2007-08-24 10:22:53 +08:00
|
|
|
//===--- CGExprAgg.cpp - Emit LLVM Code from Aggregate Expressions --------===//
|
2007-08-11 04:13:28 +08:00
|
|
|
//
|
2019-01-19 16:50:56 +08:00
|
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
2007-08-11 04:13:28 +08:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// This contains code to emit Aggregate Expr nodes as LLVM code.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#include "CodeGenFunction.h"
|
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
Reviewers: rsmith, aaron.ballman, majnemer, rnk, compnerd, rjmccall
Reviewed By: rjmccall
Subscribers: rjmccall, rsmith, aaron.ballman, junbuml, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D45476
llvm-svn: 331677
2018-05-08 05:07:10 +08:00
|
|
|
#include "CGCXXABI.h"
|
2009-07-08 09:18:33 +08:00
|
|
|
#include "CGObjCRuntime.h"
|
2012-12-04 17:13:33 +08:00
|
|
|
#include "CodeGenModule.h"
|
2018-02-19 17:49:11 +08:00
|
|
|
#include "ConstantEmitter.h"
|
2008-08-11 13:00:27 +08:00
|
|
|
#include "clang/AST/ASTContext.h"
|
2009-04-17 08:06:03 +08:00
|
|
|
#include "clang/AST/DeclCXX.h"
|
2012-02-17 16:42:25 +08:00
|
|
|
#include "clang/AST/DeclTemplate.h"
|
2008-08-11 13:00:27 +08:00
|
|
|
#include "clang/AST/StmtVisitor.h"
|
2013-01-02 19:45:17 +08:00
|
|
|
#include "llvm/IR/Constants.h"
|
|
|
|
#include "llvm/IR/Function.h"
|
|
|
|
#include "llvm/IR/GlobalVariable.h"
|
|
|
|
#include "llvm/IR/Intrinsics.h"
|
2018-03-11 07:06:31 +08:00
|
|
|
#include "llvm/IR/IntrinsicInst.h"
|
2007-08-11 04:13:28 +08:00
|
|
|
using namespace clang;
|
|
|
|
using namespace CodeGen;
|
2007-08-11 08:04:45 +08:00
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Aggregate Expression Emitter
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
namespace {
|
2009-11-29 03:45:26 +08:00
|
|
|
class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
|
2007-08-21 12:25:47 +08:00
|
|
|
CodeGenFunction &CGF;
|
2008-11-01 09:53:16 +08:00
|
|
|
CGBuilderTy &Builder;
|
2010-09-15 18:14:12 +08:00
|
|
|
AggValueSlot Dest;
|
2015-06-08 18:23:49 +08:00
|
|
|
bool IsResultUnused;
|
2010-05-22 09:48:05 +08:00
|
|
|
|
2010-09-15 18:14:12 +08:00
|
|
|
AggValueSlot EnsureSlot(QualType T) {
|
|
|
|
if (!Dest.isIgnored()) return Dest;
|
|
|
|
return CGF.CreateAggTemp(T, "agg.tmp.ensured");
|
2010-05-22 09:48:05 +08:00
|
|
|
}
|
2012-07-03 07:58:38 +08:00
|
|
|
void EnsureDest(QualType T) {
|
|
|
|
if (!Dest.isIgnored()) return;
|
|
|
|
Dest = CGF.CreateAggTemp(T, "agg.tmp.ensured");
|
|
|
|
}
|
2010-05-23 06:13:32 +08:00
|
|
|
|
2018-03-10 09:11:17 +08:00
|
|
|
// Calls `Fn` with a valid return value slot, potentially creating a temporary
|
|
|
|
// to do so. If a temporary is created, an appropriate copy into `Dest` will
|
2018-03-11 07:06:31 +08:00
|
|
|
// be emitted, as will lifetime markers.
|
2018-03-10 09:11:17 +08:00
|
|
|
//
|
|
|
|
// The given function should take a ReturnValueSlot, and return an RValue that
|
|
|
|
// points to said slot.
|
|
|
|
void withReturnValueSlot(const Expr *E,
|
|
|
|
llvm::function_ref<RValue(ReturnValueSlot)> Fn);
|
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
public:
|
2015-06-08 18:23:49 +08:00
|
|
|
AggExprEmitter(CodeGenFunction &cgf, AggValueSlot Dest, bool IsResultUnused)
|
|
|
|
: CGF(cgf), Builder(CGF.Builder), Dest(Dest),
|
|
|
|
IsResultUnused(IsResultUnused) { }
|
2007-08-21 12:25:47 +08:00
|
|
|
|
2007-08-21 12:59:27 +08:00
|
|
|
//===--------------------------------------------------------------------===//
|
|
|
|
// Utilities
|
|
|
|
//===--------------------------------------------------------------------===//
|
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
/// EmitAggLoadOfLValue - Given an expression with aggregate type that
|
|
|
|
/// represents a value lvalue, this method emits the address of the lvalue,
|
|
|
|
/// then loads the result into DestPtr.
|
|
|
|
void EmitAggLoadOfLValue(const Expr *E);
|
2008-05-20 01:51:16 +08:00
|
|
|
|
2018-02-28 15:15:55 +08:00
|
|
|
enum ExprValueKind {
|
|
|
|
EVK_RValue,
|
|
|
|
EVK_NonRValue
|
|
|
|
};
|
|
|
|
|
2009-05-24 04:28:01 +08:00
|
|
|
/// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
|
2018-02-28 15:15:55 +08:00
|
|
|
/// SrcIsRValue is true if source comes from an RValue.
|
|
|
|
void EmitFinalDestCopy(QualType type, const LValue &src,
|
|
|
|
ExprValueKind SrcValueKind = EVK_NonRValue);
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
void EmitFinalDestCopy(QualType type, RValue src);
|
2012-07-03 07:58:38 +08:00
|
|
|
void EmitCopy(QualType type, const AggValueSlot &dest,
|
|
|
|
const AggValueSlot &src);
|
2009-05-24 04:28:01 +08:00
|
|
|
|
2011-08-26 07:04:34 +08:00
|
|
|
void EmitMoveFromReturnSlot(const Expr *E, RValue Src);
|
2010-05-23 06:13:32 +08:00
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
void EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
|
2018-02-19 17:49:11 +08:00
|
|
|
QualType ArrayQTy, InitListExpr *E);
|
2012-02-17 16:42:25 +08:00
|
|
|
|
2011-08-26 04:40:09 +08:00
|
|
|
AggValueSlot::NeedsGCBarriers_t needsGC(QualType T) {
|
2012-03-11 15:00:24 +08:00
|
|
|
if (CGF.getLangOpts().getGC() && TypeRequiresGCollection(T))
|
2011-08-26 04:40:09 +08:00
|
|
|
return AggValueSlot::NeedsGCBarriers;
|
|
|
|
return AggValueSlot::DoesNotNeedGCBarriers;
|
|
|
|
}
|
|
|
|
|
2010-05-23 06:13:32 +08:00
|
|
|
bool TypeRequiresGCollection(QualType T);
|
|
|
|
|
2007-08-21 12:59:27 +08:00
|
|
|
//===--------------------------------------------------------------------===//
|
|
|
|
// Visitor Methods
|
|
|
|
//===--------------------------------------------------------------------===//
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2015-01-18 09:48:19 +08:00
|
|
|
void Visit(Expr *E) {
|
DebugInfo: Use the preferred location rather than the start location for expression line info
This causes things like assignment to refer to the '=' rather than the
LHS when attributing the store instruction, for example.
There were essentially 3 options for this:
* The beginning of an expression (this was the behavior prior to this
commit). This meant that stepping through subexpressions would bounce
around from subexpressions back to the start of the outer expression,
etc. (eg: x + y + z would go x, y, x, z, x (the repeated 'x's would be
where the actual addition occurred)).
* The end of an expression. This seems to be what GCC does /mostly/, and
certainly this for function calls. This has the advantage that
progress is always 'forwards' (never jumping backwards - except for
independent subexpressions if they're evaluated in interesting orders,
etc). "x + y + z" would go "x y z" with the additions occurring at y
and z after the respective loads.
The problem with this is that the user would still have to think
fairly hard about precedence to realize which subexpression is being
evaluated or which operator overload is being called in, say, an asan
backtrace.
* The preferred location or 'exprloc'. In this case you get sort of what
you'd expect, though it's a bit confusing in its own way due to going
'backwards'. In this case the locations would be: "x y + z +" in
lovely postfix arithmetic order. But this does mean that if the op+
were an operator overload, say, and in a backtrace, the backtrace will
point to the exact '+' that's being called, not to the end of one of
its operands.
(actually the operator overload case doesn't work yet for other reasons,
but that's being fixed - but this at least gets scalar/complex
assignments and other plain operators right)
llvm-svn: 227027
2015-01-25 09:19:10 +08:00
|
|
|
ApplyDebugLocation DL(CGF, E);
|
2015-01-18 09:48:19 +08:00
|
|
|
StmtVisitor<AggExprEmitter>::Visit(E);
|
|
|
|
}
|
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
void VisitStmt(Stmt *S) {
|
2008-08-16 08:56:44 +08:00
|
|
|
CGF.ErrorUnsupported(S, "aggregate expression");
|
2007-08-21 12:25:47 +08:00
|
|
|
}
|
|
|
|
void VisitParenExpr(ParenExpr *PE) { Visit(PE->getSubExpr()); }
|
2011-04-15 08:35:48 +08:00
|
|
|
void VisitGenericSelectionExpr(GenericSelectionExpr *GE) {
|
|
|
|
Visit(GE->getResultExpr());
|
|
|
|
}
|
2017-03-26 10:18:05 +08:00
|
|
|
void VisitCoawaitExpr(CoawaitExpr *E) {
|
|
|
|
CGF.EmitCoawaitExpr(*E, Dest, IsResultUnused);
|
|
|
|
}
|
|
|
|
void VisitCoyieldExpr(CoyieldExpr *E) {
|
|
|
|
CGF.EmitCoyieldExpr(*E, Dest, IsResultUnused);
|
|
|
|
}
|
|
|
|
void VisitUnaryCoawait(UnaryOperator *E) { Visit(E->getSubExpr()); }
|
2009-01-27 17:03:41 +08:00
|
|
|
void VisitUnaryExtension(UnaryOperator *E) { Visit(E->getSubExpr()); }
|
2011-07-15 13:09:51 +08:00
|
|
|
void VisitSubstNonTypeTemplateParmExpr(SubstNonTypeTemplateParmExpr *E) {
|
|
|
|
return Visit(E->getReplacement());
|
|
|
|
}
|
2007-08-21 12:25:47 +08:00
|
|
|
|
2018-11-09 08:41:36 +08:00
|
|
|
void VisitConstantExpr(ConstantExpr *E) {
|
|
|
|
return Visit(E->getSubExpr());
|
|
|
|
}
|
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
// l-values.
|
2017-08-25 18:07:00 +08:00
|
|
|
void VisitDeclRefExpr(DeclRefExpr *E) { EmitAggLoadOfLValue(E); }
|
2007-12-14 10:04:12 +08:00
|
|
|
void VisitMemberExpr(MemberExpr *ME) { EmitAggLoadOfLValue(ME); }
|
|
|
|
void VisitUnaryDeref(UnaryOperator *E) { EmitAggLoadOfLValue(E); }
|
2010-01-05 02:47:06 +08:00
|
|
|
void VisitStringLiteral(StringLiteral *E) { EmitAggLoadOfLValue(E); }
|
2011-06-17 12:59:12 +08:00
|
|
|
void VisitCompoundLiteralExpr(CompoundLiteralExpr *E);
|
2007-12-14 10:04:12 +08:00
|
|
|
void VisitArraySubscriptExpr(ArraySubscriptExpr *E) {
|
|
|
|
EmitAggLoadOfLValue(E);
|
|
|
|
}
|
2009-04-22 07:00:09 +08:00
|
|
|
void VisitPredefinedExpr(const PredefinedExpr *E) {
|
2009-09-09 23:08:12 +08:00
|
|
|
EmitAggLoadOfLValue(E);
|
2009-04-22 07:00:09 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
// Operators.
|
2009-08-08 07:22:37 +08:00
|
|
|
void VisitCastExpr(CastExpr *E);
|
2007-11-01 06:04:46 +08:00
|
|
|
void VisitCallExpr(const CallExpr *E);
|
2007-09-01 06:54:14 +08:00
|
|
|
void VisitStmtExpr(const StmtExpr *E);
|
2007-08-21 12:25:47 +08:00
|
|
|
void VisitBinaryOperator(const BinaryOperator *BO);
|
2009-10-23 06:57:31 +08:00
|
|
|
void VisitPointerToDataMemberBinaryOperator(const BinaryOperator *BO);
|
2007-08-21 12:43:17 +08:00
|
|
|
void VisitBinAssign(const BinaryOperator *E);
|
2008-05-20 15:56:31 +08:00
|
|
|
void VisitBinComma(const BinaryOperator *E);
|
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
Reviewers: rsmith, aaron.ballman, majnemer, rnk, compnerd, rjmccall
Reviewed By: rjmccall
Subscribers: rjmccall, rsmith, aaron.ballman, junbuml, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D45476
llvm-svn: 331677
2018-05-08 05:07:10 +08:00
|
|
|
void VisitBinCmp(const BinaryOperator *E);
|
2007-08-21 12:25:47 +08:00
|
|
|
|
2008-06-25 01:04:18 +08:00
|
|
|
void VisitObjCMessageExpr(ObjCMessageExpr *E);
|
2008-08-23 18:51:21 +08:00
|
|
|
void VisitObjCIvarRefExpr(ObjCIvarRefExpr *E) {
|
|
|
|
EmitAggLoadOfLValue(E);
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2015-06-10 08:27:52 +08:00
|
|
|
void VisitDesignatedInitUpdateExpr(DesignatedInitUpdateExpr *E);
|
2011-02-17 18:25:35 +08:00
|
|
|
void VisitAbstractConditionalOperator(const AbstractConditionalOperator *CO);
|
2009-07-09 02:33:14 +08:00
|
|
|
void VisitChooseExpr(const ChooseExpr *CE);
|
2007-10-27 01:44:44 +08:00
|
|
|
void VisitInitListExpr(InitListExpr *E);
|
2016-12-14 09:32:13 +08:00
|
|
|
void VisitArrayInitLoopExpr(const ArrayInitLoopExpr *E,
|
|
|
|
llvm::Value *outerBegin = nullptr);
|
2009-12-16 14:57:54 +08:00
|
|
|
void VisitImplicitValueInitExpr(ImplicitValueInitExpr *E);
|
2015-06-10 08:27:52 +08:00
|
|
|
void VisitNoInitExpr(NoInitExpr *E) { } // Do nothing.
|
2008-04-08 12:40:51 +08:00
|
|
|
void VisitCXXDefaultArgExpr(CXXDefaultArgExpr *DAE) {
|
Implement __builtin_LINE() et. al. to support source location capture.
Summary:
This patch implements the source location builtins `__builtin_LINE(), `__builtin_FUNCTION()`, `__builtin_FILE()` and `__builtin_COLUMN()`. These builtins are needed to implement [`std::experimental::source_location`](https://rawgit.com/cplusplus/fundamentals-ts/v2/main.html#reflection.src_loc.creation).
With the exception of `__builtin_COLUMN`, GCC also implements these builtins, and Clangs behavior is intended to match as closely as possible.
Reviewers: rsmith, joerg, aaron.ballman, bogner, majnemer, shafik, martong
Reviewed By: rsmith
Subscribers: rnkovacs, loskutov, riccibruno, mgorny, kunitoki, alexr, majnemer, hfinkel, cfe-commits
Differential Revision: https://reviews.llvm.org/D37035
llvm-svn: 360937
2019-05-17 05:04:15 +08:00
|
|
|
CodeGenFunction::CXXDefaultArgExprScope Scope(CGF, DAE);
|
2008-04-08 12:40:51 +08:00
|
|
|
Visit(DAE->getExpr());
|
|
|
|
}
|
2013-04-21 06:23:05 +08:00
|
|
|
void VisitCXXDefaultInitExpr(CXXDefaultInitExpr *DIE) {
|
Implement __builtin_LINE() et. al. to support source location capture.
Summary:
This patch implements the source location builtins `__builtin_LINE(), `__builtin_FUNCTION()`, `__builtin_FILE()` and `__builtin_COLUMN()`. These builtins are needed to implement [`std::experimental::source_location`](https://rawgit.com/cplusplus/fundamentals-ts/v2/main.html#reflection.src_loc.creation).
With the exception of `__builtin_COLUMN`, GCC also implements these builtins, and Clangs behavior is intended to match as closely as possible.
Reviewers: rsmith, joerg, aaron.ballman, bogner, majnemer, shafik, martong
Reviewed By: rsmith
Subscribers: rnkovacs, loskutov, riccibruno, mgorny, kunitoki, alexr, majnemer, hfinkel, cfe-commits
Differential Revision: https://reviews.llvm.org/D37035
llvm-svn: 360937
2019-05-17 05:04:15 +08:00
|
|
|
CodeGenFunction::CXXDefaultInitExprScope Scope(CGF, DIE);
|
2013-04-21 06:23:05 +08:00
|
|
|
Visit(DIE->getExpr());
|
|
|
|
}
|
2009-05-31 07:23:33 +08:00
|
|
|
void VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *E);
|
2009-05-04 01:47:16 +08:00
|
|
|
void VisitCXXConstructExpr(const CXXConstructExpr *E);
|
P0136R1, DR1573, DR1645, DR1715, DR1736, DR1903, DR1941, DR1959, DR1991:
Replace inheriting constructors implementation with new approach, voted into
C++ last year as a DR against C++11.
Instead of synthesizing a set of derived class constructors for each inherited
base class constructor, we make the constructors of the base class visible to
constructor lookup in the derived class, using the normal rules for
using-declarations.
For constructors, UsingShadowDecl now has a ConstructorUsingShadowDecl derived
class that tracks the requisite additional information. We create shadow
constructors (not found by name lookup) in the derived class to model the
actual initialization, and have a new expression node,
CXXInheritedCtorInitExpr, to model the initialization of a base class from such
a constructor. (This initialization is special because it performs real perfect
forwarding of arguments.)
In cases where argument forwarding is not possible (for inalloca calls,
variadic calls, and calls with callee parameter cleanup), the shadow inheriting
constructor is not emitted and instead we directly emit the initialization code
into the caller of the inherited constructor.
Note that this new model is not perfectly compatible with the old model in some
corner cases. In particular:
* if B inherits a private constructor from A, and C uses that constructor to
construct a B, then we previously required that A befriends B and B
befriends C, but the new rules require A to befriend C directly, and
* if a derived class has its own constructors (and so its implicit default
constructor is suppressed), it may still inherit a default constructor from
a base class
llvm-svn: 274049
2016-06-29 03:03:57 +08:00
|
|
|
void VisitCXXInheritedCtorInitExpr(const CXXInheritedCtorInitExpr *E);
|
2012-02-09 11:32:31 +08:00
|
|
|
void VisitLambdaExpr(LambdaExpr *E);
|
2013-06-13 06:31:48 +08:00
|
|
|
void VisitCXXStdInitializerListExpr(CXXStdInitializerListExpr *E);
|
2010-12-06 16:20:24 +08:00
|
|
|
void VisitExprWithCleanups(ExprWithCleanups *E);
|
2010-07-08 14:14:04 +08:00
|
|
|
void VisitCXXScalarValueInitExpr(CXXScalarValueInitExpr *E);
|
2009-11-18 08:40:12 +08:00
|
|
|
void VisitCXXTypeidExpr(CXXTypeidExpr *E) { EmitAggLoadOfLValue(E); }
|
2011-06-22 01:03:29 +08:00
|
|
|
void VisitMaterializeTemporaryExpr(MaterializeTemporaryExpr *E);
|
2011-02-16 16:02:54 +08:00
|
|
|
void VisitOpaqueValueExpr(OpaqueValueExpr *E);
|
|
|
|
|
2011-11-06 17:01:30 +08:00
|
|
|
void VisitPseudoObjectExpr(PseudoObjectExpr *E) {
|
|
|
|
if (E->isGLValue()) {
|
|
|
|
LValue LV = CGF.EmitPseudoObjectLValue(E);
|
2012-07-03 07:58:38 +08:00
|
|
|
return EmitFinalDestCopy(E->getType(), LV);
|
2011-11-06 17:01:30 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
CGF.EmitPseudoObjectRValue(E, EnsureSlot(E->getType()));
|
|
|
|
}
|
|
|
|
|
2008-05-27 23:51:49 +08:00
|
|
|
void VisitVAArgExpr(VAArgExpr *E);
|
2008-04-05 02:42:16 +08:00
|
|
|
|
2012-03-30 01:37:10 +08:00
|
|
|
void EmitInitializationToLValue(Expr *E, LValue Address);
|
2011-06-16 12:16:24 +08:00
|
|
|
void EmitNullInitializationToLValue(LValue Address);
|
2007-08-21 12:25:47 +08:00
|
|
|
// case Expr::ChooseExprClass:
|
2009-12-10 03:24:08 +08:00
|
|
|
void VisitCXXThrowExpr(const CXXThrowExpr *E) { CGF.EmitCXXThrowExpr(E); }
|
2011-10-11 10:20:01 +08:00
|
|
|
void VisitAtomicExpr(AtomicExpr *E) {
|
2015-11-10 03:56:35 +08:00
|
|
|
RValue Res = CGF.EmitAtomicExpr(E);
|
|
|
|
EmitFinalDestCopy(E->getType(), Res);
|
2011-10-11 10:20:01 +08:00
|
|
|
}
|
2007-08-21 12:25:47 +08:00
|
|
|
};
|
|
|
|
} // end anonymous namespace.
|
|
|
|
|
2007-08-21 12:59:27 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Utilities
|
|
|
|
//===----------------------------------------------------------------------===//
|
2007-08-21 12:25:47 +08:00
|
|
|
|
2007-08-11 08:04:45 +08:00
|
|
|
/// EmitAggLoadOfLValue - Given an expression with aggregate type that
|
|
|
|
/// represents a value lvalue, this method emits the address of the lvalue,
|
|
|
|
/// then loads the result into DestPtr.
|
2007-08-21 12:25:47 +08:00
|
|
|
void AggExprEmitter::EmitAggLoadOfLValue(const Expr *E) {
|
|
|
|
LValue LV = CGF.EmitLValue(E);
|
2013-03-08 05:37:17 +08:00
|
|
|
|
|
|
|
// If the type of the l-value is atomic, then do an atomic load.
|
2015-02-14 09:35:12 +08:00
|
|
|
if (LV.getType()->isAtomicType() || CGF.LValueIsSuitableForInlineAtomic(LV)) {
|
2013-10-02 10:29:49 +08:00
|
|
|
CGF.EmitAtomicLoad(LV, E->getExprLoc(), Dest);
|
2013-03-08 05:37:17 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitFinalDestCopy(E->getType(), LV);
|
2009-05-24 04:28:01 +08:00
|
|
|
}
|
|
|
|
|
2018-05-09 09:00:01 +08:00
|
|
|
/// True if the given aggregate type requires special GC API calls.
|
2010-05-23 06:13:32 +08:00
|
|
|
bool AggExprEmitter::TypeRequiresGCollection(QualType T) {
|
|
|
|
// Only record types have members that might require garbage collection.
|
|
|
|
const RecordType *RecordTy = T->getAs<RecordType>();
|
|
|
|
if (!RecordTy) return false;
|
|
|
|
|
|
|
|
// Don't mess with non-trivial C++ types.
|
|
|
|
RecordDecl *Record = RecordTy->getDecl();
|
|
|
|
if (isa<CXXRecordDecl>(Record) &&
|
2012-11-16 08:53:38 +08:00
|
|
|
(cast<CXXRecordDecl>(Record)->hasNonTrivialCopyConstructor() ||
|
2010-05-23 06:13:32 +08:00
|
|
|
!cast<CXXRecordDecl>(Record)->hasTrivialDestructor()))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// Check whether the type has an object member.
|
|
|
|
return Record->hasObjectMember();
|
|
|
|
}
|
|
|
|
|
2018-03-10 09:11:17 +08:00
|
|
|
void AggExprEmitter::withReturnValueSlot(
|
|
|
|
const Expr *E, llvm::function_ref<RValue(ReturnValueSlot)> EmitCall) {
|
|
|
|
QualType RetTy = E->getType();
|
|
|
|
bool RequiresDestruction =
|
|
|
|
Dest.isIgnored() &&
|
|
|
|
RetTy.isDestructedType() == QualType::DK_nontrivial_c_struct;
|
|
|
|
|
|
|
|
// If it makes no observable difference, save a memcpy + temporary.
|
|
|
|
//
|
|
|
|
// We need to always provide our own temporary if destruction is required.
|
|
|
|
// Otherwise, EmitCall will emit its own, notice that it's "unused", and end
|
|
|
|
// its lifetime before we have the chance to emit a proper destructor call.
|
|
|
|
bool UseTemp = Dest.isPotentiallyAliased() || Dest.requiresGCollection() ||
|
|
|
|
(RequiresDestruction && !Dest.getAddress().isValid());
|
|
|
|
|
|
|
|
Address RetAddr = Address::invalid();
|
2018-05-17 19:16:35 +08:00
|
|
|
Address RetAllocaAddr = Address::invalid();
|
2018-03-11 07:06:31 +08:00
|
|
|
|
|
|
|
EHScopeStack::stable_iterator LifetimeEndBlock;
|
|
|
|
llvm::Value *LifetimeSizePtr = nullptr;
|
|
|
|
llvm::IntrinsicInst *LifetimeStartInst = nullptr;
|
2018-03-10 09:11:17 +08:00
|
|
|
if (!UseTemp) {
|
|
|
|
RetAddr = Dest.getAddress();
|
|
|
|
} else {
|
2018-05-17 19:16:35 +08:00
|
|
|
RetAddr = CGF.CreateMemTemp(RetTy, "tmp", &RetAllocaAddr);
|
2018-03-10 09:11:17 +08:00
|
|
|
uint64_t Size =
|
|
|
|
CGF.CGM.getDataLayout().getTypeAllocSize(CGF.ConvertTypeForMem(RetTy));
|
2018-05-17 19:16:35 +08:00
|
|
|
LifetimeSizePtr = CGF.EmitLifetimeStart(Size, RetAllocaAddr.getPointer());
|
2018-03-11 07:06:31 +08:00
|
|
|
if (LifetimeSizePtr) {
|
|
|
|
LifetimeStartInst =
|
|
|
|
cast<llvm::IntrinsicInst>(std::prev(Builder.GetInsertPoint()));
|
|
|
|
assert(LifetimeStartInst->getIntrinsicID() ==
|
|
|
|
llvm::Intrinsic::lifetime_start &&
|
|
|
|
"Last insertion wasn't a lifetime.start?");
|
|
|
|
|
2018-03-10 09:11:17 +08:00
|
|
|
CGF.pushFullExprCleanup<CodeGenFunction::CallLifetimeEnd>(
|
2018-05-17 19:16:35 +08:00
|
|
|
NormalEHLifetimeMarker, RetAllocaAddr, LifetimeSizePtr);
|
2018-03-11 07:06:31 +08:00
|
|
|
LifetimeEndBlock = CGF.EHStack.stable_begin();
|
|
|
|
}
|
2010-06-16 06:44:06 +08:00
|
|
|
}
|
2011-08-26 07:04:34 +08:00
|
|
|
|
2018-03-10 09:11:17 +08:00
|
|
|
RValue Src =
|
|
|
|
EmitCall(ReturnValueSlot(RetAddr, Dest.isVolatile(), IsResultUnused));
|
|
|
|
|
|
|
|
if (RequiresDestruction)
|
|
|
|
CGF.pushDestroy(RetTy.isDestructedType(), Src.getAggregateAddress(), RetTy);
|
|
|
|
|
2018-03-11 07:06:31 +08:00
|
|
|
if (!UseTemp)
|
|
|
|
return;
|
|
|
|
|
|
|
|
assert(Dest.getPointer() != Src.getAggregatePointer());
|
|
|
|
EmitFinalDestCopy(E->getType(), Src);
|
|
|
|
|
|
|
|
if (!RequiresDestruction && LifetimeStartInst) {
|
|
|
|
// If there's no dtor to run, the copy was the last use of our temporary.
|
|
|
|
// Since we're not guaranteed to be in an ExprWithCleanups, clean up
|
|
|
|
// eagerly.
|
|
|
|
CGF.DeactivateCleanupBlock(LifetimeEndBlock, LifetimeStartInst);
|
2018-05-17 19:16:35 +08:00
|
|
|
CGF.EmitLifetimeEnd(LifetimeSizePtr, RetAllocaAddr.getPointer());
|
2018-03-10 09:11:17 +08:00
|
|
|
}
|
2010-05-23 06:13:32 +08:00
|
|
|
}
|
|
|
|
|
2009-05-24 04:28:01 +08:00
|
|
|
/// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
void AggExprEmitter::EmitFinalDestCopy(QualType type, RValue src) {
|
2012-07-03 07:58:38 +08:00
|
|
|
assert(src.isAggregate() && "value must be aggregate value!");
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
LValue srcLV = CGF.MakeAddrLValue(src.getAggregateAddress(), type);
|
2018-02-28 15:15:55 +08:00
|
|
|
EmitFinalDestCopy(type, srcLV, EVK_RValue);
|
2012-07-03 07:58:38 +08:00
|
|
|
}
|
2009-05-24 04:28:01 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
/// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
|
2018-02-28 15:15:55 +08:00
|
|
|
void AggExprEmitter::EmitFinalDestCopy(QualType type, const LValue &src,
|
|
|
|
ExprValueKind SrcValueKind) {
|
2010-09-15 18:14:12 +08:00
|
|
|
// If Dest is ignored, then we're evaluating an aggregate expression
|
2012-07-03 07:58:38 +08:00
|
|
|
// in a context that doesn't care about the result. Note that loads
|
|
|
|
// from volatile l-values force the existence of a non-ignored
|
|
|
|
// destination.
|
|
|
|
if (Dest.isIgnored())
|
|
|
|
return;
|
2010-10-23 06:05:03 +08:00
|
|
|
|
2018-02-28 15:15:55 +08:00
|
|
|
// Copy non-trivial C structs here.
|
|
|
|
LValue DstLV = CGF.MakeAddrLValue(
|
|
|
|
Dest.getAddress(), Dest.isVolatile() ? type.withVolatile() : type);
|
|
|
|
|
|
|
|
if (SrcValueKind == EVK_RValue) {
|
|
|
|
if (type.isNonTrivialToPrimitiveDestructiveMove() == QualType::PCK_Struct) {
|
|
|
|
if (Dest.isPotentiallyAliased())
|
|
|
|
CGF.callCStructMoveAssignmentOperator(DstLV, src);
|
|
|
|
else
|
|
|
|
CGF.callCStructMoveConstructor(DstLV, src);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (type.isNonTrivialToPrimitiveCopy() == QualType::PCK_Struct) {
|
|
|
|
if (Dest.isPotentiallyAliased())
|
|
|
|
CGF.callCStructCopyAssignmentOperator(DstLV, src);
|
|
|
|
else
|
|
|
|
CGF.callCStructCopyConstructor(DstLV, src);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
AggValueSlot srcAgg =
|
|
|
|
AggValueSlot::forLValue(src, AggValueSlot::IsDestructed,
|
2018-04-06 04:52:58 +08:00
|
|
|
needsGC(type), AggValueSlot::IsAliased,
|
|
|
|
AggValueSlot::MayOverlap);
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitCopy(type, Dest, srcAgg);
|
|
|
|
}
|
2007-08-11 08:04:45 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
/// Perform a copy from the source into the destination.
|
|
|
|
///
|
|
|
|
/// \param type - the type of the aggregate being copied; qualifiers are
|
|
|
|
/// ignored
|
|
|
|
void AggExprEmitter::EmitCopy(QualType type, const AggValueSlot &dest,
|
|
|
|
const AggValueSlot &src) {
|
|
|
|
if (dest.requiresGCollection()) {
|
2018-04-06 04:52:58 +08:00
|
|
|
CharUnits sz = dest.getPreferredSize(CGF.getContext(), type);
|
2012-07-03 07:58:38 +08:00
|
|
|
llvm::Value *size = llvm::ConstantInt::get(CGF.SizeTy, sz.getQuantity());
|
2009-09-01 03:33:16 +08:00
|
|
|
CGF.CGM.getObjCRuntime().EmitGCMemmoveCollectable(CGF,
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
dest.getAddress(),
|
|
|
|
src.getAddress(),
|
2012-07-03 07:58:38 +08:00
|
|
|
size);
|
2009-09-01 03:33:16 +08:00
|
|
|
return;
|
|
|
|
}
|
2009-05-24 04:28:01 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
// If the result of the assignment is used, copy the LHS there also.
|
|
|
|
// It's volatile if either side is. Use the minimum alignment of
|
|
|
|
// the two sides.
|
2018-01-25 22:21:55 +08:00
|
|
|
LValue DestLV = CGF.MakeAddrLValue(dest.getAddress(), type);
|
|
|
|
LValue SrcLV = CGF.MakeAddrLValue(src.getAddress(), type);
|
2018-04-06 04:52:58 +08:00
|
|
|
CGF.EmitAggregateCopy(DestLV, SrcLV, type, dest.mayOverlap(),
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
dest.isVolatile() || src.isVolatile());
|
2007-08-11 08:04:45 +08:00
|
|
|
}
|
|
|
|
|
2018-05-09 09:00:01 +08:00
|
|
|
/// Emit the initializer for a std::initializer_list initialized with a
|
2012-02-17 16:42:25 +08:00
|
|
|
/// real initializer list.
|
2013-06-13 06:31:48 +08:00
|
|
|
void
|
|
|
|
AggExprEmitter::VisitCXXStdInitializerListExpr(CXXStdInitializerListExpr *E) {
|
|
|
|
// Emit an array containing the elements. The array is externally destructed
|
|
|
|
// if the std::initializer_list object is.
|
|
|
|
ASTContext &Ctx = CGF.getContext();
|
|
|
|
LValue Array = CGF.EmitLValue(E->getSubExpr());
|
|
|
|
assert(Array.isSimple() && "initializer_list array not a simple lvalue");
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Address ArrayPtr = Array.getAddress();
|
2013-06-13 06:31:48 +08:00
|
|
|
|
|
|
|
const ConstantArrayType *ArrayType =
|
|
|
|
Ctx.getAsConstantArrayType(E->getSubExpr()->getType());
|
|
|
|
assert(ArrayType && "std::initializer_list constructed from non-array");
|
|
|
|
|
|
|
|
// FIXME: Perform the checks on the field types in SemaInit.
|
|
|
|
RecordDecl *Record = E->getType()->castAs<RecordType>()->getDecl();
|
|
|
|
RecordDecl::field_iterator Field = Record->field_begin();
|
|
|
|
if (Field == Record->field_end()) {
|
|
|
|
CGF.ErrorUnsupported(E, "weird std::initializer_list");
|
2012-02-26 04:51:13 +08:00
|
|
|
return;
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Start pointer.
|
2013-06-13 06:31:48 +08:00
|
|
|
if (!Field->getType()->isPointerType() ||
|
|
|
|
!Ctx.hasSameType(Field->getType()->getPointeeType(),
|
|
|
|
ArrayType->getElementType())) {
|
|
|
|
CGF.ErrorUnsupported(E, "weird std::initializer_list");
|
2012-02-26 04:51:13 +08:00
|
|
|
return;
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
2013-06-13 06:31:48 +08:00
|
|
|
|
|
|
|
AggValueSlot Dest = EnsureSlot(E->getType());
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
LValue DestLV = CGF.MakeAddrLValue(Dest.getAddress(), E->getType());
|
2013-06-13 06:31:48 +08:00
|
|
|
LValue Start = CGF.EmitLValueForFieldInitialization(DestLV, *Field);
|
|
|
|
llvm::Value *Zero = llvm::ConstantInt::get(CGF.PtrDiffTy, 0);
|
|
|
|
llvm::Value *IdxStart[] = { Zero, Zero };
|
|
|
|
llvm::Value *ArrayStart =
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Builder.CreateInBoundsGEP(ArrayPtr.getPointer(), IdxStart, "arraystart");
|
2013-06-13 06:31:48 +08:00
|
|
|
CGF.EmitStoreThroughLValue(RValue::get(ArrayStart), Start);
|
|
|
|
++Field;
|
|
|
|
|
|
|
|
if (Field == Record->field_end()) {
|
|
|
|
CGF.ErrorUnsupported(E, "weird std::initializer_list");
|
2012-02-26 04:51:13 +08:00
|
|
|
return;
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
2013-06-13 06:31:48 +08:00
|
|
|
|
|
|
|
llvm::Value *Size = Builder.getInt(ArrayType->getSize());
|
|
|
|
LValue EndOrLength = CGF.EmitLValueForFieldInitialization(DestLV, *Field);
|
|
|
|
if (Field->getType()->isPointerType() &&
|
|
|
|
Ctx.hasSameType(Field->getType()->getPointeeType(),
|
|
|
|
ArrayType->getElementType())) {
|
2012-02-17 16:42:25 +08:00
|
|
|
// End pointer.
|
2013-06-13 06:31:48 +08:00
|
|
|
llvm::Value *IdxEnd[] = { Zero, Size };
|
|
|
|
llvm::Value *ArrayEnd =
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Builder.CreateInBoundsGEP(ArrayPtr.getPointer(), IdxEnd, "arrayend");
|
2013-06-13 06:31:48 +08:00
|
|
|
CGF.EmitStoreThroughLValue(RValue::get(ArrayEnd), EndOrLength);
|
|
|
|
} else if (Ctx.hasSameType(Field->getType(), Ctx.getSizeType())) {
|
2012-02-17 16:42:25 +08:00
|
|
|
// Length.
|
2013-06-13 06:31:48 +08:00
|
|
|
CGF.EmitStoreThroughLValue(RValue::get(Size), EndOrLength);
|
2012-02-17 16:42:25 +08:00
|
|
|
} else {
|
2013-06-13 06:31:48 +08:00
|
|
|
CGF.ErrorUnsupported(E, "weird std::initializer_list");
|
2012-02-26 04:51:13 +08:00
|
|
|
return;
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-05-09 09:00:01 +08:00
|
|
|
/// Determine if E is a trivial array filler, that is, one that is
|
2014-06-14 07:04:49 +08:00
|
|
|
/// equivalent to zero-initialization.
|
|
|
|
static bool isTrivialFiller(Expr *E) {
|
|
|
|
if (!E)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
if (isa<ImplicitValueInitExpr>(E))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
if (auto *ILE = dyn_cast<InitListExpr>(E)) {
|
|
|
|
if (ILE->getNumInits())
|
|
|
|
return false;
|
|
|
|
return isTrivialFiller(ILE->getArrayFiller());
|
|
|
|
}
|
|
|
|
|
|
|
|
if (auto *Cons = dyn_cast_or_null<CXXConstructExpr>(E))
|
|
|
|
return Cons->getConstructor()->isDefaultConstructor() &&
|
|
|
|
Cons->getConstructor()->isTrivial();
|
|
|
|
|
|
|
|
// FIXME: Are there other cases where we can avoid emitting an initializer?
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2018-05-09 09:00:01 +08:00
|
|
|
/// Emit initialization of an array from an initializer list.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
void AggExprEmitter::EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
|
2018-02-19 17:49:11 +08:00
|
|
|
QualType ArrayQTy, InitListExpr *E) {
|
2012-02-17 16:42:25 +08:00
|
|
|
uint64_t NumInitElements = E->getNumInits();
|
|
|
|
|
|
|
|
uint64_t NumArrayElements = AType->getNumElements();
|
|
|
|
assert(NumInitElements <= NumArrayElements);
|
|
|
|
|
2018-02-19 17:49:11 +08:00
|
|
|
QualType elementType =
|
|
|
|
CGF.getContext().getAsArrayType(ArrayQTy)->getElementType();
|
|
|
|
|
2012-02-17 16:42:25 +08:00
|
|
|
// DestPtr is an array*. Construct an elementType* by drilling
|
|
|
|
// down a level.
|
|
|
|
llvm::Value *zero = llvm::ConstantInt::get(CGF.SizeTy, 0);
|
|
|
|
llvm::Value *indices[] = { zero, zero };
|
|
|
|
llvm::Value *begin =
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Builder.CreateInBoundsGEP(DestPtr.getPointer(), indices, "arrayinit.begin");
|
|
|
|
|
|
|
|
CharUnits elementSize = CGF.getContext().getTypeSizeInChars(elementType);
|
|
|
|
CharUnits elementAlign =
|
|
|
|
DestPtr.getAlignment().alignmentOfArrayElement(elementSize);
|
2012-02-17 16:42:25 +08:00
|
|
|
|
2018-02-19 17:49:11 +08:00
|
|
|
// 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;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-02-17 16:42:25 +08:00
|
|
|
// Exception safety requires us to destroy all the
|
|
|
|
// already-constructed members if an initializer throws.
|
|
|
|
// For that, we'll need an EH cleanup.
|
|
|
|
QualType::DestructionKind dtorKind = elementType.isDestructedType();
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Address endOfInit = Address::invalid();
|
2012-02-17 16:42:25 +08:00
|
|
|
EHScopeStack::stable_iterator cleanup;
|
2014-05-21 13:09:00 +08:00
|
|
|
llvm::Instruction *cleanupDominator = nullptr;
|
2012-02-17 16:42:25 +08:00
|
|
|
if (CGF.needsEHCleanup(dtorKind)) {
|
|
|
|
// In principle we could tell the cleanup where we are more
|
|
|
|
// directly, but the control flow can get so varied here that it
|
|
|
|
// would actually be quite complex. Therefore we go through an
|
|
|
|
// alloca.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
endOfInit = CGF.CreateTempAlloca(begin->getType(), CGF.getPointerAlign(),
|
2012-02-17 16:42:25 +08:00
|
|
|
"arrayinit.endOfInit");
|
|
|
|
cleanupDominator = Builder.CreateStore(begin, endOfInit);
|
|
|
|
CGF.pushIrregularPartialArrayCleanup(begin, endOfInit, elementType,
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
elementAlign,
|
2012-02-17 16:42:25 +08:00
|
|
|
CGF.getDestroyer(dtorKind));
|
|
|
|
cleanup = CGF.EHStack.stable_begin();
|
|
|
|
|
|
|
|
// Otherwise, remember that we didn't need a cleanup.
|
|
|
|
} else {
|
|
|
|
dtorKind = QualType::DK_none;
|
|
|
|
}
|
|
|
|
|
|
|
|
llvm::Value *one = llvm::ConstantInt::get(CGF.SizeTy, 1);
|
|
|
|
|
|
|
|
// The 'current element to initialize'. The invariants on this
|
|
|
|
// variable are complicated. Essentially, after each iteration of
|
|
|
|
// the loop, it points to the last initialized element, except
|
|
|
|
// that it points to the beginning of the array before any
|
|
|
|
// elements have been initialized.
|
|
|
|
llvm::Value *element = begin;
|
|
|
|
|
|
|
|
// Emit the explicit initializers.
|
|
|
|
for (uint64_t i = 0; i != NumInitElements; ++i) {
|
|
|
|
// Advance to the next element.
|
|
|
|
if (i > 0) {
|
|
|
|
element = Builder.CreateInBoundsGEP(element, one, "arrayinit.element");
|
|
|
|
|
|
|
|
// Tell the cleanup that it needs to destroy up to this
|
|
|
|
// element. TODO: some of these stores can be trivially
|
|
|
|
// observed to be unnecessary.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (endOfInit.isValid()) Builder.CreateStore(element, endOfInit);
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
LValue elementLV =
|
|
|
|
CGF.MakeAddrLValue(Address(element, elementAlign), elementType);
|
2013-06-13 06:31:48 +08:00
|
|
|
EmitInitializationToLValue(E->getInit(i), elementLV);
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Check whether there's a non-trivial array-fill expression.
|
|
|
|
Expr *filler = E->getArrayFiller();
|
2014-06-14 07:04:49 +08:00
|
|
|
bool hasTrivialFiller = isTrivialFiller(filler);
|
2012-02-17 16:42:25 +08:00
|
|
|
|
|
|
|
// Any remaining elements need to be zero-initialized, possibly
|
|
|
|
// using the filler expression. We can skip this if the we're
|
|
|
|
// emitting to zeroed memory.
|
|
|
|
if (NumInitElements != NumArrayElements &&
|
|
|
|
!(Dest.isZeroed() && hasTrivialFiller &&
|
|
|
|
CGF.getTypes().isZeroInitializable(elementType))) {
|
|
|
|
|
|
|
|
// Use an actual loop. This is basically
|
|
|
|
// do { *array++ = filler; } while (array != end);
|
|
|
|
|
|
|
|
// Advance to the start of the rest of the array.
|
|
|
|
if (NumInitElements) {
|
|
|
|
element = Builder.CreateInBoundsGEP(element, one, "arrayinit.start");
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (endOfInit.isValid()) Builder.CreateStore(element, endOfInit);
|
2012-02-17 16:42:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Compute the end of the array.
|
|
|
|
llvm::Value *end = Builder.CreateInBoundsGEP(begin,
|
|
|
|
llvm::ConstantInt::get(CGF.SizeTy, NumArrayElements),
|
|
|
|
"arrayinit.end");
|
|
|
|
|
|
|
|
llvm::BasicBlock *entryBB = Builder.GetInsertBlock();
|
|
|
|
llvm::BasicBlock *bodyBB = CGF.createBasicBlock("arrayinit.body");
|
|
|
|
|
|
|
|
// Jump into the body.
|
|
|
|
CGF.EmitBlock(bodyBB);
|
|
|
|
llvm::PHINode *currentElement =
|
|
|
|
Builder.CreatePHI(element->getType(), 2, "arrayinit.cur");
|
|
|
|
currentElement->addIncoming(element, entryBB);
|
|
|
|
|
|
|
|
// Emit the actual filler expression.
|
2017-05-12 02:58:24 +08:00
|
|
|
{
|
|
|
|
// C++1z [class.temporary]p5:
|
|
|
|
// when a default constructor is called to initialize an element of
|
|
|
|
// an array with no corresponding initializer [...] the destruction of
|
|
|
|
// every temporary created in a default argument is sequenced before
|
|
|
|
// the construction of the next array element, if any
|
|
|
|
CodeGenFunction::RunCleanupsScope CleanupsScope(CGF);
|
|
|
|
LValue elementLV =
|
|
|
|
CGF.MakeAddrLValue(Address(currentElement, elementAlign), elementType);
|
|
|
|
if (filler)
|
|
|
|
EmitInitializationToLValue(filler, elementLV);
|
|
|
|
else
|
|
|
|
EmitNullInitializationToLValue(elementLV);
|
|
|
|
}
|
2012-02-17 16:42:25 +08:00
|
|
|
|
|
|
|
// Move on to the next element.
|
|
|
|
llvm::Value *nextElement =
|
|
|
|
Builder.CreateInBoundsGEP(currentElement, one, "arrayinit.next");
|
|
|
|
|
|
|
|
// Tell the EH cleanup that we finished with the last element.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (endOfInit.isValid()) Builder.CreateStore(nextElement, endOfInit);
|
2012-02-17 16:42:25 +08:00
|
|
|
|
|
|
|
// Leave the loop if we're done.
|
|
|
|
llvm::Value *done = Builder.CreateICmpEQ(nextElement, end,
|
|
|
|
"arrayinit.done");
|
|
|
|
llvm::BasicBlock *endBB = CGF.createBasicBlock("arrayinit.end");
|
|
|
|
Builder.CreateCondBr(done, endBB, bodyBB);
|
|
|
|
currentElement->addIncoming(nextElement, Builder.GetInsertBlock());
|
|
|
|
|
|
|
|
CGF.EmitBlock(endBB);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Leave the partial-array cleanup if we entered one.
|
|
|
|
if (dtorKind) CGF.DeactivateCleanupBlock(cleanup, cleanupDominator);
|
|
|
|
}
|
|
|
|
|
2007-08-21 12:59:27 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Visitor Methods
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2011-06-22 01:03:29 +08:00
|
|
|
void AggExprEmitter::VisitMaterializeTemporaryExpr(MaterializeTemporaryExpr *E){
|
|
|
|
Visit(E->GetTemporaryExpr());
|
|
|
|
}
|
|
|
|
|
2011-02-16 16:02:54 +08:00
|
|
|
void AggExprEmitter::VisitOpaqueValueExpr(OpaqueValueExpr *e) {
|
2018-03-20 09:47:58 +08:00
|
|
|
// If this is a unique OVE, just visit its source expression.
|
|
|
|
if (e->isUnique())
|
|
|
|
Visit(e->getSourceExpr());
|
|
|
|
else
|
|
|
|
EmitFinalDestCopy(e->getType(), CGF.getOrCreateOpaqueLValueMapping(e));
|
2011-02-16 16:02:54 +08:00
|
|
|
}
|
|
|
|
|
2011-06-17 12:59:12 +08:00
|
|
|
void
|
|
|
|
AggExprEmitter::VisitCompoundLiteralExpr(CompoundLiteralExpr *E) {
|
2013-03-08 05:36:54 +08:00
|
|
|
if (Dest.isPotentiallyAliased() &&
|
|
|
|
E->getType().isPODType(CGF.getContext())) {
|
2011-06-18 00:37:20 +08:00
|
|
|
// For a POD type, just emit a load of the lvalue + a copy, because our
|
|
|
|
// compound literal might alias the destination.
|
|
|
|
EmitAggLoadOfLValue(E);
|
|
|
|
return;
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2011-06-17 12:59:12 +08:00
|
|
|
AggValueSlot Slot = EnsureSlot(E->getType());
|
|
|
|
CGF.EmitAggExpr(E->getInitializer(), Slot);
|
|
|
|
}
|
|
|
|
|
2013-03-08 05:37:17 +08:00
|
|
|
/// Attempt to look through various unimportant expressions to find a
|
|
|
|
/// cast of the given kind.
|
|
|
|
static Expr *findPeephole(Expr *op, CastKind kind) {
|
|
|
|
while (true) {
|
|
|
|
op = op->IgnoreParens();
|
|
|
|
if (CastExpr *castE = dyn_cast<CastExpr>(op)) {
|
|
|
|
if (castE->getCastKind() == kind)
|
|
|
|
return castE->getSubExpr();
|
|
|
|
if (castE->getCastKind() == CK_NoOp)
|
|
|
|
continue;
|
|
|
|
}
|
2014-05-21 13:09:00 +08:00
|
|
|
return nullptr;
|
2013-03-08 05:37:17 +08:00
|
|
|
}
|
|
|
|
}
|
2011-06-17 12:59:12 +08:00
|
|
|
|
2009-08-08 07:22:37 +08:00
|
|
|
void AggExprEmitter::VisitCastExpr(CastExpr *E) {
|
2015-10-20 12:24:12 +08:00
|
|
|
if (const auto *ECE = dyn_cast<ExplicitCastExpr>(E))
|
|
|
|
CGF.CGM.EmitExplicitCastExprType(ECE, &CGF);
|
2009-09-29 09:23:39 +08:00
|
|
|
switch (E->getCastKind()) {
|
2011-04-11 10:03:26 +08:00
|
|
|
case CK_Dynamic: {
|
2012-08-24 08:54:33 +08:00
|
|
|
// FIXME: Can this actually happen? We have no test coverage for it.
|
2010-05-15 05:31:02 +08:00
|
|
|
assert(isa<CXXDynamicCastExpr>(E) && "CK_Dynamic without a dynamic_cast?");
|
2012-08-24 08:54:33 +08:00
|
|
|
LValue LV = CGF.EmitCheckedLValue(E->getSubExpr(),
|
2012-09-08 10:08:36 +08:00
|
|
|
CodeGenFunction::TCK_Load);
|
2010-05-15 05:31:02 +08:00
|
|
|
// FIXME: Do we also need to handle property references here?
|
|
|
|
if (LV.isSimple())
|
|
|
|
CGF.EmitDynamicCast(LV.getAddress(), cast<CXXDynamicCastExpr>(E));
|
|
|
|
else
|
|
|
|
CGF.CGM.ErrorUnsupported(E, "non-simple lvalue dynamic_cast");
|
2018-06-21 01:19:40 +08:00
|
|
|
|
2010-09-15 18:14:12 +08:00
|
|
|
if (!Dest.isIgnored())
|
|
|
|
CGF.CGM.ErrorUnsupported(E, "lvalue dynamic_cast with a destination");
|
2010-05-15 05:31:02 +08:00
|
|
|
break;
|
|
|
|
}
|
2018-06-21 01:19:40 +08:00
|
|
|
|
2010-08-25 19:45:40 +08:00
|
|
|
case CK_ToUnion: {
|
2015-05-21 05:59:25 +08:00
|
|
|
// Evaluate even if the destination is ignored.
|
|
|
|
if (Dest.isIgnored()) {
|
|
|
|
CGF.EmitAnyExpr(E->getSubExpr(), AggValueSlot::ignored(),
|
|
|
|
/*ignoreResult=*/true);
|
|
|
|
break;
|
|
|
|
}
|
2011-04-13 06:02:02 +08:00
|
|
|
|
2009-08-08 07:22:37 +08:00
|
|
|
// GCC union extension
|
2010-08-21 11:15:20 +08:00
|
|
|
QualType Ty = E->getSubExpr()->getType();
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Address CastPtr =
|
|
|
|
Builder.CreateElementBitCast(Dest.getAddress(), CGF.ConvertType(Ty));
|
2011-06-16 12:16:24 +08:00
|
|
|
EmitInitializationToLValue(E->getSubExpr(),
|
2012-03-30 01:37:10 +08:00
|
|
|
CGF.MakeAddrLValue(CastPtr, Ty));
|
2009-09-29 09:23:39 +08:00
|
|
|
break;
|
2009-01-16 04:14:33 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2019-07-03 02:28:13 +08:00
|
|
|
case CK_LValueToRValueBitCast: {
|
|
|
|
if (Dest.isIgnored()) {
|
|
|
|
CGF.EmitAnyExpr(E->getSubExpr(), AggValueSlot::ignored(),
|
|
|
|
/*ignoreResult=*/true);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
LValue SourceLV = CGF.EmitLValue(E->getSubExpr());
|
|
|
|
Address SourceAddress =
|
|
|
|
Builder.CreateElementBitCast(SourceLV.getAddress(), CGF.Int8Ty);
|
|
|
|
Address DestAddress =
|
|
|
|
Builder.CreateElementBitCast(Dest.getAddress(), CGF.Int8Ty);
|
|
|
|
llvm::Value *SizeVal = llvm::ConstantInt::get(
|
|
|
|
CGF.SizeTy,
|
|
|
|
CGF.getContext().getTypeSizeInChars(E->getType()).getQuantity());
|
|
|
|
Builder.CreateMemCpy(DestAddress, SourceAddress, SizeVal);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2010-08-25 19:45:40 +08:00
|
|
|
case CK_DerivedToBase:
|
|
|
|
case CK_BaseToDerived:
|
|
|
|
case CK_UncheckedDerivedToBase: {
|
2011-09-23 13:06:16 +08:00
|
|
|
llvm_unreachable("cannot perform hierarchy conversion in EmitAggExpr: "
|
2010-05-22 13:17:18 +08:00
|
|
|
"should have been unpacked before we got here");
|
|
|
|
}
|
|
|
|
|
2013-03-08 05:37:17 +08:00
|
|
|
case CK_NonAtomicToAtomic:
|
|
|
|
case CK_AtomicToNonAtomic: {
|
|
|
|
bool isToAtomic = (E->getCastKind() == CK_NonAtomicToAtomic);
|
|
|
|
|
|
|
|
// Determine the atomic and value types.
|
|
|
|
QualType atomicType = E->getSubExpr()->getType();
|
|
|
|
QualType valueType = E->getType();
|
|
|
|
if (isToAtomic) std::swap(atomicType, valueType);
|
|
|
|
|
|
|
|
assert(atomicType->isAtomicType());
|
|
|
|
assert(CGF.getContext().hasSameUnqualifiedType(valueType,
|
|
|
|
atomicType->castAs<AtomicType>()->getValueType()));
|
|
|
|
|
|
|
|
// Just recurse normally if we're ignoring the result or the
|
|
|
|
// atomic type doesn't change representation.
|
|
|
|
if (Dest.isIgnored() || !CGF.CGM.isPaddedAtomicType(atomicType)) {
|
|
|
|
return Visit(E->getSubExpr());
|
|
|
|
}
|
|
|
|
|
|
|
|
CastKind peepholeTarget =
|
|
|
|
(isToAtomic ? CK_AtomicToNonAtomic : CK_NonAtomicToAtomic);
|
|
|
|
|
|
|
|
// These two cases are reverses of each other; try to peephole them.
|
|
|
|
if (Expr *op = findPeephole(E->getSubExpr(), peepholeTarget)) {
|
|
|
|
assert(CGF.getContext().hasSameUnqualifiedType(op->getType(),
|
|
|
|
E->getType()) &&
|
|
|
|
"peephole significantly changed types?");
|
|
|
|
return Visit(op);
|
|
|
|
}
|
|
|
|
|
|
|
|
// If we're converting an r-value of non-atomic type to an r-value
|
2013-07-11 09:32:21 +08:00
|
|
|
// of atomic type, just emit directly into the relevant sub-object.
|
2013-03-08 05:37:17 +08:00
|
|
|
if (isToAtomic) {
|
2013-07-11 09:32:21 +08:00
|
|
|
AggValueSlot valueDest = Dest;
|
|
|
|
if (!valueDest.isIgnored() && CGF.CGM.isPaddedAtomicType(atomicType)) {
|
2018-04-06 23:14:32 +08:00
|
|
|
// Zero-initialize. (Strictly speaking, we only need to initialize
|
2013-07-11 09:32:21 +08:00
|
|
|
// the padding at the end, but this is simpler.)
|
|
|
|
if (!Dest.isZeroed())
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
CGF.EmitNullInitialization(Dest.getAddress(), atomicType);
|
2013-07-11 09:32:21 +08:00
|
|
|
|
|
|
|
// Build a GEP to refer to the subobject.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Address valueAddr =
|
2019-02-10 06:22:28 +08:00
|
|
|
CGF.Builder.CreateStructGEP(valueDest.getAddress(), 0);
|
2013-07-11 09:32:21 +08:00
|
|
|
valueDest = AggValueSlot::forAddr(valueAddr,
|
|
|
|
valueDest.getQualifiers(),
|
|
|
|
valueDest.isExternallyDestructed(),
|
|
|
|
valueDest.requiresGCollection(),
|
|
|
|
valueDest.isPotentiallyAliased(),
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::DoesNotOverlap,
|
2013-07-11 09:32:21 +08:00
|
|
|
AggValueSlot::IsZeroed);
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2013-07-11 10:28:36 +08:00
|
|
|
CGF.EmitAggExpr(E->getSubExpr(), valueDest);
|
2013-03-08 05:37:17 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Otherwise, we're converting an atomic type to a non-atomic type.
|
2013-07-11 09:32:21 +08:00
|
|
|
// Make an atomic temporary, emit into that, and then copy the value out.
|
2013-03-08 05:37:17 +08:00
|
|
|
AggValueSlot atomicSlot =
|
|
|
|
CGF.CreateAggTemp(atomicType, "atomic-to-nonatomic.temp");
|
|
|
|
CGF.EmitAggExpr(E->getSubExpr(), atomicSlot);
|
|
|
|
|
2019-02-10 06:22:28 +08:00
|
|
|
Address valueAddr = Builder.CreateStructGEP(atomicSlot.getAddress(), 0);
|
2013-03-08 05:37:17 +08:00
|
|
|
RValue rvalue = RValue::getAggregate(valueAddr, atomicSlot.isVolatile());
|
|
|
|
return EmitFinalDestCopy(valueType, rvalue);
|
|
|
|
}
|
2019-04-04 18:48:36 +08:00
|
|
|
case CK_AddressSpaceConversion:
|
|
|
|
return Visit(E->getSubExpr());
|
2013-03-08 05:37:17 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
case CK_LValueToRValue:
|
|
|
|
// If we're loading from a volatile type, force the destination
|
|
|
|
// into existence.
|
|
|
|
if (E->getSubExpr()->getType().isVolatileQualified()) {
|
|
|
|
EnsureDest(E->getType());
|
|
|
|
return Visit(E->getSubExpr());
|
|
|
|
}
|
2013-03-08 05:37:17 +08:00
|
|
|
|
2017-12-20 06:06:11 +08:00
|
|
|
LLVM_FALLTHROUGH;
|
2012-07-03 07:58:38 +08:00
|
|
|
|
2019-04-04 18:48:36 +08:00
|
|
|
|
2010-08-25 19:45:40 +08:00
|
|
|
case CK_NoOp:
|
|
|
|
case CK_UserDefinedConversion:
|
|
|
|
case CK_ConstructorConversion:
|
2009-09-29 09:23:39 +08:00
|
|
|
assert(CGF.getContext().hasSameUnqualifiedType(E->getSubExpr()->getType(),
|
|
|
|
E->getType()) &&
|
|
|
|
"Implicit cast types must be compatible");
|
|
|
|
Visit(E->getSubExpr());
|
|
|
|
break;
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2010-08-25 19:45:40 +08:00
|
|
|
case CK_LValueBitCast:
|
2010-12-01 12:43:34 +08:00
|
|
|
llvm_unreachable("should not be emitting lvalue bitcast as rvalue");
|
2011-04-07 16:22:57 +08:00
|
|
|
|
2010-12-01 12:43:34 +08:00
|
|
|
case CK_Dependent:
|
|
|
|
case CK_BitCast:
|
|
|
|
case CK_ArrayToPointerDecay:
|
|
|
|
case CK_FunctionToPointerDecay:
|
|
|
|
case CK_NullToPointer:
|
|
|
|
case CK_NullToMemberPointer:
|
|
|
|
case CK_BaseToDerivedMemberPointer:
|
|
|
|
case CK_DerivedToBaseMemberPointer:
|
|
|
|
case CK_MemberPointerToBoolean:
|
2012-02-15 09:22:51 +08:00
|
|
|
case CK_ReinterpretMemberPointer:
|
2010-12-01 12:43:34 +08:00
|
|
|
case CK_IntegralToPointer:
|
|
|
|
case CK_PointerToIntegral:
|
|
|
|
case CK_PointerToBoolean:
|
|
|
|
case CK_ToVoid:
|
|
|
|
case CK_VectorSplat:
|
|
|
|
case CK_IntegralCast:
|
2016-01-13 09:52:39 +08:00
|
|
|
case CK_BooleanToSignedIntegral:
|
2010-12-01 12:43:34 +08:00
|
|
|
case CK_IntegralToBoolean:
|
|
|
|
case CK_IntegralToFloating:
|
|
|
|
case CK_FloatingToIntegral:
|
|
|
|
case CK_FloatingToBoolean:
|
|
|
|
case CK_FloatingCast:
|
2011-09-09 13:25:32 +08:00
|
|
|
case CK_CPointerToObjCPointerCast:
|
|
|
|
case CK_BlockPointerToObjCPointerCast:
|
2010-12-01 12:43:34 +08:00
|
|
|
case CK_AnyPointerToBlockPointerCast:
|
|
|
|
case CK_ObjCObjectLValueCast:
|
|
|
|
case CK_FloatingRealToComplex:
|
|
|
|
case CK_FloatingComplexToReal:
|
|
|
|
case CK_FloatingComplexToBoolean:
|
|
|
|
case CK_FloatingComplexCast:
|
|
|
|
case CK_FloatingComplexToIntegralComplex:
|
|
|
|
case CK_IntegralRealToComplex:
|
|
|
|
case CK_IntegralComplexToReal:
|
|
|
|
case CK_IntegralComplexToBoolean:
|
|
|
|
case CK_IntegralComplexCast:
|
|
|
|
case CK_IntegralComplexToFloatingComplex:
|
2011-09-10 14:18:15 +08:00
|
|
|
case CK_ARCProduceObject:
|
|
|
|
case CK_ARCConsumeObject:
|
|
|
|
case CK_ARCReclaimReturnedObject:
|
|
|
|
case CK_ARCExtendBlockObject:
|
2012-02-22 13:02:47 +08:00
|
|
|
case CK_CopyAndAutoreleaseBlockObject:
|
2012-08-31 08:14:07 +08:00
|
|
|
case CK_BuiltinFnToFnPtr:
|
2018-10-23 23:19:20 +08:00
|
|
|
case CK_ZeroToOCLOpaqueType:
|
2019-04-04 18:48:36 +08:00
|
|
|
|
2016-07-29 03:26:30 +08:00
|
|
|
case CK_IntToOCLSampler:
|
2018-10-16 00:07:02 +08:00
|
|
|
case CK_FixedPointCast:
|
2018-10-24 01:55:35 +08:00
|
|
|
case CK_FixedPointToBoolean:
|
2019-03-06 08:28:43 +08:00
|
|
|
case CK_FixedPointToIntegral:
|
|
|
|
case CK_IntegralToFixedPoint:
|
2010-12-01 12:43:34 +08:00
|
|
|
llvm_unreachable("cast kind invalid for aggregate types");
|
2009-09-29 09:23:39 +08:00
|
|
|
}
|
2008-01-14 14:28:57 +08:00
|
|
|
}
|
|
|
|
|
2008-07-27 06:37:01 +08:00
|
|
|
void AggExprEmitter::VisitCallExpr(const CallExpr *E) {
|
2015-02-26 01:36:15 +08:00
|
|
|
if (E->getCallReturnType(CGF.getContext())->isReferenceType()) {
|
2009-05-28 00:45:02 +08:00
|
|
|
EmitAggLoadOfLValue(E);
|
|
|
|
return;
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2018-03-10 09:11:17 +08:00
|
|
|
withReturnValueSlot(E, [&](ReturnValueSlot Slot) {
|
|
|
|
return CGF.EmitCallExpr(E, Slot);
|
|
|
|
});
|
2008-01-31 13:38:29 +08:00
|
|
|
}
|
2008-07-27 06:37:01 +08:00
|
|
|
|
|
|
|
void AggExprEmitter::VisitObjCMessageExpr(ObjCMessageExpr *E) {
|
2018-03-10 09:11:17 +08:00
|
|
|
withReturnValueSlot(E, [&](ReturnValueSlot Slot) {
|
|
|
|
return CGF.EmitObjCMessageExpr(E, Slot);
|
|
|
|
});
|
2008-06-25 01:04:18 +08:00
|
|
|
}
|
2008-01-31 13:38:29 +08:00
|
|
|
|
2008-07-27 06:37:01 +08:00
|
|
|
void AggExprEmitter::VisitBinComma(const BinaryOperator *E) {
|
2010-12-05 10:00:02 +08:00
|
|
|
CGF.EmitIgnoredExpr(E->getLHS());
|
2010-09-15 18:14:12 +08:00
|
|
|
Visit(E->getRHS());
|
2008-05-20 15:56:31 +08:00
|
|
|
}
|
|
|
|
|
2007-09-01 06:54:14 +08:00
|
|
|
void AggExprEmitter::VisitStmtExpr(const StmtExpr *E) {
|
2011-01-26 12:00:11 +08:00
|
|
|
CodeGenFunction::StmtExprEvaluation eval(CGF);
|
2010-09-15 18:14:12 +08:00
|
|
|
CGF.EmitCompoundStmt(*E->getSubStmt(), true, Dest);
|
2007-09-01 06:54:14 +08:00
|
|
|
}
|
|
|
|
|
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
Reviewers: rsmith, aaron.ballman, majnemer, rnk, compnerd, rjmccall
Reviewed By: rjmccall
Subscribers: rjmccall, rsmith, aaron.ballman, junbuml, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D45476
llvm-svn: 331677
2018-05-08 05:07:10 +08:00
|
|
|
enum CompareKind {
|
|
|
|
CK_Less,
|
|
|
|
CK_Greater,
|
|
|
|
CK_Equal,
|
|
|
|
};
|
|
|
|
|
|
|
|
static llvm::Value *EmitCompare(CGBuilderTy &Builder, CodeGenFunction &CGF,
|
|
|
|
const BinaryOperator *E, llvm::Value *LHS,
|
|
|
|
llvm::Value *RHS, CompareKind Kind,
|
|
|
|
const char *NameSuffix = "") {
|
|
|
|
QualType ArgTy = E->getLHS()->getType();
|
|
|
|
if (const ComplexType *CT = ArgTy->getAs<ComplexType>())
|
|
|
|
ArgTy = CT->getElementType();
|
|
|
|
|
|
|
|
if (const auto *MPT = ArgTy->getAs<MemberPointerType>()) {
|
|
|
|
assert(Kind == CK_Equal &&
|
|
|
|
"member pointers may only be compared for equality");
|
|
|
|
return CGF.CGM.getCXXABI().EmitMemberPointerComparison(
|
|
|
|
CGF, LHS, RHS, MPT, /*IsInequality*/ false);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Compute the comparison instructions for the specified comparison kind.
|
|
|
|
struct CmpInstInfo {
|
|
|
|
const char *Name;
|
|
|
|
llvm::CmpInst::Predicate FCmp;
|
|
|
|
llvm::CmpInst::Predicate SCmp;
|
|
|
|
llvm::CmpInst::Predicate UCmp;
|
|
|
|
};
|
|
|
|
CmpInstInfo InstInfo = [&]() -> CmpInstInfo {
|
|
|
|
using FI = llvm::FCmpInst;
|
|
|
|
using II = llvm::ICmpInst;
|
|
|
|
switch (Kind) {
|
|
|
|
case CK_Less:
|
|
|
|
return {"cmp.lt", FI::FCMP_OLT, II::ICMP_SLT, II::ICMP_ULT};
|
|
|
|
case CK_Greater:
|
|
|
|
return {"cmp.gt", FI::FCMP_OGT, II::ICMP_SGT, II::ICMP_UGT};
|
|
|
|
case CK_Equal:
|
|
|
|
return {"cmp.eq", FI::FCMP_OEQ, II::ICMP_EQ, II::ICMP_EQ};
|
|
|
|
}
|
2018-05-08 17:40:32 +08:00
|
|
|
llvm_unreachable("Unrecognised CompareKind enum");
|
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
Reviewers: rsmith, aaron.ballman, majnemer, rnk, compnerd, rjmccall
Reviewed By: rjmccall
Subscribers: rjmccall, rsmith, aaron.ballman, junbuml, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D45476
llvm-svn: 331677
2018-05-08 05:07:10 +08:00
|
|
|
}();
|
|
|
|
|
|
|
|
if (ArgTy->hasFloatingRepresentation())
|
|
|
|
return Builder.CreateFCmp(InstInfo.FCmp, LHS, RHS,
|
|
|
|
llvm::Twine(InstInfo.Name) + NameSuffix);
|
|
|
|
if (ArgTy->isIntegralOrEnumerationType() || ArgTy->isPointerType()) {
|
|
|
|
auto Inst =
|
|
|
|
ArgTy->hasSignedIntegerRepresentation() ? InstInfo.SCmp : InstInfo.UCmp;
|
|
|
|
return Builder.CreateICmp(Inst, LHS, RHS,
|
|
|
|
llvm::Twine(InstInfo.Name) + NameSuffix);
|
|
|
|
}
|
|
|
|
|
|
|
|
llvm_unreachable("unsupported aggregate binary expression should have "
|
|
|
|
"already been handled");
|
|
|
|
}
|
|
|
|
|
|
|
|
void AggExprEmitter::VisitBinCmp(const BinaryOperator *E) {
|
|
|
|
using llvm::BasicBlock;
|
|
|
|
using llvm::PHINode;
|
|
|
|
using llvm::Value;
|
|
|
|
assert(CGF.getContext().hasSameType(E->getLHS()->getType(),
|
|
|
|
E->getRHS()->getType()));
|
|
|
|
const ComparisonCategoryInfo &CmpInfo =
|
|
|
|
CGF.getContext().CompCategories.getInfoForType(E->getType());
|
|
|
|
assert(CmpInfo.Record->isTriviallyCopyable() &&
|
|
|
|
"cannot copy non-trivially copyable aggregate");
|
|
|
|
|
|
|
|
QualType ArgTy = E->getLHS()->getType();
|
|
|
|
|
|
|
|
// TODO: Handle comparing these types.
|
|
|
|
if (ArgTy->isVectorType())
|
|
|
|
return CGF.ErrorUnsupported(
|
|
|
|
E, "aggregate three-way comparison with vector arguments");
|
|
|
|
if (!ArgTy->isIntegralOrEnumerationType() && !ArgTy->isRealFloatingType() &&
|
|
|
|
!ArgTy->isNullPtrType() && !ArgTy->isPointerType() &&
|
|
|
|
!ArgTy->isMemberPointerType() && !ArgTy->isAnyComplexType()) {
|
2018-05-08 08:52:19 +08:00
|
|
|
return CGF.ErrorUnsupported(E, "aggregate three-way comparison");
|
[C++2a] Implement operator<=> CodeGen and ExprConstant
Summary:
This patch tackles long hanging fruit for the builtin operator<=> expressions. It is currently needs some cleanup before landing, but I want to get some initial feedback.
The main changes are:
* Lookup, build, and store the required standard library types and expressions in `ASTContext`. By storing them in ASTContext we don't need to store (and duplicate) the required expressions in the BinaryOperator AST nodes.
* Implement [expr.spaceship] checking, including diagnosing narrowing conversions.
* Implement `ExprConstant` for builtin spaceship operators.
* Implement builitin operator<=> support in `CodeGenAgg`. Initially I emitted the required comparisons using `ScalarExprEmitter::VisitBinaryOperator`, but this caused the operand expressions to be emitted once for every required cmp.
* Implement [builtin.over] with modifications to support the intent of P0946R0. See the note on `BuiltinOperatorOverloadBuilder::addThreeWayArithmeticOverloads` for more information about the workaround.
Reviewers: rsmith, aaron.ballman, majnemer, rnk, compnerd, rjmccall
Reviewed By: rjmccall
Subscribers: rjmccall, rsmith, aaron.ballman, junbuml, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D45476
llvm-svn: 331677
2018-05-08 05:07:10 +08:00
|
|
|
}
|
|
|
|
bool IsComplex = ArgTy->isAnyComplexType();
|
|
|
|
|
|
|
|
// Evaluate the operands to the expression and extract their values.
|
|
|
|
auto EmitOperand = [&](Expr *E) -> std::pair<Value *, Value *> {
|
|
|
|
RValue RV = CGF.EmitAnyExpr(E);
|
|
|
|
if (RV.isScalar())
|
|
|
|
return {RV.getScalarVal(), nullptr};
|
|
|
|
if (RV.isAggregate())
|
|
|
|
return {RV.getAggregatePointer(), nullptr};
|
|
|
|
assert(RV.isComplex());
|
|
|
|
return RV.getComplexVal();
|
|
|
|
};
|
|
|
|
auto LHSValues = EmitOperand(E->getLHS()),
|
|
|
|
RHSValues = EmitOperand(E->getRHS());
|
|
|
|
|
|
|
|
auto EmitCmp = [&](CompareKind K) {
|
|
|
|
Value *Cmp = EmitCompare(Builder, CGF, E, LHSValues.first, RHSValues.first,
|
|
|
|
K, IsComplex ? ".r" : "");
|
|
|
|
if (!IsComplex)
|
|
|
|
return Cmp;
|
|
|
|
assert(K == CompareKind::CK_Equal);
|
|
|
|
Value *CmpImag = EmitCompare(Builder, CGF, E, LHSValues.second,
|
|
|
|
RHSValues.second, K, ".i");
|
|
|
|
return Builder.CreateAnd(Cmp, CmpImag, "and.eq");
|
|
|
|
};
|
|
|
|
auto EmitCmpRes = [&](const ComparisonCategoryInfo::ValueInfo *VInfo) {
|
|
|
|
return Builder.getInt(VInfo->getIntValue());
|
|
|
|
};
|
|
|
|
|
|
|
|
Value *Select;
|
|
|
|
if (ArgTy->isNullPtrType()) {
|
|
|
|
Select = EmitCmpRes(CmpInfo.getEqualOrEquiv());
|
|
|
|
} else if (CmpInfo.isEquality()) {
|
|
|
|
Select = Builder.CreateSelect(
|
|
|
|
EmitCmp(CK_Equal), EmitCmpRes(CmpInfo.getEqualOrEquiv()),
|
|
|
|
EmitCmpRes(CmpInfo.getNonequalOrNonequiv()), "sel.eq");
|
|
|
|
} else if (!CmpInfo.isPartial()) {
|
|
|
|
Value *SelectOne =
|
|
|
|
Builder.CreateSelect(EmitCmp(CK_Less), EmitCmpRes(CmpInfo.getLess()),
|
|
|
|
EmitCmpRes(CmpInfo.getGreater()), "sel.lt");
|
|
|
|
Select = Builder.CreateSelect(EmitCmp(CK_Equal),
|
|
|
|
EmitCmpRes(CmpInfo.getEqualOrEquiv()),
|
|
|
|
SelectOne, "sel.eq");
|
|
|
|
} else {
|
|
|
|
Value *SelectEq = Builder.CreateSelect(
|
|
|
|
EmitCmp(CK_Equal), EmitCmpRes(CmpInfo.getEqualOrEquiv()),
|
|
|
|
EmitCmpRes(CmpInfo.getUnordered()), "sel.eq");
|
|
|
|
Value *SelectGT = Builder.CreateSelect(EmitCmp(CK_Greater),
|
|
|
|
EmitCmpRes(CmpInfo.getGreater()),
|
|
|
|
SelectEq, "sel.gt");
|
|
|
|
Select = Builder.CreateSelect(
|
|
|
|
EmitCmp(CK_Less), EmitCmpRes(CmpInfo.getLess()), SelectGT, "sel.lt");
|
|
|
|
}
|
|
|
|
// Create the return value in the destination slot.
|
|
|
|
EnsureDest(E->getType());
|
|
|
|
LValue DestLV = CGF.MakeAddrLValue(Dest.getAddress(), E->getType());
|
|
|
|
|
|
|
|
// Emit the address of the first (and only) field in the comparison category
|
|
|
|
// type, and initialize it from the constant integer value selected above.
|
|
|
|
LValue FieldLV = CGF.EmitLValueForFieldInitialization(
|
|
|
|
DestLV, *CmpInfo.Record->field_begin());
|
|
|
|
CGF.EmitStoreThroughLValue(RValue::get(Select), FieldLV, /*IsInit*/ true);
|
|
|
|
|
|
|
|
// All done! The result is in the Dest slot.
|
|
|
|
}
|
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
void AggExprEmitter::VisitBinaryOperator(const BinaryOperator *E) {
|
2010-08-25 19:45:40 +08:00
|
|
|
if (E->getOpcode() == BO_PtrMemD || E->getOpcode() == BO_PtrMemI)
|
2009-10-23 06:57:31 +08:00
|
|
|
VisitPointerToDataMemberBinaryOperator(E);
|
|
|
|
else
|
|
|
|
CGF.ErrorUnsupported(E, "aggregate binary expression");
|
|
|
|
}
|
|
|
|
|
|
|
|
void AggExprEmitter::VisitPointerToDataMemberBinaryOperator(
|
|
|
|
const BinaryOperator *E) {
|
|
|
|
LValue LV = CGF.EmitPointerToDataMemberBinaryExpr(E);
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitFinalDestCopy(E->getType(), LV);
|
|
|
|
}
|
|
|
|
|
|
|
|
/// Is the value of the given expression possibly a reference to or
|
|
|
|
/// into a __block variable?
|
|
|
|
static bool isBlockVarRef(const Expr *E) {
|
|
|
|
// Make sure we look through parens.
|
|
|
|
E = E->IgnoreParens();
|
|
|
|
|
|
|
|
// Check for a direct reference to a __block variable.
|
|
|
|
if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
|
|
|
|
const VarDecl *var = dyn_cast<VarDecl>(DRE->getDecl());
|
|
|
|
return (var && var->hasAttr<BlocksAttr>());
|
|
|
|
}
|
|
|
|
|
|
|
|
// More complicated stuff.
|
|
|
|
|
|
|
|
// Binary operators.
|
|
|
|
if (const BinaryOperator *op = dyn_cast<BinaryOperator>(E)) {
|
|
|
|
// For an assignment or pointer-to-member operation, just care
|
|
|
|
// about the LHS.
|
|
|
|
if (op->isAssignmentOp() || op->isPtrMemOp())
|
|
|
|
return isBlockVarRef(op->getLHS());
|
|
|
|
|
|
|
|
// For a comma, just care about the RHS.
|
|
|
|
if (op->getOpcode() == BO_Comma)
|
|
|
|
return isBlockVarRef(op->getRHS());
|
|
|
|
|
|
|
|
// FIXME: pointer arithmetic?
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// Check both sides of a conditional operator.
|
|
|
|
} else if (const AbstractConditionalOperator *op
|
|
|
|
= dyn_cast<AbstractConditionalOperator>(E)) {
|
|
|
|
return isBlockVarRef(op->getTrueExpr())
|
|
|
|
|| isBlockVarRef(op->getFalseExpr());
|
|
|
|
|
|
|
|
// OVEs are required to support BinaryConditionalOperators.
|
|
|
|
} else if (const OpaqueValueExpr *op
|
|
|
|
= dyn_cast<OpaqueValueExpr>(E)) {
|
|
|
|
if (const Expr *src = op->getSourceExpr())
|
|
|
|
return isBlockVarRef(src);
|
|
|
|
|
|
|
|
// Casts are necessary to get things like (*(int*)&var) = foo().
|
|
|
|
// We don't really care about the kind of cast here, except
|
|
|
|
// we don't want to look through l2r casts, because it's okay
|
|
|
|
// to get the *value* in a __block variable.
|
|
|
|
} else if (const CastExpr *cast = dyn_cast<CastExpr>(E)) {
|
|
|
|
if (cast->getCastKind() == CK_LValueToRValue)
|
|
|
|
return false;
|
|
|
|
return isBlockVarRef(cast->getSubExpr());
|
|
|
|
|
|
|
|
// Handle unary operators. Again, just aggressively look through
|
|
|
|
// it, ignoring the operation.
|
|
|
|
} else if (const UnaryOperator *uop = dyn_cast<UnaryOperator>(E)) {
|
|
|
|
return isBlockVarRef(uop->getSubExpr());
|
|
|
|
|
|
|
|
// Look into the base of a field access.
|
|
|
|
} else if (const MemberExpr *mem = dyn_cast<MemberExpr>(E)) {
|
|
|
|
return isBlockVarRef(mem->getBase());
|
|
|
|
|
|
|
|
// Look into the base of a subscript.
|
|
|
|
} else if (const ArraySubscriptExpr *sub = dyn_cast<ArraySubscriptExpr>(E)) {
|
|
|
|
return isBlockVarRef(sub->getBase());
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
2007-08-21 12:59:27 +08:00
|
|
|
}
|
|
|
|
|
2007-08-21 12:43:17 +08:00
|
|
|
void AggExprEmitter::VisitBinAssign(const BinaryOperator *E) {
|
2008-02-11 09:09:17 +08:00
|
|
|
// For an assignment to work, the value on the right has
|
|
|
|
// to be compatible with the value on the left.
|
2009-05-29 07:04:00 +08:00
|
|
|
assert(CGF.getContext().hasSameUnqualifiedType(E->getLHS()->getType(),
|
|
|
|
E->getRHS()->getType())
|
2008-02-11 09:09:17 +08:00
|
|
|
&& "Invalid assignment");
|
2010-12-06 14:10:02 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
// If the LHS might be a __block variable, and the RHS can
|
|
|
|
// potentially cause a block copy, we need to evaluate the RHS first
|
|
|
|
// so that the assignment goes the right place.
|
|
|
|
// This is pretty semantically fragile.
|
|
|
|
if (isBlockVarRef(E->getLHS()) &&
|
|
|
|
E->getRHS()->HasSideEffects(CGF.getContext())) {
|
|
|
|
// Ensure that we have a destination, and evaluate the RHS into that.
|
|
|
|
EnsureDest(E->getRHS()->getType());
|
|
|
|
Visit(E->getRHS());
|
|
|
|
|
|
|
|
// Now emit the LHS and copy into it.
|
2012-10-10 03:52:38 +08:00
|
|
|
LValue LHS = CGF.EmitCheckedLValue(E->getLHS(), CodeGenFunction::TCK_Store);
|
2012-07-03 07:58:38 +08:00
|
|
|
|
2013-03-08 05:37:17 +08:00
|
|
|
// That copy is an atomic copy if the LHS is atomic.
|
2015-02-14 09:35:12 +08:00
|
|
|
if (LHS.getType()->isAtomicType() ||
|
|
|
|
CGF.LValueIsSuitableForInlineAtomic(LHS)) {
|
2013-03-08 05:37:17 +08:00
|
|
|
CGF.EmitAtomicStore(Dest.asRValue(), LHS, /*isInit*/ false);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitCopy(E->getLHS()->getType(),
|
|
|
|
AggValueSlot::forLValue(LHS, AggValueSlot::IsDestructed,
|
|
|
|
needsGC(E->getLHS()->getType()),
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::IsAliased,
|
|
|
|
AggValueSlot::MayOverlap),
|
2012-07-03 07:58:38 +08:00
|
|
|
Dest);
|
|
|
|
return;
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
LValue LHS = CGF.EmitLValue(E->getLHS());
|
2007-08-11 08:04:45 +08:00
|
|
|
|
2013-03-08 05:37:17 +08:00
|
|
|
// If we have an atomic type, evaluate into the destination and then
|
|
|
|
// do an atomic copy.
|
2015-02-14 09:35:12 +08:00
|
|
|
if (LHS.getType()->isAtomicType() ||
|
|
|
|
CGF.LValueIsSuitableForInlineAtomic(LHS)) {
|
2013-03-08 05:37:17 +08:00
|
|
|
EnsureDest(E->getRHS()->getType());
|
|
|
|
Visit(E->getRHS());
|
|
|
|
CGF.EmitAtomicStore(Dest.asRValue(), LHS, /*isInit*/ false);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2011-11-07 11:59:57 +08:00
|
|
|
// Codegen the RHS so that it stores directly into the LHS.
|
|
|
|
AggValueSlot LHSSlot =
|
2018-07-31 03:24:48 +08:00
|
|
|
AggValueSlot::forLValue(LHS, AggValueSlot::IsDestructed,
|
2011-11-07 11:59:57 +08:00
|
|
|
needsGC(E->getLHS()->getType()),
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::IsAliased,
|
|
|
|
AggValueSlot::MayOverlap);
|
2013-01-26 07:57:05 +08:00
|
|
|
// A non-volatile aggregate destination might have volatile member.
|
|
|
|
if (!LHSSlot.isVolatile() &&
|
|
|
|
CGF.hasVolatileMember(E->getLHS()->getType()))
|
|
|
|
LHSSlot.setVolatile(true);
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
CGF.EmitAggExpr(E->getRHS(), LHSSlot);
|
|
|
|
|
|
|
|
// Copy into the destination if the assignment isn't ignored.
|
|
|
|
EmitFinalDestCopy(E->getType(), LHS);
|
2007-08-11 08:04:45 +08:00
|
|
|
}
|
|
|
|
|
2011-02-17 18:25:35 +08:00
|
|
|
void AggExprEmitter::
|
|
|
|
VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
|
2008-11-13 09:38:36 +08:00
|
|
|
llvm::BasicBlock *LHSBlock = CGF.createBasicBlock("cond.true");
|
|
|
|
llvm::BasicBlock *RHSBlock = CGF.createBasicBlock("cond.false");
|
|
|
|
llvm::BasicBlock *ContBlock = CGF.createBasicBlock("cond.end");
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2011-02-17 18:25:35 +08:00
|
|
|
// Bind the common expression if necessary.
|
2012-01-07 04:42:20 +08:00
|
|
|
CodeGenFunction::OpaqueValueMapping binding(CGF, E);
|
2011-02-17 18:25:35 +08:00
|
|
|
|
2011-01-26 12:00:11 +08:00
|
|
|
CodeGenFunction::ConditionalEvaluation eval(CGF);
|
2015-04-24 07:06:47 +08:00
|
|
|
CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock,
|
|
|
|
CGF.getProfileCount(E));
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-11-17 08:07:33 +08:00
|
|
|
// Save whether the destination's lifetime is externally managed.
|
2011-08-26 16:02:37 +08:00
|
|
|
bool isExternallyDestructed = Dest.isExternallyDestructed();
|
2007-08-11 08:04:45 +08:00
|
|
|
|
2011-01-26 12:00:11 +08:00
|
|
|
eval.begin(CGF);
|
|
|
|
CGF.EmitBlock(LHSBlock);
|
2015-04-24 07:06:47 +08:00
|
|
|
CGF.incrementProfileCounter(E);
|
2011-02-17 18:25:35 +08:00
|
|
|
Visit(E->getTrueExpr());
|
2011-01-26 12:00:11 +08:00
|
|
|
eval.end(CGF);
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2011-01-26 12:00:11 +08:00
|
|
|
assert(CGF.HaveInsertPoint() && "expression evaluation ended with no IP!");
|
|
|
|
CGF.Builder.CreateBr(ContBlock);
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-11-17 08:07:33 +08:00
|
|
|
// If the result of an agg expression is unused, then the emission
|
|
|
|
// of the LHS might need to create a destination slot. That's fine
|
|
|
|
// with us, and we can safely emit the RHS into the same slot, but
|
2011-08-26 16:02:37 +08:00
|
|
|
// we shouldn't claim that it's already being destructed.
|
|
|
|
Dest.setExternallyDestructed(isExternallyDestructed);
|
2010-11-17 08:07:33 +08:00
|
|
|
|
2011-01-26 12:00:11 +08:00
|
|
|
eval.begin(CGF);
|
|
|
|
CGF.EmitBlock(RHSBlock);
|
2011-02-17 18:25:35 +08:00
|
|
|
Visit(E->getFalseExpr());
|
2011-01-26 12:00:11 +08:00
|
|
|
eval.end(CGF);
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2007-08-21 12:25:47 +08:00
|
|
|
CGF.EmitBlock(ContBlock);
|
2007-08-11 08:04:45 +08:00
|
|
|
}
|
2007-08-21 12:59:27 +08:00
|
|
|
|
2009-07-09 02:33:14 +08:00
|
|
|
void AggExprEmitter::VisitChooseExpr(const ChooseExpr *CE) {
|
2013-07-20 08:40:58 +08:00
|
|
|
Visit(CE->getChosenSubExpr());
|
2009-07-09 02:33:14 +08:00
|
|
|
}
|
|
|
|
|
2008-05-27 23:51:49 +08:00
|
|
|
void AggExprEmitter::VisitVAArgExpr(VAArgExpr *VE) {
|
2015-09-18 04:55:33 +08:00
|
|
|
Address ArgValue = Address::invalid();
|
|
|
|
Address ArgPtr = CGF.EmitVAArg(VE, ArgValue);
|
2008-11-04 13:30:00 +08:00
|
|
|
|
Default vaarg lowering should support indirect struct types.
Fixes PR11517 for SPARC.
On most targets, clang lowers va_arg itself, eschewing the use of the
llvm vaarg instruction. This is necessary (at least for now) as the type
argument to the vaarg instruction cannot represent all the ABI
information that is needed to support complex calling conventions.
However, on targets with a simpler varrags ABIs, the LLVM instruction
can work just fine, and clang can simply lower to it. Unfortunately,
even on such targets, vaarg with a struct argument would fail, because
the default lowering to vaarg was naive: it didn't take into account the
ABI attribute computed by classifyArgumentType. In particular, for the
DefaultABIInfo, structs are supposed to be passed indirectly and so
llvm's vaarg instruction should be emitted with a pointer argument.
Now, vaarg instruction emission is able to use computed ABIArgInfo for
the provided argument type, which allows the default ABI support to work
for structs too.
I haven't touched the EmitVAArg implementation for PPC32_SVR4 or XCore,
although I believe both are now redundant, and could be switched over to
use the default implementation as well.
Differential Revision: http://reviews.llvm.org/D16154
llvm-svn: 261717
2016-02-24 10:59:33 +08:00
|
|
|
// If EmitVAArg fails, emit an error.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (!ArgPtr.isValid()) {
|
Default vaarg lowering should support indirect struct types.
Fixes PR11517 for SPARC.
On most targets, clang lowers va_arg itself, eschewing the use of the
llvm vaarg instruction. This is necessary (at least for now) as the type
argument to the vaarg instruction cannot represent all the ABI
information that is needed to support complex calling conventions.
However, on targets with a simpler varrags ABIs, the LLVM instruction
can work just fine, and clang can simply lower to it. Unfortunately,
even on such targets, vaarg with a struct argument would fail, because
the default lowering to vaarg was naive: it didn't take into account the
ABI attribute computed by classifyArgumentType. In particular, for the
DefaultABIInfo, structs are supposed to be passed indirectly and so
llvm's vaarg instruction should be emitted with a pointer argument.
Now, vaarg instruction emission is able to use computed ABIArgInfo for
the provided argument type, which allows the default ABI support to work
for structs too.
I haven't touched the EmitVAArg implementation for PPC32_SVR4 or XCore,
although I believe both are now redundant, and could be switched over to
use the default implementation as well.
Differential Revision: http://reviews.llvm.org/D16154
llvm-svn: 261717
2016-02-24 10:59:33 +08:00
|
|
|
CGF.ErrorUnsupported(VE, "aggregate va_arg expression");
|
2009-01-10 05:09:38 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitFinalDestCopy(VE->getType(), CGF.MakeAddrLValue(ArgPtr, VE->getType()));
|
2008-05-27 23:51:49 +08:00
|
|
|
}
|
|
|
|
|
2009-05-31 07:23:33 +08:00
|
|
|
void AggExprEmitter::VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *E) {
|
2010-09-15 18:14:12 +08:00
|
|
|
// Ensure that we have a slot, but if we already do, remember
|
2011-08-26 16:02:37 +08:00
|
|
|
// whether it was externally destructed.
|
|
|
|
bool wasExternallyDestructed = Dest.isExternallyDestructed();
|
2012-07-03 07:58:38 +08:00
|
|
|
EnsureDest(E->getType());
|
2011-08-26 16:02:37 +08:00
|
|
|
|
|
|
|
// We're going to push a destructor if there isn't already one.
|
|
|
|
Dest.setExternallyDestructed();
|
2010-09-15 18:14:12 +08:00
|
|
|
|
|
|
|
Visit(E->getSubExpr());
|
|
|
|
|
2011-08-26 16:02:37 +08:00
|
|
|
// Push that destructor we promised.
|
|
|
|
if (!wasExternallyDestructed)
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
CGF.EmitCXXTemporary(E->getTemporary(), E->getType(), Dest.getAddress());
|
2009-05-31 07:23:33 +08:00
|
|
|
}
|
|
|
|
|
2009-04-17 08:06:03 +08:00
|
|
|
void
|
2009-05-04 01:47:16 +08:00
|
|
|
AggExprEmitter::VisitCXXConstructExpr(const CXXConstructExpr *E) {
|
2010-09-15 18:14:12 +08:00
|
|
|
AggValueSlot Slot = EnsureSlot(E->getType());
|
|
|
|
CGF.EmitCXXConstructExpr(E, Slot);
|
2009-05-19 12:48:36 +08:00
|
|
|
}
|
|
|
|
|
P0136R1, DR1573, DR1645, DR1715, DR1736, DR1903, DR1941, DR1959, DR1991:
Replace inheriting constructors implementation with new approach, voted into
C++ last year as a DR against C++11.
Instead of synthesizing a set of derived class constructors for each inherited
base class constructor, we make the constructors of the base class visible to
constructor lookup in the derived class, using the normal rules for
using-declarations.
For constructors, UsingShadowDecl now has a ConstructorUsingShadowDecl derived
class that tracks the requisite additional information. We create shadow
constructors (not found by name lookup) in the derived class to model the
actual initialization, and have a new expression node,
CXXInheritedCtorInitExpr, to model the initialization of a base class from such
a constructor. (This initialization is special because it performs real perfect
forwarding of arguments.)
In cases where argument forwarding is not possible (for inalloca calls,
variadic calls, and calls with callee parameter cleanup), the shadow inheriting
constructor is not emitted and instead we directly emit the initialization code
into the caller of the inherited constructor.
Note that this new model is not perfectly compatible with the old model in some
corner cases. In particular:
* if B inherits a private constructor from A, and C uses that constructor to
construct a B, then we previously required that A befriends B and B
befriends C, but the new rules require A to befriend C directly, and
* if a derived class has its own constructors (and so its implicit default
constructor is suppressed), it may still inherit a default constructor from
a base class
llvm-svn: 274049
2016-06-29 03:03:57 +08:00
|
|
|
void AggExprEmitter::VisitCXXInheritedCtorInitExpr(
|
|
|
|
const CXXInheritedCtorInitExpr *E) {
|
|
|
|
AggValueSlot Slot = EnsureSlot(E->getType());
|
|
|
|
CGF.EmitInheritedCXXConstructorCall(
|
|
|
|
E->getConstructor(), E->constructsVBase(), Slot.getAddress(),
|
|
|
|
E->inheritedFromVBase(), E);
|
|
|
|
}
|
|
|
|
|
2012-02-09 11:32:31 +08:00
|
|
|
void
|
|
|
|
AggExprEmitter::VisitLambdaExpr(LambdaExpr *E) {
|
|
|
|
AggValueSlot Slot = EnsureSlot(E->getType());
|
2019-01-18 06:05:50 +08:00
|
|
|
LValue SlotLV = CGF.MakeAddrLValue(Slot.getAddress(), E->getType());
|
|
|
|
|
|
|
|
// We'll need to enter cleanup scopes in case any of the element
|
|
|
|
// initializers throws an exception.
|
|
|
|
SmallVector<EHScopeStack::stable_iterator, 16> Cleanups;
|
|
|
|
llvm::Instruction *CleanupDominator = nullptr;
|
|
|
|
|
|
|
|
CXXRecordDecl::field_iterator CurField = E->getLambdaClass()->field_begin();
|
|
|
|
for (LambdaExpr::const_capture_init_iterator i = E->capture_init_begin(),
|
|
|
|
e = E->capture_init_end();
|
|
|
|
i != e; ++i, ++CurField) {
|
|
|
|
// Emit initialization
|
|
|
|
LValue LV = CGF.EmitLValueForFieldInitialization(SlotLV, *CurField);
|
|
|
|
if (CurField->hasCapturedVLAType()) {
|
|
|
|
CGF.EmitLambdaVLACapture(CurField->getCapturedVLAType(), LV);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
EmitInitializationToLValue(*i, LV);
|
|
|
|
|
|
|
|
// Push a destructor if necessary.
|
|
|
|
if (QualType::DestructionKind DtorKind =
|
|
|
|
CurField->getType().isDestructedType()) {
|
|
|
|
assert(LV.isSimple());
|
|
|
|
if (CGF.needsEHCleanup(DtorKind)) {
|
|
|
|
if (!CleanupDominator)
|
|
|
|
CleanupDominator = CGF.Builder.CreateAlignedLoad(
|
|
|
|
CGF.Int8Ty,
|
|
|
|
llvm::Constant::getNullValue(CGF.Int8PtrTy),
|
|
|
|
CharUnits::One()); // placeholder
|
|
|
|
|
|
|
|
CGF.pushDestroy(EHCleanup, LV.getAddress(), CurField->getType(),
|
|
|
|
CGF.getDestroyer(DtorKind), false);
|
|
|
|
Cleanups.push_back(CGF.EHStack.stable_begin());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Deactivate all the partial cleanups in reverse order, which
|
|
|
|
// generally means popping them.
|
|
|
|
for (unsigned i = Cleanups.size(); i != 0; --i)
|
|
|
|
CGF.DeactivateCleanupBlock(Cleanups[i-1], CleanupDominator);
|
|
|
|
|
|
|
|
// Destroy the placeholder if we made one.
|
|
|
|
if (CleanupDominator)
|
|
|
|
CleanupDominator->eraseFromParent();
|
2012-02-09 11:32:31 +08:00
|
|
|
}
|
|
|
|
|
2010-12-06 16:20:24 +08:00
|
|
|
void AggExprEmitter::VisitExprWithCleanups(ExprWithCleanups *E) {
|
2011-11-10 16:15:53 +08:00
|
|
|
CGF.enterFullExpression(E);
|
|
|
|
CodeGenFunction::RunCleanupsScope cleanups(CGF);
|
|
|
|
Visit(E->getSubExpr());
|
2009-04-17 08:06:03 +08:00
|
|
|
}
|
|
|
|
|
2010-07-08 14:14:04 +08:00
|
|
|
void AggExprEmitter::VisitCXXScalarValueInitExpr(CXXScalarValueInitExpr *E) {
|
2010-09-15 18:14:12 +08:00
|
|
|
QualType T = E->getType();
|
|
|
|
AggValueSlot Slot = EnsureSlot(T);
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
EmitNullInitializationToLValue(CGF.MakeAddrLValue(Slot.getAddress(), T));
|
2009-12-16 14:57:54 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void AggExprEmitter::VisitImplicitValueInitExpr(ImplicitValueInitExpr *E) {
|
2010-09-15 18:14:12 +08:00
|
|
|
QualType T = E->getType();
|
|
|
|
AggValueSlot Slot = EnsureSlot(T);
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
EmitNullInitializationToLValue(CGF.MakeAddrLValue(Slot.getAddress(), T));
|
2009-10-18 23:18:11 +08:00
|
|
|
}
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
/// isSimpleZero - If emitting this value will obviously just cause a store of
|
|
|
|
/// zero to memory, return true. This can return false if uncertain, so it just
|
|
|
|
/// handles simple cases.
|
|
|
|
static bool isSimpleZero(const Expr *E, CodeGenFunction &CGF) {
|
2011-04-15 08:35:48 +08:00
|
|
|
E = E->IgnoreParens();
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// 0
|
|
|
|
if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E))
|
|
|
|
return IL->getValue() == 0;
|
|
|
|
// +0.0
|
|
|
|
if (const FloatingLiteral *FL = dyn_cast<FloatingLiteral>(E))
|
|
|
|
return FL->getValue().isPosZero();
|
|
|
|
// int()
|
|
|
|
if ((isa<ImplicitValueInitExpr>(E) || isa<CXXScalarValueInitExpr>(E)) &&
|
|
|
|
CGF.getTypes().isZeroInitializable(E->getType()))
|
|
|
|
return true;
|
|
|
|
// (int*)0 - Null pointer expressions.
|
|
|
|
if (const CastExpr *ICE = dyn_cast<CastExpr>(E))
|
2016-12-15 16:09:08 +08:00
|
|
|
return ICE->getCastKind() == CK_NullToPointer &&
|
2019-06-15 01:46:38 +08:00
|
|
|
CGF.getTypes().isPointerZeroInitializable(E->getType()) &&
|
|
|
|
!E->HasSideEffects(CGF.getContext());
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// '\0'
|
|
|
|
if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E))
|
|
|
|
return CL->getValue() == 0;
|
2018-07-31 03:24:48 +08:00
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// Otherwise, hard case: conservatively return false.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2018-07-31 03:24:48 +08:00
|
|
|
void
|
2013-10-02 10:29:49 +08:00
|
|
|
AggExprEmitter::EmitInitializationToLValue(Expr *E, LValue LV) {
|
2011-06-16 12:16:24 +08:00
|
|
|
QualType type = LV.getType();
|
2009-05-29 23:46:01 +08:00
|
|
|
// FIXME: Ignore result?
|
2008-04-05 02:42:16 +08:00
|
|
|
// FIXME: Are initializers affected by volatile?
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
if (Dest.isZeroed() && isSimpleZero(E, CGF)) {
|
|
|
|
// Storing "i32 0" to a zero'd memory location is a noop.
|
2013-03-08 05:37:08 +08:00
|
|
|
return;
|
2012-12-21 11:17:28 +08:00
|
|
|
} else if (isa<ImplicitValueInitExpr>(E) || isa<CXXScalarValueInitExpr>(E)) {
|
2013-03-08 05:37:08 +08:00
|
|
|
return EmitNullInitializationToLValue(LV);
|
2015-06-10 08:27:52 +08:00
|
|
|
} else if (isa<NoInitExpr>(E)) {
|
|
|
|
// Do nothing.
|
|
|
|
return;
|
2011-06-16 12:16:24 +08:00
|
|
|
} else if (type->isReferenceType()) {
|
2013-06-13 07:38:09 +08:00
|
|
|
RValue RV = CGF.EmitReferenceBindingToExpr(E);
|
2013-03-08 05:37:08 +08:00
|
|
|
return CGF.EmitStoreThroughLValue(RV, LV);
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2013-03-08 05:37:08 +08:00
|
|
|
switch (CGF.getEvaluationKind(type)) {
|
|
|
|
case TEK_Complex:
|
|
|
|
CGF.EmitComplexExprIntoLValue(E, LV, /*isInit*/ true);
|
|
|
|
return;
|
|
|
|
case TEK_Aggregate:
|
2011-08-26 04:40:09 +08:00
|
|
|
CGF.EmitAggExpr(E, AggValueSlot::forLValue(LV,
|
|
|
|
AggValueSlot::IsDestructed,
|
|
|
|
AggValueSlot::DoesNotNeedGCBarriers,
|
2011-08-26 07:04:34 +08:00
|
|
|
AggValueSlot::IsNotAliased,
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::MayOverlap,
|
2011-06-16 12:16:24 +08:00
|
|
|
Dest.isZeroed()));
|
2013-03-08 05:37:08 +08:00
|
|
|
return;
|
|
|
|
case TEK_Scalar:
|
|
|
|
if (LV.isSimple()) {
|
2014-05-21 13:09:00 +08:00
|
|
|
CGF.EmitScalarInit(E, /*D=*/nullptr, LV, /*Captured=*/false);
|
2013-03-08 05:37:08 +08:00
|
|
|
} else {
|
|
|
|
CGF.EmitStoreThroughLValue(RValue::get(CGF.EmitScalarExpr(E)), LV);
|
|
|
|
}
|
|
|
|
return;
|
2008-04-05 02:42:16 +08:00
|
|
|
}
|
2013-03-08 05:37:08 +08:00
|
|
|
llvm_unreachable("bad evaluation kind");
|
2008-04-05 02:42:16 +08:00
|
|
|
}
|
2008-02-19 06:44:02 +08:00
|
|
|
|
2011-06-16 12:16:24 +08:00
|
|
|
void AggExprEmitter::EmitNullInitializationToLValue(LValue lv) {
|
|
|
|
QualType type = lv.getType();
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// If the destination slot is already zeroed out before the aggregate is
|
|
|
|
// copied into it, we don't have to emit any zeros here.
|
2011-06-16 12:16:24 +08:00
|
|
|
if (Dest.isZeroed() && CGF.getTypes().isZeroInitializable(type))
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
return;
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2013-03-08 05:37:08 +08:00
|
|
|
if (CGF.hasScalarEvaluationKind(type)) {
|
2012-12-21 11:17:28 +08:00
|
|
|
// For non-aggregates, we can store the appropriate null constant.
|
|
|
|
llvm::Value *null = CGF.CGM.EmitNullConstant(type);
|
2012-02-22 13:38:59 +08:00
|
|
|
// Note that the following is not equivalent to
|
|
|
|
// EmitStoreThroughBitfieldLValue for ARC types.
|
2012-02-25 07:53:49 +08:00
|
|
|
if (lv.isBitField()) {
|
2012-02-22 13:38:59 +08:00
|
|
|
CGF.EmitStoreThroughBitfieldLValue(RValue::get(null), lv);
|
2012-02-25 07:53:49 +08:00
|
|
|
} else {
|
|
|
|
assert(lv.isSimple());
|
|
|
|
CGF.EmitStoreOfScalar(null, lv, /* isInitialization */ true);
|
|
|
|
}
|
2008-04-05 02:42:16 +08:00
|
|
|
} else {
|
|
|
|
// There's a potential optimization opportunity in combining
|
|
|
|
// memsets; that would be easy for arrays, but relatively
|
|
|
|
// difficult for structures with the current code.
|
2011-06-16 12:16:24 +08:00
|
|
|
CGF.EmitNullInitialization(lv.getAddress(), lv.getType());
|
2008-04-05 02:42:16 +08:00
|
|
|
}
|
|
|
|
}
|
2008-02-19 06:44:02 +08:00
|
|
|
|
2008-04-05 02:42:16 +08:00
|
|
|
void AggExprEmitter::VisitInitListExpr(InitListExpr *E) {
|
2008-12-02 09:17:45 +08:00
|
|
|
#if 0
|
2009-12-04 09:30:56 +08:00
|
|
|
// FIXME: Assess perf here? Figure out what cases are worth optimizing here
|
|
|
|
// (Length of globals? Chunks of zeroed-out space?).
|
2008-12-02 09:17:45 +08:00
|
|
|
//
|
2009-05-16 15:57:57 +08:00
|
|
|
// If we can, prefer a copy from a global; this is a lot less code for long
|
|
|
|
// globals, and it's easier for the current optimizers to analyze.
|
2009-12-04 09:30:56 +08:00
|
|
|
if (llvm::Constant* C = CGF.CGM.EmitConstantExpr(E, E->getType(), &CGF)) {
|
2008-11-30 10:11:09 +08:00
|
|
|
llvm::GlobalVariable* GV =
|
2009-12-04 09:30:56 +08:00
|
|
|
new llvm::GlobalVariable(CGF.CGM.getModule(), C->getType(), true,
|
|
|
|
llvm::GlobalValue::InternalLinkage, C, "");
|
2012-07-03 07:58:38 +08:00
|
|
|
EmitFinalDestCopy(E->getType(), CGF.MakeAddrLValue(GV, E->getType()));
|
2008-11-30 10:11:09 +08:00
|
|
|
return;
|
|
|
|
}
|
2008-12-02 09:17:45 +08:00
|
|
|
#endif
|
2010-09-06 08:11:41 +08:00
|
|
|
if (E->hadArrayRangeDesignator())
|
2009-01-30 03:42:23 +08:00
|
|
|
CGF.ErrorUnsupported(E, "GNU array range designator extension");
|
|
|
|
|
2016-12-07 07:52:28 +08:00
|
|
|
if (E->isTransparent())
|
|
|
|
return Visit(E->getInit(0));
|
|
|
|
|
2013-05-24 05:54:14 +08:00
|
|
|
AggValueSlot Dest = EnsureSlot(E->getType());
|
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
LValue DestLV = CGF.MakeAddrLValue(Dest.getAddress(), E->getType());
|
2010-09-15 18:14:12 +08:00
|
|
|
|
2008-04-05 02:42:16 +08:00
|
|
|
// Handle initialization of an array.
|
|
|
|
if (E->getType()->isArrayType()) {
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
auto AType = cast<llvm::ArrayType>(Dest.getAddress().getElementType());
|
2018-02-19 17:49:11 +08:00
|
|
|
EmitArrayInit(Dest.getAddress(), AType, E->getType(), E);
|
2008-02-19 06:44:02 +08:00
|
|
|
return;
|
2008-04-05 02:42:16 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-04-05 02:42:16 +08:00
|
|
|
assert(E->getType()->isRecordType() && "Only support structs/unions here!");
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-04-05 02:42:16 +08:00
|
|
|
// Do struct initialization; this code just sets each individual member
|
|
|
|
// to the approprate value. This makes bitfield support automatic;
|
|
|
|
// the disadvantage is that the generated code is more difficult for
|
|
|
|
// the optimizer, especially with bitfields.
|
|
|
|
unsigned NumInitElements = E->getNumInits();
|
2011-07-12 03:35:02 +08:00
|
|
|
RecordDecl *record = E->getType()->castAs<RecordType>()->getDecl();
|
2013-04-21 06:23:05 +08:00
|
|
|
|
2016-03-09 06:17:41 +08:00
|
|
|
// We'll need to enter cleanup scopes in case any of the element
|
|
|
|
// initializers throws an exception.
|
|
|
|
SmallVector<EHScopeStack::stable_iterator, 16> cleanups;
|
|
|
|
llvm::Instruction *cleanupDominator = nullptr;
|
2019-07-26 04:59:48 +08:00
|
|
|
auto addCleanup = [&](const EHScopeStack::stable_iterator &cleanup) {
|
|
|
|
cleanups.push_back(cleanup);
|
|
|
|
if (!cleanupDominator) // create placeholder once needed
|
|
|
|
cleanupDominator = CGF.Builder.CreateAlignedLoad(
|
|
|
|
CGF.Int8Ty, llvm::Constant::getNullValue(CGF.Int8PtrTy),
|
|
|
|
CharUnits::One());
|
|
|
|
};
|
2016-03-09 06:17:41 +08:00
|
|
|
|
|
|
|
unsigned curInitIndex = 0;
|
|
|
|
|
|
|
|
// Emit initialization of base classes.
|
|
|
|
if (auto *CXXRD = dyn_cast<CXXRecordDecl>(record)) {
|
|
|
|
assert(E->getNumInits() >= CXXRD->getNumBases() &&
|
|
|
|
"missing initializer for base class");
|
|
|
|
for (auto &Base : CXXRD->bases()) {
|
|
|
|
assert(!Base.isVirtual() && "should not see vbases here");
|
|
|
|
auto *BaseRD = Base.getType()->getAsCXXRecordDecl();
|
|
|
|
Address V = CGF.GetAddressOfDirectBaseInCompleteClass(
|
|
|
|
Dest.getAddress(), CXXRD, BaseRD,
|
|
|
|
/*isBaseVirtual*/ false);
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot AggSlot = AggValueSlot::forAddr(
|
|
|
|
V, Qualifiers(),
|
|
|
|
AggValueSlot::IsDestructed,
|
|
|
|
AggValueSlot::DoesNotNeedGCBarriers,
|
|
|
|
AggValueSlot::IsNotAliased,
|
2019-06-21 04:56:20 +08:00
|
|
|
CGF.getOverlapForBaseInit(CXXRD, BaseRD, Base.isVirtual()));
|
2016-03-09 06:17:41 +08:00
|
|
|
CGF.EmitAggExpr(E->getInit(curInitIndex++), AggSlot);
|
|
|
|
|
|
|
|
if (QualType::DestructionKind dtorKind =
|
|
|
|
Base.getType().isDestructedType()) {
|
|
|
|
CGF.pushDestroy(dtorKind, V, Base.getType());
|
2019-07-26 04:59:48 +08:00
|
|
|
addCleanup(CGF.EHStack.stable_begin());
|
2016-03-09 06:17:41 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-04-21 06:23:05 +08:00
|
|
|
// Prepare a 'this' for CXXDefaultInitExprs.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
CodeGenFunction::FieldConstructionScope FCS(CGF, Dest.getAddress());
|
2013-04-21 06:23:05 +08:00
|
|
|
|
2011-07-12 03:35:02 +08:00
|
|
|
if (record->isUnion()) {
|
2009-01-30 00:53:55 +08:00
|
|
|
// Only initialize one field of a union. The field itself is
|
|
|
|
// specified by the initializer list.
|
|
|
|
if (!E->getInitializedFieldInUnion()) {
|
|
|
|
// Empty union; we have nothing to do.
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2009-01-30 00:53:55 +08:00
|
|
|
#ifndef NDEBUG
|
|
|
|
// Make sure that it's really an empty and not a failure of
|
|
|
|
// semantic analysis.
|
2014-03-09 04:12:42 +08:00
|
|
|
for (const auto *Field : record->fields())
|
2009-01-30 00:53:55 +08:00
|
|
|
assert(Field->isUnnamedBitfield() && "Only unnamed bitfields allowed");
|
|
|
|
#endif
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// FIXME: volatility
|
|
|
|
FieldDecl *Field = E->getInitializedFieldInUnion();
|
|
|
|
|
2012-04-16 11:54:45 +08:00
|
|
|
LValue FieldLoc = CGF.EmitLValueForFieldInitialization(DestLV, Field);
|
2009-01-30 00:53:55 +08:00
|
|
|
if (NumInitElements) {
|
|
|
|
// Store the initializer into the field
|
2012-03-30 01:37:10 +08:00
|
|
|
EmitInitializationToLValue(E->getInit(0), FieldLoc);
|
2009-01-30 00:53:55 +08:00
|
|
|
} else {
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// Default-initialize to null.
|
2011-06-16 12:16:24 +08:00
|
|
|
EmitNullInitializationToLValue(FieldLoc);
|
2009-01-30 00:53:55 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return;
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-04-05 02:42:16 +08:00
|
|
|
// Here we iterate over the fields; this makes it simpler to both
|
|
|
|
// default-initialize fields and skip over unnamed fields.
|
2014-03-09 04:12:42 +08:00
|
|
|
for (const auto *field : record->fields()) {
|
2011-07-12 03:35:02 +08:00
|
|
|
// We're done once we hit the flexible array member.
|
|
|
|
if (field->getType()->isIncompleteArrayType())
|
2008-12-12 00:49:14 +08:00
|
|
|
break;
|
|
|
|
|
2011-07-12 03:35:02 +08:00
|
|
|
// Always skip anonymous bitfields.
|
|
|
|
if (field->isUnnamedBitfield())
|
2008-04-05 02:42:16 +08:00
|
|
|
continue;
|
2009-01-29 07:36:17 +08:00
|
|
|
|
2011-07-12 03:35:02 +08:00
|
|
|
// We're done if we reach the end of the explicit initializers, we
|
|
|
|
// have a zeroed object, and the rest of the fields are
|
|
|
|
// zero-initializable.
|
|
|
|
if (curInitIndex == NumInitElements && Dest.isZeroed() &&
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
CGF.getTypes().isZeroInitializable(E->getType()))
|
|
|
|
break;
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2012-04-16 11:54:45 +08:00
|
|
|
|
2014-03-09 04:12:42 +08:00
|
|
|
LValue LV = CGF.EmitLValueForFieldInitialization(DestLV, field);
|
2009-05-28 03:54:11 +08:00
|
|
|
// We never generate write-barries for initialized fields.
|
2011-07-12 03:35:02 +08:00
|
|
|
LV.setNonGC(true);
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2011-07-12 03:35:02 +08:00
|
|
|
if (curInitIndex < NumInitElements) {
|
2010-03-09 05:08:07 +08:00
|
|
|
// Store the initializer into the field.
|
2012-03-30 01:37:10 +08:00
|
|
|
EmitInitializationToLValue(E->getInit(curInitIndex++), LV);
|
2008-04-05 02:42:16 +08:00
|
|
|
} else {
|
2017-03-30 22:13:19 +08:00
|
|
|
// We're out of initializers; default-initialize to null
|
2011-07-12 03:35:02 +08:00
|
|
|
EmitNullInitializationToLValue(LV);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Push a destructor if necessary.
|
|
|
|
// FIXME: if we have an array of structures, all explicitly
|
|
|
|
// initialized, we can end up pushing a linear number of cleanups.
|
|
|
|
bool pushedCleanup = false;
|
|
|
|
if (QualType::DestructionKind dtorKind
|
|
|
|
= field->getType().isDestructedType()) {
|
|
|
|
assert(LV.isSimple());
|
|
|
|
if (CGF.needsEHCleanup(dtorKind)) {
|
|
|
|
CGF.pushDestroy(EHCleanup, LV.getAddress(), field->getType(),
|
|
|
|
CGF.getDestroyer(dtorKind), false);
|
2019-07-26 04:59:48 +08:00
|
|
|
addCleanup(CGF.EHStack.stable_begin());
|
2011-07-12 03:35:02 +08:00
|
|
|
pushedCleanup = true;
|
|
|
|
}
|
2008-04-05 02:42:16 +08:00
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// If the GEP didn't get used because of a dead zero init or something
|
|
|
|
// else, clean it up for -O0 builds and general tidiness.
|
2018-07-31 03:24:48 +08:00
|
|
|
if (!pushedCleanup && LV.isSimple())
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
if (llvm::GetElementPtrInst *GEP =
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
dyn_cast<llvm::GetElementPtrInst>(LV.getPointer()))
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
if (GEP->use_empty())
|
|
|
|
GEP->eraseFromParent();
|
2008-02-20 03:27:31 +08:00
|
|
|
}
|
2011-07-12 03:35:02 +08:00
|
|
|
|
|
|
|
// Deactivate all the partial cleanups in reverse order, which
|
|
|
|
// generally means popping them.
|
2019-07-26 04:59:48 +08:00
|
|
|
assert((cleanupDominator || cleanups.empty()) &&
|
|
|
|
"Missing cleanupDominator before deactivating cleanup blocks");
|
2011-07-12 03:35:02 +08:00
|
|
|
for (unsigned i = cleanups.size(); i != 0; --i)
|
2011-11-10 18:43:54 +08:00
|
|
|
CGF.DeactivateCleanupBlock(cleanups[i-1], cleanupDominator);
|
|
|
|
|
|
|
|
// Destroy the placeholder if we made one.
|
|
|
|
if (cleanupDominator)
|
|
|
|
cleanupDominator->eraseFromParent();
|
2007-10-27 01:44:44 +08:00
|
|
|
}
|
|
|
|
|
2016-12-14 09:32:13 +08:00
|
|
|
void AggExprEmitter::VisitArrayInitLoopExpr(const ArrayInitLoopExpr *E,
|
|
|
|
llvm::Value *outerBegin) {
|
2016-12-12 10:53:20 +08:00
|
|
|
// Emit the common subexpression.
|
|
|
|
CodeGenFunction::OpaqueValueMapping binding(CGF, E->getCommonExpr());
|
|
|
|
|
|
|
|
Address destPtr = EnsureSlot(E->getType()).getAddress();
|
|
|
|
uint64_t numElements = E->getArraySize().getZExtValue();
|
|
|
|
|
|
|
|
if (!numElements)
|
|
|
|
return;
|
|
|
|
|
|
|
|
// destPtr is an array*. Construct an elementType* by drilling down a level.
|
|
|
|
llvm::Value *zero = llvm::ConstantInt::get(CGF.SizeTy, 0);
|
|
|
|
llvm::Value *indices[] = {zero, zero};
|
|
|
|
llvm::Value *begin = Builder.CreateInBoundsGEP(destPtr.getPointer(), indices,
|
|
|
|
"arrayinit.begin");
|
|
|
|
|
2016-12-14 09:32:13 +08:00
|
|
|
// Prepare to special-case multidimensional array initialization: we avoid
|
|
|
|
// emitting multiple destructor loops in that case.
|
|
|
|
if (!outerBegin)
|
|
|
|
outerBegin = begin;
|
|
|
|
ArrayInitLoopExpr *InnerLoop = dyn_cast<ArrayInitLoopExpr>(E->getSubExpr());
|
|
|
|
|
2016-12-14 08:03:17 +08:00
|
|
|
QualType elementType =
|
|
|
|
CGF.getContext().getAsArrayType(E->getType())->getElementType();
|
2016-12-12 10:53:20 +08:00
|
|
|
CharUnits elementSize = CGF.getContext().getTypeSizeInChars(elementType);
|
|
|
|
CharUnits elementAlign =
|
|
|
|
destPtr.getAlignment().alignmentOfArrayElement(elementSize);
|
|
|
|
|
|
|
|
llvm::BasicBlock *entryBB = Builder.GetInsertBlock();
|
|
|
|
llvm::BasicBlock *bodyBB = CGF.createBasicBlock("arrayinit.body");
|
|
|
|
|
|
|
|
// Jump into the body.
|
|
|
|
CGF.EmitBlock(bodyBB);
|
|
|
|
llvm::PHINode *index =
|
|
|
|
Builder.CreatePHI(zero->getType(), 2, "arrayinit.index");
|
|
|
|
index->addIncoming(zero, entryBB);
|
|
|
|
llvm::Value *element = Builder.CreateInBoundsGEP(begin, index);
|
|
|
|
|
2016-12-14 08:03:17 +08:00
|
|
|
// Prepare for a cleanup.
|
|
|
|
QualType::DestructionKind dtorKind = elementType.isDestructedType();
|
|
|
|
EHScopeStack::stable_iterator cleanup;
|
2016-12-14 09:32:13 +08:00
|
|
|
if (CGF.needsEHCleanup(dtorKind) && !InnerLoop) {
|
|
|
|
if (outerBegin->getType() != element->getType())
|
|
|
|
outerBegin = Builder.CreateBitCast(outerBegin, element->getType());
|
|
|
|
CGF.pushRegularPartialArrayCleanup(outerBegin, element, elementType,
|
|
|
|
elementAlign,
|
|
|
|
CGF.getDestroyer(dtorKind));
|
2016-12-14 08:03:17 +08:00
|
|
|
cleanup = CGF.EHStack.stable_begin();
|
|
|
|
} else {
|
|
|
|
dtorKind = QualType::DK_none;
|
|
|
|
}
|
2016-12-12 10:53:20 +08:00
|
|
|
|
|
|
|
// Emit the actual filler expression.
|
|
|
|
{
|
2016-12-14 08:03:17 +08:00
|
|
|
// Temporaries created in an array initialization loop are destroyed
|
|
|
|
// at the end of each iteration.
|
|
|
|
CodeGenFunction::RunCleanupsScope CleanupsScope(CGF);
|
2016-12-12 10:53:20 +08:00
|
|
|
CodeGenFunction::ArrayInitLoopExprScope Scope(CGF, index);
|
|
|
|
LValue elementLV =
|
|
|
|
CGF.MakeAddrLValue(Address(element, elementAlign), elementType);
|
2016-12-14 09:32:13 +08:00
|
|
|
|
|
|
|
if (InnerLoop) {
|
|
|
|
// If the subexpression is an ArrayInitLoopExpr, share its cleanup.
|
|
|
|
auto elementSlot = AggValueSlot::forLValue(
|
|
|
|
elementLV, AggValueSlot::IsDestructed,
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::DoesNotNeedGCBarriers,
|
|
|
|
AggValueSlot::IsNotAliased,
|
|
|
|
AggValueSlot::DoesNotOverlap);
|
2016-12-14 09:32:13 +08:00
|
|
|
AggExprEmitter(CGF, elementSlot, false)
|
|
|
|
.VisitArrayInitLoopExpr(InnerLoop, outerBegin);
|
|
|
|
} else
|
|
|
|
EmitInitializationToLValue(E->getSubExpr(), elementLV);
|
2016-12-12 10:53:20 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Move on to the next element.
|
|
|
|
llvm::Value *nextIndex = Builder.CreateNUWAdd(
|
|
|
|
index, llvm::ConstantInt::get(CGF.SizeTy, 1), "arrayinit.next");
|
|
|
|
index->addIncoming(nextIndex, Builder.GetInsertBlock());
|
|
|
|
|
|
|
|
// Leave the loop if we're done.
|
|
|
|
llvm::Value *done = Builder.CreateICmpEQ(
|
|
|
|
nextIndex, llvm::ConstantInt::get(CGF.SizeTy, numElements),
|
|
|
|
"arrayinit.done");
|
|
|
|
llvm::BasicBlock *endBB = CGF.createBasicBlock("arrayinit.end");
|
|
|
|
Builder.CreateCondBr(done, endBB, bodyBB);
|
|
|
|
|
|
|
|
CGF.EmitBlock(endBB);
|
|
|
|
|
|
|
|
// Leave the partial-array cleanup if we entered one.
|
2016-12-14 08:03:17 +08:00
|
|
|
if (dtorKind)
|
|
|
|
CGF.DeactivateCleanupBlock(cleanup, index);
|
2016-12-12 10:53:20 +08:00
|
|
|
}
|
|
|
|
|
2015-06-10 08:27:52 +08:00
|
|
|
void AggExprEmitter::VisitDesignatedInitUpdateExpr(DesignatedInitUpdateExpr *E) {
|
|
|
|
AggValueSlot Dest = EnsureSlot(E->getType());
|
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
LValue DestLV = CGF.MakeAddrLValue(Dest.getAddress(), E->getType());
|
2015-06-10 08:27:52 +08:00
|
|
|
EmitInitializationToLValue(E->getBase(), DestLV);
|
|
|
|
VisitInitListExpr(E->getUpdater());
|
|
|
|
}
|
|
|
|
|
2007-08-21 12:59:27 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Entry Points into this File
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
/// GetNumNonZeroBytesInInit - Get an approximate count of the number of
|
|
|
|
/// non-zero bytes that will be stored when outputting the initializer for the
|
|
|
|
/// specified initializer expression.
|
2011-04-25 01:17:56 +08:00
|
|
|
static CharUnits GetNumNonZeroBytesInInit(const Expr *E, CodeGenFunction &CGF) {
|
2011-04-15 08:35:48 +08:00
|
|
|
E = E->IgnoreParens();
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
|
|
|
|
// 0 and 0.0 won't require any non-zero stores!
|
2011-04-25 01:17:56 +08:00
|
|
|
if (isSimpleZero(E, CGF)) return CharUnits::Zero();
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
|
|
|
|
// If this is an initlist expr, sum up the size of sizes of the (present)
|
|
|
|
// elements. If this is something weird, assume the whole thing is non-zero.
|
|
|
|
const InitListExpr *ILE = dyn_cast<InitListExpr>(E);
|
2018-06-13 10:06:28 +08:00
|
|
|
while (ILE && ILE->isTransparent())
|
|
|
|
ILE = dyn_cast<InitListExpr>(ILE->getInit(0));
|
2014-05-21 13:09:00 +08:00
|
|
|
if (!ILE || !CGF.getTypes().isZeroInitializable(ILE->getType()))
|
2011-04-25 01:17:56 +08:00
|
|
|
return CGF.getContext().getTypeSizeInChars(E->getType());
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2010-12-03 02:29:00 +08:00
|
|
|
// InitListExprs for structs have to be handled carefully. If there are
|
|
|
|
// reference members, we need to consider the size of the reference, not the
|
|
|
|
// referencee. InitListExprs for unions and arrays can't have references.
|
2010-12-03 06:52:04 +08:00
|
|
|
if (const RecordType *RT = E->getType()->getAs<RecordType>()) {
|
|
|
|
if (!RT->isUnionType()) {
|
2019-10-02 05:50:30 +08:00
|
|
|
RecordDecl *SD = RT->getDecl();
|
2011-04-25 01:17:56 +08:00
|
|
|
CharUnits NumNonZeroBytes = CharUnits::Zero();
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2010-12-03 06:52:04 +08:00
|
|
|
unsigned ILEElement = 0;
|
2016-03-09 06:17:41 +08:00
|
|
|
if (auto *CXXRD = dyn_cast<CXXRecordDecl>(SD))
|
2016-03-09 07:16:16 +08:00
|
|
|
while (ILEElement != CXXRD->getNumBases())
|
2016-03-09 06:17:41 +08:00
|
|
|
NumNonZeroBytes +=
|
|
|
|
GetNumNonZeroBytesInInit(ILE->getInit(ILEElement++), CGF);
|
2014-03-09 04:12:42 +08:00
|
|
|
for (const auto *Field : SD->fields()) {
|
2010-12-03 06:52:04 +08:00
|
|
|
// We're done once we hit the flexible array member or run out of
|
|
|
|
// InitListExpr elements.
|
|
|
|
if (Field->getType()->isIncompleteArrayType() ||
|
|
|
|
ILEElement == ILE->getNumInits())
|
|
|
|
break;
|
|
|
|
if (Field->isUnnamedBitfield())
|
|
|
|
continue;
|
|
|
|
|
|
|
|
const Expr *E = ILE->getInit(ILEElement++);
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2010-12-03 06:52:04 +08:00
|
|
|
// Reference values are always non-null and have the width of a pointer.
|
|
|
|
if (Field->getType()->isReferenceType())
|
2011-04-25 01:17:56 +08:00
|
|
|
NumNonZeroBytes += CGF.getContext().toCharUnitsFromBits(
|
2013-04-17 06:48:15 +08:00
|
|
|
CGF.getTarget().getPointerWidth(0));
|
2010-12-03 06:52:04 +08:00
|
|
|
else
|
|
|
|
NumNonZeroBytes += GetNumNonZeroBytesInInit(E, CGF);
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2010-12-03 06:52:04 +08:00
|
|
|
return NumNonZeroBytes;
|
2010-12-03 02:29:00 +08:00
|
|
|
}
|
|
|
|
}
|
2018-07-31 03:24:48 +08:00
|
|
|
|
|
|
|
|
2011-04-25 01:17:56 +08:00
|
|
|
CharUnits NumNonZeroBytes = CharUnits::Zero();
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
for (unsigned i = 0, e = ILE->getNumInits(); i != e; ++i)
|
|
|
|
NumNonZeroBytes += GetNumNonZeroBytesInInit(ILE->getInit(i), CGF);
|
|
|
|
return NumNonZeroBytes;
|
|
|
|
}
|
|
|
|
|
|
|
|
/// CheckAggExprForMemSetUse - If the initializer is large and has a lot of
|
|
|
|
/// zeros in it, emit a memset and avoid storing the individual zeros.
|
|
|
|
///
|
|
|
|
static void CheckAggExprForMemSetUse(AggValueSlot &Slot, const Expr *E,
|
|
|
|
CodeGenFunction &CGF) {
|
|
|
|
// If the slot is already known to be zeroed, nothing to do. Don't mess with
|
|
|
|
// volatile stores.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (Slot.isZeroed() || Slot.isVolatile() || !Slot.getAddress().isValid())
|
2014-05-21 13:09:00 +08:00
|
|
|
return;
|
2011-04-29 06:57:55 +08:00
|
|
|
|
|
|
|
// C++ objects with a user-declared constructor don't need zero'ing.
|
2012-11-02 06:30:59 +08:00
|
|
|
if (CGF.getLangOpts().CPlusPlus)
|
2011-04-29 06:57:55 +08:00
|
|
|
if (const RecordType *RT = CGF.getContext()
|
|
|
|
.getBaseElementType(E->getType())->getAs<RecordType>()) {
|
|
|
|
const CXXRecordDecl *RD = cast<CXXRecordDecl>(RT->getDecl());
|
|
|
|
if (RD->hasUserDeclaredConstructor())
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// If the type is 16-bytes or smaller, prefer individual stores over memset.
|
2018-04-06 04:52:58 +08:00
|
|
|
CharUnits Size = Slot.getPreferredSize(CGF.getContext(), E->getType());
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (Size <= CharUnits::fromQuantity(16))
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
// Check to see if over 3/4 of the initializer are known to be zero. If so,
|
|
|
|
// we prefer to emit memset + individual stores for the rest.
|
2011-04-25 01:25:32 +08:00
|
|
|
CharUnits NumNonZeroBytes = GetNumNonZeroBytesInInit(E, CGF);
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (NumNonZeroBytes*4 > Size)
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
return;
|
2018-07-31 03:24:48 +08:00
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// Okay, it seems like a good idea to use an initial memset, emit the call.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
llvm::Constant *SizeVal = CGF.Builder.getInt64(Size.getQuantity());
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
|
2018-07-31 03:24:48 +08:00
|
|
|
Address Loc = Slot.getAddress();
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Loc = CGF.Builder.CreateElementBitCast(Loc, CGF.Int8Ty);
|
|
|
|
CGF.Builder.CreateMemSet(Loc, CGF.Builder.getInt8(0), SizeVal, false);
|
2018-07-31 03:24:48 +08:00
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// Tell the AggExprEmitter that the slot is known zero.
|
|
|
|
Slot.setZeroed();
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-05-27 02:57:45 +08:00
|
|
|
/// EmitAggExpr - Emit the computation of the specified expression of aggregate
|
|
|
|
/// type. The result is computed into DestPtr. Note that if DestPtr is null,
|
|
|
|
/// the value of the aggregate expression is not needed. If VolatileDest is
|
|
|
|
/// true, DestPtr cannot be 0.
|
2012-07-03 07:58:38 +08:00
|
|
|
void CodeGenFunction::EmitAggExpr(const Expr *E, AggValueSlot Slot) {
|
2013-03-08 05:37:08 +08:00
|
|
|
assert(E && hasAggregateEvaluationKind(E->getType()) &&
|
2007-08-21 12:59:27 +08:00
|
|
|
"Invalid aggregate expression to emit");
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
assert((Slot.getAddress().isValid() || Slot.isIgnored()) &&
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
"slot has bits but no address");
|
2009-09-09 23:08:12 +08:00
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
// Optimize the slot if possible.
|
|
|
|
CheckAggExprForMemSetUse(Slot, E, *this);
|
2018-07-31 03:24:48 +08:00
|
|
|
|
2015-06-08 18:23:49 +08:00
|
|
|
AggExprEmitter(*this, Slot, Slot.isIgnored()).Visit(const_cast<Expr*>(E));
|
2007-08-21 12:59:27 +08:00
|
|
|
}
|
2008-09-10 04:49:46 +08:00
|
|
|
|
2010-02-06 03:38:31 +08:00
|
|
|
LValue CodeGenFunction::EmitAggExprToLValue(const Expr *E) {
|
2013-03-08 05:37:08 +08:00
|
|
|
assert(hasAggregateEvaluationKind(E->getType()) && "Invalid argument!");
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
Address Temp = CreateMemTemp(E->getType());
|
2010-08-21 11:15:20 +08:00
|
|
|
LValue LV = MakeAddrLValue(Temp, E->getType());
|
2011-08-26 04:40:09 +08:00
|
|
|
EmitAggExpr(E, AggValueSlot::forLValue(LV, AggValueSlot::IsNotDestructed,
|
2011-08-26 15:31:35 +08:00
|
|
|
AggValueSlot::DoesNotNeedGCBarriers,
|
2018-04-06 04:52:58 +08:00
|
|
|
AggValueSlot::IsNotAliased,
|
|
|
|
AggValueSlot::DoesNotOverlap));
|
2010-08-21 11:15:20 +08:00
|
|
|
return LV;
|
2010-02-06 03:38:31 +08:00
|
|
|
}
|
|
|
|
|
2019-06-21 04:44:45 +08:00
|
|
|
AggValueSlot::Overlap_t
|
2019-06-21 04:56:20 +08:00
|
|
|
CodeGenFunction::getOverlapForFieldInit(const FieldDecl *FD) {
|
2019-06-21 04:44:45 +08:00
|
|
|
if (!FD->hasAttr<NoUniqueAddressAttr>() || !FD->getType()->isRecordType())
|
|
|
|
return AggValueSlot::DoesNotOverlap;
|
|
|
|
|
|
|
|
// If the field lies entirely within the enclosing class's nvsize, its tail
|
|
|
|
// padding cannot overlap any already-initialized object. (The only subobjects
|
|
|
|
// with greater addresses that might already be initialized are vbases.)
|
|
|
|
const RecordDecl *ClassRD = FD->getParent();
|
|
|
|
const ASTRecordLayout &Layout = getContext().getASTRecordLayout(ClassRD);
|
|
|
|
if (Layout.getFieldOffset(FD->getFieldIndex()) +
|
|
|
|
getContext().getTypeSize(FD->getType()) <=
|
|
|
|
(uint64_t)getContext().toBits(Layout.getNonVirtualSize()))
|
|
|
|
return AggValueSlot::DoesNotOverlap;
|
|
|
|
|
|
|
|
// The tail padding may contain values we need to preserve.
|
|
|
|
return AggValueSlot::MayOverlap;
|
|
|
|
}
|
|
|
|
|
2019-06-21 04:56:20 +08:00
|
|
|
AggValueSlot::Overlap_t CodeGenFunction::getOverlapForBaseInit(
|
2018-04-06 04:52:58 +08:00
|
|
|
const CXXRecordDecl *RD, const CXXRecordDecl *BaseRD, bool IsVirtual) {
|
2019-06-21 04:44:45 +08:00
|
|
|
// If the most-derived object is a field declared with [[no_unique_address]],
|
|
|
|
// the tail padding of any virtual base could be reused for other subobjects
|
|
|
|
// of that field's class.
|
2018-04-06 04:52:58 +08:00
|
|
|
if (IsVirtual)
|
2019-06-21 04:44:45 +08:00
|
|
|
return AggValueSlot::MayOverlap;
|
2018-04-06 04:52:58 +08:00
|
|
|
|
|
|
|
// If the base class is laid out entirely within the nvsize of the derived
|
|
|
|
// class, its tail padding cannot yet be initialized, so we can issue
|
|
|
|
// stores at the full width of the base class.
|
|
|
|
const ASTRecordLayout &Layout = getContext().getASTRecordLayout(RD);
|
|
|
|
if (Layout.getBaseClassOffset(BaseRD) +
|
|
|
|
getContext().getASTRecordLayout(BaseRD).getSize() <=
|
|
|
|
Layout.getNonVirtualSize())
|
|
|
|
return AggValueSlot::DoesNotOverlap;
|
|
|
|
|
|
|
|
// The tail padding may contain values we need to preserve.
|
|
|
|
return AggValueSlot::MayOverlap;
|
|
|
|
}
|
|
|
|
|
|
|
|
void CodeGenFunction::EmitAggregateCopy(LValue Dest, LValue Src, QualType Ty,
|
|
|
|
AggValueSlot::Overlap_t MayOverlap,
|
|
|
|
bool isVolatile) {
|
2012-03-30 01:37:10 +08:00
|
|
|
assert(!Ty->isAnyComplexType() && "Shouldn't happen for complex");
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2018-01-25 22:21:55 +08:00
|
|
|
Address DestPtr = Dest.getAddress();
|
|
|
|
Address SrcPtr = Src.getAddress();
|
|
|
|
|
2012-11-02 06:30:59 +08:00
|
|
|
if (getLangOpts().CPlusPlus) {
|
2012-03-30 01:37:10 +08:00
|
|
|
if (const RecordType *RT = Ty->getAs<RecordType>()) {
|
|
|
|
CXXRecordDecl *Record = cast<CXXRecordDecl>(RT->getDecl());
|
2018-07-31 03:24:48 +08:00
|
|
|
assert((Record->hasTrivialCopyConstructor() ||
|
2012-03-30 01:37:10 +08:00
|
|
|
Record->hasTrivialCopyAssignment() ||
|
|
|
|
Record->hasTrivialMoveConstructor() ||
|
2015-04-30 03:26:57 +08:00
|
|
|
Record->hasTrivialMoveAssignment() ||
|
|
|
|
Record->isUnion()) &&
|
2012-11-16 08:53:38 +08:00
|
|
|
"Trying to aggregate-copy a type without a trivial copy/move "
|
2010-05-20 23:39:01 +08:00
|
|
|
"constructor or assignment operator");
|
2012-03-30 01:37:10 +08:00
|
|
|
// Ignore empty classes in C++.
|
|
|
|
if (Record->isEmpty())
|
2010-05-03 09:20:20 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
2018-01-25 22:21:55 +08:00
|
|
|
|
2012-03-30 01:37:10 +08:00
|
|
|
// Aggregate assignment turns into llvm.memcpy. This is almost valid per
|
|
|
|
// C99 6.5.16.1p3, which states "If the value being stored in an object is
|
|
|
|
// read from another object that overlaps in anyway the storage of the first
|
|
|
|
// object, then the overlap shall be exact and the two objects shall have
|
|
|
|
// qualified or unqualified versions of a compatible type."
|
|
|
|
//
|
|
|
|
// memcpy is not defined if the source and destination pointers are exactly
|
|
|
|
// equal, but other compilers do this optimization, and almost every memcpy
|
|
|
|
// implementation handles this case safely. If there is a libc that does not
|
|
|
|
// safely handle this, we can add a target hook.
|
|
|
|
|
2018-04-06 04:52:58 +08:00
|
|
|
// Get data size info for this aggregate. Don't copy the tail padding if this
|
|
|
|
// might be a potentially-overlapping subobject, since the tail padding might
|
|
|
|
// be occupied by a different object. Otherwise, copying it is fine.
|
2012-09-30 20:43:37 +08:00
|
|
|
std::pair<CharUnits, CharUnits> TypeInfo;
|
2018-04-06 04:52:58 +08:00
|
|
|
if (MayOverlap)
|
2012-09-30 20:43:37 +08:00
|
|
|
TypeInfo = getContext().getTypeInfoDataSizeInChars(Ty);
|
|
|
|
else
|
|
|
|
TypeInfo = getContext().getTypeInfoInChars(Ty);
|
2012-03-30 01:37:10 +08:00
|
|
|
|
2015-05-20 11:46:04 +08:00
|
|
|
llvm::Value *SizeVal = nullptr;
|
|
|
|
if (TypeInfo.first.isZero()) {
|
|
|
|
// But note that getTypeInfo returns 0 for a VLA.
|
|
|
|
if (auto *VAT = dyn_cast_or_null<VariableArrayType>(
|
|
|
|
getContext().getAsArrayType(Ty))) {
|
|
|
|
QualType BaseEltTy;
|
|
|
|
SizeVal = emitArrayLength(VAT, BaseEltTy, DestPtr);
|
2018-04-06 04:52:58 +08:00
|
|
|
TypeInfo = getContext().getTypeInfoInChars(BaseEltTy);
|
2015-05-20 11:46:04 +08:00
|
|
|
assert(!TypeInfo.first.isZero());
|
|
|
|
SizeVal = Builder.CreateNUWMul(
|
|
|
|
SizeVal,
|
|
|
|
llvm::ConstantInt::get(SizeTy, TypeInfo.first.getQuantity()));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (!SizeVal) {
|
|
|
|
SizeVal = llvm::ConstantInt::get(SizeTy, TypeInfo.first.getQuantity());
|
|
|
|
}
|
2012-03-30 01:37:10 +08:00
|
|
|
|
|
|
|
// FIXME: If we have a volatile struct, the optimizer can remove what might
|
|
|
|
// appear to be `extra' memory ops:
|
|
|
|
//
|
|
|
|
// volatile struct { int i; } a, b;
|
|
|
|
//
|
|
|
|
// int main() {
|
|
|
|
// a = b;
|
|
|
|
// a = b;
|
|
|
|
// }
|
|
|
|
//
|
|
|
|
// we need to use a different call here. We use isVolatile to indicate when
|
|
|
|
// either the source or the destination is volatile.
|
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
DestPtr = Builder.CreateElementBitCast(DestPtr, Int8Ty);
|
|
|
|
SrcPtr = Builder.CreateElementBitCast(SrcPtr, Int8Ty);
|
2010-04-04 11:10:52 +08:00
|
|
|
|
2011-06-16 07:02:42 +08:00
|
|
|
// Don't do any of the memmove_collectable tests if GC isn't set.
|
2012-03-11 15:00:24 +08:00
|
|
|
if (CGM.getLangOpts().getGC() == LangOptions::NonGC) {
|
2011-06-16 07:02:42 +08:00
|
|
|
// fall through
|
2012-03-30 01:37:10 +08:00
|
|
|
} else if (const RecordType *RecordTy = Ty->getAs<RecordType>()) {
|
|
|
|
RecordDecl *Record = RecordTy->getDecl();
|
|
|
|
if (Record->hasObjectMember()) {
|
2018-07-31 03:24:48 +08:00
|
|
|
CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, DestPtr, SrcPtr,
|
2012-03-30 01:37:10 +08:00
|
|
|
SizeVal);
|
2010-06-16 06:44:06 +08:00
|
|
|
return;
|
|
|
|
}
|
2012-03-30 01:37:10 +08:00
|
|
|
} else if (Ty->isArrayType()) {
|
|
|
|
QualType BaseType = getContext().getBaseElementType(Ty);
|
|
|
|
if (const RecordType *RecordTy = BaseType->getAs<RecordType>()) {
|
|
|
|
if (RecordTy->getDecl()->hasObjectMember()) {
|
2018-07-31 03:24:48 +08:00
|
|
|
CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, DestPtr, SrcPtr,
|
2012-03-30 01:37:10 +08:00
|
|
|
SizeVal);
|
2010-06-16 06:44:06 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2012-09-29 05:58:29 +08:00
|
|
|
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
auto Inst = Builder.CreateMemCpy(DestPtr, SrcPtr, SizeVal, isVolatile);
|
|
|
|
|
2012-09-29 05:58:29 +08:00
|
|
|
// Determine the metadata to describe the position of any padding in this
|
|
|
|
// memcpy, as well as the TBAA tags for the members of the struct, in case
|
|
|
|
// the optimizer wishes to expand it in to scalar memory operations.
|
Compute and preserve alignment more faithfully in IR-generation.
Introduce an Address type to bundle a pointer value with an
alignment. Introduce APIs on CGBuilderTy to work with Address
values. Change core APIs on CGF/CGM to traffic in Address where
appropriate. Require alignments to be non-zero. Update a ton
of code to compute and propagate alignment information.
As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.
The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned. Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay. I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.
Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.
We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment. In particular,
field access now uses alignmentAtOffset instead of min.
Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs. For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint. That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.
ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments. In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments. That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.
I partially punted on applying this work to CGBuiltin. Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.
llvm-svn: 246985
2015-09-08 16:05:57 +08:00
|
|
|
if (llvm::MDNode *TBAAStructTag = CGM.getTBAAStructInfo(Ty))
|
|
|
|
Inst->setMetadata(llvm::LLVMContext::MD_tbaa_struct, TBAAStructTag);
|
2018-01-25 22:21:55 +08:00
|
|
|
|
|
|
|
if (CGM.getCodeGenOpts().NewStructPathTBAA) {
|
|
|
|
TBAAAccessInfo TBAAInfo = CGM.mergeTBAAInfoForMemoryTransfer(
|
|
|
|
Dest.getTBAAInfo(), Src.getTBAAInfo());
|
|
|
|
CGM.DecorateInstructionWithTBAA(Inst, TBAAInfo);
|
|
|
|
}
|
2008-09-10 04:49:46 +08:00
|
|
|
}
|