blob: 373d2990aed248e57818c34ac4b30adc33aedf3c [file] [log] [blame]
// 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 "mem_pass.h"
#include "basic_block.h"
#include "cfa.h"
#include "dominator_analysis.h"
#include "ir_context.h"
#include "iterator.h"
namespace spvtools {
namespace opt {
namespace {
const uint32_t kCopyObjectOperandInIdx = 0;
const uint32_t kLoadPtrIdInIdx = 0;
const uint32_t kLoopMergeMergeBlockIdInIdx = 0;
const uint32_t kStorePtrIdInIdx = 0;
const uint32_t kStoreValIdInIdx = 1;
const uint32_t kTypePointerStorageClassInIdx = 0;
const uint32_t kTypePointerTypeIdInIdx = 1;
const uint32_t kVariableInitIdInIdx = 1;
} // namespace
bool MemPass::IsBaseTargetType(const ir::Instruction* typeInst) const {
switch (typeInst->opcode()) {
case SpvOpTypeInt:
case SpvOpTypeFloat:
case SpvOpTypeBool:
case SpvOpTypeVector:
case SpvOpTypeMatrix:
case SpvOpTypeImage:
case SpvOpTypeSampler:
case SpvOpTypeSampledImage:
case SpvOpTypePointer:
return true;
default:
break;
}
return false;
}
bool MemPass::IsTargetType(const ir::Instruction* typeInst) const {
if (IsBaseTargetType(typeInst)) return true;
if (typeInst->opcode() == SpvOpTypeArray) {
if (!IsTargetType(
get_def_use_mgr()->GetDef(typeInst->GetSingleWordOperand(1)))) {
return false;
}
return true;
}
if (typeInst->opcode() != SpvOpTypeStruct) return false;
// All struct members must be math type
return typeInst->WhileEachInId([this](const uint32_t* tid) {
ir::Instruction* compTypeInst = get_def_use_mgr()->GetDef(*tid);
if (!IsTargetType(compTypeInst)) return false;
return true;
});
}
bool MemPass::IsNonPtrAccessChain(const SpvOp opcode) const {
return opcode == SpvOpAccessChain || opcode == SpvOpInBoundsAccessChain;
}
bool MemPass::IsPtr(uint32_t ptrId) {
uint32_t varId = ptrId;
ir::Instruction* ptrInst = get_def_use_mgr()->GetDef(varId);
while (ptrInst->opcode() == SpvOpCopyObject) {
varId = ptrInst->GetSingleWordInOperand(kCopyObjectOperandInIdx);
ptrInst = get_def_use_mgr()->GetDef(varId);
}
const SpvOp op = ptrInst->opcode();
if (op == SpvOpVariable || IsNonPtrAccessChain(op)) return true;
if (op != SpvOpFunctionParameter) return false;
const uint32_t varTypeId = ptrInst->type_id();
const ir::Instruction* varTypeInst = get_def_use_mgr()->GetDef(varTypeId);
return varTypeInst->opcode() == SpvOpTypePointer;
}
ir::Instruction* MemPass::GetPtr(uint32_t ptrId, uint32_t* varId) {
*varId = ptrId;
ir::Instruction* ptrInst = get_def_use_mgr()->GetDef(*varId);
ir::Instruction* varInst;
if (ptrInst->opcode() != SpvOpVariable &&
ptrInst->opcode() != SpvOpFunctionParameter) {
varInst = ptrInst->GetBaseAddress();
} else {
varInst = ptrInst;
}
if (varInst->opcode() == SpvOpVariable) {
*varId = varInst->result_id();
} else {
*varId = 0;
}
while (ptrInst->opcode() == SpvOpCopyObject) {
uint32_t temp = ptrInst->GetSingleWordInOperand(0);
ptrInst = get_def_use_mgr()->GetDef(temp);
}
return ptrInst;
}
ir::Instruction* MemPass::GetPtr(ir::Instruction* ip, uint32_t* varId) {
const SpvOp op = ip->opcode();
assert(op == SpvOpStore || op == SpvOpLoad);
const uint32_t ptrId = ip->GetSingleWordInOperand(
op == SpvOpStore ? kStorePtrIdInIdx : kLoadPtrIdInIdx);
return GetPtr(ptrId, varId);
}
bool MemPass::HasOnlyNamesAndDecorates(uint32_t id) const {
return get_def_use_mgr()->WhileEachUser(id, [this](ir::Instruction* user) {
SpvOp op = user->opcode();
if (op != SpvOpName && !IsNonTypeDecorate(op)) {
return false;
}
return true;
});
}
void MemPass::KillAllInsts(ir::BasicBlock* bp, bool killLabel) {
bp->ForEachInst([this, killLabel](ir::Instruction* ip) {
if (killLabel || ip->opcode() != SpvOpLabel) {
context()->KillInst(ip);
}
});
}
bool MemPass::HasLoads(uint32_t varId) const {
return !get_def_use_mgr()->WhileEachUser(varId, [this](
ir::Instruction* user) {
SpvOp op = user->opcode();
// TODO(): The following is slightly conservative. Could be
// better handling of non-store/name.
if (IsNonPtrAccessChain(op) || op == SpvOpCopyObject) {
if (HasLoads(user->result_id())) {
return false;
}
} else if (op != SpvOpStore && op != SpvOpName && !IsNonTypeDecorate(op)) {
return false;
}
return true;
});
}
bool MemPass::IsLiveVar(uint32_t varId) const {
const ir::Instruction* varInst = get_def_use_mgr()->GetDef(varId);
// assume live if not a variable eg. function parameter
if (varInst->opcode() != SpvOpVariable) return true;
// non-function scope vars are live
const uint32_t varTypeId = varInst->type_id();
const ir::Instruction* varTypeInst = get_def_use_mgr()->GetDef(varTypeId);
if (varTypeInst->GetSingleWordInOperand(kTypePointerStorageClassInIdx) !=
SpvStorageClassFunction)
return true;
// test if variable is loaded from
return HasLoads(varId);
}
bool MemPass::IsLiveStore(ir::Instruction* storeInst) {
// get store's variable
uint32_t varId;
(void)GetPtr(storeInst, &varId);
if (varId == 0) {
// If we do not know which variable we are accessing, assume the store is
// live.
return true;
}
return IsLiveVar(varId);
}
void MemPass::AddStores(uint32_t ptr_id, std::queue<ir::Instruction*>* insts) {
get_def_use_mgr()->ForEachUser(ptr_id, [this, insts](ir::Instruction* user) {
SpvOp op = user->opcode();
if (IsNonPtrAccessChain(op)) {
AddStores(user->result_id(), insts);
} else if (op == SpvOpStore) {
insts->push(user);
}
});
}
void MemPass::DCEInst(ir::Instruction* inst,
const function<void(ir::Instruction*)>& call_back) {
std::queue<ir::Instruction*> deadInsts;
deadInsts.push(inst);
while (!deadInsts.empty()) {
ir::Instruction* di = deadInsts.front();
// Don't delete labels
if (di->opcode() == SpvOpLabel) {
deadInsts.pop();
continue;
}
// Remember operands
std::set<uint32_t> ids;
di->ForEachInId([&ids](uint32_t* iid) { ids.insert(*iid); });
uint32_t varId = 0;
// Remember variable if dead load
if (di->opcode() == SpvOpLoad) (void)GetPtr(di, &varId);
if (call_back) {
call_back(di);
}
context()->KillInst(di);
// For all operands with no remaining uses, add their instruction
// to the dead instruction queue.
for (auto id : ids)
if (HasOnlyNamesAndDecorates(id))
deadInsts.push(get_def_use_mgr()->GetDef(id));
// if a load was deleted and it was the variable's
// last load, add all its stores to dead queue
if (varId != 0 && !IsLiveVar(varId)) AddStores(varId, &deadInsts);
deadInsts.pop();
}
}
MemPass::MemPass() {}
bool MemPass::HasOnlySupportedRefs(uint32_t varId) {
if (supported_ref_vars_.find(varId) != supported_ref_vars_.end()) return true;
return get_def_use_mgr()->WhileEachUser(varId, [this](ir::Instruction* user) {
SpvOp op = user->opcode();
if (op != SpvOpStore && op != SpvOpLoad && op != SpvOpName &&
!IsNonTypeDecorate(op)) {
return false;
}
return true;
});
}
void MemPass::InitSSARewrite(ir::Function* func) {
// Clear collections.
seen_target_vars_.clear();
seen_non_target_vars_.clear();
visitedBlocks_.clear();
type2undefs_.clear();
supported_ref_vars_.clear();
block_defs_map_.clear();
phis_to_patch_.clear();
dominator_ = context()->GetDominatorAnalysis(func, *cfg());
// Collect target (and non-) variable sets. Remove variables with
// non-load/store refs from target variable set
for (auto& blk : *func) {
for (auto& inst : blk) {
switch (inst.opcode()) {
case SpvOpStore:
case SpvOpLoad: {
uint32_t varId;
(void)GetPtr(&inst, &varId);
if (!IsTargetVar(varId)) break;
if (HasOnlySupportedRefs(varId)) break;
seen_non_target_vars_.insert(varId);
seen_target_vars_.erase(varId);
} break;
default:
break;
}
}
}
}
bool MemPass::IsLiveAfter(uint32_t var_id, uint32_t label) const {
// For now, return very conservative result: true. This will result in
// correct, but possibly usused, phi code to be generated. A subsequent
// DCE pass should eliminate this code.
// TODO(greg-lunarg): Return more accurate information
(void)var_id;
(void)label;
return true;
}
uint32_t MemPass::Type2Undef(uint32_t type_id) {
const auto uitr = type2undefs_.find(type_id);
if (uitr != type2undefs_.end()) return uitr->second;
const uint32_t undefId = TakeNextId();
std::unique_ptr<ir::Instruction> undef_inst(
new ir::Instruction(context(), SpvOpUndef, type_id, undefId, {}));
get_def_use_mgr()->AnalyzeInstDefUse(&*undef_inst);
get_module()->AddGlobalValue(std::move(undef_inst));
type2undefs_[type_id] = undefId;
return undefId;
}
void MemPass::CollectLiveVars(uint32_t block_label,
std::map<uint32_t, uint32_t>* live_vars) {
// Walk up the dominator chain starting at |block_label| looking for variables
// defined at each block in the chain. Since we are only interested for the
// most recent value for each live variable, we only add a <variable, value>
// pair to |live_vars| if this is the first time we find the variable in the
// chain.
for (ir::BasicBlock* block = cfg()->block(block_label); block != nullptr;
block = dominator_->ImmediateDominator(block)) {
for (const auto& var_val : block_defs_map_[block->id()]) {
auto live_vars_it = live_vars->find(var_val.first);
if (live_vars_it == live_vars->end()) live_vars->insert(var_val);
}
}
}
uint32_t MemPass::GetCurrentValue(uint32_t var_id, uint32_t block_label) {
// Walk up the dominator chain starting at |block_label| looking for the
// current value of variable |var_id|. The first block we find containing a
// definition for |var_id| is the one we are interested in.
for (ir::BasicBlock* block = cfg()->block(block_label); block != nullptr;
block = dominator_->ImmediateDominator(block)) {
const auto& block_defs = block_defs_map_[block->id()];
const auto& var_val_it = block_defs.find(var_id);
if (var_val_it != block_defs.end()) return var_val_it->second;
}
return 0;
}
void MemPass::SSABlockInitLoopHeader(
std::list<ir::BasicBlock*>::iterator block_itr) {
const uint32_t label = (*block_itr)->id();
// Determine the back-edge label.
uint32_t backLabel = 0;
for (uint32_t predLabel : cfg()->preds(label))
if (visitedBlocks_.find(predLabel) == visitedBlocks_.end()) {
assert(backLabel == 0);
backLabel = predLabel;
break;
}
assert(backLabel != 0);
// Determine merge block.
auto mergeInst = (*block_itr)->end();
--mergeInst;
--mergeInst;
uint32_t mergeLabel =
mergeInst->GetSingleWordInOperand(kLoopMergeMergeBlockIdInIdx);
// Collect all live variables and a default value for each across all
// non-backedge predecesors. Must be ordered map because phis are
// generated based on order and test results will otherwise vary across
// platforms.
std::map<uint32_t, uint32_t> liveVars;
for (uint32_t predLabel : cfg()->preds(label)) {
CollectLiveVars(predLabel, &liveVars);
}
// Add all stored variables in loop. Set their default value id to zero.
for (auto bi = block_itr; (*bi)->id() != mergeLabel; ++bi) {
ir::BasicBlock* bp = *bi;
for (auto ii = bp->begin(); ii != bp->end(); ++ii) {
if (ii->opcode() != SpvOpStore) {
continue;
}
uint32_t varId;
(void)GetPtr(&*ii, &varId);
if (!IsTargetVar(varId)) {
continue;
}
liveVars[varId] = 0;
}
}
// Insert phi for all live variables that require them. All variables
// defined in loop require a phi. Otherwise all variables
// with differing predecessor values require a phi.
auto insertItr = (*block_itr)->begin();
for (auto var_val : liveVars) {
const uint32_t varId = var_val.first;
if (!IsLiveAfter(varId, label)) {
continue;
}
const uint32_t val0Id = var_val.second;
bool needsPhi = false;
if (val0Id != 0) {
for (uint32_t predLabel : cfg()->preds(label)) {
// Skip back edge predecessor.
if (predLabel == backLabel) continue;
uint32_t current_value = GetCurrentValue(varId, predLabel);
// Missing (undef) values always cause difference with (defined) value
if (current_value == 0) {
needsPhi = true;
break;
}
if (current_value != val0Id) {
needsPhi = true;
break;
}
}
} else {
needsPhi = true;
}
// If val is the same for all predecessors, enter it in map
if (!needsPhi) {
block_defs_map_[label].insert(var_val);
continue;
}
// Val differs across predecessors. Add phi op to block and
// add its result id to the map. For back edge predecessor,
// use the variable id. We will patch this after visiting back
// edge predecessor. For predecessors that do not define a value,
// use undef.
std::vector<ir::Operand> phi_in_operands;
uint32_t typeId = GetPointeeTypeId(get_def_use_mgr()->GetDef(varId));
for (uint32_t predLabel : cfg()->preds(label)) {
uint32_t valId;
if (predLabel == backLabel) {
valId = varId;
} else {
uint32_t current_value = GetCurrentValue(varId, predLabel);
if (current_value == 0)
valId = Type2Undef(typeId);
else
valId = current_value;
}
phi_in_operands.push_back(
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {valId}});
phi_in_operands.push_back(
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {predLabel}});
}
const uint32_t phiId = TakeNextId();
std::unique_ptr<ir::Instruction> newPhi(new ir::Instruction(
context(), SpvOpPhi, typeId, phiId, phi_in_operands));
// The only phis requiring patching are the ones we create.
phis_to_patch_.insert(phiId);
// Only analyze the phi define now; analyze the phi uses after the
// phi backedge predecessor value is patched.
get_def_use_mgr()->AnalyzeInstDef(&*newPhi);
context()->set_instr_block(&*newPhi, *block_itr);
insertItr = insertItr.InsertBefore(std::move(newPhi));
++insertItr;
block_defs_map_[label].insert({varId, phiId});
}
}
void MemPass::SSABlockInitMultiPred(ir::BasicBlock* block_ptr) {
const uint32_t label = block_ptr->id();
// Collect all live variables and a default value for each across all
// predecesors. Must be ordered map because phis are generated based on
// order and test results will otherwise vary across platforms.
std::map<uint32_t, uint32_t> liveVars;
for (uint32_t predLabel : cfg()->preds(label)) {
assert(visitedBlocks_.find(predLabel) != visitedBlocks_.end());
CollectLiveVars(predLabel, &liveVars);
}
// For each live variable, look for a difference in values across
// predecessors that would require a phi and insert one.
auto insertItr = block_ptr->begin();
for (auto var_val : liveVars) {
const uint32_t varId = var_val.first;
if (!IsLiveAfter(varId, label)) continue;
const uint32_t val0Id = var_val.second;
bool differs = false;
for (uint32_t predLabel : cfg()->preds(label)) {
uint32_t current_value = GetCurrentValue(varId, predLabel);
// Missing values cause a difference because we'll need to create an
// undef for that predecessor.
if (current_value == 0) {
differs = true;
break;
}
if (current_value != val0Id) {
differs = true;
break;
}
}
// If val is the same for all predecessors, enter it in map
if (!differs) {
block_defs_map_[label].insert(var_val);
continue;
}
// Val differs across predecessors. Add phi op to block and add its result
// id to the map.
std::vector<ir::Operand> phi_in_operands;
const uint32_t typeId = GetPointeeTypeId(get_def_use_mgr()->GetDef(varId));
for (uint32_t predLabel : cfg()->preds(label)) {
uint32_t current_value = GetCurrentValue(varId, predLabel);
// If variable not defined on this path, use undef
const uint32_t valId =
(current_value > 0) ? current_value : Type2Undef(typeId);
phi_in_operands.push_back(
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {valId}});
phi_in_operands.push_back(
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {predLabel}});
}
const uint32_t phiId = TakeNextId();
std::unique_ptr<ir::Instruction> newPhi(new ir::Instruction(
context(), SpvOpPhi, typeId, phiId, phi_in_operands));
get_def_use_mgr()->AnalyzeInstDefUse(&*newPhi);
context()->set_instr_block(&*newPhi, block_ptr);
insertItr = insertItr.InsertBefore(std::move(newPhi));
++insertItr;
block_defs_map_[label].insert({varId, phiId});
}
}
void MemPass::SSABlockInit(std::list<ir::BasicBlock*>::iterator block_itr) {
const size_t numPreds = cfg()->preds((*block_itr)->id()).size();
if (numPreds == 0) return;
if ((*block_itr)->IsLoopHeader())
SSABlockInitLoopHeader(block_itr);
else
SSABlockInitMultiPred(*block_itr);
}
bool MemPass::IsTargetVar(uint32_t varId) {
if (varId == 0) {
return false;
}
if (seen_non_target_vars_.find(varId) != seen_non_target_vars_.end())
return false;
if (seen_target_vars_.find(varId) != seen_target_vars_.end()) return true;
const ir::Instruction* varInst = get_def_use_mgr()->GetDef(varId);
if (varInst->opcode() != SpvOpVariable) return false;
const uint32_t varTypeId = varInst->type_id();
const ir::Instruction* varTypeInst = get_def_use_mgr()->GetDef(varTypeId);
if (varTypeInst->GetSingleWordInOperand(kTypePointerStorageClassInIdx) !=
SpvStorageClassFunction) {
seen_non_target_vars_.insert(varId);
return false;
}
const uint32_t varPteTypeId =
varTypeInst->GetSingleWordInOperand(kTypePointerTypeIdInIdx);
ir::Instruction* varPteTypeInst = get_def_use_mgr()->GetDef(varPteTypeId);
if (!IsTargetType(varPteTypeInst)) {
seen_non_target_vars_.insert(varId);
return false;
}
seen_target_vars_.insert(varId);
return true;
}
void MemPass::PatchPhis(uint32_t header_id, uint32_t back_id) {
ir::BasicBlock* header = cfg()->block(header_id);
auto phiItr = header->begin();
for (; phiItr->opcode() == SpvOpPhi; ++phiItr) {
// Only patch phis that we created in a loop header.
// There might be other phis unrelated to our optimizations.
if (0 == phis_to_patch_.count(phiItr->result_id())) continue;
// Find phi operand index for back edge
uint32_t cnt = 0;
uint32_t idx = phiItr->NumInOperands();
phiItr->ForEachInId([&cnt, &back_id, &idx](uint32_t* iid) {
if (cnt % 2 == 1 && *iid == back_id) idx = cnt - 1;
++cnt;
});
assert(idx != phiItr->NumInOperands());
// Replace temporary phi operand with variable's value in backedge block
// map. Use undef if variable not in map.
const uint32_t varId = phiItr->GetSingleWordInOperand(idx);
uint32_t current_value = GetCurrentValue(varId, back_id);
uint32_t valId =
(current_value > 0)
? current_value
: Type2Undef(GetPointeeTypeId(get_def_use_mgr()->GetDef(varId)));
phiItr->SetInOperand(idx, {valId});
// Analyze uses now that they are complete
get_def_use_mgr()->AnalyzeInstUse(&*phiItr);
}
}
Pass::Status MemPass::InsertPhiInstructions(ir::Function* func) {
// TODO(dnovillo) the current Phi placement mechanism assumes structured
// control-flow. This should be generalized
// (https://github.com/KhronosGroup/SPIRV-Tools/issues/893).
assert(context()->get_feature_mgr()->HasCapability(SpvCapabilityShader) &&
"This only works on structured control flow");
// Initialize the data structures used to insert Phi instructions.
InitSSARewrite(func);
// Process all blocks in structured order. This is just one way (the
// simplest?) to make sure all predecessors blocks are processed before
// a block itself.
std::list<ir::BasicBlock*> structuredOrder;
cfg()->ComputeStructuredOrder(func, cfg()->pseudo_entry_block(),
&structuredOrder);
for (auto bi = structuredOrder.begin(); bi != structuredOrder.end(); ++bi) {
// Skip pseudo entry block
if (cfg()->IsPseudoEntryBlock(*bi)) {
continue;
}
// Process all stores and loads of targeted variables.
SSABlockInit(bi);
ir::BasicBlock* bp = *bi;
const uint32_t label = bp->id();
ir::Instruction* inst = &*bp->begin();
while (inst) {
ir::Instruction* next_instruction = inst->NextNode();
switch (inst->opcode()) {
case SpvOpStore: {
uint32_t varId;
(void)GetPtr(inst, &varId);
if (!IsTargetVar(varId)) break;
// Register new stored value for the variable
block_defs_map_[label][varId] =
inst->GetSingleWordInOperand(kStoreValIdInIdx);
} break;
case SpvOpVariable: {
// Treat initialized OpVariable like an OpStore
if (inst->NumInOperands() < 2) break;
uint32_t varId = inst->result_id();
if (!IsTargetVar(varId)) break;
// Register new stored value for the variable
block_defs_map_[label][varId] =
inst->GetSingleWordInOperand(kVariableInitIdInIdx);
} break;
case SpvOpLoad: {
uint32_t varId;
(void)GetPtr(inst, &varId);
if (!IsTargetVar(varId)) break;
uint32_t replId = GetCurrentValue(varId, label);
// If the variable is not defined, use undef.
if (replId == 0) {
replId =
Type2Undef(GetPointeeTypeId(get_def_use_mgr()->GetDef(varId)));
}
// Replace load's id with the last stored value id for variable
// and delete load. Kill any names or decorates using id before
// replacing to prevent incorrect replacement in those instructions.
const uint32_t loadId = inst->result_id();
context()->KillNamesAndDecorates(loadId);
(void)context()->ReplaceAllUsesWith(loadId, replId);
context()->KillInst(inst);
} break;
default:
break;
}
inst = next_instruction;
}
visitedBlocks_.insert(label);
// Look for successor backedge and patch phis in loop header
// if found.
uint32_t header = 0;
const auto* const_bp = bp;
const_bp->ForEachSuccessorLabel([&header, this](uint32_t succ) {
if (visitedBlocks_.find(succ) == visitedBlocks_.end()) return;
assert(header == 0);
header = succ;
});
if (header != 0) PatchPhis(header, label);
}
return Status::SuccessWithChange;
}
// Remove all |phi| operands coming from unreachable blocks (i.e., blocks not in
// |reachable_blocks|). There are two types of removal that this function can
// perform:
//
// 1- Any operand that comes directly from an unreachable block is completely
// removed. Since the block is unreachable, the edge between the unreachable
// block and the block holding |phi| has been removed.
//
// 2- Any operand that comes via a live block and was defined at an unreachable
// block gets its value replaced with an OpUndef value. Since the argument
// was generated in an unreachable block, it no longer exists, so it cannot
// be referenced. However, since the value does not reach |phi| directly
// from the unreachable block, the operand cannot be removed from |phi|.
// Therefore, we replace the argument value with OpUndef.
//
// For example, in the switch() below, assume that we want to remove the
// argument with value %11 coming from block %41.
//
// [ ... ]
// %41 = OpLabel <--- Unreachable block
// %11 = OpLoad %int %y
// [ ... ]
// OpSelectionMerge %16 None
// OpSwitch %12 %16 10 %13 13 %14 18 %15
// %13 = OpLabel
// OpBranch %16
// %14 = OpLabel
// OpStore %outparm %int_14
// OpBranch %16
// %15 = OpLabel
// OpStore %outparm %int_15
// OpBranch %16
// %16 = OpLabel
// %30 = OpPhi %int %11 %41 %int_42 %13 %11 %14 %11 %15
//
// Since %41 is now an unreachable block, the first operand of |phi| needs to
// be removed completely. But the operands (%11 %14) and (%11 %15) cannot be
// removed because %14 and %15 are reachable blocks. Since %11 no longer exist,
// in those arguments, we replace all references to %11 with an OpUndef value.
// This results in |phi| looking like:
//
// %50 = OpUndef %int
// [ ... ]
// %30 = OpPhi %int %int_42 %13 %50 %14 %50 %15
void MemPass::RemovePhiOperands(
ir::Instruction* phi,
std::unordered_set<ir::BasicBlock*> reachable_blocks) {
std::vector<ir::Operand> keep_operands;
uint32_t type_id = 0;
// The id of an undefined value we've generated.
uint32_t undef_id = 0;
// Traverse all the operands in |phi|. Build the new operand vector by adding
// all the original operands from |phi| except the unwanted ones.
for (uint32_t i = 0; i < phi->NumOperands();) {
if (i < 2) {
// The first two arguments are always preserved.
keep_operands.push_back(phi->GetOperand(i));
++i;
continue;
}
// The remaining Phi arguments come in pairs. Index 'i' contains the
// variable id, index 'i + 1' is the originating block id.
assert(i % 2 == 0 && i < phi->NumOperands() - 1 &&
"malformed Phi arguments");
ir::BasicBlock* in_block = cfg()->block(phi->GetSingleWordOperand(i + 1));
if (reachable_blocks.find(in_block) == reachable_blocks.end()) {
// If the incoming block is unreachable, remove both operands as this
// means that the |phi| has lost an incoming edge.
i += 2;
continue;
}
// In all other cases, the operand must be kept but may need to be changed.
uint32_t arg_id = phi->GetSingleWordOperand(i);
ir::Instruction* arg_def_instr = get_def_use_mgr()->GetDef(arg_id);
ir::BasicBlock* def_block = context()->get_instr_block(arg_def_instr);
if (def_block &&
reachable_blocks.find(def_block) == reachable_blocks.end()) {
// If the current |phi| argument was defined in an unreachable block, it
// means that this |phi| argument is no longer defined. Replace it with
// |undef_id|.
if (!undef_id) {
type_id = arg_def_instr->type_id();
undef_id = Type2Undef(type_id);
}
keep_operands.push_back(
ir::Operand(spv_operand_type_t::SPV_OPERAND_TYPE_ID, {undef_id}));
} else {
// Otherwise, the argument comes from a reachable block or from no block
// at all (meaning that it was defined in the global section of the
// program). In both cases, keep the argument intact.
keep_operands.push_back(phi->GetOperand(i));
}
keep_operands.push_back(phi->GetOperand(i + 1));
i += 2;
}
context()->ForgetUses(phi);
phi->ReplaceOperands(keep_operands);
context()->AnalyzeUses(phi);
}
void MemPass::RemoveBlock(ir::Function::iterator* bi) {
auto& rm_block = **bi;
// Remove instructions from the block.
rm_block.ForEachInst([&rm_block, this](ir::Instruction* inst) {
// Note that we do not kill the block label instruction here. The label
// instruction is needed to identify the block, which is needed by the
// removal of phi operands.
if (inst != rm_block.GetLabelInst()) {
context()->KillInst(inst);
}
});
// Remove the label instruction last.
auto label = rm_block.GetLabelInst();
context()->KillInst(label);
*bi = bi->Erase();
}
bool MemPass::RemoveUnreachableBlocks(ir::Function* func) {
bool modified = false;
// Mark reachable all blocks reachable from the function's entry block.
std::unordered_set<ir::BasicBlock*> reachable_blocks;
std::unordered_set<ir::BasicBlock*> visited_blocks;
std::queue<ir::BasicBlock*> worklist;
reachable_blocks.insert(func->entry().get());
// Initially mark the function entry point as reachable.
worklist.push(func->entry().get());
auto mark_reachable = [&reachable_blocks, &visited_blocks, &worklist,
this](uint32_t label_id) {
auto successor = cfg()->block(label_id);
if (visited_blocks.count(successor) == 0) {
reachable_blocks.insert(successor);
worklist.push(successor);
visited_blocks.insert(successor);
}
};
// Transitively mark all blocks reachable from the entry as reachable.
while (!worklist.empty()) {
ir::BasicBlock* block = worklist.front();
worklist.pop();
// All the successors of a live block are also live.
static_cast<const ir::BasicBlock*>(block)->ForEachSuccessorLabel(
mark_reachable);
// All the Merge and ContinueTarget blocks of a live block are also live.
block->ForMergeAndContinueLabel(mark_reachable);
}
// Update operands of Phi nodes that reference unreachable blocks.
for (auto& block : *func) {
// If the block is about to be removed, don't bother updating its
// Phi instructions.
if (reachable_blocks.count(&block) == 0) {
continue;
}
// If the block is reachable and has Phi instructions, remove all
// operands from its Phi instructions that reference unreachable blocks.
// If the block has no Phi instructions, this is a no-op.
block.ForEachPhiInst([&reachable_blocks, this](ir::Instruction* phi) {
RemovePhiOperands(phi, reachable_blocks);
});
}
// Erase unreachable blocks.
for (auto ebi = func->begin(); ebi != func->end();) {
if (reachable_blocks.count(&*ebi) == 0) {
RemoveBlock(&ebi);
modified = true;
} else {
++ebi;
}
}
return modified;
}
bool MemPass::CFGCleanup(ir::Function* func) {
bool modified = false;
modified |= RemoveUnreachableBlocks(func);
return modified;
}
} // namespace opt
} // namespace spvtools