1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 02:33:06 +01:00
llvm-mirror/lib/IR/Function.cpp

1934 lines
68 KiB
C++
Raw Normal View History

//===- Function.cpp - Implement the Global object classes -----------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
2001-06-06 22:29:01 +02:00
//
// This file implements the Function class for the IR library.
2001-06-06 22:29:01 +02:00
//
//===----------------------------------------------------------------------===//
#include "llvm/IR/Function.h"
#include "SymbolTableListTraitsImpl.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/IR/AbstractCallSite.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/InstIterator.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/IntrinsicsAArch64.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/IntrinsicsARM.h"
#include "llvm/IR/IntrinsicsBPF.h"
#include "llvm/IR/IntrinsicsHexagon.h"
#include "llvm/IR/IntrinsicsMips.h"
#include "llvm/IR/IntrinsicsNVPTX.h"
#include "llvm/IR/IntrinsicsPowerPC.h"
#include "llvm/IR/IntrinsicsR600.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/IR/IntrinsicsS390.h"
#include "llvm/IR/IntrinsicsVE.h"
#include "llvm/IR/IntrinsicsWebAssembly.h"
#include "llvm/IR/IntrinsicsX86.h"
#include "llvm/IR/IntrinsicsXCore.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/SymbolTableListTraits.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Use.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/IR/ValueSymbolTable.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ErrorHandling.h"
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <cstdint>
#include <cstring>
#include <string>
using namespace llvm;
using ProfileCount = Function::ProfileCount;
// Explicit instantiations of SymbolTableListTraits since some of the methods
// are not in the public header file...
template class llvm::SymbolTableListTraits<BasicBlock>;
2001-06-06 22:29:01 +02:00
static cl::opt<unsigned> NonGlobalValueMaxNameSize(
"non-global-value-max-name-size", cl::Hidden, cl::init(1024),
cl::desc("Maximum size for the name of non-global values."));
//===----------------------------------------------------------------------===//
// Argument Implementation
//===----------------------------------------------------------------------===//
Argument::Argument(Type *Ty, const Twine &Name, Function *Par, unsigned ArgNo)
: Value(Ty, Value::ArgumentVal), Parent(Par), ArgNo(ArgNo) {
setName(Name);
}
void Argument::setParent(Function *parent) {
Parent = parent;
}
bool Argument::hasNonNullAttr(bool AllowUndefOrPoison) const {
if (!getType()->isPointerTy()) return false;
if (getParent()->hasParamAttribute(getArgNo(), Attribute::NonNull) &&
(AllowUndefOrPoison ||
getParent()->hasParamAttribute(getArgNo(), Attribute::NoUndef)))
return true;
else if (getDereferenceableBytes() > 0 &&
!NullPointerIsDefined(getParent(),
getType()->getPointerAddressSpace()))
return true;
return false;
}
bool Argument::hasByValAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::ByVal);
}
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
bool Argument::hasByRefAttr() const {
if (!getType()->isPointerTy())
return false;
return hasAttribute(Attribute::ByRef);
}
bool Argument::hasSwiftSelfAttr() const {
return getParent()->hasParamAttribute(getArgNo(), Attribute::SwiftSelf);
}
bool Argument::hasSwiftErrorAttr() const {
return getParent()->hasParamAttribute(getArgNo(), Attribute::SwiftError);
}
bool Argument::hasInAllocaAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::InAlloca);
}
Reland [X86] Codegen for preallocated See https://reviews.llvm.org/D74651 for the preallocated IR constructs and LangRef changes. In X86TargetLowering::LowerCall(), if a call is preallocated, record each argument's offset from the stack pointer and the total stack adjustment. Associate the call Value with an integer index. Store the info in X86MachineFunctionInfo with the integer index as the key. This adds two new target independent ISDOpcodes and two new target dependent Opcodes corresponding to @llvm.call.preallocated.{setup,arg}. The setup ISelDAG node takes in a chain and outputs a chain and a SrcValue of the preallocated call Value. It is lowered to a target dependent node with the SrcValue replaced with the integer index key by looking in X86MachineFunctionInfo. In X86TargetLowering::EmitInstrWithCustomInserter() this is lowered to an %esp adjustment, the exact amount determined by looking in X86MachineFunctionInfo with the integer index key. The arg ISelDAG node takes in a chain, a SrcValue of the preallocated call Value, and the arg index int constant. It produces a chain and the pointer fo the arg. It is lowered to a target dependent node with the SrcValue replaced with the integer index key by looking in X86MachineFunctionInfo. In X86TargetLowering::EmitInstrWithCustomInserter() this is lowered to a lea of the stack pointer plus an offset determined by looking in X86MachineFunctionInfo with the integer index key. Force any function containing a preallocated call to use the frame pointer. Does not yet handle a setup without a call, or a conditional call. Does not yet handle musttail. That requires a LangRef change first. Tried to look at all references to inalloca and see if they apply to preallocated. I've made preallocated versions of tests testing inalloca whenever possible and when they make sense (e.g. not alloca related, inalloca edge cases). Aside from the tests added here, I checked that this codegen produces correct code for something like ``` struct A { A(); A(A&&); ~A(); }; void bar() { foo(foo(foo(foo(foo(A(), 4), 5), 6), 7), 8); } ``` by replacing the inalloca version of the .ll file with the appropriate preallocated code. Running the executable produces the same results as using the current inalloca implementation. Reverted due to unexpectedly passing tests, added REQUIRES: asserts for reland. Subscribers: hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D77689
2020-03-16 20:32:36 +01:00
bool Argument::hasPreallocatedAttr() const {
if (!getType()->isPointerTy())
return false;
return hasAttribute(Attribute::Preallocated);
}
bool Argument::hasPassPointeeByValueCopyAttr() const {
if (!getType()->isPointerTy()) return false;
AttributeList Attrs = getParent()->getAttributes();
return Attrs.hasParamAttribute(getArgNo(), Attribute::ByVal) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::InAlloca) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::Preallocated);
}
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
bool Argument::hasPointeeInMemoryValueAttr() const {
if (!getType()->isPointerTy())
return false;
AttributeList Attrs = getParent()->getAttributes();
return Attrs.hasParamAttribute(getArgNo(), Attribute::ByVal) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::StructRet) ||
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
Attrs.hasParamAttribute(getArgNo(), Attribute::InAlloca) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::Preallocated) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::ByRef);
}
/// For a byval, sret, inalloca, or preallocated parameter, get the in-memory
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
/// parameter type.
static Type *getMemoryParamAllocType(AttributeSet ParamAttrs, Type *ArgTy) {
// FIXME: All the type carrying attributes are mutually exclusive, so there
// should be a single query to get the stored type that handles any of them.
if (Type *ByValTy = ParamAttrs.getByValType())
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
return ByValTy;
if (Type *ByRefTy = ParamAttrs.getByRefType())
return ByRefTy;
if (Type *PreAllocTy = ParamAttrs.getPreallocatedType())
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
return PreAllocTy;
if (Type *InAllocaTy = ParamAttrs.getInAllocaType())
return InAllocaTy;
if (Type *SRetTy = ParamAttrs.getStructRetType())
return SRetTy;
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
return nullptr;
}
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
uint64_t Argument::getPassPointeeByValueCopySize(const DataLayout &DL) const {
AttributeSet ParamAttrs =
getParent()->getAttributes().getParamAttributes(getArgNo());
if (Type *MemTy = getMemoryParamAllocType(ParamAttrs, getType()))
return DL.getTypeAllocSize(MemTy);
return 0;
}
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
Type *Argument::getPointeeInMemoryValueType() const {
AttributeSet ParamAttrs =
getParent()->getAttributes().getParamAttributes(getArgNo());
return getMemoryParamAllocType(ParamAttrs, getType());
}
unsigned Argument::getParamAlignment() const {
assert(getType()->isPointerTy() && "Only pointers have alignments");
return getParent()->getParamAlignment(getArgNo());
}
MaybeAlign Argument::getParamAlign() const {
assert(getType()->isPointerTy() && "Only pointers have alignments");
return getParent()->getParamAlign(getArgNo());
}
[clang][AArch64] Correctly align HFA arguments when passed on the stack When we pass a AArch64 Homogeneous Floating-Point Aggregate (HFA) argument with increased alignment requirements, for example struct S { __attribute__ ((__aligned__(16))) double v[4]; }; Clang uses `[4 x double]` for the parameter, which is passed on the stack at alignment 8, whereas it should be at alignment 16, following Rule C.4 in AAPCS (https://github.com/ARM-software/abi-aa/blob/master/aapcs64/aapcs64.rst#642parameter-passing-rules) Currently we don't have a way to express in LLVM IR the alignment requirements of the function arguments. The align attribute is applicable to pointers only, and only for some special ways of passing arguments (e..g byval). When implementing AAPCS32/AAPCS64, clang resorts to dubious hacks of coercing to types, which naturally have the needed alignment. We don't have enough types to cover all the cases, though. This patch introduces a new use of the stackalign attribute to control stack slot alignment, when and if an argument is passed in memory. The attribute align is left as an optimizer hint - it still applies to pointer types only and pertains to the content of the pointer, whereas the alignment of the pointer itself is determined by the stackalign attribute. For byval arguments, the stackalign attribute assumes the role, previously perfomed by align, falling back to align if stackalign` is absent. On the clang side, when passing arguments using the "direct" style (cf. `ABIArgInfo::Kind`), now we can optionally specify an alignment, which is emitted as the new `stackalign` attribute. Patch by Momchil Velikov and Lucas Prates. Differential Revision: https://reviews.llvm.org/D98794
2021-04-15 20:58:54 +02:00
MaybeAlign Argument::getParamStackAlign() const {
return getParent()->getParamStackAlign(getArgNo());
}
Type *Argument::getParamByValType() const {
assert(getType()->isPointerTy() && "Only pointers have byval types");
return getParent()->getParamByValType(getArgNo());
}
Type *Argument::getParamStructRetType() const {
assert(getType()->isPointerTy() && "Only pointers have sret types");
return getParent()->getParamStructRetType(getArgNo());
}
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
Type *Argument::getParamByRefType() const {
assert(getType()->isPointerTy() && "Only pointers have byref types");
IR: Define byref parameter attribute This allows tracking the in-memory type of a pointer argument to a function for ABI purposes. This is essentially a stripped down version of byval to remove some of the stack-copy implications in its definition. This includes the base IR changes, and some tests for places where it should be treated similarly to byval. Codegen support will be in a future patch. My original attempt at solving some of these problems was to repurpose byval with a different address space from the stack. However, it is technically permitted for the callee to introduce a write to the argument, although nothing does this in reality. There is also talk of removing and replacing the byval attribute, so a new attribute would need to take its place anyway. This is intended avoid some optimization issues with the current handling of aggregate arguments, as well as fixes inflexibilty in how frontends can specify the kernel ABI. The most honest representation of the amdgpu_kernel convention is to expose all kernel arguments as loads from constant memory. Today, these are raw, SSA Argument values and codegen is responsible for turning these into loads. Background: There currently isn't a satisfactory way to represent how arguments for the amdgpu_kernel calling convention are passed. In reality, arguments are passed in a single, flat, constant memory buffer implicitly passed to the function. It is also illegal to call this function in the IR, and this is only ever invoked by a driver of some kind. It does not make sense to have a stack passed parameter in this context as is implied by byval. It is never valid to write to the kernel arguments, as this would corrupt the inputs seen by other dispatches of the kernel. These argumets are also not in the same address space as the stack, so a copy is needed to an alloca. From a source C-like language, the kernel parameters are invisible. Semantically, a copy is always required from the constant argument memory to a mutable variable. The current clang calling convention lowering emits raw values, including aggregates into the function argument list, since using byval would not make sense. This has some unfortunate consequences for the optimizer. In the aggregate case, we end up with an aggregate store to alloca, which both SROA and instcombine turn into a store of each aggregate field. The optimizer never pieces this back together to see that this is really just a copy from constant memory, so we end up stuck with expensive stack usage. This also means the backend dictates the alignment of arguments, and arbitrarily picks the LLVM IR ABI type alignment. By allowing an explicit alignment, frontends can make better decisions. For example, there's real no advantage to an aligment higher than 4, so a frontend could choose to compact the argument layout. Similarly, there is a high penalty to using an alignment lower than 4, so a frontend could opt into more padding for small arguments. Another design consideration is when it is appropriate to expose the fact that these arguments are all really passed in adjacent memory. Currently we have a late IR optimization pass in codegen to rewrite the kernel argument values into explicit loads to enable vectorization. In most programs, unrelated argument loads can be merged together. However, exposing this property directly from the frontend has some disadvantages. We still need a way to track the original argument sizes and alignments to report to the driver. I find using some side-channel, metadata mechanism to track this unappealing. If the kernel arguments were exposed as a single buffer to begin with, alias analysis would be unaware that the padding bits betewen arguments are meaningless. Another family of problems is there are still some gaps in replacing all of the available parameter attributes with metadata equivalents once lowered to loads. The immediate plan is to start using this new attribute to handle all aggregate argumets for kernels. Long term, it makes sense to migrate all kernel arguments, including scalars, to be passed indirectly in the same manner. Additional context is in D79744.
2020-06-05 22:58:47 +02:00
return getParent()->getParamByRefType(getArgNo());
}
Type *Argument::getParamInAllocaType() const {
assert(getType()->isPointerTy() && "Only pointers have inalloca types");
return getParent()->getParamInAllocaType(getArgNo());
}
uint64_t Argument::getDereferenceableBytes() const {
assert(getType()->isPointerTy() &&
"Only pointers have dereferenceable bytes");
return getParent()->getParamDereferenceableBytes(getArgNo());
}
uint64_t Argument::getDereferenceableOrNullBytes() const {
assert(getType()->isPointerTy() &&
"Only pointers have dereferenceable bytes");
return getParent()->getParamDereferenceableOrNullBytes(getArgNo());
}
bool Argument::hasNestAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::Nest);
}
bool Argument::hasNoAliasAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::NoAlias);
}
bool Argument::hasNoCaptureAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::NoCapture);
}
bool Argument::hasNoFreeAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::NoFree);
}
bool Argument::hasStructRetAttr() const {
if (!getType()->isPointerTy()) return false;
return hasAttribute(Attribute::StructRet);
}
bool Argument::hasInRegAttr() const {
return hasAttribute(Attribute::InReg);
}
bool Argument::hasReturnedAttr() const {
return hasAttribute(Attribute::Returned);
}
bool Argument::hasZExtAttr() const {
return hasAttribute(Attribute::ZExt);
}
bool Argument::hasSExtAttr() const {
return hasAttribute(Attribute::SExt);
}
bool Argument::onlyReadsMemory() const {
AttributeList Attrs = getParent()->getAttributes();
return Attrs.hasParamAttribute(getArgNo(), Attribute::ReadOnly) ||
Attrs.hasParamAttribute(getArgNo(), Attribute::ReadNone);
}
void Argument::addAttrs(AttrBuilder &B) {
AttributeList AL = getParent()->getAttributes();
AL = AL.addParamAttributes(Parent->getContext(), getArgNo(), B);
getParent()->setAttributes(AL);
}
void Argument::addAttr(Attribute::AttrKind Kind) {
getParent()->addParamAttr(getArgNo(), Kind);
}
void Argument::addAttr(Attribute Attr) {
getParent()->addParamAttr(getArgNo(), Attr);
}
void Argument::removeAttr(Attribute::AttrKind Kind) {
getParent()->removeParamAttr(getArgNo(), Kind);
}
void Argument::removeAttrs(const AttrBuilder &B) {
AttributeList AL = getParent()->getAttributes();
AL = AL.removeParamAttributes(Parent->getContext(), getArgNo(), B);
getParent()->setAttributes(AL);
}
bool Argument::hasAttribute(Attribute::AttrKind Kind) const {
return getParent()->hasParamAttribute(getArgNo(), Kind);
}
Attribute Argument::getAttribute(Attribute::AttrKind Kind) const {
return getParent()->getParamAttribute(getArgNo(), Kind);
}
//===----------------------------------------------------------------------===//
// Helper Methods in Function
//===----------------------------------------------------------------------===//
LLVMContext &Function::getContext() const {
return getType()->getContext();
}
unsigned Function::getInstructionCount() const {
unsigned NumInstrs = 0;
for (const BasicBlock &BB : BasicBlocks)
NumInstrs += std::distance(BB.instructionsWithoutDebug().begin(),
BB.instructionsWithoutDebug().end());
return NumInstrs;
}
Function *Function::Create(FunctionType *Ty, LinkageTypes Linkage,
const Twine &N, Module &M) {
return Create(Ty, Linkage, M.getDataLayout().getProgramAddressSpace(), N, &M);
}
[IR][sanitizer] Set nounwind on module ctor/dtor, additionally set uwtable if -fasynchronous-unwind-tables On ELF targets, if a function has uwtable or personality, or does not have nounwind (`needsUnwindTableEntry`), it marks that `.eh_frame` is needed in the module. Then, a function gets `.eh_frame` if `needsUnwindTableEntry` or `-g[123]` is specified. (i.e. If -g[123], every function gets `.eh_frame`. This behavior is strange but that is the status quo on GCC and Clang.) Let's take asan as an example. Other sanitizers are similar. `asan.module_[cd]tor` has no attribute. `needsUnwindTableEntry` returns true, so every function gets `.eh_frame` if `-g[123]` is specified. This is the root cause that `-fno-exceptions -fno-asynchronous-unwind-tables -g` produces .debug_frame while `-fno-exceptions -fno-asynchronous-unwind-tables -g -fsanitize=address` produces .eh_frame. This patch * sets the nounwind attribute on sanitizer module ctor/dtor. * let Clang emit a module flag metadata "uwtable" for -fasynchronous-unwind-tables. If "uwtable" is set, sanitizer module ctor/dtor additionally get the uwtable attribute. The "uwtable" mechanism is generic: synthesized functions not cloned/specialized from existing ones should consider `Function::createWithDefaultAttr` instead of `Function::create` if they want to get some default attributes which have more of module semantics. Other candidates: "frame-pointer" (https://github.com/ClangBuiltLinux/linux/issues/955 https://github.com/ClangBuiltLinux/linux/issues/1238), dso_local, etc. Differential Revision: https://reviews.llvm.org/D100251
2021-04-22 00:58:20 +02:00
Function *Function::createWithDefaultAttr(FunctionType *Ty,
LinkageTypes Linkage,
unsigned AddrSpace, const Twine &N,
Module *M) {
auto *F = new Function(Ty, Linkage, AddrSpace, N, M);
AttrBuilder B;
[IR][sanitizer] Set nounwind on module ctor/dtor, additionally set uwtable if -fasynchronous-unwind-tables On ELF targets, if a function has uwtable or personality, or does not have nounwind (`needsUnwindTableEntry`), it marks that `.eh_frame` is needed in the module. Then, a function gets `.eh_frame` if `needsUnwindTableEntry` or `-g[123]` is specified. (i.e. If -g[123], every function gets `.eh_frame`. This behavior is strange but that is the status quo on GCC and Clang.) Let's take asan as an example. Other sanitizers are similar. `asan.module_[cd]tor` has no attribute. `needsUnwindTableEntry` returns true, so every function gets `.eh_frame` if `-g[123]` is specified. This is the root cause that `-fno-exceptions -fno-asynchronous-unwind-tables -g` produces .debug_frame while `-fno-exceptions -fno-asynchronous-unwind-tables -g -fsanitize=address` produces .eh_frame. This patch * sets the nounwind attribute on sanitizer module ctor/dtor. * let Clang emit a module flag metadata "uwtable" for -fasynchronous-unwind-tables. If "uwtable" is set, sanitizer module ctor/dtor additionally get the uwtable attribute. The "uwtable" mechanism is generic: synthesized functions not cloned/specialized from existing ones should consider `Function::createWithDefaultAttr` instead of `Function::create` if they want to get some default attributes which have more of module semantics. Other candidates: "frame-pointer" (https://github.com/ClangBuiltLinux/linux/issues/955 https://github.com/ClangBuiltLinux/linux/issues/1238), dso_local, etc. Differential Revision: https://reviews.llvm.org/D100251
2021-04-22 00:58:20 +02:00
if (M->getUwtable())
B.addAttribute(Attribute::UWTable);
switch (M->getFramePointer()) {
case FramePointerKind::None:
// 0 ("none") is the default.
break;
case FramePointerKind::NonLeaf:
B.addAttribute("frame-pointer", "non-leaf");
break;
case FramePointerKind::All:
B.addAttribute("frame-pointer", "all");
break;
}
F->addAttributes(AttributeList::FunctionIndex, B);
[IR][sanitizer] Set nounwind on module ctor/dtor, additionally set uwtable if -fasynchronous-unwind-tables On ELF targets, if a function has uwtable or personality, or does not have nounwind (`needsUnwindTableEntry`), it marks that `.eh_frame` is needed in the module. Then, a function gets `.eh_frame` if `needsUnwindTableEntry` or `-g[123]` is specified. (i.e. If -g[123], every function gets `.eh_frame`. This behavior is strange but that is the status quo on GCC and Clang.) Let's take asan as an example. Other sanitizers are similar. `asan.module_[cd]tor` has no attribute. `needsUnwindTableEntry` returns true, so every function gets `.eh_frame` if `-g[123]` is specified. This is the root cause that `-fno-exceptions -fno-asynchronous-unwind-tables -g` produces .debug_frame while `-fno-exceptions -fno-asynchronous-unwind-tables -g -fsanitize=address` produces .eh_frame. This patch * sets the nounwind attribute on sanitizer module ctor/dtor. * let Clang emit a module flag metadata "uwtable" for -fasynchronous-unwind-tables. If "uwtable" is set, sanitizer module ctor/dtor additionally get the uwtable attribute. The "uwtable" mechanism is generic: synthesized functions not cloned/specialized from existing ones should consider `Function::createWithDefaultAttr` instead of `Function::create` if they want to get some default attributes which have more of module semantics. Other candidates: "frame-pointer" (https://github.com/ClangBuiltLinux/linux/issues/955 https://github.com/ClangBuiltLinux/linux/issues/1238), dso_local, etc. Differential Revision: https://reviews.llvm.org/D100251
2021-04-22 00:58:20 +02:00
return F;
}
void Function::removeFromParent() {
getParent()->getFunctionList().remove(getIterator());
}
void Function::eraseFromParent() {
getParent()->getFunctionList().erase(getIterator());
}
//===----------------------------------------------------------------------===//
// Function Implementation
//===----------------------------------------------------------------------===//
static unsigned computeAddrSpace(unsigned AddrSpace, Module *M) {
// If AS == -1 and we are passed a valid module pointer we place the function
// in the program address space. Otherwise we default to AS0.
if (AddrSpace == static_cast<unsigned>(-1))
return M ? M->getDataLayout().getProgramAddressSpace() : 0;
return AddrSpace;
}
Function::Function(FunctionType *Ty, LinkageTypes Linkage, unsigned AddrSpace,
const Twine &name, Module *ParentModule)
: GlobalObject(Ty, Value::FunctionVal,
OperandTraits<Function>::op_begin(this), 0, Linkage, name,
computeAddrSpace(AddrSpace, ParentModule)),
NumArgs(Ty->getNumParams()) {
assert(FunctionType::isValidReturnType(getReturnType()) &&
"invalid return type");
setGlobalObjectSubClassData(0);
// We only need a symbol table for a function if the context keeps value names
if (!getContext().shouldDiscardValueNames())
SymTab = std::make_unique<ValueSymbolTable>(NonGlobalValueMaxNameSize);
// If the function has arguments, mark them as lazily built.
if (Ty->getNumParams())
setValueSubclassData(1); // Set the "has lazy arguments" bit.
if (ParentModule)
ParentModule->getFunctionList().push_back(this);
HasLLVMReservedName = getName().startswith("llvm.");
// Ensure intrinsics have the right parameter attributes.
// Note, the IntID field will have been set in Value::setName if this function
// name is a valid intrinsic ID.
if (IntID)
setAttributes(Intrinsic::getAttributes(getContext(), IntID));
2001-06-06 22:29:01 +02:00
}
Function::~Function() {
dropAllReferences(); // After this it is safe to delete instructions.
2001-06-06 22:29:01 +02:00
// Delete all of the method arguments and unlink from symbol table...
if (Arguments)
clearArguments();
// Remove the function from the on-the-side GC table.
clearGC();
2001-06-06 22:29:01 +02:00
}
void Function::BuildLazyArguments() const {
// Create the arguments vector, all arguments start out unnamed.
auto *FT = getFunctionType();
if (NumArgs > 0) {
Arguments = std::allocator<Argument>().allocate(NumArgs);
for (unsigned i = 0, e = NumArgs; i != e; ++i) {
Type *ArgTy = FT->getParamType(i);
assert(!ArgTy->isVoidTy() && "Cannot have void typed arguments!");
new (Arguments + i) Argument(ArgTy, "", const_cast<Function *>(this), i);
}
}
// Clear the lazy arguments bit.
unsigned SDC = getSubclassDataFromValue();
SDC &= ~(1 << 0);
const_cast<Function*>(this)->setValueSubclassData(SDC);
assert(!hasLazyArguments());
}
static MutableArrayRef<Argument> makeArgArray(Argument *Args, size_t Count) {
return MutableArrayRef<Argument>(Args, Count);
}
bool Function::isConstrainedFPIntrinsic() const {
switch (getIntrinsicID()) {
#define INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC) \
case Intrinsic::INTRINSIC:
#include "llvm/IR/ConstrainedOps.def"
return true;
#undef INSTRUCTION
default:
return false;
}
}
void Function::clearArguments() {
for (Argument &A : makeArgArray(Arguments, NumArgs)) {
A.setName("");
A.~Argument();
}
std::allocator<Argument>().deallocate(Arguments, NumArgs);
Arguments = nullptr;
}
void Function::stealArgumentListFrom(Function &Src) {
assert(isDeclaration() && "Expected no references to current arguments");
// Drop the current arguments, if any, and set the lazy argument bit.
if (!hasLazyArguments()) {
assert(llvm::all_of(makeArgArray(Arguments, NumArgs),
[](const Argument &A) { return A.use_empty(); }) &&
"Expected arguments to be unused in declaration");
clearArguments();
setValueSubclassData(getSubclassDataFromValue() | (1 << 0));
}
// Nothing to steal if Src has lazy arguments.
if (Src.hasLazyArguments())
return;
// Steal arguments from Src, and fix the lazy argument bits.
assert(arg_size() == Src.arg_size());
Arguments = Src.Arguments;
Src.Arguments = nullptr;
for (Argument &A : makeArgArray(Arguments, NumArgs)) {
// FIXME: This does the work of transferNodesFromList inefficiently.
SmallString<128> Name;
if (A.hasName())
Name = A.getName();
if (!Name.empty())
A.setName("");
A.setParent(this);
if (!Name.empty())
A.setName(Name);
}
setValueSubclassData(getSubclassDataFromValue() & ~(1 << 0));
assert(!hasLazyArguments());
Src.setValueSubclassData(Src.getSubclassDataFromValue() | (1 << 0));
}
2001-06-06 22:29:01 +02:00
// dropAllReferences() - This function causes all the subinstructions to "let
// go" of all references that they are maintaining. This allows one to
// 'delete' a whole class at a time, even though there may be circular
// references... first all references are dropped, and all use counts go to
2003-10-10 19:54:14 +02:00
// zero. Then everything is deleted for real. Note that no operations are
// valid on an object that has "dropped all references", except operator
2001-06-06 22:29:01 +02:00
// delete.
//
void Function::dropAllReferences() {
setIsMaterializable(false);
for (BasicBlock &BB : *this)
BB.dropAllReferences();
// Delete all basic blocks. They are now unused, except possibly by
// blockaddresses, but BasicBlock's destructor takes care of those.
while (!BasicBlocks.empty())
BasicBlocks.begin()->eraseFromParent();
// Drop uses of any optional data (real or placeholder).
if (getNumOperands()) {
User::dropAllReferences();
setNumHungOffUseOperands(0);
setValueSubclassData(getSubclassDataFromValue() & ~0xe);
}
// Metadata is stored in a side-table.
clearMetadata();
2001-06-06 22:29:01 +02:00
}
void Function::addAttribute(unsigned i, Attribute::AttrKind Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.addAttribute(getContext(), i, Kind);
setAttributes(PAL);
}
void Function::addAttribute(unsigned i, Attribute Attr) {
AttributeList PAL = getAttributes();
PAL = PAL.addAttribute(getContext(), i, Attr);
setAttributes(PAL);
}
void Function::addAttributes(unsigned i, const AttrBuilder &Attrs) {
AttributeList PAL = getAttributes();
PAL = PAL.addAttributes(getContext(), i, Attrs);
setAttributes(PAL);
}
void Function::addParamAttr(unsigned ArgNo, Attribute::AttrKind Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.addParamAttribute(getContext(), ArgNo, Kind);
setAttributes(PAL);
}
void Function::addParamAttr(unsigned ArgNo, Attribute Attr) {
AttributeList PAL = getAttributes();
PAL = PAL.addParamAttribute(getContext(), ArgNo, Attr);
setAttributes(PAL);
}
void Function::addParamAttrs(unsigned ArgNo, const AttrBuilder &Attrs) {
AttributeList PAL = getAttributes();
PAL = PAL.addParamAttributes(getContext(), ArgNo, Attrs);
setAttributes(PAL);
}
void Function::removeAttribute(unsigned i, Attribute::AttrKind Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.removeAttribute(getContext(), i, Kind);
setAttributes(PAL);
}
void Function::removeAttribute(unsigned i, StringRef Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.removeAttribute(getContext(), i, Kind);
setAttributes(PAL);
}
void Function::removeAttributes(unsigned i, const AttrBuilder &Attrs) {
AttributeList PAL = getAttributes();
PAL = PAL.removeAttributes(getContext(), i, Attrs);
setAttributes(PAL);
}
void Function::removeParamAttr(unsigned ArgNo, Attribute::AttrKind Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.removeParamAttribute(getContext(), ArgNo, Kind);
setAttributes(PAL);
}
void Function::removeParamAttr(unsigned ArgNo, StringRef Kind) {
AttributeList PAL = getAttributes();
PAL = PAL.removeParamAttribute(getContext(), ArgNo, Kind);
setAttributes(PAL);
}
void Function::removeParamAttrs(unsigned ArgNo, const AttrBuilder &Attrs) {
AttributeList PAL = getAttributes();
PAL = PAL.removeParamAttributes(getContext(), ArgNo, Attrs);
setAttributes(PAL);
}
void Function::addDereferenceableAttr(unsigned i, uint64_t Bytes) {
AttributeList PAL = getAttributes();
PAL = PAL.addDereferenceableAttr(getContext(), i, Bytes);
setAttributes(PAL);
}
void Function::addDereferenceableParamAttr(unsigned ArgNo, uint64_t Bytes) {
AttributeList PAL = getAttributes();
PAL = PAL.addDereferenceableParamAttr(getContext(), ArgNo, Bytes);
setAttributes(PAL);
}
void Function::addDereferenceableOrNullAttr(unsigned i, uint64_t Bytes) {
AttributeList PAL = getAttributes();
PAL = PAL.addDereferenceableOrNullAttr(getContext(), i, Bytes);
setAttributes(PAL);
}
void Function::addDereferenceableOrNullParamAttr(unsigned ArgNo,
uint64_t Bytes) {
AttributeList PAL = getAttributes();
PAL = PAL.addDereferenceableOrNullParamAttr(getContext(), ArgNo, Bytes);
setAttributes(PAL);
}
DenormalMode Function::getDenormalMode(const fltSemantics &FPType) const {
if (&FPType == &APFloat::IEEEsingle()) {
Attribute Attr = getFnAttribute("denormal-fp-math-f32");
StringRef Val = Attr.getValueAsString();
if (!Val.empty())
return parseDenormalFPAttribute(Val);
// If the f32 variant of the attribute isn't specified, try to use the
// generic one.
}
Attribute Attr = getFnAttribute("denormal-fp-math");
return parseDenormalFPAttribute(Attr.getValueAsString());
}
const std::string &Function::getGC() const {
assert(hasGC() && "Function has no collector");
return getContext().getGC(*this);
}
void Function::setGC(std::string Str) {
setValueSubclassDataBit(14, !Str.empty());
getContext().setGC(*this, std::move(Str));
}
void Function::clearGC() {
if (!hasGC())
return;
getContext().deleteGC(*this);
setValueSubclassDataBit(14, false);
}
[Inline] prevent inlining on stack protector mismatch It's common for code that manipulates the stack via inline assembly or that has to set up its own stack canary (such as the Linux kernel) would like to avoid stack protectors in certain functions. In this case, we've been bitten by numerous bugs where a callee with a stack protector is inlined into an attribute((no_stack_protector)) caller, which generally breaks the caller's assumptions about not having a stack protector. LTO exacerbates the issue. While developers can avoid this by putting all no_stack_protector functions in one translation unit together and compiling those with -fno-stack-protector, it's generally not very ergonomic or as ergonomic as a function attribute, and still doesn't work for LTO. See also: https://lore.kernel.org/linux-pm/20200915172658.1432732-1-rkir@google.com/ https://lore.kernel.org/lkml/20200918201436.2932360-30-samitolvanen@google.com/T/#u SSP attributes can be ordered by strength. Weakest to strongest, they are: ssp, sspstrong, sspreq. Callees with differing SSP attributes may be inlined into each other, and the strongest attribute will be applied to the caller. (No change) After this change: * A callee with no SSP attributes will no longer be inlined into a caller with SSP attributes. * The reverse is also true: a callee with an SSP attribute will not be inlined into a caller with no SSP attributes. * The alwaysinline attribute overrides these rules. Functions that get synthesized by the compiler may not get inlined as a result if they are not created with the same stack protector function attribute as their callers. Alternative approach to https://reviews.llvm.org/D87956. Fixes pr/47479. Signed-off-by: Nick Desaulniers <ndesaulniers@google.com> Reviewed By: rnk, MaskRay Differential Revision: https://reviews.llvm.org/D91816
2020-12-02 19:44:35 +01:00
bool Function::hasStackProtectorFnAttr() const {
return hasFnAttribute(Attribute::StackProtect) ||
hasFnAttribute(Attribute::StackProtectStrong) ||
hasFnAttribute(Attribute::StackProtectReq);
}
/// Copy all additional attributes (those not needed to create a Function) from
/// the Function Src to this one.
void Function::copyAttributesFrom(const Function *Src) {
GlobalObject::copyAttributesFrom(Src);
setCallingConv(Src->getCallingConv());
setAttributes(Src->getAttributes());
if (Src->hasGC())
setGC(Src->getGC());
else
clearGC();
if (Src->hasPersonalityFn())
setPersonalityFn(Src->getPersonalityFn());
if (Src->hasPrefixData())
setPrefixData(Src->getPrefixData());
if (Src->hasPrologueData())
setPrologueData(Src->getPrologueData());
}
/// Table of string intrinsic names indexed by enum value.
static const char * const IntrinsicNameTable[] = {
"not_intrinsic",
#define GET_INTRINSIC_NAME_TABLE
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_INTRINSIC_NAME_TABLE
};
/// Table of per-target intrinsic name tables.
#define GET_INTRINSIC_TARGET_DATA
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_INTRINSIC_TARGET_DATA
bool Function::isTargetIntrinsic(Intrinsic::ID IID) {
return IID > TargetInfos[0].Count;
}
bool Function::isTargetIntrinsic() const {
return isTargetIntrinsic(IntID);
}
/// Find the segment of \c IntrinsicNameTable for intrinsics with the same
/// target as \c Name, or the generic table if \c Name is not target specific.
///
/// Returns the relevant slice of \c IntrinsicNameTable
static ArrayRef<const char *> findTargetSubtable(StringRef Name) {
assert(Name.startswith("llvm."));
ArrayRef<IntrinsicTargetInfo> Targets(TargetInfos);
// Drop "llvm." and take the first dotted component. That will be the target
// if this is target specific.
StringRef Target = Name.drop_front(5).split('.').first;
auto It = partition_point(
Targets, [=](const IntrinsicTargetInfo &TI) { return TI.Name < Target; });
// We've either found the target or just fall back to the generic set, which
// is always first.
const auto &TI = It != Targets.end() && It->Name == Target ? *It : Targets[0];
return makeArrayRef(&IntrinsicNameTable[1] + TI.Offset, TI.Count);
}
/// This does the actual lookup of an intrinsic ID which
/// matches the given function name.
Intrinsic::ID Function::lookupIntrinsicID(StringRef Name) {
ArrayRef<const char *> NameTable = findTargetSubtable(Name);
int Idx = Intrinsic::lookupLLVMIntrinsicByName(NameTable, Name);
if (Idx == -1)
return Intrinsic::not_intrinsic;
// Intrinsic IDs correspond to the location in IntrinsicNameTable, but we have
// an index into a sub-table.
int Adjust = NameTable.data() - IntrinsicNameTable;
Intrinsic::ID ID = static_cast<Intrinsic::ID>(Idx + Adjust);
// If the intrinsic is not overloaded, require an exact match. If it is
// overloaded, require either exact or prefix match.
const auto MatchSize = strlen(NameTable[Idx]);
assert(Name.size() >= MatchSize && "Expected either exact or prefix match");
bool IsExactMatch = Name.size() == MatchSize;
return IsExactMatch || Intrinsic::isOverloaded(ID) ? ID
: Intrinsic::not_intrinsic;
}
void Function::recalculateIntrinsicID() {
StringRef Name = getName();
if (!Name.startswith("llvm.")) {
HasLLVMReservedName = false;
IntID = Intrinsic::not_intrinsic;
return;
}
HasLLVMReservedName = true;
IntID = lookupIntrinsicID(Name);
}
/// Returns a stable mangling for the type specified for use in the name
/// mangling scheme used by 'any' types in intrinsic signatures. The mangling
/// of named types is simply their name. Manglings for unnamed types consist
/// of a prefix ('p' for pointers, 'a' for arrays, 'f_' for functions)
/// combined with the mangling of their component types. A vararg function
/// type will have a suffix of 'vararg'. Since function types can contain
/// other function types, we close a function type mangling with suffix 'f'
/// which can't be confused with it's prefix. This ensures we don't have
/// collisions between two unrelated function types. Otherwise, you might
/// parse ffXX as f(fXX) or f(fX)X. (X is a placeholder for any other type.)
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
/// The HasUnnamedType boolean is set if an unnamed type was encountered,
/// indicating that extra care must be taken to ensure a unique name.
static std::string getMangledTypeStr(Type *Ty, bool &HasUnnamedType) {
std::string Result;
if (PointerType *PTyp = dyn_cast<PointerType>(Ty)) {
Result += "p" + utostr(PTyp->getAddressSpace());
// Opaque pointer doesn't have pointee type information, so we just mangle
// address space for opaque pointer.
if (!PTyp->isOpaque())
Result += getMangledTypeStr(PTyp->getElementType(), HasUnnamedType);
} else if (ArrayType *ATyp = dyn_cast<ArrayType>(Ty)) {
Result += "a" + utostr(ATyp->getNumElements()) +
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
getMangledTypeStr(ATyp->getElementType(), HasUnnamedType);
} else if (StructType *STyp = dyn_cast<StructType>(Ty)) {
if (!STyp->isLiteral()) {
Result += "s_";
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
if (STyp->hasName())
Result += STyp->getName();
else
HasUnnamedType = true;
} else {
Result += "sl_";
for (auto Elem : STyp->elements())
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
Result += getMangledTypeStr(Elem, HasUnnamedType);
}
// Ensure nested structs are distinguishable.
Result += "s";
} else if (FunctionType *FT = dyn_cast<FunctionType>(Ty)) {
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
Result += "f_" + getMangledTypeStr(FT->getReturnType(), HasUnnamedType);
for (size_t i = 0; i < FT->getNumParams(); i++)
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
Result += getMangledTypeStr(FT->getParamType(i), HasUnnamedType);
if (FT->isVarArg())
Result += "vararg";
// Ensure nested function types are distinguishable.
Result += "f";
} else if (VectorType *VTy = dyn_cast<VectorType>(Ty)) {
ElementCount EC = VTy->getElementCount();
if (EC.isScalable())
Result += "nx";
Result += "v" + utostr(EC.getKnownMinValue()) +
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
getMangledTypeStr(VTy->getElementType(), HasUnnamedType);
} else if (Ty) {
switch (Ty->getTypeID()) {
default: llvm_unreachable("Unhandled type");
case Type::VoidTyID: Result += "isVoid"; break;
case Type::MetadataTyID: Result += "Metadata"; break;
case Type::HalfTyID: Result += "f16"; break;
case Type::BFloatTyID: Result += "bf16"; break;
case Type::FloatTyID: Result += "f32"; break;
case Type::DoubleTyID: Result += "f64"; break;
case Type::X86_FP80TyID: Result += "f80"; break;
case Type::FP128TyID: Result += "f128"; break;
case Type::PPC_FP128TyID: Result += "ppcf128"; break;
case Type::X86_MMXTyID: Result += "x86mmx"; break;
case Type::X86_AMXTyID: Result += "x86amx"; break;
case Type::IntegerTyID:
Result += "i" + utostr(cast<IntegerType>(Ty)->getBitWidth());
break;
}
}
return Result;
}
StringRef Intrinsic::getBaseName(ID id) {
assert(id < num_intrinsics && "Invalid intrinsic ID!");
return IntrinsicNameTable[id];
}
StringRef Intrinsic::getName(ID id) {
assert(id < num_intrinsics && "Invalid intrinsic ID!");
assert(!Intrinsic::isOverloaded(id) &&
"This version of getName does not support overloading");
return getBaseName(id);
}
static std::string getIntrinsicNameImpl(Intrinsic::ID Id, ArrayRef<Type *> Tys,
Module *M, FunctionType *FT,
bool EarlyModuleCheck) {
assert(Id < Intrinsic::num_intrinsics && "Invalid intrinsic ID!");
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
assert((Tys.empty() || Intrinsic::isOverloaded(Id)) &&
"This version of getName is for overloaded intrinsics only");
(void)EarlyModuleCheck;
assert((!EarlyModuleCheck || M ||
!any_of(Tys, [](Type *T) { return isa<PointerType>(T); })) &&
"Intrinsic overloading on pointer types need to provide a Module");
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
bool HasUnnamedType = false;
std::string Result(Intrinsic::getBaseName(Id));
for (Type *Ty : Tys)
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
Result += "." + getMangledTypeStr(Ty, HasUnnamedType);
if (HasUnnamedType) {
assert(M && "unnamed types need a module");
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
if (!FT)
FT = Intrinsic::getType(M->getContext(), Id, Tys);
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
else
assert((FT == Intrinsic::getType(M->getContext(), Id, Tys)) &&
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
"Provided FunctionType must match arguments");
return M->getUniqueIntrinsicName(Result, Id, FT);
}
return Result;
}
std::string Intrinsic::getName(ID Id, ArrayRef<Type *> Tys, Module *M,
FunctionType *FT) {
assert(M && "We need to have a Module");
return getIntrinsicNameImpl(Id, Tys, M, FT, true);
}
std::string Intrinsic::getNameNoUnnamedTypes(ID Id, ArrayRef<Type *> Tys) {
return getIntrinsicNameImpl(Id, Tys, nullptr, nullptr, false);
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
}
/// IIT_Info - These are enumerators that describe the entries returned by the
/// getIntrinsicInfoTableEntries function.
///
/// NOTE: This must be kept in synch with the copy in TblGen/IntrinsicEmitter!
enum IIT_Info {
// Common values should be encoded with 0-15.
IIT_Done = 0,
IIT_I1 = 1,
IIT_I8 = 2,
IIT_I16 = 3,
IIT_I32 = 4,
IIT_I64 = 5,
IIT_F16 = 6,
IIT_F32 = 7,
IIT_F64 = 8,
IIT_V2 = 9,
IIT_V4 = 10,
IIT_V8 = 11,
IIT_V16 = 12,
IIT_V32 = 13,
IIT_PTR = 14,
IIT_ARG = 15,
// Values from 16+ are only encodable with the inefficient encoding.
IIT_V64 = 16,
IIT_MMX = 17,
IIT_TOKEN = 18,
IIT_METADATA = 19,
IIT_EMPTYSTRUCT = 20,
IIT_STRUCT2 = 21,
IIT_STRUCT3 = 22,
IIT_STRUCT4 = 23,
IIT_STRUCT5 = 24,
IIT_EXTEND_ARG = 25,
IIT_TRUNC_ARG = 26,
IIT_ANYPTR = 27,
IIT_V1 = 28,
IIT_VARARG = 29,
IIT_HALF_VEC_ARG = 30,
IIT_SAME_VEC_WIDTH_ARG = 31,
IIT_PTR_TO_ARG = 32,
IIT_PTR_TO_ELT = 33,
IIT_VEC_OF_ANYPTRS_TO_ELT = 34,
IIT_I128 = 35,
IIT_V512 = 36,
IIT_V1024 = 37,
IIT_STRUCT6 = 38,
IIT_STRUCT7 = 39,
IIT_STRUCT8 = 40,
IIT_F128 = 41,
IIT_VEC_ELEMENT = 42,
IIT_SCALABLE_VEC = 43,
IIT_SUBDIVIDE2_ARG = 44,
IIT_SUBDIVIDE4_ARG = 45,
2020-02-07 16:30:31 +01:00
IIT_VEC_OF_BITCASTS_TO_INT = 46,
IIT_V128 = 47,
IIT_BF16 = 48,
IIT_STRUCT9 = 49,
IIT_V256 = 50,
IIT_AMX = 51
};
static void DecodeIITType(unsigned &NextElt, ArrayRef<unsigned char> Infos,
IIT_Info LastInfo,
SmallVectorImpl<Intrinsic::IITDescriptor> &OutputTable) {
using namespace Intrinsic;
bool IsScalableVector = (LastInfo == IIT_SCALABLE_VEC);
IIT_Info Info = IIT_Info(Infos[NextElt++]);
unsigned StructElts = 2;
switch (Info) {
case IIT_Done:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Void, 0));
return;
case IIT_VARARG:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::VarArg, 0));
return;
case IIT_MMX:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::MMX, 0));
return;
case IIT_AMX:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::AMX, 0));
return;
case IIT_TOKEN:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Token, 0));
return;
case IIT_METADATA:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Metadata, 0));
return;
case IIT_F16:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Half, 0));
return;
case IIT_BF16:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::BFloat, 0));
return;
case IIT_F32:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Float, 0));
return;
case IIT_F64:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Double, 0));
return;
case IIT_F128:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Quad, 0));
return;
case IIT_I1:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer, 1));
return;
case IIT_I8:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer, 8));
return;
case IIT_I16:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer,16));
return;
case IIT_I32:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer, 32));
return;
case IIT_I64:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer, 64));
return;
case IIT_I128:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Integer, 128));
return;
case IIT_V1:
OutputTable.push_back(IITDescriptor::getVector(1, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V2:
OutputTable.push_back(IITDescriptor::getVector(2, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V4:
OutputTable.push_back(IITDescriptor::getVector(4, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V8:
OutputTable.push_back(IITDescriptor::getVector(8, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V16:
OutputTable.push_back(IITDescriptor::getVector(16, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V32:
OutputTable.push_back(IITDescriptor::getVector(32, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V64:
OutputTable.push_back(IITDescriptor::getVector(64, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
2020-02-07 16:30:31 +01:00
case IIT_V128:
OutputTable.push_back(IITDescriptor::getVector(128, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
2020-02-07 16:30:31 +01:00
return;
case IIT_V256:
OutputTable.push_back(IITDescriptor::getVector(256, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V512:
OutputTable.push_back(IITDescriptor::getVector(512, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_V1024:
OutputTable.push_back(IITDescriptor::getVector(1024, IsScalableVector));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_PTR:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Pointer, 0));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
case IIT_ANYPTR: { // [ANYPTR addrspace, subtype]
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Pointer,
Infos[NextElt++]));
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
}
case IIT_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Argument, ArgInfo));
return;
}
case IIT_EXTEND_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::ExtendArgument,
ArgInfo));
return;
}
case IIT_TRUNC_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::TruncArgument,
ArgInfo));
return;
}
case IIT_HALF_VEC_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::HalfVecArgument,
ArgInfo));
return;
}
case IIT_SAME_VEC_WIDTH_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::SameVecWidthArgument,
ArgInfo));
return;
}
case IIT_PTR_TO_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::PtrToArgument,
ArgInfo));
return;
}
case IIT_PTR_TO_ELT: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::PtrToElt, ArgInfo));
return;
}
case IIT_VEC_OF_ANYPTRS_TO_ELT: {
unsigned short ArgNo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
unsigned short RefNo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(
IITDescriptor::get(IITDescriptor::VecOfAnyPtrsToElt, ArgNo, RefNo));
Masked Gather and Scatter Intrinsics. Gather and Scatter are new introduced intrinsics, comming after recently implemented masked load and store. This is the first patch for Gather and Scatter intrinsics. It includes only the syntax, parsing and verification. Gather and Scatter intrinsics allow to perform multiple memory accesses (read/write) in one vector instruction. The intrinsics are not target specific and will have the following syntax: Gather: declare <16 x i32> @llvm.masked.gather.v16i32(<16 x i32*> <vector of ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x i32> <passthru>) declare <8 x float> @llvm.masked.gather.v8f32(<8 x float*><vector of ptrs>, i32 <alignment>, <8 x i1> <mask>, <8 x float><passthru>) Scatter: declare void @llvm.masked.scatter.v8i32(<8 x i32><vector value to be stored> , <8 x i32*><vector of ptrs> , i32 <alignment>, <8 x i1> <mask>) declare void @llvm.masked.scatter.v16i32(<16 x i32> <vector value to be stored> , <16 x i32*> <vector of ptrs>, i32 <alignment>, <16 x i1><mask> ) Vector of ptrs - a set of source/destination addresses, to load/store the value. Mask - switches on/off vector lanes to prevent memory access for switched-off lanes vector of ptrs, value and mask should have the same vector width. These are code examples where gather / scatter should be used and will allow function vectorization ;void foo1(int * restrict A, int * restrict B, int * restrict C) { ; for (int i=0; i<SIZE; i++) { ; A[i] = B[C[i]]; ; } ;} ;void foo3(int * restrict A, int * restrict B) { ; for (int i=0; i<SIZE; i++) { ; A[B[i]] = i+5; ; } ;} Tests will come in the following patches, with CodeGen and Vectorizer. http://reviews.llvm.org/D7433 llvm-svn: 228521
2015-02-08 09:27:19 +01:00
return;
}
case IIT_EMPTYSTRUCT:
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Struct, 0));
return;
case IIT_STRUCT9: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT8: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT7: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT6: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT5: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT4: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT3: ++StructElts; LLVM_FALLTHROUGH;
case IIT_STRUCT2: {
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Struct,StructElts));
for (unsigned i = 0; i != StructElts; ++i)
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
}
case IIT_SUBDIVIDE2_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Subdivide2Argument,
ArgInfo));
return;
}
case IIT_SUBDIVIDE4_ARG: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::Subdivide4Argument,
ArgInfo));
return;
}
case IIT_VEC_ELEMENT: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::VecElementArgument,
ArgInfo));
return;
}
case IIT_SCALABLE_VEC: {
DecodeIITType(NextElt, Infos, Info, OutputTable);
return;
}
case IIT_VEC_OF_BITCASTS_TO_INT: {
unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
OutputTable.push_back(IITDescriptor::get(IITDescriptor::VecOfBitcastsToInt,
ArgInfo));
return;
}
}
llvm_unreachable("unhandled");
}
#define GET_INTRINSIC_GENERATOR_GLOBAL
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_INTRINSIC_GENERATOR_GLOBAL
void Intrinsic::getIntrinsicInfoTableEntries(ID id,
SmallVectorImpl<IITDescriptor> &T){
// Check to see if the intrinsic's type was expressible by the table.
unsigned TableVal = IIT_Table[id-1];
// Decode the TableVal into an array of IITValues.
SmallVector<unsigned char, 8> IITValues;
ArrayRef<unsigned char> IITEntries;
unsigned NextElt = 0;
if ((TableVal >> 31) != 0) {
// This is an offset into the IIT_LongEncodingTable.
IITEntries = IIT_LongEncodingTable;
// Strip sentinel bit.
NextElt = (TableVal << 1) >> 1;
} else {
// Decode the TableVal into an array of IITValues. If the entry was encoded
// into a single word in the table itself, decode it now.
do {
IITValues.push_back(TableVal & 0xF);
TableVal >>= 4;
} while (TableVal);
IITEntries = IITValues;
NextElt = 0;
}
// Okay, decode the table into the output vector of IITDescriptors.
DecodeIITType(NextElt, IITEntries, IIT_Done, T);
while (NextElt != IITEntries.size() && IITEntries[NextElt] != 0)
DecodeIITType(NextElt, IITEntries, IIT_Done, T);
}
static Type *DecodeFixedType(ArrayRef<Intrinsic::IITDescriptor> &Infos,
ArrayRef<Type*> Tys, LLVMContext &Context) {
using namespace Intrinsic;
IITDescriptor D = Infos.front();
Infos = Infos.slice(1);
switch (D.Kind) {
case IITDescriptor::Void: return Type::getVoidTy(Context);
case IITDescriptor::VarArg: return Type::getVoidTy(Context);
case IITDescriptor::MMX: return Type::getX86_MMXTy(Context);
case IITDescriptor::AMX: return Type::getX86_AMXTy(Context);
case IITDescriptor::Token: return Type::getTokenTy(Context);
case IITDescriptor::Metadata: return Type::getMetadataTy(Context);
case IITDescriptor::Half: return Type::getHalfTy(Context);
case IITDescriptor::BFloat: return Type::getBFloatTy(Context);
case IITDescriptor::Float: return Type::getFloatTy(Context);
case IITDescriptor::Double: return Type::getDoubleTy(Context);
case IITDescriptor::Quad: return Type::getFP128Ty(Context);
case IITDescriptor::Integer:
return IntegerType::get(Context, D.Integer_Width);
case IITDescriptor::Vector:
return VectorType::get(DecodeFixedType(Infos, Tys, Context),
D.Vector_Width);
case IITDescriptor::Pointer:
return PointerType::get(DecodeFixedType(Infos, Tys, Context),
D.Pointer_AddressSpace);
case IITDescriptor::Struct: {
SmallVector<Type *, 8> Elts;
for (unsigned i = 0, e = D.Struct_NumElements; i != e; ++i)
Elts.push_back(DecodeFixedType(Infos, Tys, Context));
return StructType::get(Context, Elts);
}
case IITDescriptor::Argument:
return Tys[D.getArgumentNumber()];
case IITDescriptor::ExtendArgument: {
Type *Ty = Tys[D.getArgumentNumber()];
if (VectorType *VTy = dyn_cast<VectorType>(Ty))
return VectorType::getExtendedElementVectorType(VTy);
return IntegerType::get(Context, 2 * cast<IntegerType>(Ty)->getBitWidth());
}
case IITDescriptor::TruncArgument: {
Type *Ty = Tys[D.getArgumentNumber()];
if (VectorType *VTy = dyn_cast<VectorType>(Ty))
return VectorType::getTruncatedElementVectorType(VTy);
IntegerType *ITy = cast<IntegerType>(Ty);
assert(ITy->getBitWidth() % 2 == 0);
return IntegerType::get(Context, ITy->getBitWidth() / 2);
}
case IITDescriptor::Subdivide2Argument:
case IITDescriptor::Subdivide4Argument: {
Type *Ty = Tys[D.getArgumentNumber()];
VectorType *VTy = dyn_cast<VectorType>(Ty);
assert(VTy && "Expected an argument of Vector Type");
int SubDivs = D.Kind == IITDescriptor::Subdivide2Argument ? 1 : 2;
return VectorType::getSubdividedVectorType(VTy, SubDivs);
}
case IITDescriptor::HalfVecArgument:
return VectorType::getHalfElementsVectorType(cast<VectorType>(
Tys[D.getArgumentNumber()]));
case IITDescriptor::SameVecWidthArgument: {
Type *EltTy = DecodeFixedType(Infos, Tys, Context);
Type *Ty = Tys[D.getArgumentNumber()];
if (auto *VTy = dyn_cast<VectorType>(Ty))
return VectorType::get(EltTy, VTy->getElementCount());
return EltTy;
}
case IITDescriptor::PtrToArgument: {
Type *Ty = Tys[D.getArgumentNumber()];
return PointerType::getUnqual(Ty);
}
case IITDescriptor::PtrToElt: {
Type *Ty = Tys[D.getArgumentNumber()];
VectorType *VTy = dyn_cast<VectorType>(Ty);
if (!VTy)
llvm_unreachable("Expected an argument of Vector Type");
Type *EltTy = VTy->getElementType();
return PointerType::getUnqual(EltTy);
}
case IITDescriptor::VecElementArgument: {
Type *Ty = Tys[D.getArgumentNumber()];
if (VectorType *VTy = dyn_cast<VectorType>(Ty))
return VTy->getElementType();
llvm_unreachable("Expected an argument of Vector Type");
}
case IITDescriptor::VecOfBitcastsToInt: {
Type *Ty = Tys[D.getArgumentNumber()];
VectorType *VTy = dyn_cast<VectorType>(Ty);
assert(VTy && "Expected an argument of Vector Type");
return VectorType::getInteger(VTy);
}
case IITDescriptor::VecOfAnyPtrsToElt:
// Return the overloaded type (which determines the pointers address space)
return Tys[D.getOverloadArgNumber()];
}
llvm_unreachable("unhandled");
}
FunctionType *Intrinsic::getType(LLVMContext &Context,
ID id, ArrayRef<Type*> Tys) {
SmallVector<IITDescriptor, 8> Table;
getIntrinsicInfoTableEntries(id, Table);
ArrayRef<IITDescriptor> TableRef = Table;
Type *ResultTy = DecodeFixedType(TableRef, Tys, Context);
SmallVector<Type*, 8> ArgTys;
while (!TableRef.empty())
ArgTys.push_back(DecodeFixedType(TableRef, Tys, Context));
// DecodeFixedType returns Void for IITDescriptor::Void and IITDescriptor::VarArg
// If we see void type as the type of the last argument, it is vararg intrinsic
if (!ArgTys.empty() && ArgTys.back()->isVoidTy()) {
ArgTys.pop_back();
return FunctionType::get(ResultTy, ArgTys, true);
}
return FunctionType::get(ResultTy, ArgTys, false);
}
bool Intrinsic::isOverloaded(ID id) {
#define GET_INTRINSIC_OVERLOAD_TABLE
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_INTRINSIC_OVERLOAD_TABLE
}
bool Intrinsic::isLeaf(ID id) {
switch (id) {
default:
return true;
case Intrinsic::experimental_gc_statepoint:
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
return false;
}
}
/// This defines the "Intrinsic::getAttributes(ID id)" method.
#define GET_INTRINSIC_ATTRIBUTES
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_INTRINSIC_ATTRIBUTES
Function *Intrinsic::getDeclaration(Module *M, ID id, ArrayRef<Type*> Tys) {
// There can never be multiple globals with the same name of different types,
// because intrinsics must be a specific type.
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
auto *FT = getType(M->getContext(), id, Tys);
[opaque pointer types] Add a FunctionCallee wrapper type, and use it. Recommit r352791 after tweaking DerivedTypes.h slightly, so that gcc doesn't choke on it, hopefully. Original Message: The FunctionCallee type is effectively a {FunctionType*,Value*} pair, and is a useful convenience to enable code to continue passing the result of getOrInsertFunction() through to EmitCall, even once pointer types lose their pointee-type. Then: - update the CallInst/InvokeInst instruction creation functions to take a Callee, - modify getOrInsertFunction to return FunctionCallee, and - update all callers appropriately. One area of particular note is the change to the sanitizer code. Previously, they had been casting the result of `getOrInsertFunction` to a `Function*` via `checkSanitizerInterfaceFunction`, and storing that. That would report an error if someone had already inserted a function declaraction with a mismatching signature. However, in general, LLVM allows for such mismatches, as `getOrInsertFunction` will automatically insert a bitcast if needed. As part of this cleanup, cause the sanitizer code to do the same. (It will call its functions using the expected signature, however they may have been declared.) Finally, in a small number of locations, callers of `getOrInsertFunction` actually were expecting/requiring that a brand new function was being created. In such cases, I've switched them to Function::Create instead. Differential Revision: https://reviews.llvm.org/D57315 llvm-svn: 352827
2019-02-01 03:28:03 +01:00
return cast<Function>(
Support intrinsic overloading on unnamed types This patch adds support for intrinsic overloading on unnamed types. This fixes PR38117 and PR48340 and will also be needed for the Full Restrict Patches (D68484). The main problem is that the intrinsic overloading name mangling is using 's_s' for unnamed types. This can result in identical intrinsic mangled names for different function prototypes. This patch changes this by adding a '.XXXXX' to the intrinsic mangled name when at least one of the types is based on an unnamed type, ensuring that we get a unique name. Implementation details: - The mapping is created on demand and kept in Module. - It also checks for existing clashes and recycles potentially existing prototypes and declarations. - Because of extra data in Module, Intrinsic::getName needs an extra Module* argument and, for speed, an optional FunctionType* argument. - I still kept the original two-argument 'Intrinsic::getName' around which keeps the original behavior (providing the base name). -- Main reason is that I did not want to change the LLVMIntrinsicGetName version, as I don't know how acceptable such a change is -- The current situation already has a limitation. So that should not get worse with this patch. - Intrinsic::getDeclaration and the verifier are now using the new version. Other notes: - As far as I see, this should not suffer from stability issues. The count is only added for prototypes depending on at least one anonymous struct - The initial count starts from 0 for each intrinsic mangled name. - In case of name clashes, existing prototypes are remembered and reused when that makes sense. Reviewed By: fhahn Differential Revision: https://reviews.llvm.org/D91250
2021-03-19 14:34:25 +01:00
M->getOrInsertFunction(Tys.empty() ? getName(id)
: getName(id, Tys, M, FT),
[opaque pointer types] Add a FunctionCallee wrapper type, and use it. Recommit r352791 after tweaking DerivedTypes.h slightly, so that gcc doesn't choke on it, hopefully. Original Message: The FunctionCallee type is effectively a {FunctionType*,Value*} pair, and is a useful convenience to enable code to continue passing the result of getOrInsertFunction() through to EmitCall, even once pointer types lose their pointee-type. Then: - update the CallInst/InvokeInst instruction creation functions to take a Callee, - modify getOrInsertFunction to return FunctionCallee, and - update all callers appropriately. One area of particular note is the change to the sanitizer code. Previously, they had been casting the result of `getOrInsertFunction` to a `Function*` via `checkSanitizerInterfaceFunction`, and storing that. That would report an error if someone had already inserted a function declaraction with a mismatching signature. However, in general, LLVM allows for such mismatches, as `getOrInsertFunction` will automatically insert a bitcast if needed. As part of this cleanup, cause the sanitizer code to do the same. (It will call its functions using the expected signature, however they may have been declared.) Finally, in a small number of locations, callers of `getOrInsertFunction` actually were expecting/requiring that a brand new function was being created. In such cases, I've switched them to Function::Create instead. Differential Revision: https://reviews.llvm.org/D57315 llvm-svn: 352827
2019-02-01 03:28:03 +01:00
getType(M->getContext(), id, Tys))
.getCallee());
}
// This defines the "Intrinsic::getIntrinsicForGCCBuiltin()" method.
#define GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN
// This defines the "Intrinsic::getIntrinsicForMSBuiltin()" method.
#define GET_LLVM_INTRINSIC_FOR_MS_BUILTIN
#include "llvm/IR/IntrinsicImpl.inc"
#undef GET_LLVM_INTRINSIC_FOR_MS_BUILTIN
using DeferredIntrinsicMatchPair =
std::pair<Type *, ArrayRef<Intrinsic::IITDescriptor>>;
static bool matchIntrinsicType(
Type *Ty, ArrayRef<Intrinsic::IITDescriptor> &Infos,
SmallVectorImpl<Type *> &ArgTys,
SmallVectorImpl<DeferredIntrinsicMatchPair> &DeferredChecks,
bool IsDeferredCheck) {
using namespace Intrinsic;
// If we ran out of descriptors, there are too many arguments.
if (Infos.empty()) return true;
// Do this before slicing off the 'front' part
auto InfosRef = Infos;
auto DeferCheck = [&DeferredChecks, &InfosRef](Type *T) {
DeferredChecks.emplace_back(T, InfosRef);
return false;
};
IITDescriptor D = Infos.front();
Infos = Infos.slice(1);
switch (D.Kind) {
case IITDescriptor::Void: return !Ty->isVoidTy();
case IITDescriptor::VarArg: return true;
case IITDescriptor::MMX: return !Ty->isX86_MMXTy();
case IITDescriptor::AMX: return !Ty->isX86_AMXTy();
case IITDescriptor::Token: return !Ty->isTokenTy();
case IITDescriptor::Metadata: return !Ty->isMetadataTy();
case IITDescriptor::Half: return !Ty->isHalfTy();
case IITDescriptor::BFloat: return !Ty->isBFloatTy();
case IITDescriptor::Float: return !Ty->isFloatTy();
case IITDescriptor::Double: return !Ty->isDoubleTy();
case IITDescriptor::Quad: return !Ty->isFP128Ty();
case IITDescriptor::Integer: return !Ty->isIntegerTy(D.Integer_Width);
case IITDescriptor::Vector: {
VectorType *VT = dyn_cast<VectorType>(Ty);
return !VT || VT->getElementCount() != D.Vector_Width ||
matchIntrinsicType(VT->getElementType(), Infos, ArgTys,
DeferredChecks, IsDeferredCheck);
}
case IITDescriptor::Pointer: {
PointerType *PT = dyn_cast<PointerType>(Ty);
if (!PT || PT->getAddressSpace() != D.Pointer_AddressSpace)
return true;
if (!PT->isOpaque())
return matchIntrinsicType(PT->getElementType(), Infos, ArgTys,
DeferredChecks, IsDeferredCheck);
// If typed pointers are supported, do not allow using opaque pointer in
// place of fixed pointer type. This would make the intrinsic signature
// non-unique.
if (Ty->getContext().supportsTypedPointers())
return true;
// Consume IIT descriptors relating to the pointer element type.
while (Infos.front().Kind == IITDescriptor::Pointer)
Infos = Infos.slice(1);
Infos = Infos.slice(1);
return false;
}
case IITDescriptor::Struct: {
StructType *ST = dyn_cast<StructType>(Ty);
if (!ST || ST->getNumElements() != D.Struct_NumElements)
return true;
for (unsigned i = 0, e = D.Struct_NumElements; i != e; ++i)
if (matchIntrinsicType(ST->getElementType(i), Infos, ArgTys,
DeferredChecks, IsDeferredCheck))
return true;
return false;
}
case IITDescriptor::Argument:
// If this is the second occurrence of an argument,
// verify that the later instance matches the previous instance.
if (D.getArgumentNumber() < ArgTys.size())
return Ty != ArgTys[D.getArgumentNumber()];
if (D.getArgumentNumber() > ArgTys.size() ||
D.getArgumentKind() == IITDescriptor::AK_MatchType)
return IsDeferredCheck || DeferCheck(Ty);
assert(D.getArgumentNumber() == ArgTys.size() && !IsDeferredCheck &&
"Table consistency error");
ArgTys.push_back(Ty);
switch (D.getArgumentKind()) {
case IITDescriptor::AK_Any: return false; // Success
case IITDescriptor::AK_AnyInteger: return !Ty->isIntOrIntVectorTy();
case IITDescriptor::AK_AnyFloat: return !Ty->isFPOrFPVectorTy();
case IITDescriptor::AK_AnyVector: return !isa<VectorType>(Ty);
case IITDescriptor::AK_AnyPointer: return !isa<PointerType>(Ty);
default: break;
}
llvm_unreachable("all argument kinds not covered");
case IITDescriptor::ExtendArgument: {
// If this is a forward reference, defer the check for later.
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
Type *NewTy = ArgTys[D.getArgumentNumber()];
if (VectorType *VTy = dyn_cast<VectorType>(NewTy))
NewTy = VectorType::getExtendedElementVectorType(VTy);
else if (IntegerType *ITy = dyn_cast<IntegerType>(NewTy))
NewTy = IntegerType::get(ITy->getContext(), 2 * ITy->getBitWidth());
else
return true;
return Ty != NewTy;
}
case IITDescriptor::TruncArgument: {
// If this is a forward reference, defer the check for later.
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
Type *NewTy = ArgTys[D.getArgumentNumber()];
if (VectorType *VTy = dyn_cast<VectorType>(NewTy))
NewTy = VectorType::getTruncatedElementVectorType(VTy);
else if (IntegerType *ITy = dyn_cast<IntegerType>(NewTy))
NewTy = IntegerType::get(ITy->getContext(), ITy->getBitWidth() / 2);
else
return true;
return Ty != NewTy;
}
case IITDescriptor::HalfVecArgument:
// If this is a forward reference, defer the check for later.
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
return !isa<VectorType>(ArgTys[D.getArgumentNumber()]) ||
VectorType::getHalfElementsVectorType(
cast<VectorType>(ArgTys[D.getArgumentNumber()])) != Ty;
case IITDescriptor::SameVecWidthArgument: {
if (D.getArgumentNumber() >= ArgTys.size()) {
// Defer check and subsequent check for the vector element type.
Infos = Infos.slice(1);
return IsDeferredCheck || DeferCheck(Ty);
}
auto *ReferenceType = dyn_cast<VectorType>(ArgTys[D.getArgumentNumber()]);
auto *ThisArgType = dyn_cast<VectorType>(Ty);
// Both must be vectors of the same number of elements or neither.
if ((ReferenceType != nullptr) != (ThisArgType != nullptr))
return true;
Type *EltTy = Ty;
if (ThisArgType) {
if (ReferenceType->getElementCount() !=
ThisArgType->getElementCount())
return true;
EltTy = ThisArgType->getElementType();
}
return matchIntrinsicType(EltTy, Infos, ArgTys, DeferredChecks,
IsDeferredCheck);
}
case IITDescriptor::PtrToArgument: {
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
Type * ReferenceType = ArgTys[D.getArgumentNumber()];
PointerType *ThisArgType = dyn_cast<PointerType>(Ty);
return (!ThisArgType || ThisArgType->getElementType() != ReferenceType);
}
case IITDescriptor::PtrToElt: {
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
VectorType * ReferenceType =
dyn_cast<VectorType> (ArgTys[D.getArgumentNumber()]);
PointerType *ThisArgType = dyn_cast<PointerType>(Ty);
if (!ThisArgType || !ReferenceType)
return true;
if (!ThisArgType->isOpaque())
return ThisArgType->getElementType() != ReferenceType->getElementType();
// If typed pointers are supported, do not allow opaque pointer to ensure
// uniqueness.
return Ty->getContext().supportsTypedPointers();
}
case IITDescriptor::VecOfAnyPtrsToElt: {
unsigned RefArgNumber = D.getRefArgNumber();
if (RefArgNumber >= ArgTys.size()) {
if (IsDeferredCheck)
return true;
// If forward referencing, already add the pointer-vector type and
// defer the checks for later.
ArgTys.push_back(Ty);
return DeferCheck(Ty);
}
if (!IsDeferredCheck){
assert(D.getOverloadArgNumber() == ArgTys.size() &&
"Table consistency error");
ArgTys.push_back(Ty);
}
// Verify the overloaded type "matches" the Ref type.
// i.e. Ty is a vector with the same width as Ref.
// Composed of pointers to the same element type as Ref.
auto *ReferenceType = dyn_cast<VectorType>(ArgTys[RefArgNumber]);
auto *ThisArgVecTy = dyn_cast<VectorType>(Ty);
if (!ThisArgVecTy || !ReferenceType ||
(ReferenceType->getElementCount() != ThisArgVecTy->getElementCount()))
return true;
PointerType *ThisArgEltTy =
dyn_cast<PointerType>(ThisArgVecTy->getElementType());
if (!ThisArgEltTy)
return true;
return !ThisArgEltTy->isOpaqueOrPointeeTypeMatches(
ReferenceType->getElementType());
}
case IITDescriptor::VecElementArgument: {
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck ? true : DeferCheck(Ty);
auto *ReferenceType = dyn_cast<VectorType>(ArgTys[D.getArgumentNumber()]);
return !ReferenceType || Ty != ReferenceType->getElementType();
}
case IITDescriptor::Subdivide2Argument:
case IITDescriptor::Subdivide4Argument: {
// If this is a forward reference, defer the check for later.
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
Type *NewTy = ArgTys[D.getArgumentNumber()];
if (auto *VTy = dyn_cast<VectorType>(NewTy)) {
int SubDivs = D.Kind == IITDescriptor::Subdivide2Argument ? 1 : 2;
NewTy = VectorType::getSubdividedVectorType(VTy, SubDivs);
return Ty != NewTy;
}
return true;
}
case IITDescriptor::VecOfBitcastsToInt: {
if (D.getArgumentNumber() >= ArgTys.size())
return IsDeferredCheck || DeferCheck(Ty);
auto *ReferenceType = dyn_cast<VectorType>(ArgTys[D.getArgumentNumber()]);
auto *ThisArgVecTy = dyn_cast<VectorType>(Ty);
if (!ThisArgVecTy || !ReferenceType)
return true;
return ThisArgVecTy != VectorType::getInteger(ReferenceType);
}
}
llvm_unreachable("unhandled");
}
Intrinsic::MatchIntrinsicTypesResult
Intrinsic::matchIntrinsicSignature(FunctionType *FTy,
ArrayRef<Intrinsic::IITDescriptor> &Infos,
SmallVectorImpl<Type *> &ArgTys) {
SmallVector<DeferredIntrinsicMatchPair, 2> DeferredChecks;
if (matchIntrinsicType(FTy->getReturnType(), Infos, ArgTys, DeferredChecks,
false))
return MatchIntrinsicTypes_NoMatchRet;
unsigned NumDeferredReturnChecks = DeferredChecks.size();
for (auto Ty : FTy->params())
if (matchIntrinsicType(Ty, Infos, ArgTys, DeferredChecks, false))
return MatchIntrinsicTypes_NoMatchArg;
for (unsigned I = 0, E = DeferredChecks.size(); I != E; ++I) {
DeferredIntrinsicMatchPair &Check = DeferredChecks[I];
if (matchIntrinsicType(Check.first, Check.second, ArgTys, DeferredChecks,
true))
return I < NumDeferredReturnChecks ? MatchIntrinsicTypes_NoMatchRet
: MatchIntrinsicTypes_NoMatchArg;
}
return MatchIntrinsicTypes_Match;
}
bool
Intrinsic::matchIntrinsicVarArg(bool isVarArg,
ArrayRef<Intrinsic::IITDescriptor> &Infos) {
// If there are no descriptors left, then it can't be a vararg.
if (Infos.empty())
return isVarArg;
// There should be only one descriptor remaining at this point.
if (Infos.size() != 1)
return true;
// Check and verify the descriptor.
IITDescriptor D = Infos.front();
Infos = Infos.slice(1);
if (D.Kind == IITDescriptor::VarArg)
return !isVarArg;
return true;
}
bool Intrinsic::getIntrinsicSignature(Function *F,
SmallVectorImpl<Type *> &ArgTys) {
Intrinsic::ID ID = F->getIntrinsicID();
if (!ID)
return false;
SmallVector<Intrinsic::IITDescriptor, 8> Table;
getIntrinsicInfoTableEntries(ID, Table);
ArrayRef<Intrinsic::IITDescriptor> TableRef = Table;
if (Intrinsic::matchIntrinsicSignature(F->getFunctionType(), TableRef,
ArgTys) !=
Intrinsic::MatchIntrinsicTypesResult::MatchIntrinsicTypes_Match) {
return false;
}
if (Intrinsic::matchIntrinsicVarArg(F->getFunctionType()->isVarArg(),
TableRef))
return false;
return true;
}
Optional<Function *> Intrinsic::remangleIntrinsicFunction(Function *F) {
SmallVector<Type *, 4> ArgTys;
if (!getIntrinsicSignature(F, ArgTys))
return None;
Intrinsic::ID ID = F->getIntrinsicID();
StringRef Name = F->getName();
std::string WantedName =
Intrinsic::getName(ID, ArgTys, F->getParent(), F->getFunctionType());
if (Name == WantedName)
return None;
Function *NewDecl = [&] {
if (auto *ExistingGV = F->getParent()->getNamedValue(WantedName)) {
if (auto *ExistingF = dyn_cast<Function>(ExistingGV))
if (ExistingF->getFunctionType() == F->getFunctionType())
return ExistingF;
// The name already exists, but is not a function or has the wrong
// prototype. Make place for the new one by renaming the old version.
// Either this old version will be removed later on or the module is
// invalid and we'll get an error.
ExistingGV->setName(WantedName + ".renamed");
}
return Intrinsic::getDeclaration(F->getParent(), ID, ArgTys);
}();
NewDecl->setCallingConv(F->getCallingConv());
assert(NewDecl->getFunctionType() == F->getFunctionType() &&
"Shouldn't change the signature");
return NewDecl;
}
/// hasAddressTaken - returns true if there are any uses of this function
/// other than direct calls or invokes to it. Optionally ignores callback
/// uses, assume like pointer annotation calls, and references in llvm.used
/// and llvm.compiler.used variables.
bool Function::hasAddressTaken(const User **PutOffender,
bool IgnoreCallbackUses,
bool IgnoreAssumeLikeCalls,
bool IgnoreLLVMUsed) const {
[C++11] Add range based accessors for the Use-Def chain of a Value. This requires a number of steps. 1) Move value_use_iterator into the Value class as an implementation detail 2) Change it to actually be a *Use* iterator rather than a *User* iterator. 3) Add an adaptor which is a User iterator that always looks through the Use to the User. 4) Wrap these in Value::use_iterator and Value::user_iterator typedefs. 5) Add the range adaptors as Value::uses() and Value::users(). 6) Update *all* of the callers to correctly distinguish between whether they wanted a use_iterator (and to explicitly dig out the User when needed), or a user_iterator which makes the Use itself totally opaque. Because #6 requires churning essentially everything that walked the Use-Def chains, I went ahead and added all of the range adaptors and switched them to range-based loops where appropriate. Also because the renaming requires at least churning every line of code, it didn't make any sense to split these up into multiple commits -- all of which would touch all of the same lies of code. The result is still not quite optimal. The Value::use_iterator is a nice regular iterator, but Value::user_iterator is an iterator over User*s rather than over the User objects themselves. As a consequence, it fits a bit awkwardly into the range-based world and it has the weird extra-dereferencing 'operator->' that so many of our iterators have. I think this could be fixed by providing something which transforms a range of T&s into a range of T*s, but that *can* be separated into another patch, and it isn't yet 100% clear whether this is the right move. However, this change gets us most of the benefit and cleans up a substantial amount of code around Use and User. =] llvm-svn: 203364
2014-03-09 04:16:01 +01:00
for (const Use &U : uses()) {
const User *FU = U.getUser();
if (isa<BlockAddress>(FU))
continue;
if (IgnoreCallbackUses) {
AbstractCallSite ACS(&U);
if (ACS && ACS.isCallbackCall())
continue;
}
const auto *Call = dyn_cast<CallBase>(FU);
if (!Call) {
if (IgnoreAssumeLikeCalls) {
if (const auto *FI = dyn_cast<Instruction>(FU)) {
if (FI->isCast() && !FI->user_empty() &&
llvm::all_of(FU->users(), [](const User *U) {
if (const auto *I = dyn_cast<IntrinsicInst>(U))
return I->isAssumeLikeIntrinsic();
return false;
}))
continue;
}
}
if (IgnoreLLVMUsed && !FU->user_empty()) {
const User *FUU = FU;
if (isa<BitCastOperator>(FU) && FU->hasOneUse() &&
!FU->user_begin()->user_empty())
FUU = *FU->user_begin();
if (llvm::all_of(FUU->users(), [](const User *U) {
if (const auto *GV = dyn_cast<GlobalVariable>(U))
return GV->hasName() &&
(GV->getName().equals("llvm.compiler.used") ||
GV->getName().equals("llvm.used"));
return false;
}))
continue;
}
if (PutOffender)
*PutOffender = FU;
return true;
}
if (!Call->isCallee(&U)) {
if (PutOffender)
*PutOffender = FU;
return true;
}
}
return false;
}
bool Function::isDefTriviallyDead() const {
// Check the linkage
if (!hasLinkOnceLinkage() && !hasLocalLinkage() &&
!hasAvailableExternallyLinkage())
return false;
// Check if the function is used by anything other than a blockaddress.
[C++11] Add range based accessors for the Use-Def chain of a Value. This requires a number of steps. 1) Move value_use_iterator into the Value class as an implementation detail 2) Change it to actually be a *Use* iterator rather than a *User* iterator. 3) Add an adaptor which is a User iterator that always looks through the Use to the User. 4) Wrap these in Value::use_iterator and Value::user_iterator typedefs. 5) Add the range adaptors as Value::uses() and Value::users(). 6) Update *all* of the callers to correctly distinguish between whether they wanted a use_iterator (and to explicitly dig out the User when needed), or a user_iterator which makes the Use itself totally opaque. Because #6 requires churning essentially everything that walked the Use-Def chains, I went ahead and added all of the range adaptors and switched them to range-based loops where appropriate. Also because the renaming requires at least churning every line of code, it didn't make any sense to split these up into multiple commits -- all of which would touch all of the same lies of code. The result is still not quite optimal. The Value::use_iterator is a nice regular iterator, but Value::user_iterator is an iterator over User*s rather than over the User objects themselves. As a consequence, it fits a bit awkwardly into the range-based world and it has the weird extra-dereferencing 'operator->' that so many of our iterators have. I think this could be fixed by providing something which transforms a range of T&s into a range of T*s, but that *can* be separated into another patch, and it isn't yet 100% clear whether this is the right move. However, this change gets us most of the benefit and cleans up a substantial amount of code around Use and User. =] llvm-svn: 203364
2014-03-09 04:16:01 +01:00
for (const User *U : users())
if (!isa<BlockAddress>(U))
return false;
return true;
}
/// callsFunctionThatReturnsTwice - Return true if the function has a call to
/// setjmp or other function that gcc recognizes as "returning twice".
bool Function::callsFunctionThatReturnsTwice() const {
for (const Instruction &I : instructions(this))
if (const auto *Call = dyn_cast<CallBase>(&I))
if (Call->hasFnAttr(Attribute::ReturnsTwice))
return true;
return false;
}
Constant *Function::getPersonalityFn() const {
assert(hasPersonalityFn() && getNumOperands());
return cast<Constant>(Op<0>());
}
void Function::setPersonalityFn(Constant *Fn) {
setHungoffOperand<0>(Fn);
setValueSubclassDataBit(3, Fn != nullptr);
}
Constant *Function::getPrefixData() const {
assert(hasPrefixData() && getNumOperands());
return cast<Constant>(Op<1>());
}
void Function::setPrefixData(Constant *PrefixData) {
setHungoffOperand<1>(PrefixData);
setValueSubclassDataBit(1, PrefixData != nullptr);
}
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
Constant *Function::getPrologueData() const {
assert(hasPrologueData() && getNumOperands());
return cast<Constant>(Op<2>());
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
}
void Function::setPrologueData(Constant *PrologueData) {
setHungoffOperand<2>(PrologueData);
setValueSubclassDataBit(2, PrologueData != nullptr);
}
void Function::allocHungoffUselist() {
// If we've already allocated a uselist, stop here.
if (getNumOperands())
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
return;
allocHungoffUses(3, /*IsPhi=*/ false);
setNumHungOffUseOperands(3);
// Initialize the uselist with placeholder operands to allow traversal.
auto *CPN = ConstantPointerNull::get(Type::getInt1PtrTy(getContext(), 0));
Op<0>().set(CPN);
Op<1>().set(CPN);
Op<2>().set(CPN);
}
template <int Idx>
void Function::setHungoffOperand(Constant *C) {
if (C) {
allocHungoffUselist();
Op<Idx>().set(C);
} else if (getNumOperands()) {
Op<Idx>().set(
ConstantPointerNull::get(Type::getInt1PtrTy(getContext(), 0)));
}
}
void Function::setValueSubclassDataBit(unsigned Bit, bool On) {
assert(Bit < 16 && "SubclassData contains only 16 bits");
if (On)
setValueSubclassData(getSubclassDataFromValue() | (1 << Bit));
else
setValueSubclassData(getSubclassDataFromValue() & ~(1 << Bit));
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
}
void Function::setEntryCount(ProfileCount Count,
const DenseSet<GlobalValue::GUID> *S) {
assert(Count.hasValue());
#if !defined(NDEBUG)
auto PrevCount = getEntryCount();
assert(!PrevCount.hasValue() || PrevCount.getType() == Count.getType());
#endif
auto ImportGUIDs = getImportGUIDs();
if (S == nullptr && ImportGUIDs.size())
S = &ImportGUIDs;
MDBuilder MDB(getContext());
setMetadata(
LLVMContext::MD_prof,
MDB.createFunctionEntryCount(Count.getCount(), Count.isSynthetic(), S));
}
void Function::setEntryCount(uint64_t Count, Function::ProfileCountType Type,
const DenseSet<GlobalValue::GUID> *Imports) {
setEntryCount(ProfileCount(Count, Type), Imports);
}
ProfileCount Function::getEntryCount(bool AllowSynthetic) const {
MDNode *MD = getMetadata(LLVMContext::MD_prof);
if (MD && MD->getOperand(0))
if (MDString *MDS = dyn_cast<MDString>(MD->getOperand(0))) {
if (MDS->getString().equals("function_entry_count")) {
ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(1));
uint64_t Count = CI->getValue().getZExtValue();
// A value of -1 is used for SamplePGO when there were no samples.
// Treat this the same as unknown.
if (Count == (uint64_t)-1)
return ProfileCount::getInvalid();
return ProfileCount(Count, PCT_Real);
} else if (AllowSynthetic &&
MDS->getString().equals("synthetic_function_entry_count")) {
ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(1));
uint64_t Count = CI->getValue().getZExtValue();
return ProfileCount(Count, PCT_Synthetic);
}
}
return ProfileCount::getInvalid();
}
DenseSet<GlobalValue::GUID> Function::getImportGUIDs() const {
DenseSet<GlobalValue::GUID> R;
if (MDNode *MD = getMetadata(LLVMContext::MD_prof))
if (MDString *MDS = dyn_cast<MDString>(MD->getOperand(0)))
if (MDS->getString().equals("function_entry_count"))
for (unsigned i = 2; i < MD->getNumOperands(); i++)
R.insert(mdconst::extract<ConstantInt>(MD->getOperand(i))
->getValue()
.getZExtValue());
return R;
}
void Function::setSectionPrefix(StringRef Prefix) {
MDBuilder MDB(getContext());
setMetadata(LLVMContext::MD_section_prefix,
MDB.createFunctionSectionPrefix(Prefix));
}
Optional<StringRef> Function::getSectionPrefix() const {
if (MDNode *MD = getMetadata(LLVMContext::MD_section_prefix)) {
assert(cast<MDString>(MD->getOperand(0))
->getString()
.equals("function_section_prefix") &&
"Metadata not match");
return cast<MDString>(MD->getOperand(1))->getString();
}
return None;
}
bool Function::nullPointerIsDefined() const {
return hasFnAttribute(Attribute::NullPointerIsValid);
}
bool llvm::NullPointerIsDefined(const Function *F, unsigned AS) {
if (F && F->nullPointerIsDefined())
return true;
if (AS != 0)
return true;
return false;
}