mirror of
https://github.com/KhronosGroup/SPIRV-Tools
synced 2024-11-27 13:50:07 +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.
105 lines
3.7 KiB
C++
105 lines
3.7 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_ssa_elim_pass.h"
|
|
|
|
#include "cfa.h"
|
|
#include "iterator.h"
|
|
|
|
namespace spvtools {
|
|
namespace opt {
|
|
|
|
void LocalMultiStoreElimPass::Initialize(ir::IRContext* c) {
|
|
InitializeProcessing(c);
|
|
|
|
// Initialize extension whitelist
|
|
InitExtensions();
|
|
};
|
|
|
|
bool LocalMultiStoreElimPass::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 LocalMultiStoreElimPass::ProcessImpl() {
|
|
// Assumes all control flow structured.
|
|
// TODO(greg-lunarg): Do SSA rewrite for non-structured control flow
|
|
if (!context()->get_feature_mgr()->HasCapability(SpvCapabilityShader))
|
|
return Status::SuccessWithoutChange;
|
|
// Assumes relaxed logical addressing only (see instruction.h)
|
|
// TODO(greg-lunarg): Add support for physical addressing
|
|
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;
|
|
// Do not process if any disallowed extensions are enabled
|
|
if (!AllExtensionsSupported()) return Status::SuccessWithoutChange;
|
|
// Process functions
|
|
ProcessFunction pfn = [this](ir::Function* fp) {
|
|
return InsertPhiInstructions(fp);
|
|
};
|
|
bool modified = ProcessEntryPointCallTree(pfn, get_module());
|
|
return modified ? Status::SuccessWithChange : Status::SuccessWithoutChange;
|
|
}
|
|
|
|
LocalMultiStoreElimPass::LocalMultiStoreElimPass() {}
|
|
|
|
Pass::Status LocalMultiStoreElimPass::Process(ir::IRContext* c) {
|
|
Initialize(c);
|
|
return ProcessImpl();
|
|
}
|
|
|
|
void LocalMultiStoreElimPass::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
|