1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
|
#pragma once
#include <torch/csrc/Export.h>
#include <torch/csrc/jit/codegen/fuser/arg_spec.h>
#include <torch/csrc/jit/codegen/fuser/partition_desc.h>
#include <torch/csrc/jit/codegen/fuser/tensor_desc.h>
#include <torch/csrc/jit/ir/ir.h>
#include <string>
#include <vector>
namespace torch::jit::fuser {
// Creates a CPU or CUDA kernel for the given graph.
// Returns the C++ or CUDA string implementing the kernel.
TORCH_API std::string generateKernel(
const std::string& name,
const Graph& graph,
const std::vector<std::pair<const Value*, const std::optional<TensorDesc>>>&
inputs,
const std::vector<std::pair<const Value*, const TensorDesc>>& outputs,
const bool use_cuda);
} // namespace torch::jit::fuser
|
