SPIRV-Tools/source/opt/inline_pass.cpp

696 lines
28 KiB
C++
Raw Normal View History

// Copyright (c) 2017 The Khronos Group Inc.
// Copyright (c) 2017 Valve Corporation
// Copyright (c) 2017 LunarG 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/inline_pass.h"
#include <unordered_set>
#include <utility>
#include "source/cfa.h"
#include "source/util/make_unique.h"
// Indices of operands in SPIR-V instructions
static const int kSpvFunctionCallFunctionId = 2;
static const int kSpvFunctionCallArgumentId = 3;
static const int kSpvReturnValueId = 0;
static const int kSpvLoopMergeMergeBlockId = 0;
static const int kSpvLoopMergeContinueTargetIdInIdx = 1;
namespace spvtools {
namespace opt {
uint32_t InlinePass::AddPointerToType(uint32_t type_id,
SpvStorageClass storage_class) {
uint32_t resultId = TakeNextId();
std::unique_ptr<Instruction> type_inst(
new Instruction(context(), SpvOpTypePointer, 0, resultId,
{{spv_operand_type_t::SPV_OPERAND_TYPE_STORAGE_CLASS,
{uint32_t(storage_class)}},
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {type_id}}}));
context()->AddType(std::move(type_inst));
analysis::Type* pointeeTy;
std::unique_ptr<analysis::Pointer> pointerTy;
std::tie(pointeeTy, pointerTy) =
context()->get_type_mgr()->GetTypeAndPointerType(type_id,
SpvStorageClassFunction);
context()->get_type_mgr()->RegisterType(resultId, *pointerTy);
return resultId;
}
void InlinePass::AddBranch(uint32_t label_id,
std::unique_ptr<BasicBlock>* block_ptr) {
std::unique_ptr<Instruction> newBranch(
new Instruction(context(), SpvOpBranch, 0, 0,
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {label_id}}}));
(*block_ptr)->AddInstruction(std::move(newBranch));
}
void InlinePass::AddBranchCond(uint32_t cond_id, uint32_t true_id,
uint32_t false_id,
std::unique_ptr<BasicBlock>* block_ptr) {
std::unique_ptr<Instruction> newBranch(
new Instruction(context(), SpvOpBranchConditional, 0, 0,
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {cond_id}},
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {true_id}},
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {false_id}}}));
(*block_ptr)->AddInstruction(std::move(newBranch));
}
void InlinePass::AddLoopMerge(uint32_t merge_id, uint32_t continue_id,
std::unique_ptr<BasicBlock>* block_ptr) {
std::unique_ptr<Instruction> newLoopMerge(new 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(), SpvOpLoopMerge, 0, 0,
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {merge_id}},
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {continue_id}},
{spv_operand_type_t::SPV_OPERAND_TYPE_LOOP_CONTROL, {0}}}));
(*block_ptr)->AddInstruction(std::move(newLoopMerge));
}
void InlinePass::AddStore(uint32_t ptr_id, uint32_t val_id,
std::unique_ptr<BasicBlock>* block_ptr) {
std::unique_ptr<Instruction> newStore(
new Instruction(context(), SpvOpStore, 0, 0,
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {ptr_id}},
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {val_id}}}));
(*block_ptr)->AddInstruction(std::move(newStore));
}
void InlinePass::AddLoad(uint32_t type_id, uint32_t resultId, uint32_t ptr_id,
std::unique_ptr<BasicBlock>* block_ptr) {
std::unique_ptr<Instruction> newLoad(
new Instruction(context(), SpvOpLoad, type_id, resultId,
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {ptr_id}}}));
(*block_ptr)->AddInstruction(std::move(newLoad));
}
std::unique_ptr<Instruction> InlinePass::NewLabel(uint32_t label_id) {
std::unique_ptr<Instruction> newLabel(
new Instruction(context(), SpvOpLabel, 0, label_id, {}));
return newLabel;
}
uint32_t InlinePass::GetFalseId() {
if (false_id_ != 0) return false_id_;
false_id_ = get_module()->GetGlobalValue(SpvOpConstantFalse);
if (false_id_ != 0) return false_id_;
uint32_t boolId = get_module()->GetGlobalValue(SpvOpTypeBool);
if (boolId == 0) {
boolId = TakeNextId();
get_module()->AddGlobalValue(SpvOpTypeBool, boolId, 0);
}
false_id_ = TakeNextId();
get_module()->AddGlobalValue(SpvOpConstantFalse, false_id_, boolId);
return false_id_;
}
void InlinePass::MapParams(
Function* calleeFn, BasicBlock::iterator call_inst_itr,
std::unordered_map<uint32_t, uint32_t>* callee2caller) {
int param_idx = 0;
calleeFn->ForEachParam([&call_inst_itr, &param_idx,
&callee2caller](const Instruction* cpi) {
const uint32_t pid = cpi->result_id();
(*callee2caller)[pid] = call_inst_itr->GetSingleWordOperand(
kSpvFunctionCallArgumentId + param_idx);
++param_idx;
});
}
void InlinePass::CloneAndMapLocals(
Function* calleeFn, std::vector<std::unique_ptr<Instruction>>* new_vars,
std::unordered_map<uint32_t, uint32_t>* callee2caller) {
auto callee_block_itr = calleeFn->begin();
auto callee_var_itr = callee_block_itr->begin();
while (callee_var_itr->opcode() == SpvOp::SpvOpVariable) {
std::unique_ptr<Instruction> var_inst(callee_var_itr->Clone(context()));
uint32_t newId = TakeNextId();
get_decoration_mgr()->CloneDecorations(callee_var_itr->result_id(), newId);
var_inst->SetResultId(newId);
(*callee2caller)[callee_var_itr->result_id()] = newId;
new_vars->push_back(std::move(var_inst));
++callee_var_itr;
}
}
uint32_t InlinePass::CreateReturnVar(
Function* calleeFn, std::vector<std::unique_ptr<Instruction>>* new_vars) {
uint32_t returnVarId = 0;
const uint32_t calleeTypeId = calleeFn->type_id();
analysis::Type* calleeType = context()->get_type_mgr()->GetType(calleeTypeId);
if (calleeType->AsVoid() == nullptr) {
// Find or create ptr to callee return type.
uint32_t returnVarTypeId = context()->get_type_mgr()->FindPointerToType(
calleeTypeId, SpvStorageClassFunction);
if (returnVarTypeId == 0)
returnVarTypeId = AddPointerToType(calleeTypeId, SpvStorageClassFunction);
// Add return var to new function scope variables.
returnVarId = TakeNextId();
std::unique_ptr<Instruction> var_inst(
new Instruction(context(), SpvOpVariable, returnVarTypeId, returnVarId,
{{spv_operand_type_t::SPV_OPERAND_TYPE_STORAGE_CLASS,
{SpvStorageClassFunction}}}));
new_vars->push_back(std::move(var_inst));
}
get_decoration_mgr()->CloneDecorations(calleeFn->result_id(), returnVarId);
return returnVarId;
}
bool InlinePass::IsSameBlockOp(const Instruction* inst) const {
return inst->opcode() == SpvOpSampledImage || inst->opcode() == SpvOpImage;
}
void InlinePass::CloneSameBlockOps(
std::unique_ptr<Instruction>* inst,
std::unordered_map<uint32_t, uint32_t>* postCallSB,
std::unordered_map<uint32_t, Instruction*>* preCallSB,
std::unique_ptr<BasicBlock>* block_ptr) {
(*inst)->ForEachInId(
[&postCallSB, &preCallSB, &block_ptr, this](uint32_t* iid) {
const auto mapItr = (*postCallSB).find(*iid);
if (mapItr == (*postCallSB).end()) {
const auto mapItr2 = (*preCallSB).find(*iid);
if (mapItr2 != (*preCallSB).end()) {
// Clone pre-call same-block ops, map result id.
const Instruction* inInst = mapItr2->second;
std::unique_ptr<Instruction> sb_inst(inInst->Clone(context()));
CloneSameBlockOps(&sb_inst, postCallSB, preCallSB, block_ptr);
const uint32_t rid = sb_inst->result_id();
const uint32_t nid = this->TakeNextId();
get_decoration_mgr()->CloneDecorations(rid, nid);
sb_inst->SetResultId(nid);
(*postCallSB)[rid] = nid;
*iid = nid;
(*block_ptr)->AddInstruction(std::move(sb_inst));
}
} else {
// Reset same-block op operand.
*iid = mapItr->second;
}
});
}
void InlinePass::GenInlineCode(
std::vector<std::unique_ptr<BasicBlock>>* new_blocks,
std::vector<std::unique_ptr<Instruction>>* new_vars,
BasicBlock::iterator call_inst_itr,
UptrVectorIterator<BasicBlock> call_block_itr) {
// Map from all ids in the callee to their equivalent id in the caller
// as callee instructions are copied into caller.
std::unordered_map<uint32_t, uint32_t> callee2caller;
// Pre-call same-block insts
std::unordered_map<uint32_t, Instruction*> preCallSB;
// Post-call same-block op ids
std::unordered_map<uint32_t, uint32_t> postCallSB;
// Invalidate the def-use chains. They are not kept up to date while
// inlining. However, certain calls try to keep them up-to-date if they are
// valid. These operations can fail.
context()->InvalidateAnalyses(IRContext::kAnalysisDefUse);
Function* calleeFn = id2function_[call_inst_itr->GetSingleWordOperand(
kSpvFunctionCallFunctionId)];
// Check for multiple returns in the callee.
auto fi = multi_return_funcs_.find(calleeFn->result_id());
const bool multiReturn = fi != multi_return_funcs_.end();
// Map parameters to actual arguments.
MapParams(calleeFn, call_inst_itr, &callee2caller);
// Define caller local variables for all callee variables and create map to
// them.
CloneAndMapLocals(calleeFn, new_vars, &callee2caller);
// Create return var if needed.
uint32_t returnVarId = CreateReturnVar(calleeFn, new_vars);
// Create set of callee result ids. Used to detect forward references
std::unordered_set<uint32_t> callee_result_ids;
calleeFn->ForEachInst([&callee_result_ids](const Instruction* cpi) {
const uint32_t rid = cpi->result_id();
if (rid != 0) callee_result_ids.insert(rid);
});
// If the caller is in a single-block loop, and the callee has multiple
// blocks, then the normal inlining logic will place the OpLoopMerge in
// the last of several blocks in the loop. Instead, it should be placed
// at the end of the first block. First determine if the caller is in a
// single block loop. We'll wait to move the OpLoopMerge until the end
// of the regular inlining logic, and only if necessary.
bool caller_is_single_block_loop = false;
bool caller_is_loop_header = false;
if (auto* loop_merge = call_block_itr->GetLoopMergeInst()) {
caller_is_loop_header = true;
caller_is_single_block_loop =
call_block_itr->id() ==
loop_merge->GetSingleWordInOperand(kSpvLoopMergeContinueTargetIdInIdx);
}
bool callee_begins_with_structured_header =
(*(calleeFn->begin())).GetMergeInst() != nullptr;
// Clone and map callee code. Copy caller block code to beginning of
// first block and end of last block.
bool prevInstWasReturn = false;
uint32_t singleTripLoopHeaderId = 0;
uint32_t singleTripLoopContinueId = 0;
uint32_t returnLabelId = 0;
bool multiBlocks = false;
const uint32_t calleeTypeId = calleeFn->type_id();
// new_blk_ptr is a new basic block in the caller. New instructions are
// written to it. It is created when we encounter the OpLabel
// of the first callee block. It is appended to new_blocks only when
// it is complete.
std::unique_ptr<BasicBlock> new_blk_ptr;
calleeFn->ForEachInst([&new_blocks, &callee2caller, &call_block_itr,
&call_inst_itr, &new_blk_ptr, &prevInstWasReturn,
&returnLabelId, &returnVarId, caller_is_loop_header,
callee_begins_with_structured_header, &calleeTypeId,
&multiBlocks, &postCallSB, &preCallSB, multiReturn,
&singleTripLoopHeaderId, &singleTripLoopContinueId,
&callee_result_ids, this](const Instruction* cpi) {
switch (cpi->opcode()) {
case SpvOpFunction:
case SpvOpFunctionParameter:
// Already processed
break;
case SpvOpVariable:
if (cpi->NumInOperands() == 2) {
assert(callee2caller.count(cpi->result_id()) &&
"Expected the variable to have already been mapped.");
uint32_t new_var_id = callee2caller.at(cpi->result_id());
// The initializer must be a constant or global value. No mapped
// should be used.
uint32_t val_id = cpi->GetSingleWordInOperand(1);
AddStore(new_var_id, val_id, &new_blk_ptr);
}
break;
case SpvOpUnreachable:
case SpvOpKill: {
// Generate a return label so that we split the block with the function
// call. Copy the terminator into the new block.
if (returnLabelId == 0) returnLabelId = this->TakeNextId();
std::unique_ptr<Instruction> terminator(
new Instruction(context(), cpi->opcode(), 0, 0, {}));
new_blk_ptr->AddInstruction(std::move(terminator));
break;
}
case SpvOpLabel: {
// If previous instruction was early return, insert branch
// instruction to return block.
if (prevInstWasReturn) {
if (returnLabelId == 0) returnLabelId = this->TakeNextId();
AddBranch(returnLabelId, &new_blk_ptr);
prevInstWasReturn = false;
}
// Finish current block (if it exists) and get label for next block.
uint32_t labelId;
bool firstBlock = false;
if (new_blk_ptr != nullptr) {
new_blocks->push_back(std::move(new_blk_ptr));
// If result id is already mapped, use it, otherwise get a new
// one.
const uint32_t rid = cpi->result_id();
const auto mapItr = callee2caller.find(rid);
labelId = (mapItr != callee2caller.end()) ? mapItr->second
: this->TakeNextId();
} else {
// First block needs to use label of original block
// but map callee label in case of phi reference.
labelId = call_block_itr->id();
callee2caller[cpi->result_id()] = labelId;
firstBlock = true;
}
// Create first/next block.
new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(labelId));
if (firstBlock) {
// Copy contents of original caller block up to call instruction.
for (auto cii = call_block_itr->begin(); cii != call_inst_itr;
cii = call_block_itr->begin()) {
Instruction* inst = &*cii;
inst->RemoveFromList();
std::unique_ptr<Instruction> cp_inst(inst);
// Remember same-block ops for possible regeneration.
if (IsSameBlockOp(&*cp_inst)) {
auto* sb_inst_ptr = cp_inst.get();
preCallSB[cp_inst->result_id()] = sb_inst_ptr;
}
new_blk_ptr->AddInstruction(std::move(cp_inst));
}
if (caller_is_loop_header && callee_begins_with_structured_header) {
// We can't place both the caller's merge instruction and another
// merge instruction in the same block. So split the calling block.
// Insert an unconditional branch to a new guard block. Later,
// once we know the ID of the last block, we will move the caller's
// OpLoopMerge from the last generated block into the first block.
// We also wait to avoid invalidating various iterators.
const auto guard_block_id = this->TakeNextId();
AddBranch(guard_block_id, &new_blk_ptr);
new_blocks->push_back(std::move(new_blk_ptr));
// Start the next block.
new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(guard_block_id));
// Reset the mapping of the callee's entry block to point to
// the guard block. Do this so we can fix up phis later on to
// satisfy dominance.
callee2caller[cpi->result_id()] = guard_block_id;
}
// If callee has multiple returns, insert a header block for
// single-trip loop that will encompass callee code. Start postheader
// block.
//
// Note: Consider the following combination:
// - the caller is a single block loop
// - the callee does not begin with a structure header
// - the callee has multiple returns.
// We still need to split the caller block and insert a guard block.
// But we only need to do it once. We haven't done it yet, but the
// single-trip loop header will serve the same purpose.
if (multiReturn) {
singleTripLoopHeaderId = this->TakeNextId();
AddBranch(singleTripLoopHeaderId, &new_blk_ptr);
new_blocks->push_back(std::move(new_blk_ptr));
new_blk_ptr =
MakeUnique<BasicBlock>(NewLabel(singleTripLoopHeaderId));
returnLabelId = this->TakeNextId();
singleTripLoopContinueId = this->TakeNextId();
AddLoopMerge(returnLabelId, singleTripLoopContinueId, &new_blk_ptr);
uint32_t postHeaderId = this->TakeNextId();
AddBranch(postHeaderId, &new_blk_ptr);
new_blocks->push_back(std::move(new_blk_ptr));
new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(postHeaderId));
multiBlocks = true;
// Reset the mapping of the callee's entry block to point to
// the post-header block. Do this so we can fix up phis later
// on to satisfy dominance.
callee2caller[cpi->result_id()] = postHeaderId;
}
} else {
multiBlocks = true;
}
} break;
case SpvOpReturnValue: {
// Store return value to return variable.
assert(returnVarId != 0);
uint32_t valId = cpi->GetInOperand(kSpvReturnValueId).words[0];
const auto mapItr = callee2caller.find(valId);
if (mapItr != callee2caller.end()) {
valId = mapItr->second;
}
AddStore(returnVarId, valId, &new_blk_ptr);
// Remember we saw a return; if followed by a label, will need to
// insert branch.
prevInstWasReturn = true;
} break;
case SpvOpReturn: {
// Remember we saw a return; if followed by a label, will need to
// insert branch.
prevInstWasReturn = true;
} break;
case SpvOpFunctionEnd: {
// If there was an early return, we generated a return label id
// for it. Now we have to generate the return block with that Id.
if (returnLabelId != 0) {
// If previous instruction was return, insert branch instruction
// to return block.
if (prevInstWasReturn) AddBranch(returnLabelId, &new_blk_ptr);
if (multiReturn) {
// If we generated a loop header to for the single-trip loop
// to accommodate multiple returns, insert the continue
// target block now, with a false branch back to the loop header.
new_blocks->push_back(std::move(new_blk_ptr));
new_blk_ptr =
MakeUnique<BasicBlock>(NewLabel(singleTripLoopContinueId));
AddBranchCond(GetFalseId(), singleTripLoopHeaderId, returnLabelId,
&new_blk_ptr);
}
// Generate the return block.
new_blocks->push_back(std::move(new_blk_ptr));
new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(returnLabelId));
multiBlocks = true;
}
// Load return value into result id of call, if it exists.
if (returnVarId != 0) {
const uint32_t resId = call_inst_itr->result_id();
assert(resId != 0);
AddLoad(calleeTypeId, resId, returnVarId, &new_blk_ptr);
}
// Copy remaining instructions from caller block.
for (Instruction* inst = call_inst_itr->NextNode(); inst;
inst = call_inst_itr->NextNode()) {
inst->RemoveFromList();
std::unique_ptr<Instruction> cp_inst(inst);
// If multiple blocks generated, regenerate any same-block
// instruction that has not been seen in this last block.
if (multiBlocks) {
CloneSameBlockOps(&cp_inst, &postCallSB, &preCallSB, &new_blk_ptr);
// Remember same-block ops in this block.
if (IsSameBlockOp(&*cp_inst)) {
const uint32_t rid = cp_inst->result_id();
postCallSB[rid] = rid;
}
}
new_blk_ptr->AddInstruction(std::move(cp_inst));
}
// Finalize inline code.
new_blocks->push_back(std::move(new_blk_ptr));
} break;
default: {
// Copy callee instruction and remap all input Ids.
std::unique_ptr<Instruction> cp_inst(cpi->Clone(context()));
cp_inst->ForEachInId([&callee2caller, &callee_result_ids,
this](uint32_t* iid) {
const auto mapItr = callee2caller.find(*iid);
if (mapItr != callee2caller.end()) {
*iid = mapItr->second;
} else if (callee_result_ids.find(*iid) != callee_result_ids.end()) {
// Forward reference. Allocate a new id, map it,
// use it and check for it when remapping result ids
const uint32_t nid = this->TakeNextId();
callee2caller[*iid] = nid;
*iid = nid;
}
});
// If result id is non-zero, remap it. If already mapped, use mapped
// value, else use next id.
const uint32_t rid = cp_inst->result_id();
if (rid != 0) {
const auto mapItr = callee2caller.find(rid);
uint32_t nid;
if (mapItr != callee2caller.end()) {
nid = mapItr->second;
} else {
nid = this->TakeNextId();
callee2caller[rid] = nid;
}
cp_inst->SetResultId(nid);
get_decoration_mgr()->CloneDecorations(rid, nid);
}
new_blk_ptr->AddInstruction(std::move(cp_inst));
} break;
}
});
if (caller_is_loop_header && (new_blocks->size() > 1)) {
// Move the OpLoopMerge from the last block back to the first, where
// it belongs.
auto& first = new_blocks->front();
auto& last = new_blocks->back();
assert(first != last);
// Insert a modified copy of the loop merge into the first block.
auto loop_merge_itr = last->tail();
--loop_merge_itr;
assert(loop_merge_itr->opcode() == SpvOpLoopMerge);
std::unique_ptr<Instruction> cp_inst(loop_merge_itr->Clone(context()));
if (caller_is_single_block_loop) {
// Also, update its continue target to point to the last block.
cp_inst->SetInOperand(kSpvLoopMergeContinueTargetIdInIdx, {last->id()});
}
first->tail().InsertBefore(std::move(cp_inst));
// Remove the loop merge from the last block.
loop_merge_itr->RemoveFromList();
delete &*loop_merge_itr;
}
// Update block map given replacement blocks.
for (auto& blk : *new_blocks) {
id2block_[blk->id()] = &*blk;
}
}
bool InlinePass::IsInlinableFunctionCall(const Instruction* inst) {
if (inst->opcode() != SpvOp::SpvOpFunctionCall) return false;
const uint32_t calleeFnId =
inst->GetSingleWordOperand(kSpvFunctionCallFunctionId);
const auto ci = inlinable_.find(calleeFnId);
return ci != inlinable_.cend();
}
void InlinePass::UpdateSucceedingPhis(
std::vector<std::unique_ptr<BasicBlock>>& new_blocks) {
const auto firstBlk = new_blocks.begin();
const auto lastBlk = new_blocks.end() - 1;
const uint32_t firstId = (*firstBlk)->id();
const uint32_t lastId = (*lastBlk)->id();
const BasicBlock& const_last_block = *lastBlk->get();
const_last_block.ForEachSuccessorLabel(
[&firstId, &lastId, this](const uint32_t succ) {
BasicBlock* sbp = this->id2block_[succ];
sbp->ForEachPhiInst([&firstId, &lastId](Instruction* phi) {
phi->ForEachInId([&firstId, &lastId](uint32_t* id) {
if (*id == firstId) *id = lastId;
});
});
});
}
bool InlinePass::HasMultipleReturns(Function* func) {
bool seenReturn = false;
bool multipleReturns = false;
for (auto& blk : *func) {
auto terminal_ii = blk.cend();
--terminal_ii;
if (terminal_ii->opcode() == SpvOpReturn ||
terminal_ii->opcode() == SpvOpReturnValue) {
if (seenReturn) {
multipleReturns = true;
break;
}
seenReturn = true;
}
}
return multipleReturns;
}
void InlinePass::ComputeStructuredSuccessors(Function* func) {
// If header, make merge block first successor.
for (auto& blk : *func) {
uint32_t mbid = blk.MergeBlockIdIfAny();
if (mbid != 0) {
block2structured_succs_[&blk].push_back(id2block_[mbid]);
}
// Add true successors.
const auto& const_blk = blk;
const_blk.ForEachSuccessorLabel([&blk, this](const uint32_t sbid) {
block2structured_succs_[&blk].push_back(id2block_[sbid]);
});
}
}
InlinePass::GetBlocksFunction InlinePass::StructuredSuccessorsFunction() {
return [this](const BasicBlock* block) {
return &(block2structured_succs_[block]);
};
}
bool InlinePass::HasNoReturnInLoop(Function* func) {
// If control not structured, do not do loop/return analysis
// TODO: Analyze returns in non-structured control flow
if (!context()->get_feature_mgr()->HasCapability(SpvCapabilityShader))
return false;
// Compute structured block order. This order has the property
// that dominators are before all blocks they dominate and merge blocks
// are after all blocks that are in the control constructs of their header.
ComputeStructuredSuccessors(func);
auto ignore_block = [](cbb_ptr) {};
auto ignore_edge = [](cbb_ptr, cbb_ptr) {};
std::list<const BasicBlock*> structuredOrder;
CFA<BasicBlock>::DepthFirstTraversal(
&*func->begin(), StructuredSuccessorsFunction(), ignore_block,
[&](cbb_ptr b) { structuredOrder.push_front(b); }, ignore_edge);
// Search for returns in loops. Only need to track outermost loop
bool return_in_loop = false;
uint32_t outerLoopMergeId = 0;
for (auto& blk : structuredOrder) {
// Exiting current outer loop
if (blk->id() == outerLoopMergeId) outerLoopMergeId = 0;
// Return block
auto terminal_ii = blk->cend();
--terminal_ii;
if (terminal_ii->opcode() == SpvOpReturn ||
terminal_ii->opcode() == SpvOpReturnValue) {
if (outerLoopMergeId != 0) {
return_in_loop = true;
break;
}
} else if (terminal_ii != blk->cbegin()) {
auto merge_ii = terminal_ii;
--merge_ii;
// Entering outermost loop
if (merge_ii->opcode() == SpvOpLoopMerge && outerLoopMergeId == 0)
outerLoopMergeId =
merge_ii->GetSingleWordOperand(kSpvLoopMergeMergeBlockId);
}
}
return !return_in_loop;
}
void InlinePass::AnalyzeReturns(Function* func) {
// Look for multiple returns
if (!HasMultipleReturns(func)) {
no_return_in_loop_.insert(func->result_id());
return;
}
multi_return_funcs_.insert(func->result_id());
// If multiple returns, see if any are in a loop
if (HasNoReturnInLoop(func)) no_return_in_loop_.insert(func->result_id());
}
bool InlinePass::IsInlinableFunction(Function* func) {
// We can only inline a function if it has blocks.
if (func->cbegin() == func->cend()) return false;
// Do not inline functions with returns in loops. Currently early return
// functions are inlined by wrapping them in a one trip loop and implementing
// the returns as a branch to the loop's merge block. However, this can only
// done validly if the return was not in a loop in the original function.
// Also remember functions with multiple (early) returns.
AnalyzeReturns(func);
return no_return_in_loop_.find(func->result_id()) !=
no_return_in_loop_.cend();
}
void InlinePass::InitializeInline() {
false_id_ = 0;
// clear collections
id2function_.clear();
id2block_.clear();
block2structured_succs_.clear();
inlinable_.clear();
no_return_in_loop_.clear();
multi_return_funcs_.clear();
for (auto& fn : *get_module()) {
// Initialize function and block maps.
id2function_[fn.result_id()] = &fn;
for (auto& blk : fn) {
id2block_[blk.id()] = &blk;
}
// Compute inlinability
if (IsInlinableFunction(&fn)) inlinable_.insert(fn.result_id());
}
}
InlinePass::InlinePass() {}
} // namespace opt
} // namespace spvtools