mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-06 12:20:52 +01:00
533cb9530e
Summary: Sub-step of my attempt to split up the torch_cuda library, as it is huge. Please look at https://github.com/pytorch/pytorch/issues/49050 for details on the split and which files are in which target. This PR introduces two new macros for Windows DLL purposes, TORCH_CUDA_CPP_API and TORCH_CUDA_CU_API. Both are defined as TORCH_CUDA_API for the time being. Pull Request resolved: https://github.com/pytorch/pytorch/pull/50627 Reviewed By: mruberry Differential Revision: D25955441 Pulled By: janeyx99 fbshipit-source-id: ff226026833b8fb2fb7c77df6f2d6c824f006869
48 lines
1.4 KiB
C++
48 lines
1.4 KiB
C++
#pragma once
|
|
|
|
#include <torch/csrc/WindowsTorchApiMacro.h>
|
|
#include <torch/csrc/jit/ir/ir.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 isReductionNode(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);
|
|
|
|
// 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
|