[[][[]}}3 x i32] [i32 2222, i32 2222, i32 0], [3 x i32] [i32 2222, i32 2222, i32 3333]],
+// CHECK-DAG: [1 x %struct.M] [%struct.M { [2 x %struct.I] [%struct.I { [3 x i32] [i32 4, i32 4, i32 0] }, %struct.I { [3 x i32] [i32 4, i32 4, i32 5] }] }],
+// CHECK-DAG: [2 x [3 x i32]] {{[[][[]}}3 x i32] [i32 2222, i32 2222, i32 0], [3 x i32] [i32 2222, i32 2222, i32 3333]],
+// CHECK-DAG: [[INIT14:.*]] = private global [16 x i32] [i32 0, i32 0, i32 0, i32 0, i32 0, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 0, i32 0, i32 0, i32 0], align 4
 
 void f1() {
   // Scalars in braces.
@@ -33,8 +34,8 @@ void f3() {
 }
 
 // Constants
-// CHECK: @g3 = constant i32 10
-// CHECK: @f4.g4 = internal constant i32 12
+// CHECK-DAG: @g3 = constant i32 10
+// CHECK-DAG: @f4.g4 = internal constant i32 12
 const int g3 = 10;
 int f4() {
   static const int g4 = 12;
@@ -61,7 +62,7 @@ void f6() {
 
 
 
-// CHECK: @test7 = global{{.*}}{ i32 0, [4 x i8] c"bar\00" }
+// CHECK-DAG: @test7 = global{{.*}}{ i32 0, [4 x i8] c"bar\00" }
 // PR8217
 struct a7 {
   int  b;
@@ -151,3 +152,15 @@ void PR20473() {
   // CHECK: memcpy{{.*}}getelementptr inbounds ([3 x i8], [3 x i8]* @
   bar((char[3]) {""});
 }
+
+// Test that we initialize large member arrays by copying from a global and not
+// with a series of stores.
+struct S14 { int a[16]; };
+
+void test14(struct S14 *s14) {
+// CHECK-LABEL: @test14
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 {{.*}}, i8* align 4 {{.*}} [[INIT14]] {{.*}}, i32 64, i1 false)
+// CHECK-NOT: store
+// CHECK: ret void
+  *s14 = (struct S14) { { [5 ... 11] = 17 } };
+}