!12667 【GraphKernel】Combine the GraphKernelOptimization of Gpu and Ascend
From: @dayschan Reviewed-by: @lingyunli63 Signed-off-by:pull/12667/MERGE
mindspore-ci-bot
4 years ago
committed by
Gitee
commit
4e1931481a
@ -0,0 +1,171 @@
|
||||
/**
|
||||
* Copyright 2021 Huawei Technologies Co., Ltd
|
||||
*
|
||||
* 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 "backend/optimizer/graph_kernel/graph_kernel_optimization.h"
|
||||
|
||||
#include <vector>
|
||||
#include <string>
|
||||
#include <memory>
|
||||
|
||||
#include "ir/func_graph.h"
|
||||
#include "utils/ms_context.h"
|
||||
#include "backend/optimizer/graph_kernel/add_atomic_clean.h"
|
||||
#include "backend/optimizer/graph_kernel/add_atomic_clean_gpu.h"
|
||||
#include "backend/optimizer/graph_kernel/add_stitch_atomic_clean_gpu.h"
|
||||
#include "backend/optimizer/graph_kernel/arithmetic_simplify.h"
|
||||
#include "backend/optimizer/graph_kernel/basic_ops_fusion.h"
|
||||
#include "backend/optimizer/graph_kernel/clean_all_in_once.h"
|
||||
#include "backend/optimizer/graph_kernel/depend_formater.h"
|
||||
#include "backend/optimizer/graph_kernel/eliminate_redundant_output.h"
|
||||
#include "backend/optimizer/graph_kernel/tensor_promotion.h"
|
||||
#include "backend/optimizer/graph_kernel/graph_kernel_splitter.h"
|
||||
#include "backend/optimizer/graph_kernel/graph_kernel_expander.h"
|
||||
#include "backend/optimizer/graph_kernel/raise_reduction_precision.h"
|
||||
#include "backend/optimizer/graph_kernel/graph_kernel_cse.h"
|
||||
#include "backend/optimizer/graph_kernel/shape_ops_splitter.h"
|
||||
#include "backend/optimizer/graph_kernel/value_graph_binder.h"
|
||||
#include "backend/optimizer/graph_kernel/parallel_fusion.h"
|
||||
#include "backend/optimizer/graph_kernel/optimize_assign.h"
|
||||
#include "backend/optimizer/graph_kernel/split_assign.h"
|
||||
#include "backend/optimizer/graph_kernel/reorder_ops.h"
|
||||
#include "backend/optimizer/pass/getitem_tuple.h"
|
||||
|
||||
namespace mindspore {
|
||||
namespace opt {
|
||||
PassManagerPtr GraphKernelOptimizer::PreProcess() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage1_preprocess");
|
||||
// Change Assign(p, a, U) to Assign(Depend(p, U), a)
|
||||
pm->AddPass(std::make_shared<SplitAssign>());
|
||||
|
||||
// Move the Depend nodes to the bottom of graph
|
||||
pm->AddPass(std::make_shared<DependFormater>());
|
||||
|
||||
// Reorder TransData-Cast to Cast-TransData,
|
||||
if (is_ascend) {
|
||||
pm->AddPass(std::make_shared<ReorderOps>());
|
||||
}
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::Cluster() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage2_cluster");
|
||||
// Expand complex basic kernels to composite kernels
|
||||
pm->AddPass(std::make_shared<GraphKernelExpander>());
|
||||
|
||||
// Fuse basic kernels and composite kernels
|
||||
pm->AddPass(std::make_shared<BasicOpsFusion>());
|
||||
|
||||
// Eliminate the outputs without external user
|
||||
pm->AddPass(std::make_shared<EliminateRedundantOutput>());
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::HighLevelOpt1() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage3_highlevelopt1");
|
||||
// Replace Assign with InplaceAssign, and replace original output with overridden parameters
|
||||
pm->AddPass(std::make_shared<OptimizeAssign>());
|
||||
pm->AddPass(std::make_shared<EliminateRedundantOutput>());
|
||||
|
||||
// Cast the input of ReduceSum from float16 to float32 for higher precision*/
|
||||
pm->AddPass(std::make_shared<RaiseReductionPrecision>());
|
||||
|
||||
// Universal arithmetic simplify
|
||||
if (is_gpu) {
|
||||
pm->AddPass(std::make_shared<ArithmeticSimplify>());
|
||||
}
|
||||
|
||||
// Common subexpression elimination
|
||||
pm->AddPass(std::make_shared<GraphKernelCSE>());
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::Split() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage4_split");
|
||||
// Move the non-scalar tensor (in composite node) to parameter list
|
||||
pm->AddPass(std::make_shared<TensorPromotion>());
|
||||
|
||||
// Make certain nodes redundant so that they are used by only one user,
|
||||
// which can avoid unnecessary input-output and get better performance.
|
||||
if (is_gpu) {
|
||||
std::vector<PrimitivePtr> duplicated_ops = {prim::kPrimReshape, prim::kPrimExpandDims, prim::kPrimCast};
|
||||
pm->AddPass(std::make_shared<ShapeOpsSplitter>(duplicated_ops));
|
||||
}
|
||||
|
||||
// Split kernel according to costmodel
|
||||
pm->AddPass(std::make_shared<GraphKernelSplitter>());
|
||||
|
||||
// Eliminate the redundant node that is copied above but not handled by GraphKernelSplitter
|
||||
if (is_gpu) {
|
||||
pm->AddPass(std::make_shared<GraphKernelCSE>());
|
||||
pm->AddPass(std::make_shared<EliminateRedundantOutput>());
|
||||
}
|
||||
|
||||
// After Simplify and Splitter, a lot of redundant getitem/maketuple
|
||||
// will be exposed, use GetitemTuple Pass to delete them.
|
||||
pm->AddPass(std::make_shared<GetitemTuple>());
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::HighLevelOpt2() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage5_highlevelopt2");
|
||||
// Enable atomic add
|
||||
if (is_gpu) {
|
||||
pm->AddPass(std::make_shared<AtomicCleanInsertter>());
|
||||
pm->AddPass(std::make_shared<StitchAtomicCleanInsertter>());
|
||||
} else /* if (is_ascend) */ {
|
||||
pm->AddPass(std::make_shared<CleanAddAtomic>());
|
||||
}
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::Combine() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage6_combine");
|
||||
// Enable parallel fusion
|
||||
if (is_gpu) {
|
||||
// Prevent fake loop in parallel fusion
|
||||
pm->AddPass(std::make_shared<DependFormater>());
|
||||
// Do parallel fusion for gpu device
|
||||
pm->AddPass(std::make_shared<ParallelOpFusion>(kGPUDevice, ParallelConfig(7)));
|
||||
}
|
||||
return pm;
|
||||
}
|
||||
|
||||
PassManagerPtr GraphKernelOptimizer::PostProcess() {
|
||||
auto pm = std::make_shared<PassManager>("graphkernel_stage7_postprocess");
|
||||
// Add the new tensors to the kernel_graph
|
||||
pm->AddPass(std::make_shared<BindValueToGraph>());
|
||||
return pm;
|
||||
}
|
||||
|
||||
void GraphKernelOptimizer::Run(const KernelGraphPtr &kernel_graph) {
|
||||
auto context_ptr = MsContext::GetInstance();
|
||||
MS_EXCEPTION_IF_NULL(context_ptr);
|
||||
is_gpu = (context_ptr->get_param<std::string>(MS_CTX_DEVICE_TARGET) == kGPUDevice);
|
||||
is_ascend = (context_ptr->get_param<std::string>(MS_CTX_DEVICE_TARGET) == kAscendDevice);
|
||||
|
||||
auto optimizer = std::make_shared<GraphOptimizer>("graph_kernel_optimizer");
|
||||
optimizer->AddPassManager(PreProcess());
|
||||
optimizer->AddPassManager(Cluster());
|
||||
optimizer->AddPassManager(HighLevelOpt1());
|
||||
optimizer->AddPassManager(Split());
|
||||
optimizer->AddPassManager(HighLevelOpt2());
|
||||
optimizer->AddPassManager(Combine());
|
||||
optimizer->AddPassManager(PostProcess());
|
||||
(void)optimizer->Optimize(kernel_graph);
|
||||
}
|
||||
|
||||
void GraphKernelOptimize(const KernelGraphPtr &kernel_graph) { GraphKernelOptimizer().Run(kernel_graph); }
|
||||
} // namespace opt
|
||||
} // namespace mindspore
|
||||
@ -0,0 +1,54 @@
|
||||
/**
|
||||
* Copyright 2021 Huawei Technologies Co., Ltd
|
||||
*
|
||||
* 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.
|
||||
*/
|
||||
#ifndef MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_GRAPH_KERNEL_OPTIMIZATION_H_
|
||||
#define MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_GRAPH_KERNEL_OPTIMIZATION_H_
|
||||
|
||||
#include "ir/anf.h"
|
||||
#include "ir/func_graph.h"
|
||||
#include "backend/session/kernel_graph.h"
|
||||
#include "backend/optimizer/common/optimizer.h"
|
||||
#include "backend/optimizer/common/pass_manager.h"
|
||||
|
||||
namespace mindspore {
|
||||
namespace opt {
|
||||
class GraphKernelOptimizer {
|
||||
public:
|
||||
void Run(const KernelGraphPtr &kernel_graph);
|
||||
|
||||
private:
|
||||
// Pre-process
|
||||
PassManagerPtr PreProcess();
|
||||
// Cluster kernels
|
||||
PassManagerPtr Cluster();
|
||||
// High level optimize 1
|
||||
PassManagerPtr HighLevelOpt1();
|
||||
// Split kernels
|
||||
PassManagerPtr Split();
|
||||
// High level optimize 2
|
||||
PassManagerPtr HighLevelOpt2();
|
||||
// Combine kernels
|
||||
PassManagerPtr Combine();
|
||||
// Post-process
|
||||
PassManagerPtr PostProcess();
|
||||
|
||||
bool is_gpu{false};
|
||||
bool is_ascend{false};
|
||||
};
|
||||
|
||||
void GraphKernelOptimize(const KernelGraphPtr &kernel_graph);
|
||||
} // namespace opt
|
||||
} // namespace mindspore
|
||||
#endif // MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_GRAPH_KERNEL_OPTIMIZATION_H_
|
||||
Loading…
Reference in new issue