mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
127c9402d0
Summary:
This reverts commit 03389dc851.
Attempt again for PR: https://github.com/pytorch/pytorch/issues/63745
Fixes the windows build failure.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65137
Reviewed By: seemethere, dzhulgakov, heitorschueroff
Differential Revision: D30994556
Pulled By: malfet
fbshipit-source-id: f1925b6c5cc1a1a441a96499667c91e8dfc1b53d
56 lines
1.7 KiB
C++
56 lines
1.7 KiB
C++
#pragma once
|
|
|
|
#include <torch/csrc/WindowsTorchApiMacro.h>
|
|
#include <torch/csrc/jit/ir/ir.h>
|
|
#include <torch/csrc/jit/runtime/profiling_record.h>
|
|
|
|
#include <torch/csrc/jit/codegen/cuda/fusion.h>
|
|
|
|
/*
|
|
* This file handles Parsing PyTorch jit ir;
|
|
*
|
|
* It is used in two places:
|
|
* 1. When partitioning PyTorch jit ir to create prim::CudaFusionGroup, each
|
|
* node is queried by `isNodeParsible` to determine whether the node could
|
|
* be handled by the fuser (whether a given PyTorch jit operator should be
|
|
* merged);
|
|
* 2. lowering PyTorch jit ir to CUDA codegen ir.
|
|
* creates a `Fusion` by traversing a PyTorch jit graph.
|
|
*
|
|
* TODO: we could consider exposing API to allow custom registration of parsing
|
|
* rules for a given PyTorch jit operator.
|
|
*/
|
|
|
|
namespace torch {
|
|
namespace jit {
|
|
namespace fuser {
|
|
namespace cuda {
|
|
|
|
constexpr int kPwThreadX = 128;
|
|
constexpr int kFcdReductionThreadX = 128;
|
|
constexpr int kNonFcdReductionThreadX = 32;
|
|
constexpr int kNonFcdReductionThreadY = 32;
|
|
|
|
TORCH_CUDA_CU_API bool hasReductionNode(const Block* block);
|
|
TORCH_CUDA_CU_API bool isReductionToSizeNode(const Node* node);
|
|
TORCH_CUDA_CU_API bool isReductionNode(const Node* node);
|
|
|
|
TORCH_CUDA_CU_API bool hasNormalizationNode(const Block* block);
|
|
TORCH_CUDA_CU_API bool isNormalizationNode(const Node* node);
|
|
|
|
TORCH_CUDA_CU_API bool isElementWiseNode(const Node* node);
|
|
|
|
// returns whether or not a parsing function exists for the given node type.
|
|
TORCH_CUDA_CU_API bool isNodeParsible(const Node* node);
|
|
|
|
void InsertProfileNodes(ProfilingRecord* pr);
|
|
|
|
// lowers PyTorch jit graph to `Fusion`.
|
|
TORCH_CUDA_CU_API std::unique_ptr<Fusion> parseJitIR(
|
|
const std::shared_ptr<Graph>& graph);
|
|
|
|
} // namespace cuda
|
|
} // namespace fuser
|
|
} // namespace jit
|
|
} // namespace torch
|