SPIRV-Tools/source/opt/instruction.cpp

930 lines
28 KiB
C++
Raw Normal View History

// Copyright (c) 2016 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "source/opt/instruction.h"
#include <initializer_list>
#include "OpenCLDebugInfo100.h"
#include "source/disassemble.h"
#include "source/opt/fold.h"
#include "source/opt/ir_context.h"
#include "source/opt/reflect.h"
namespace spvtools {
namespace opt {
namespace {
// Indices used to get particular operands out of instructions using InOperand.
const uint32_t kTypeImageDimIndex = 1;
const uint32_t kLoadBaseIndex = 0;
const uint32_t kPointerTypeStorageClassIndex = 0;
const uint32_t kTypeImageSampledIndex = 5;
// Constants for OpenCL.DebugInfo.100 extension instructions.
const uint32_t kExtInstSetIdInIdx = 0;
const uint32_t kExtInstInstructionInIdx = 1;
const uint32_t kDebugScopeNumWords = 7;
const uint32_t kDebugScopeNumWordsWithoutInlinedAt = 6;
const uint32_t kDebugNoScopeNumWords = 5;
// Number of operands of an OpBranchConditional instruction
// with weights.
const uint32_t kOpBranchConditionalWithWeightsNumOperands = 5;
} // namespace
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
Instruction::Instruction(IRContext* c)
: utils::IntrusiveNodeBase<Instruction>(),
context_(c),
opcode_(SpvOpNop),
has_type_id_(false),
has_result_id_(false),
unique_id_(c->TakeNextUniqueId()),
dbg_scope_(kNoDebugScope, kNoInlinedAt) {}
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
Instruction::Instruction(IRContext* c, SpvOp op)
: utils::IntrusiveNodeBase<Instruction>(),
context_(c),
opcode_(op),
has_type_id_(false),
has_result_id_(false),
unique_id_(c->TakeNextUniqueId()),
dbg_scope_(kNoDebugScope, kNoInlinedAt) {}
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
Instruction::Instruction(IRContext* c, const spv_parsed_instruction_t& inst,
std::vector<Instruction>&& dbg_line)
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
: context_(c),
opcode_(static_cast<SpvOp>(inst.opcode)),
has_type_id_(inst.type_id != 0),
has_result_id_(inst.result_id != 0),
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
unique_id_(c->TakeNextUniqueId()),
dbg_line_insts_(std::move(dbg_line)),
dbg_scope_(kNoDebugScope, kNoInlinedAt) {
assert((!IsDebugLineInst(opcode_) || dbg_line.empty()) &&
"Op(No)Line attaching to Op(No)Line found");
for (uint32_t i = 0; i < inst.num_operands; ++i) {
const auto& current_payload = inst.operands[i];
std::vector<uint32_t> words(
inst.words + current_payload.offset,
inst.words + current_payload.offset + current_payload.num_words);
operands_.emplace_back(current_payload.type, std::move(words));
}
}
Instruction::Instruction(IRContext* c, const spv_parsed_instruction_t& inst,
const DebugScope& dbg_scope)
: context_(c),
opcode_(static_cast<SpvOp>(inst.opcode)),
has_type_id_(inst.type_id != 0),
has_result_id_(inst.result_id != 0),
unique_id_(c->TakeNextUniqueId()),
dbg_scope_(dbg_scope) {
for (uint32_t i = 0; i < inst.num_operands; ++i) {
const auto& current_payload = inst.operands[i];
std::vector<uint32_t> words(
inst.words + current_payload.offset,
inst.words + current_payload.offset + current_payload.num_words);
operands_.emplace_back(current_payload.type, std::move(words));
}
}
Instruction::Instruction(IRContext* c, SpvOp op, uint32_t ty_id,
uint32_t res_id, const OperandList& in_operands)
: utils::IntrusiveNodeBase<Instruction>(),
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
context_(c),
opcode_(op),
has_type_id_(ty_id != 0),
has_result_id_(res_id != 0),
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
unique_id_(c->TakeNextUniqueId()),
operands_(),
dbg_scope_(kNoDebugScope, kNoInlinedAt) {
if (has_type_id_) {
operands_.emplace_back(spv_operand_type_t::SPV_OPERAND_TYPE_TYPE_ID,
std::initializer_list<uint32_t>{ty_id});
}
if (has_result_id_) {
operands_.emplace_back(spv_operand_type_t::SPV_OPERAND_TYPE_RESULT_ID,
std::initializer_list<uint32_t>{res_id});
}
operands_.insert(operands_.end(), in_operands.begin(), in_operands.end());
}
Instruction::Instruction(Instruction&& that)
: utils::IntrusiveNodeBase<Instruction>(),
opcode_(that.opcode_),
has_type_id_(that.has_type_id_),
has_result_id_(that.has_result_id_),
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
unique_id_(that.unique_id_),
operands_(std::move(that.operands_)),
dbg_line_insts_(std::move(that.dbg_line_insts_)),
dbg_scope_(that.dbg_scope_) {
for (auto& i : dbg_line_insts_) {
i.dbg_scope_ = that.dbg_scope_;
}
}
Instruction& Instruction::operator=(Instruction&& that) {
opcode_ = that.opcode_;
has_type_id_ = that.has_type_id_;
has_result_id_ = that.has_result_id_;
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
unique_id_ = that.unique_id_;
operands_ = std::move(that.operands_);
dbg_line_insts_ = std::move(that.dbg_line_insts_);
dbg_scope_ = that.dbg_scope_;
return *this;
}
Instruction* Instruction::Clone(IRContext* c) const {
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
Instruction* clone = new Instruction(c);
clone->opcode_ = opcode_;
clone->has_type_id_ = has_type_id_;
clone->has_result_id_ = has_result_id_;
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
clone->unique_id_ = c->TakeNextUniqueId();
clone->operands_ = operands_;
clone->dbg_line_insts_ = dbg_line_insts_;
clone->dbg_scope_ = dbg_scope_;
return clone;
}
uint32_t Instruction::GetSingleWordOperand(uint32_t index) const {
const auto& words = GetOperand(index).words;
assert(words.size() == 1 && "expected the operand only taking one word");
return words.front();
}
uint32_t Instruction::NumInOperandWords() const {
uint32_t size = 0;
for (uint32_t i = TypeResultIdCount(); i < operands_.size(); ++i)
size += static_cast<uint32_t>(operands_[i].words.size());
return size;
}
bool Instruction::HasBranchWeights() const {
if (opcode_ == SpvOpBranchConditional &&
NumOperands() == kOpBranchConditionalWithWeightsNumOperands) {
return true;
}
return false;
}
void Instruction::ToBinaryWithoutAttachedDebugInsts(
std::vector<uint32_t>* binary) const {
const uint32_t num_words = 1 + NumOperandWords();
binary->push_back((num_words << 16) | static_cast<uint16_t>(opcode_));
for (const auto& operand : operands_) {
binary->insert(binary->end(), operand.words.begin(), operand.words.end());
}
}
void Instruction::ReplaceOperands(const OperandList& new_operands) {
operands_.clear();
operands_.insert(operands_.begin(), new_operands.begin(), new_operands.end());
}
bool Instruction::IsReadOnlyLoad() const {
if (IsLoad()) {
Instruction* address_def = GetBaseAddress();
if (!address_def) {
return false;
}
if (address_def->opcode() == SpvOpVariable) {
if (address_def->IsReadOnlyPointer()) {
return true;
}
}
if (address_def->opcode() == SpvOpLoad) {
const analysis::Type* address_type =
context()->get_type_mgr()->GetType(address_def->type_id());
if (address_type->AsSampledImage() != nullptr) {
const auto* image_type =
address_type->AsSampledImage()->image_type()->AsImage();
if (image_type->sampled() == 1) {
return true;
}
}
}
}
return false;
}
Instruction* Instruction::GetBaseAddress() const {
uint32_t base = GetSingleWordInOperand(kLoadBaseIndex);
Instruction* base_inst = context()->get_def_use_mgr()->GetDef(base);
bool done = false;
while (!done) {
switch (base_inst->opcode()) {
case SpvOpAccessChain:
case SpvOpInBoundsAccessChain:
case SpvOpPtrAccessChain:
case SpvOpInBoundsPtrAccessChain:
case SpvOpImageTexelPointer:
case SpvOpCopyObject:
// All of these instructions have the base pointer use a base pointer
// in in-operand 0.
base = base_inst->GetSingleWordInOperand(0);
base_inst = context()->get_def_use_mgr()->GetDef(base);
break;
default:
done = true;
break;
}
}
return base_inst;
}
bool Instruction::IsReadOnlyPointer() const {
if (context()->get_feature_mgr()->HasCapability(SpvCapabilityShader))
return IsReadOnlyPointerShaders();
else
return IsReadOnlyPointerKernel();
}
bool Instruction::IsVulkanStorageImage() const {
if (opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
GetSingleWordInOperand(kPointerTypeStorageClassIndex);
if (storage_class != SpvStorageClassUniformConstant) {
return false;
}
Instruction* base_type =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(1));
// Unpack the optional layer of arraying.
if (base_type->opcode() == SpvOpTypeArray ||
base_type->opcode() == SpvOpTypeRuntimeArray) {
base_type = context()->get_def_use_mgr()->GetDef(
base_type->GetSingleWordInOperand(0));
}
if (base_type->opcode() != SpvOpTypeImage) {
return false;
}
if (base_type->GetSingleWordInOperand(kTypeImageDimIndex) == SpvDimBuffer) {
return false;
}
// Check if the image is sampled. If we do not know for sure that it is,
// then assume it is a storage image.
return base_type->GetSingleWordInOperand(kTypeImageSampledIndex) != 1;
}
bool Instruction::IsVulkanSampledImage() const {
if (opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
GetSingleWordInOperand(kPointerTypeStorageClassIndex);
if (storage_class != SpvStorageClassUniformConstant) {
return false;
}
Instruction* base_type =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(1));
// Unpack the optional layer of arraying.
if (base_type->opcode() == SpvOpTypeArray ||
base_type->opcode() == SpvOpTypeRuntimeArray) {
base_type = context()->get_def_use_mgr()->GetDef(
base_type->GetSingleWordInOperand(0));
}
if (base_type->opcode() != SpvOpTypeImage) {
return false;
}
if (base_type->GetSingleWordInOperand(kTypeImageDimIndex) == SpvDimBuffer) {
return false;
}
// Check if the image is sampled. If we know for sure that it is,
// then return true.
return base_type->GetSingleWordInOperand(kTypeImageSampledIndex) == 1;
}
bool Instruction::IsVulkanStorageTexelBuffer() const {
if (opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
GetSingleWordInOperand(kPointerTypeStorageClassIndex);
if (storage_class != SpvStorageClassUniformConstant) {
return false;
}
Instruction* base_type =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(1));
// Unpack the optional layer of arraying.
if (base_type->opcode() == SpvOpTypeArray ||
base_type->opcode() == SpvOpTypeRuntimeArray) {
base_type = context()->get_def_use_mgr()->GetDef(
base_type->GetSingleWordInOperand(0));
}
if (base_type->opcode() != SpvOpTypeImage) {
return false;
}
if (base_type->GetSingleWordInOperand(kTypeImageDimIndex) != SpvDimBuffer) {
return false;
}
// Check if the image is sampled. If we do not know for sure that it is,
// then assume it is a storage texel buffer.
return base_type->GetSingleWordInOperand(kTypeImageSampledIndex) != 1;
}
bool Instruction::IsVulkanStorageBuffer() const {
// Is there a difference between a "Storage buffer" and a "dynamic storage
// buffer" in SPIR-V and do we care about the difference?
if (opcode() != SpvOpTypePointer) {
return false;
}
Instruction* base_type =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(1));
// Unpack the optional layer of arraying.
if (base_type->opcode() == SpvOpTypeArray ||
base_type->opcode() == SpvOpTypeRuntimeArray) {
base_type = context()->get_def_use_mgr()->GetDef(
base_type->GetSingleWordInOperand(0));
}
if (base_type->opcode() != SpvOpTypeStruct) {
return false;
}
uint32_t storage_class =
GetSingleWordInOperand(kPointerTypeStorageClassIndex);
if (storage_class == SpvStorageClassUniform) {
bool is_buffer_block = false;
context()->get_decoration_mgr()->ForEachDecoration(
base_type->result_id(), SpvDecorationBufferBlock,
[&is_buffer_block](const Instruction&) { is_buffer_block = true; });
return is_buffer_block;
} else if (storage_class == SpvStorageClassStorageBuffer) {
bool is_block = false;
context()->get_decoration_mgr()->ForEachDecoration(
base_type->result_id(), SpvDecorationBlock,
[&is_block](const Instruction&) { is_block = true; });
return is_block;
}
return false;
}
bool Instruction::IsVulkanUniformBuffer() const {
if (opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
GetSingleWordInOperand(kPointerTypeStorageClassIndex);
if (storage_class != SpvStorageClassUniform) {
return false;
}
Instruction* base_type =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(1));
// Unpack the optional layer of arraying.
if (base_type->opcode() == SpvOpTypeArray ||
base_type->opcode() == SpvOpTypeRuntimeArray) {
base_type = context()->get_def_use_mgr()->GetDef(
base_type->GetSingleWordInOperand(0));
}
if (base_type->opcode() != SpvOpTypeStruct) {
return false;
}
bool is_block = false;
context()->get_decoration_mgr()->ForEachDecoration(
base_type->result_id(), SpvDecorationBlock,
[&is_block](const Instruction&) { is_block = true; });
return is_block;
}
bool Instruction::IsReadOnlyPointerShaders() const {
if (type_id() == 0) {
return false;
}
Instruction* type_def = context()->get_def_use_mgr()->GetDef(type_id());
if (type_def->opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
type_def->GetSingleWordInOperand(kPointerTypeStorageClassIndex);
switch (storage_class) {
case SpvStorageClassUniformConstant:
if (!type_def->IsVulkanStorageImage() &&
!type_def->IsVulkanStorageTexelBuffer()) {
return true;
}
break;
case SpvStorageClassUniform:
if (!type_def->IsVulkanStorageBuffer()) {
return true;
}
break;
case SpvStorageClassPushConstant:
case SpvStorageClassInput:
return true;
default:
break;
}
bool is_nonwritable = false;
context()->get_decoration_mgr()->ForEachDecoration(
result_id(), SpvDecorationNonWritable,
[&is_nonwritable](const Instruction&) { is_nonwritable = true; });
return is_nonwritable;
}
bool Instruction::IsReadOnlyPointerKernel() const {
if (type_id() == 0) {
return false;
}
Instruction* type_def = context()->get_def_use_mgr()->GetDef(type_id());
if (type_def->opcode() != SpvOpTypePointer) {
return false;
}
uint32_t storage_class =
type_def->GetSingleWordInOperand(kPointerTypeStorageClassIndex);
return storage_class == SpvStorageClassUniformConstant;
}
uint32_t Instruction::GetTypeComponent(uint32_t element) const {
uint32_t subtype = 0;
switch (opcode()) {
case SpvOpTypeStruct:
subtype = GetSingleWordInOperand(element);
break;
case SpvOpTypeArray:
case SpvOpTypeRuntimeArray:
case SpvOpTypeVector:
case SpvOpTypeMatrix:
// These types all have uniform subtypes.
subtype = GetSingleWordInOperand(0u);
break;
default:
break;
}
return subtype;
}
Instruction* Instruction::InsertBefore(std::unique_ptr<Instruction>&& inst) {
inst.get()->InsertBefore(this);
return inst.release();
}
Instruction* Instruction::InsertBefore(
std::vector<std::unique_ptr<Instruction>>&& list) {
Instruction* first_node = list.front().get();
for (auto& inst : list) {
inst.release()->InsertBefore(this);
}
list.clear();
return first_node;
}
bool Instruction::IsValidBasePointer() const {
uint32_t tid = type_id();
if (tid == 0) {
return false;
}
Instruction* type = context()->get_def_use_mgr()->GetDef(tid);
if (type->opcode() != SpvOpTypePointer) {
return false;
}
auto feature_mgr = context()->get_feature_mgr();
if (feature_mgr->HasCapability(SpvCapabilityAddresses)) {
// TODO: The rules here could be more restrictive.
return true;
}
if (opcode() == SpvOpVariable || opcode() == SpvOpFunctionParameter) {
return true;
}
// With variable pointers, there are more valid base pointer objects.
// Variable pointers implicitly declares Variable pointers storage buffer.
SpvStorageClass storage_class =
static_cast<SpvStorageClass>(type->GetSingleWordInOperand(0));
if ((feature_mgr->HasCapability(SpvCapabilityVariablePointersStorageBuffer) &&
storage_class == SpvStorageClassStorageBuffer) ||
(feature_mgr->HasCapability(SpvCapabilityVariablePointers) &&
storage_class == SpvStorageClassWorkgroup)) {
switch (opcode()) {
case SpvOpPhi:
case SpvOpSelect:
case SpvOpFunctionCall:
case SpvOpConstantNull:
return true;
default:
break;
}
}
uint32_t pointee_type_id = type->GetSingleWordInOperand(1);
Instruction* pointee_type_inst =
context()->get_def_use_mgr()->GetDef(pointee_type_id);
if (pointee_type_inst->IsOpaqueType()) {
return true;
}
return false;
}
OpenCLDebugInfo100Instructions Instruction::GetOpenCL100DebugOpcode() const {
if (opcode() != SpvOpExtInst) {
return OpenCLDebugInfo100InstructionsMax;
}
if (!context()->get_feature_mgr()->GetExtInstImportId_OpenCL100DebugInfo()) {
return OpenCLDebugInfo100InstructionsMax;
}
if (GetSingleWordInOperand(kExtInstSetIdInIdx) !=
context()->get_feature_mgr()->GetExtInstImportId_OpenCL100DebugInfo()) {
return OpenCLDebugInfo100InstructionsMax;
}
return OpenCLDebugInfo100Instructions(
GetSingleWordInOperand(kExtInstInstructionInIdx));
}
bool Instruction::IsValidBaseImage() const {
uint32_t tid = type_id();
if (tid == 0) {
return false;
}
Instruction* type = context()->get_def_use_mgr()->GetDef(tid);
return (type->opcode() == SpvOpTypeImage ||
type->opcode() == SpvOpTypeSampledImage);
}
bool Instruction::IsOpaqueType() const {
if (opcode() == SpvOpTypeStruct) {
bool is_opaque = false;
ForEachInOperand([&is_opaque, this](const uint32_t* op_id) {
Instruction* type_inst = context()->get_def_use_mgr()->GetDef(*op_id);
is_opaque |= type_inst->IsOpaqueType();
});
return is_opaque;
} else if (opcode() == SpvOpTypeArray) {
uint32_t sub_type_id = GetSingleWordInOperand(0);
Instruction* sub_type_inst =
context()->get_def_use_mgr()->GetDef(sub_type_id);
return sub_type_inst->IsOpaqueType();
} else {
return opcode() == SpvOpTypeRuntimeArray ||
spvOpcodeIsBaseOpaqueType(opcode());
}
}
bool Instruction::IsFoldable() const {
return IsFoldableByFoldScalar() ||
context()->get_instruction_folder().HasConstFoldingRule(this);
}
bool Instruction::IsFoldableByFoldScalar() const {
const InstructionFolder& folder = context()->get_instruction_folder();
if (!folder.IsFoldableOpcode(opcode())) {
return false;
}
Instruction* type = context()->get_def_use_mgr()->GetDef(type_id());
if (!folder.IsFoldableType(type)) {
return false;
}
// Even if the type of the instruction is foldable, its operands may not be
// foldable (e.g., comparisons of 64bit types). Check that all operand types
// are foldable before accepting the instruction.
return WhileEachInOperand([&folder, this](const uint32_t* op_id) {
Instruction* def_inst = context()->get_def_use_mgr()->GetDef(*op_id);
Instruction* def_inst_type =
context()->get_def_use_mgr()->GetDef(def_inst->type_id());
return folder.IsFoldableType(def_inst_type);
});
}
bool Instruction::IsFloatingPointFoldingAllowed() const {
// TODO: Add the rules for kernels. For now it will be pessimistic.
// For now, do not support capabilities introduced by SPV_KHR_float_controls.
if (!context_->get_feature_mgr()->HasCapability(SpvCapabilityShader) ||
context_->get_feature_mgr()->HasCapability(SpvCapabilityDenormPreserve) ||
context_->get_feature_mgr()->HasCapability(
SpvCapabilityDenormFlushToZero) ||
context_->get_feature_mgr()->HasCapability(
SpvCapabilitySignedZeroInfNanPreserve) ||
context_->get_feature_mgr()->HasCapability(
SpvCapabilityRoundingModeRTZ) ||
context_->get_feature_mgr()->HasCapability(
SpvCapabilityRoundingModeRTE)) {
return false;
}
bool is_nocontract = false;
context_->get_decoration_mgr()->WhileEachDecoration(
result_id(), SpvDecorationNoContraction,
[&is_nocontract](const Instruction&) {
is_nocontract = true;
return false;
});
return !is_nocontract;
}
std::string Instruction::PrettyPrint(uint32_t options) const {
// Convert the module to binary.
std::vector<uint32_t> module_binary;
context()->module()->ToBinary(&module_binary, /* skip_nop = */ false);
// Convert the instruction to binary. This is used to identify the correct
// stream of words to output from the module.
std::vector<uint32_t> inst_binary;
ToBinaryWithoutAttachedDebugInsts(&inst_binary);
// Do not generate a header.
return spvInstructionBinaryToText(
context()->grammar().target_env(), inst_binary.data(), inst_binary.size(),
module_binary.data(), module_binary.size(),
options | SPV_BINARY_TO_TEXT_OPTION_NO_HEADER);
}
std::ostream& operator<<(std::ostream& str, const Instruction& inst) {
str << inst.PrettyPrint();
return str;
}
void Instruction::Dump() const {
std::cerr << "Instruction #" << unique_id() << "\n" << *this << "\n";
}
bool Instruction::IsOpcodeCodeMotionSafe() const {
switch (opcode_) {
case SpvOpNop:
case SpvOpUndef:
case SpvOpLoad:
case SpvOpAccessChain:
case SpvOpInBoundsAccessChain:
case SpvOpArrayLength:
case SpvOpVectorExtractDynamic:
case SpvOpVectorInsertDynamic:
case SpvOpVectorShuffle:
case SpvOpCompositeConstruct:
case SpvOpCompositeExtract:
case SpvOpCompositeInsert:
case SpvOpCopyObject:
case SpvOpTranspose:
case SpvOpConvertFToU:
case SpvOpConvertFToS:
case SpvOpConvertSToF:
case SpvOpConvertUToF:
case SpvOpUConvert:
case SpvOpSConvert:
case SpvOpFConvert:
case SpvOpQuantizeToF16:
case SpvOpBitcast:
case SpvOpSNegate:
case SpvOpFNegate:
case SpvOpIAdd:
case SpvOpFAdd:
case SpvOpISub:
case SpvOpFSub:
case SpvOpIMul:
case SpvOpFMul:
case SpvOpUDiv:
case SpvOpSDiv:
case SpvOpFDiv:
case SpvOpUMod:
case SpvOpSRem:
case SpvOpSMod:
case SpvOpFRem:
case SpvOpFMod:
case SpvOpVectorTimesScalar:
case SpvOpMatrixTimesScalar:
case SpvOpVectorTimesMatrix:
case SpvOpMatrixTimesVector:
case SpvOpMatrixTimesMatrix:
case SpvOpOuterProduct:
case SpvOpDot:
case SpvOpIAddCarry:
case SpvOpISubBorrow:
case SpvOpUMulExtended:
case SpvOpSMulExtended:
case SpvOpAny:
case SpvOpAll:
case SpvOpIsNan:
case SpvOpIsInf:
case SpvOpLogicalEqual:
case SpvOpLogicalNotEqual:
case SpvOpLogicalOr:
case SpvOpLogicalAnd:
case SpvOpLogicalNot:
case SpvOpSelect:
case SpvOpIEqual:
case SpvOpINotEqual:
case SpvOpUGreaterThan:
case SpvOpSGreaterThan:
case SpvOpUGreaterThanEqual:
case SpvOpSGreaterThanEqual:
case SpvOpULessThan:
case SpvOpSLessThan:
case SpvOpULessThanEqual:
case SpvOpSLessThanEqual:
case SpvOpFOrdEqual:
case SpvOpFUnordEqual:
case SpvOpFOrdNotEqual:
case SpvOpFUnordNotEqual:
case SpvOpFOrdLessThan:
case SpvOpFUnordLessThan:
case SpvOpFOrdGreaterThan:
case SpvOpFUnordGreaterThan:
case SpvOpFOrdLessThanEqual:
case SpvOpFUnordLessThanEqual:
case SpvOpFOrdGreaterThanEqual:
case SpvOpFUnordGreaterThanEqual:
case SpvOpShiftRightLogical:
case SpvOpShiftRightArithmetic:
case SpvOpShiftLeftLogical:
case SpvOpBitwiseOr:
case SpvOpBitwiseXor:
case SpvOpBitwiseAnd:
case SpvOpNot:
case SpvOpBitFieldInsert:
case SpvOpBitFieldSExtract:
case SpvOpBitFieldUExtract:
case SpvOpBitReverse:
case SpvOpBitCount:
case SpvOpSizeOf:
return true;
default:
return false;
}
}
bool Instruction::IsScalarizable() const {
if (spvOpcodeIsScalarizable(opcode())) {
return true;
}
if (opcode() == SpvOpExtInst) {
uint32_t instSetId =
context()->get_feature_mgr()->GetExtInstImportId_GLSLstd450();
if (GetSingleWordInOperand(kExtInstSetIdInIdx) == instSetId) {
switch (GetSingleWordInOperand(kExtInstInstructionInIdx)) {
case GLSLstd450Round:
case GLSLstd450RoundEven:
case GLSLstd450Trunc:
case GLSLstd450FAbs:
case GLSLstd450SAbs:
case GLSLstd450FSign:
case GLSLstd450SSign:
case GLSLstd450Floor:
case GLSLstd450Ceil:
case GLSLstd450Fract:
case GLSLstd450Radians:
case GLSLstd450Degrees:
case GLSLstd450Sin:
case GLSLstd450Cos:
case GLSLstd450Tan:
case GLSLstd450Asin:
case GLSLstd450Acos:
case GLSLstd450Atan:
case GLSLstd450Sinh:
case GLSLstd450Cosh:
case GLSLstd450Tanh:
case GLSLstd450Asinh:
case GLSLstd450Acosh:
case GLSLstd450Atanh:
case GLSLstd450Atan2:
case GLSLstd450Pow:
case GLSLstd450Exp:
case GLSLstd450Log:
case GLSLstd450Exp2:
case GLSLstd450Log2:
case GLSLstd450Sqrt:
case GLSLstd450InverseSqrt:
case GLSLstd450Modf:
case GLSLstd450FMin:
case GLSLstd450UMin:
case GLSLstd450SMin:
case GLSLstd450FMax:
case GLSLstd450UMax:
case GLSLstd450SMax:
case GLSLstd450FClamp:
case GLSLstd450UClamp:
case GLSLstd450SClamp:
case GLSLstd450FMix:
case GLSLstd450Step:
case GLSLstd450SmoothStep:
case GLSLstd450Fma:
case GLSLstd450Frexp:
case GLSLstd450Ldexp:
case GLSLstd450FindILsb:
case GLSLstd450FindSMsb:
case GLSLstd450FindUMsb:
case GLSLstd450NMin:
case GLSLstd450NMax:
case GLSLstd450NClamp:
return true;
default:
return false;
}
}
}
return false;
}
bool Instruction::IsOpcodeSafeToDelete() const {
if (context()->IsCombinatorInstruction(this)) {
return true;
}
switch (opcode()) {
case SpvOpDPdx:
case SpvOpDPdy:
case SpvOpFwidth:
case SpvOpDPdxFine:
case SpvOpDPdyFine:
case SpvOpFwidthFine:
case SpvOpDPdxCoarse:
case SpvOpDPdyCoarse:
case SpvOpFwidthCoarse:
case SpvOpImageQueryLod:
return true;
default:
return false;
}
}
bool Instruction::IsNonSemanticInstruction() const {
if (!HasResultId()) return false;
if (opcode() != SpvOpExtInst) return false;
auto import_inst =
context()->get_def_use_mgr()->GetDef(GetSingleWordInOperand(0));
std::string import_name = import_inst->GetInOperand(0).AsString();
return import_name.find("NonSemantic.") == 0;
}
void DebugScope::ToBinary(uint32_t type_id, uint32_t result_id,
uint32_t ext_set,
std::vector<uint32_t>* binary) const {
uint32_t num_words = kDebugScopeNumWords;
OpenCLDebugInfo100Instructions dbg_opcode = OpenCLDebugInfo100DebugScope;
if (GetLexicalScope() == kNoDebugScope) {
num_words = kDebugNoScopeNumWords;
dbg_opcode = OpenCLDebugInfo100DebugNoScope;
} else if (GetInlinedAt() == kNoInlinedAt) {
num_words = kDebugScopeNumWordsWithoutInlinedAt;
}
std::vector<uint32_t> operands = {
(num_words << 16) | static_cast<uint16_t>(SpvOpExtInst),
type_id,
result_id,
ext_set,
static_cast<uint32_t>(dbg_opcode),
};
binary->insert(binary->end(), operands.begin(), operands.end());
if (GetLexicalScope() != kNoDebugScope) binary->push_back(GetLexicalScope());
if (GetInlinedAt() != kNoInlinedAt) binary->push_back(GetInlinedAt());
}
} // namespace opt
} // namespace spvtools