mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
ac146c4820
Summary: 1. Added CudaFusionGuard as the custom TypeCheck for nvfuser; enabled dynamic shape support with profiling executor; 2. dropped support for legacy fuser; 3. re-enabled nvfuser tests; 4. added registration for profiling record to allow profiling on user specified nodes. Pull Request resolved: https://github.com/pytorch/pytorch/pull/46452 Reviewed By: zou3519, anjali411 Differential Revision: D24364642 Pulled By: ngimel fbshipit-source-id: daf53a9a6b6636e1ede420a3a6d0397d4a8b450b
45 lines
1.3 KiB
C++
45 lines
1.3 KiB
C++
#pragma once
|
|
|
|
#include <torch/csrc/WindowsTorchApiMacro.h>
|
|
#include <torch/csrc/jit/ir/ir.h>
|
|
|
|
/*
|
|
* This file contains APIs for cuda fuser;
|
|
*
|
|
* We use an empty static struct to hold the function pointers, which are
|
|
* registered separately. This is to support cpu-only compilation.
|
|
* Registration is done in torch/csrc/jit/codegen/cuda/register_interface.cpp
|
|
*/
|
|
|
|
namespace torch {
|
|
namespace jit {
|
|
namespace fuser {
|
|
namespace cuda {
|
|
|
|
TORCH_API std::atomic<bool>& getCudaFusionGuardMode();
|
|
|
|
// dummy struct to allow API registration
|
|
struct CudaFuserInterface {
|
|
void (*fn_compile_n_)(Node*) = nullptr;
|
|
void (*fn_run_n_s_)(const Node*, Stack&) = nullptr;
|
|
void (*fn_fuse_graph)(std::shared_ptr<Graph>&) = nullptr;
|
|
bool (*fn_can_fuse_n_)(const Node*) = nullptr;
|
|
};
|
|
|
|
// Get interface, this is used by registration and user facing API internally
|
|
C10_EXPORT CudaFuserInterface* getFuserInterface();
|
|
|
|
C10_EXPORT void compileFusionGroup(Node* fusion_node);
|
|
C10_EXPORT void runFusionGroup(const Node* fusion_node, Stack& stack);
|
|
C10_EXPORT void fuseGraph(std::shared_ptr<Graph>&);
|
|
C10_EXPORT bool canFuseNode(const Node* node);
|
|
|
|
C10_EXPORT bool complyWith(
|
|
const at::Tensor& tensor,
|
|
const c10::TensorTypePtr& guard_tensor_type);
|
|
|
|
} // namespace cuda
|
|
} // namespace fuser
|
|
} // namespace jit
|
|
} // namespace torch
|