wehub-resource-sync
74 lines
2.4 KiB
C++
74 lines
2.4 KiB
C++
// Copyright (c) 2023 CINN Authors. All Rights Reserved.
|
|
//
|
|
// 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.
|
|
|
|
#pragma once
|
|
#include <string>
|
|
|
|
#include "paddle/cinn/ir/ir.h"
|
|
#include "paddle/cinn/pass/pass.h"
|
|
|
|
namespace cinn {
|
|
namespace optim {
|
|
|
|
/**
|
|
* UpdateBufferAxisPass optimizes buffer access by formalizing indices and
|
|
* replacing redundant accesses with zero.
|
|
*
|
|
* This pass is used in `OptimizeExprGpu` and is applicable in scenarios
|
|
* where buffer accesses in shared or local GPU memory have consistent index
|
|
* expressions across the same axis. In such cases, the pass analyzes the Expr
|
|
* AST to determine if these consistent indices imply that less memory is
|
|
* needed. By setting these redundant indices to zero, the pass can help
|
|
* minimize memory usage.
|
|
*
|
|
* When applied, this pass analyzes buffer access patterns and identifies
|
|
* indices that are consistently accessed with the same expression across the
|
|
* same axis in shared or local GPU memory. It then replaces these indices with
|
|
* zero, which can lead to reduced memory allocation requirements and
|
|
* streamlined memory usage.
|
|
*
|
|
* Performance impact: This pass addresses memory optimization in GPU
|
|
* environments by potentially reducing memory allocation and improving access
|
|
* efficiency, which can enhance overall performance.
|
|
*
|
|
* Examples:
|
|
* 1. Consistent Index Access in GPU Shared Memory:
|
|
* Input IR:
|
|
* `A[i * 3][j] = ...`
|
|
* `... = A[k][j]`
|
|
* Output IR:
|
|
* `A[i * 3][0] = ...`
|
|
* `... = A[k][0]`
|
|
*
|
|
* 2. Single Dimension Access Simplified:
|
|
* Input IR:
|
|
* B[i * n + j] = ...
|
|
* ... = B[k * n + j]
|
|
* Output IR:
|
|
* B[i * n + 0] = ...
|
|
* ... = B[k * n + 0]
|
|
*/
|
|
|
|
class UpdateBufferAxisPass : public BlockPass {
|
|
public:
|
|
UpdateBufferAxisPass() : BlockPass("update_buffer_axis_pass") {}
|
|
|
|
LogicalResult Run(ir::stmt::BlockRef) override;
|
|
};
|
|
|
|
std::unique_ptr<BlockPass> CreateUpdateBufferAxisPass();
|
|
|
|
} // namespace optim
|
|
} // namespace cinn
|