Star721
270
代码
代码
Issues71
Pull Requests164
Actions
项目讨论
Wiki
分析
Star721
270
已合并
【质量加固】显式声明部分 lambda 捕获列表 #1756
小白学习者创建于 8月17日
【质量加固】显式声明部分 lambda 捕获列表 #1756
已合并
小白学习者创建于 8月17日
从wzx_0826/graph-autofusionquality-hardening-autofuse合入到cann/graph-autofusiondevelop
共 5 个文件变更+32-27
Mautofuse/att/gen_model_info/expr_gen/arg_list_reorder.cpp+9-7
@@ -249,9 +249,10 @@ bool ArgListReorder::IsReduceInputTensor(const TensorPtr &tensor, const SubAxis
249 if ((tensor == nullptr) || (tensor->data_type_size == 0U)) {249 if ((tensor == nullptr) || (tensor->data_type_size == 0U)) {
250 return false;250 return false;
251 }251 }
252- return std::any_of(tensor->dim_info.begin(), tensor->dim_info.end(), [&](const SubAxis *dim) {252+ return std::any_of(tensor->dim_info.begin(), tensor->dim_info.end(),
253- return IsSameOrRelatedAxis(dim, reduce_axis) || IsReduceOrigAxis(dim, reduce_axis_ori_axes_set);253+ [this, reduce_axis, &reduce_axis_ori_axes_set](const SubAxis *dim) {
254- });254+ return IsSameOrRelatedAxis(dim, reduce_axis) || IsReduceOrigAxis(dim, reduce_axis_ori_axes_set);
255+ });
255}256}
256 257 
257bool ArgListReorder::CollectReduceTailAxis(const TensorPtr &tensor, const SubAxis *reduce_axis,258bool ArgListReorder::CollectReduceTailAxis(const TensorPtr &tensor, const SubAxis *reduce_axis,
@@ -282,9 +283,10 @@ bool ArgListReorder::IsReduceTailTileAxis(const SubAxis *tail_axis) const {
282 if (tail_axis == nullptr) {283 if (tail_axis == nullptr) {
283 return false;284 return false;
284 }285 }
285- return std::any_of(tuning_space_->sub_axes.begin(), tuning_space_->sub_axes.end(), [&](const SubAxisPtr &axis) {286+ return std::any_of(
286- return (axis->name == tail_axis->name) && (axis->axis_type == AxisPosition::INNER) && !axis->is_bind_multi_core;287+ tuning_space_->sub_axes.begin(), tuning_space_->sub_axes.end(), [&tail_axis](const SubAxisPtr &axis) {
287- });288+ return (axis->name == tail_axis->name) && (axis->axis_type == AxisPosition::INNER) && !axis->is_bind_multi_core;
289+ });
288}290}
289 291 
290bool ArgListReorder::TryGetReduceTailTileInfo(const NodeInfo &node,292bool ArgListReorder::TryGetReduceTailTileInfo(const NodeInfo &node,
@@ -684,7 +686,7 @@ void ArgListReorder::MakeSureLoadStoreInnerestSameOrder(const std::vector<AttAxi
684 }686 }
685 687 
686 const bool overlaps_load_store = std::any_of(688 const bool overlaps_load_store = std::any_of(
687- equal_order_reduce_tail_axes_.begin(), equal_order_reduce_tail_axes_.end(), [&](const std::string &name) {689+ equal_order_reduce_tail_axes_.begin(), equal_order_reduce_tail_axes_.end(), [this](const std::string &name) {
688 return load_store_inner_most_dims_.find(name) != load_store_inner_most_dims_.end();690 return load_store_inner_most_dims_.find(name) != load_store_inner_most_dims_.end();
689 });691 });
690 if (overlaps_load_store) {692 if (overlaps_load_store) {
Mautofuse/att/generator/tiling_code_gen_impl.cpp+8-6
@@ -3847,11 +3847,13 @@ void TilingCodeGenImpl::GenPGOByCoreNumFunctionHead(size_t impl_graph_id) {
3847 3847 
3848bool TilingCodeGenImpl::TryGenPGOByCoreNumReuseTiling(size_t asc_graph_id, size_t impl_graph_id, size_t group_id,3848bool TilingCodeGenImpl::TryGenPGOByCoreNumReuseTiling(size_t asc_graph_id, size_t impl_graph_id, size_t group_id,
3849 uint32_t group_index) {3849 uint32_t group_index) {
3850- const auto iter = std::find_if(tiling_model_info_.cbegin(), tiling_model_info_.cend(), [&](const auto &model_info) {3850+ const auto iter = std::find_if(tiling_model_info_.cbegin(), tiling_model_info_.cend(),
3851- const auto &ident = model_info.schedule_group_ident;3851+ [asc_graph_id, impl_graph_id, group_id](const auto &model_info) {
3852- return ident.asc_graph_id == asc_graph_id && ident.impl_graph_id == impl_graph_id && ident.group_id == group_id &&3852+ const auto &ident = model_info.schedule_group_ident;
3853- model_info.reuse_schedule_group != nullptr && model_info.reuse_schedule_group->IsReuseGroup(ident);3853+ return ident.asc_graph_id == asc_graph_id && ident.impl_graph_id == impl_graph_id &&
3854- });3854+ ident.group_id == group_id && model_info.reuse_schedule_group != nullptr &&
3855+ model_info.reuse_schedule_group->IsReuseGroup(ident);
3856+ });
3855 if (iter == tiling_model_info_.cend()) {3857 if (iter == tiling_model_info_.cend()) {
3856 return false;3858 return false;
3857 }3859 }
@@ -5175,7 +5177,7 @@ std::pair<std::string, bool> TilingCodeGenImpl::GenConflictExprContextCode(
5175 auto input_vars = GetVarsNames(args_manager.GetInputVars());5177 auto input_vars = GetVarsNames(args_manager.GetInputVars());
5176 input_var_names.insert(input_vars.begin(), input_vars.end());5178 input_var_names.insert(input_vars.begin(), input_vars.end());
5177 }5179 }
5178- auto emit_decl = [&](const std::string &name, const std::string &src) {5180+ auto emit_decl = [&code, &declared_symbols](const std::string &name, const std::string &src) {
5179 code += " auto " + name + " = " + src + ".get_" + name + "();\n";5181 code += " auto " + name + " = " + src + ".get_" + name + "();\n";
5180 declared_symbols.insert(name);5182 declared_symbols.insert(name);
5181 };5183 };
Mautofuse/codegen/codegen.cpp+1-1
@@ -396,7 +396,7 @@ Status Codegen::GenerateForInductor(const ascir::FusedScheduledResult &fused_sch
396 fallback.tiling_lib_.DisableInductorPgo();396 fallback.tiling_lib_.DisableInductorPgo();
397 return fallback.GenerateForInductor(fused_schedule_result, result);397 return fallback.GenerateForInductor(fused_schedule_result, result);
398 }398 }
399- const auto generate_tiling_without_pgo = [&]() {399+ const auto generate_tiling_without_pgo = [this, &fused_schedule_result, &result]() {
400 if (!tiling_lib_.IsInductorPgoEnabled() || ascgen_utils::IsCubeFusedScheduled(fused_schedule_result) ||400 if (!tiling_lib_.IsInductorPgoEnabled() || ascgen_utils::IsCubeFusedScheduled(fused_schedule_result) ||
401 !ascgen_utils::IsStaticSchedResult(fused_schedule_result)) {401 !ascgen_utils::IsStaticSchedResult(fused_schedule_result)) {
402 return af::FAILED;402 return af::FAILED;
Mautofuse/optimize/static_ub_template_filter.cpp+12-11
@@ -304,16 +304,17 @@ bool FilterScheduledResult(FilterState &state, ascir::ScheduledResult &scheduled
304 const size_t before_group_size = schedule_group.impl_graphs.size();304 const size_t before_group_size = schedule_group.impl_graphs.size();
305 size_t impl_idx = 0UL;305 size_t impl_idx = 0UL;
306 schedule_group.impl_graphs.erase(306 schedule_group.impl_graphs.erase(
307- std::remove_if(schedule_group.impl_graphs.begin(), schedule_group.impl_graphs.end(),307+ std::remove_if(
308- [&](const af::AscGraph &impl_graph) {308+ schedule_group.impl_graphs.begin(), schedule_group.impl_graphs.end(),
309- const size_t current_impl_idx = impl_idx++;309+ [&schedule_group, &impl_idx, &state, node_idx, result_idx, group_idx](const af::AscGraph &impl_graph) {
310- const TemplatePosition position = {node_idx, result_idx, group_idx, current_impl_idx};310+ const size_t current_impl_idx = impl_idx++;
311- const bool drop = ShouldDropImplGraph(impl_graph, state.ub_size, position);311+ const TemplatePosition position = {node_idx, result_idx, group_idx, current_impl_idx};
312- if (drop) {312+ const bool drop = ShouldDropImplGraph(impl_graph, state.ub_size, position);
313- schedule_group.graph_name_to_score_funcs.erase(impl_graph.GetName());313+ if (drop) {
314- }314+ schedule_group.graph_name_to_score_funcs.erase(impl_graph.GetName());
315- return drop;315+ }
316- }),316+ return drop;
317+ }),
317 schedule_group.impl_graphs.end());318 schedule_group.impl_graphs.end());
318 if (before_group_size > 0UL && schedule_group.impl_graphs.empty()) {319 if (before_group_size > 0UL && schedule_group.impl_graphs.empty()) {
319 GELOGD(320 GELOGD(
@@ -333,7 +334,7 @@ af::Status FilterNodeScheduledResults(ascir::FusedScheduledResult &fused_schedul
333 FilterState state = {fused_scheduled_result, ub_size};334 FilterState state = {fused_scheduled_result, ub_size};
334 size_t result_idx = 0UL;335 size_t result_idx = 0UL;
335 scheduled_results.erase(std::remove_if(scheduled_results.begin(), scheduled_results.end(),336 scheduled_results.erase(std::remove_if(scheduled_results.begin(), scheduled_results.end(),
336- [&](ascir::ScheduledResult &scheduled_result) {337+ [&state, node_idx, &result_idx](ascir::ScheduledResult &scheduled_result) {
337 const size_t current_result_idx = result_idx++;338 const size_t current_result_idx = result_idx++;
338 return !FilterScheduledResult(state, scheduled_result, node_idx,339 return !FilterScheduledResult(state, scheduled_result, node_idx,
339 current_result_idx);340 current_result_idx);
Mautofuse/optimize/task_generator/recompute_case_generator.cpp+2-2
@@ -290,10 +290,10 @@ bool RecomputeCaseGenerator::IsRecomputableNode(ascir::HintGraph &hint_graph, af
290 auto output_tensors = node->outputs();290 auto output_tensors = node->outputs();
291 if (is_static_graph_) {291 if (is_static_graph_) {
292 return std::all_of(output_tensors.begin(), output_tensors.end(),292 return std::all_of(output_tensors.begin(), output_tensors.end(),
293- [&](af::AscTensor *&tensor) { return check_static_tensor(tensor->attr); });293+ [&check_static_tensor](af::AscTensor *&tensor) { return check_static_tensor(tensor->attr); });
294 }294 }
295 return std::all_of(output_tensors.begin(), output_tensors.end(),295 return std::all_of(output_tensors.begin(), output_tensors.end(),
296- [&](af::AscTensor *&tensor) { return check_dynamic_tensor(tensor->attr); });296+ [&check_dynamic_tensor](af::AscTensor *&tensor) { return check_dynamic_tensor(tensor->attr); });
297}297}
298 298 
299bool RecomputeCaseGenerator::IsRecomputableAlwaysBetter() const {299bool RecomputeCaseGenerator::IsRecomputableAlwaysBetter() const {