mirror of
https://github.com/KhronosGroup/SPIRV-Tools
synced 2024-11-30 15:00:06 +00:00
2cb589cc14
The algorithm used in DCEInst to remove dead code is very slow. It is fine if you only want to remove a small number of instructions, but, if you need to remove a large number of instructions, then the algorithm in ADCE is much faster. This PR removes the calls to DCEInst in the load-store removal passes and adds a pass of ADCE afterwards. A number of different iterations of the order of optimization, and I believe this is the best I could find. The results I have on 3 sets of shaders are: Legalization: Set 1: 5.39 -> 5.01 Set 2: 13.98 -> 8.38 Set 3: 98.00 -> 96.26 Performance passes: Set 1: 6.90 -> 5.23 Set 2: 10.11 -> 6.62 Set 3: 253.69 -> 253.74 Size reduction passes: Set 1: 7.16 -> 7.25 Set 2: 17.17 -> 16.81 Set 3: 112.06 -> 107.71 Note that the third set's compile time is large because of the large number of basic blocks, not so much because of the number of instructions. That is why we don't see much gain there.
212 lines
7.1 KiB
C++
212 lines
7.1 KiB
C++
// 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 "local_single_block_elim_pass.h"
|
|
|
|
#include "iterator.h"
|
|
|
|
namespace spvtools {
|
|
namespace opt {
|
|
|
|
namespace {
|
|
|
|
const uint32_t kStoreValIdInIdx = 1;
|
|
|
|
} // anonymous namespace
|
|
|
|
bool LocalSingleBlockLoadStoreElimPass::HasOnlySupportedRefs(uint32_t ptrId) {
|
|
if (supported_ref_ptrs_.find(ptrId) != supported_ref_ptrs_.end()) return true;
|
|
if (get_def_use_mgr()->WhileEachUser(ptrId, [this](ir::Instruction* user) {
|
|
SpvOp op = user->opcode();
|
|
if (IsNonPtrAccessChain(op) || op == SpvOpCopyObject) {
|
|
if (!HasOnlySupportedRefs(user->result_id())) {
|
|
return false;
|
|
}
|
|
} else if (op != SpvOpStore && op != SpvOpLoad && op != SpvOpName &&
|
|
!IsNonTypeDecorate(op)) {
|
|
return false;
|
|
}
|
|
return true;
|
|
})) {
|
|
supported_ref_ptrs_.insert(ptrId);
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
bool LocalSingleBlockLoadStoreElimPass::LocalSingleBlockLoadStoreElim(
|
|
ir::Function* func) {
|
|
// Perform local store/load and load/load elimination on each block
|
|
bool modified = false;
|
|
for (auto bi = func->begin(); bi != func->end(); ++bi) {
|
|
var2store_.clear();
|
|
var2load_.clear();
|
|
pinned_vars_.clear();
|
|
auto next = bi->begin();
|
|
for (auto ii = next; ii != bi->end(); ii = next) {
|
|
++next;
|
|
switch (ii->opcode()) {
|
|
case SpvOpStore: {
|
|
// Verify store variable is target type
|
|
uint32_t varId;
|
|
ir::Instruction* ptrInst = GetPtr(&*ii, &varId);
|
|
if (!IsTargetVar(varId)) continue;
|
|
if (!HasOnlySupportedRefs(varId)) continue;
|
|
// Register the store
|
|
if (ptrInst->opcode() == SpvOpVariable) {
|
|
// if not pinned, look for WAW
|
|
if (pinned_vars_.find(varId) == pinned_vars_.end()) {
|
|
auto si = var2store_.find(varId);
|
|
if (si != var2store_.end()) {
|
|
}
|
|
}
|
|
var2store_[varId] = &*ii;
|
|
} else {
|
|
assert(IsNonPtrAccessChain(ptrInst->opcode()));
|
|
var2store_.erase(varId);
|
|
}
|
|
pinned_vars_.erase(varId);
|
|
var2load_.erase(varId);
|
|
} break;
|
|
case SpvOpLoad: {
|
|
// Verify store variable is target type
|
|
uint32_t varId;
|
|
ir::Instruction* ptrInst = GetPtr(&*ii, &varId);
|
|
if (!IsTargetVar(varId)) continue;
|
|
if (!HasOnlySupportedRefs(varId)) continue;
|
|
// Look for previous store or load
|
|
uint32_t replId = 0;
|
|
if (ptrInst->opcode() == SpvOpVariable) {
|
|
auto si = var2store_.find(varId);
|
|
if (si != var2store_.end()) {
|
|
replId = si->second->GetSingleWordInOperand(kStoreValIdInIdx);
|
|
} else {
|
|
auto li = var2load_.find(varId);
|
|
if (li != var2load_.end()) {
|
|
replId = li->second->result_id();
|
|
}
|
|
}
|
|
}
|
|
if (replId != 0) {
|
|
// replace load's result id and delete load
|
|
context()->KillNamesAndDecorates(&*ii);
|
|
context()->ReplaceAllUsesWith(ii->result_id(), replId);
|
|
modified = true;
|
|
} else {
|
|
if (ptrInst->opcode() == SpvOpVariable)
|
|
var2load_[varId] = &*ii; // register load
|
|
pinned_vars_.insert(varId);
|
|
}
|
|
} break;
|
|
case SpvOpFunctionCall: {
|
|
// Conservatively assume all locals are redefined for now.
|
|
// TODO(): Handle more optimally
|
|
var2store_.clear();
|
|
var2load_.clear();
|
|
pinned_vars_.clear();
|
|
} break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
return modified;
|
|
}
|
|
|
|
void LocalSingleBlockLoadStoreElimPass::Initialize(ir::IRContext* c) {
|
|
InitializeProcessing(c);
|
|
|
|
// Initialize Target Type Caches
|
|
seen_target_vars_.clear();
|
|
seen_non_target_vars_.clear();
|
|
|
|
// Clear collections
|
|
supported_ref_ptrs_.clear();
|
|
|
|
// Initialize extensions whitelist
|
|
InitExtensions();
|
|
};
|
|
|
|
bool LocalSingleBlockLoadStoreElimPass::AllExtensionsSupported() const {
|
|
// If any extension not in whitelist, return false
|
|
for (auto& ei : get_module()->extensions()) {
|
|
const char* extName =
|
|
reinterpret_cast<const char*>(&ei.GetInOperand(0).words[0]);
|
|
if (extensions_whitelist_.find(extName) == extensions_whitelist_.end())
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
Pass::Status LocalSingleBlockLoadStoreElimPass::ProcessImpl() {
|
|
// Assumes relaxed logical addressing only (see instruction.h).
|
|
if (context()->get_feature_mgr()->HasCapability(SpvCapabilityAddresses))
|
|
return Status::SuccessWithoutChange;
|
|
// Do not process if module contains OpGroupDecorate. Additional
|
|
// support required in KillNamesAndDecorates().
|
|
// TODO(greg-lunarg): Add support for OpGroupDecorate
|
|
for (auto& ai : get_module()->annotations())
|
|
if (ai.opcode() == SpvOpGroupDecorate) return Status::SuccessWithoutChange;
|
|
// If any extensions in the module are not explicitly supported,
|
|
// return unmodified.
|
|
if (!AllExtensionsSupported()) return Status::SuccessWithoutChange;
|
|
// Process all entry point functions
|
|
ProcessFunction pfn = [this](ir::Function* fp) {
|
|
return LocalSingleBlockLoadStoreElim(fp);
|
|
};
|
|
bool modified = ProcessEntryPointCallTree(pfn, get_module());
|
|
return modified ? Status::SuccessWithChange : Status::SuccessWithoutChange;
|
|
}
|
|
|
|
LocalSingleBlockLoadStoreElimPass::LocalSingleBlockLoadStoreElimPass() {}
|
|
|
|
Pass::Status LocalSingleBlockLoadStoreElimPass::Process(ir::IRContext* c) {
|
|
Initialize(c);
|
|
return ProcessImpl();
|
|
}
|
|
|
|
void LocalSingleBlockLoadStoreElimPass::InitExtensions() {
|
|
extensions_whitelist_.clear();
|
|
extensions_whitelist_.insert({
|
|
"SPV_AMD_shader_explicit_vertex_parameter",
|
|
"SPV_AMD_shader_trinary_minmax",
|
|
"SPV_AMD_gcn_shader",
|
|
"SPV_KHR_shader_ballot",
|
|
"SPV_AMD_shader_ballot",
|
|
"SPV_AMD_gpu_shader_half_float",
|
|
"SPV_KHR_shader_draw_parameters",
|
|
"SPV_KHR_subgroup_vote",
|
|
"SPV_KHR_16bit_storage",
|
|
"SPV_KHR_device_group",
|
|
"SPV_KHR_multiview",
|
|
"SPV_NVX_multiview_per_view_attributes",
|
|
"SPV_NV_viewport_array2",
|
|
"SPV_NV_stereo_view_rendering",
|
|
"SPV_NV_sample_mask_override_coverage",
|
|
"SPV_NV_geometry_shader_passthrough",
|
|
"SPV_AMD_texture_gather_bias_lod",
|
|
"SPV_KHR_storage_buffer_storage_class",
|
|
// SPV_KHR_variable_pointers
|
|
// Currently do not support extended pointer expressions
|
|
"SPV_AMD_gpu_shader_int16",
|
|
"SPV_KHR_post_depth_coverage",
|
|
"SPV_KHR_shader_atomic_counter_ops",
|
|
});
|
|
}
|
|
|
|
} // namespace opt
|
|
} // namespace spvtools
|