Skip to content

N
ngraph
Commit 59df2998 authored by Fenglei Tian's avatar Fenglei Tian
Browse files
Options

fix code style, copyright, and clean unimplement functions

parent e08ee468
Hide whitespace changes
Inline Side-by-side
Showing with 55 additions and 174 deletions
src/ngraph/runtime/gpu/gpu_call_frame.cpp
View file @ 59df2998
......@@ -34,7 +34,7 @@ runtime::gpu::GPU_CallFrame::GPU_CallFrame(std::shared_ptr<GPU_ExternalFunction>
{
//Create context use driver API and make it current, the runtime call will pickup the context
//http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#interoperability-between-runtime-and-driver-apis
ngraph::runtime::gpu::Cuda_context_manager::Instance();
ngraph::runtime::gpu::CudaContextManager::Instance();
cublasStatus_t cublasStatus = cublasCreate(&m_cublas_handle);
if (cublasStatus != CUBLAS_STATUS_SUCCESS)
{
......@@ -46,7 +46,7 @@ runtime::gpu::GPU_CallFrame::GPU_CallFrame(std::shared_ptr<GPU_ExternalFunction>
throw runtime_error("cuDnn create handle failed");
}
// Pass scalars as reference on the Host
// Pass scalars as reference on the Device
cublasSetPointerMode(m_cublas_handle, CUBLAS_POINTER_MODE_HOST);
}
......@@ -60,7 +60,7 @@ void runtime::gpu::GPU_CallFrame::tensor_call(
const std::vector<std::shared_ptr<ngraph::runtime::TensorView>>& input_tvs,
const std::vector<std::shared_ptr<ngraph::runtime::TensorView>>& output_tvs)
{
// Host tensors
//Device tensors
vector<void*> inputs;
vector<void*> outputs;
......
src/ngraph/runtime/gpu/gpu_cuda_context_manager.hpp
View file @ 59df2998
......@@ -27,33 +27,33 @@ namespace ngraph
{
namespace gpu
{
class Cuda_context_manager
class CudaContextManager
{
public:
static Cuda_context_manager& Instance()
static CudaContextManager& Instance()
{
static Cuda_context_manager manager;
static CudaContextManager manager;
return manager;
}
Cuda_context_manager(Cuda_context_manager const&) = delete;
Cuda_context_manager(Cuda_context_manager&&) = delete;
Cuda_context_manager& operator=(Cuda_context_manager const&) = delete;
Cuda_context_manager& operator=(Cuda_context_manager&&) = delete;
CudaContextManager(CudaContextManager const&) = delete;
CudaContextManager(CudaContextManager&&) = delete;
CudaContextManager& operator=(CudaContextManager const&) = delete;
CudaContextManager& operator=(CudaContextManager&&) = delete;
std::shared_ptr<CUcontext> GetContext() { return context_ptr; }
protected:
Cuda_context_manager()
CudaContextManager()
{
CUDA_SAFE_CALL(cuInit(0));
CUDA_SAFE_CALL(cuDeviceGet(&cuDevice, 0));
CUDA_SAFE_CALL(cuCtxCreate(&context, 0, cuDevice));
context_ptr = std::make_shared<CUcontext>(context);
CUDA_SAFE_CALL(cuDeviceGet(&m_device, 0));
CUDA_SAFE_CALL(cuCtxCreate(&m_context, 0, m_device));
m_context_ptr = std::make_shared<CUcontext>(m_context);
}
~Cuda_context_manager() {}
CUdevice cuDevice;
CUcontext context;
std::shared_ptr<CUcontext> context_ptr;
~CudaContextManager() {}
CUdevice m_device;
CUcontext m_context;
std::shared_ptr<CUcontext> m_context_ptr;
};
}
}
......
src/ngraph/runtime/gpu/gpu_cuda_function_builder.hpp
View file @ 59df2998
......@@ -27,7 +27,7 @@ namespace ngraph
{
namespace gpu
{
class Cuda_function_builder
class CudaFunctionBuilder
{
public:
static std::shared_ptr<CUfunction> Get(const std::string& name,
......@@ -43,23 +43,17 @@ namespace ngraph
NULL, // headers
NULL)); // includeNames
nvrtcResult compileResult =
nvrtcResult compile_result =
nvrtcCompileProgram(prog, number_of_options, options);
if (compileResult != NVRTC_SUCCESS)
if (compile_result != NVRTC_SUCCESS)
{
// size_t logSize;
// NVRTC_SAFE_CALL(nvrtcGetProgramLogSize(prog, &logSize));
// char *log = new char[logSize];
// NVRTC_SAFE_CALL(nvrtcGetProgramLog(prog, log));
// std::cout << log << '\n';
// delete[] log;
throw std::runtime_error("compile error: \n" + kernel + "\n options");
}
size_t ptxSize;
NVRTC_SAFE_CALL(nvrtcGetPTXSize(prog, &ptxSize));
char* ptx = new char[ptxSize];
size_t ptx_size;
NVRTC_SAFE_CALL(nvrtcGetPTXSize(prog, &ptx_size));
char* ptx = new char[ptx_size];
NVRTC_SAFE_CALL(nvrtcGetPTX(
prog,
ptx)); // Load the generated PTX and get a handle to the parent kernel.
......
src/ngraph/runtime/gpu/gpu_cuda_function_pool.hpp
View file @ 59df2998
......@@ -27,29 +27,29 @@ namespace ngraph
{
namespace gpu
{
class Cuda_function_pool
class CudaFunctionPool
{
public:
static Cuda_function_pool& Instance()
static CudaFunctionPool& Instance()
{
static Cuda_function_pool pool;
static CudaFunctionPool pool;
return pool;
}
Cuda_function_pool(Cuda_function_pool const&) = delete;
Cuda_function_pool(Cuda_function_pool&&) = delete;
Cuda_function_pool& operator=(Cuda_function_pool const&) = delete;
Cuda_function_pool& operator=(Cuda_function_pool&&) = delete;
CudaFunctionPool(CudaFunctionPool const&) = delete;
CudaFunctionPool(CudaFunctionPool&&) = delete;
CudaFunctionPool& operator=(CudaFunctionPool const&) = delete;
CudaFunctionPool& operator=(CudaFunctionPool&&) = delete;
void Set(std::string& name, std::shared_ptr<CUfunction> function)
{
CUfunction_map.insert({name, function});
m_function_map.insert({name, function});
}
std::shared_ptr<CUfunction> Get(std::string& name)
{
auto it = CUfunction_map.find(name);
if (it != CUfunction_map.end())
auto it = m_function_map.find(name);
if (it != m_function_map.end())
{
return (*it).second;
}
......@@ -57,9 +57,9 @@ namespace ngraph
}
protected:
Cuda_function_pool() {}
~Cuda_function_pool() {}
std::unordered_map<std::string, std::shared_ptr<CUfunction>> CUfunction_map;
CudaFunctionPool() {}
~CudaFunctionPool() {}
std::unordered_map<std::string, std::shared_ptr<CUfunction>> m_function_map;
};
}
}
......
src/ngraph/runtime/gpu/gpu_cuda_kernel_emitters.cpp
View file @ 59df2998
......@@ -43,7 +43,7 @@ namespace ngraph
std::string kernel;
Cuda_kernel_builder::Get_1_element_op(name, "float", "fabsf", kernel);
Cuda_function_pool::Instance().Set(
name, Cuda_function_builder::Get("cuda_" + name, kernel, 2, opts));
name, CudaFunctionBuilder::Get("cuda_" + name, kernel, 2, opts));
}
//convert runtime ptr to driver api ptr
......@@ -66,74 +66,6 @@ namespace ngraph
0)); // arguments
CUDA_SAFE_CALL(cuCtxSynchronize()); // Retrieve and print output.
}
void emit_broadcast(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisSet& broadcast_axes)
{
}
//
// For the reference kernel this is gpud on, see ngraph/runtime/kernel/concat.hpp.
//
void emit_concat(codegen::CodeWriter& writer,
const std::string& element_type,
const std::vector<std::string>& args,
const std::string& out,
const std::vector<Shape>& in_shapes,
const Shape& out_shape,
size_t concatenation_axis)
{
}
void emit_replace_slice(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& arg1, // replacement value
const std::string& out,
const Shape& arg1_shape,
const Shape& out_shape,
const Coordinate& lower_bounds,
const Coordinate& upper_bounds,
const Strides& strides)
{
}
void emit_slice(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const Coordinate& lower_bounds,
const Coordinate& upper_bounds,
const Strides& strides)
{
}
void emit_reshape(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisVector& arg0_axis_order)
{
}
void emit_sum(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisSet& reduction_axes)
{
}
}
}
}
......
src/ngraph/runtime/gpu/gpu_cuda_kernel_emitters.hpp
View file @ 59df2998
// ----------------------------------------------------------------------------
// Copyright 2017 Nervana Systems Inc.
// 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
// ----------------------------------------------------------------------------
/*******************************************************************************
* Copyright 2017-2018 Intel Corporation
*
* 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
......@@ -28,55 +30,6 @@ namespace ngraph
namespace kernel
{
void emit_abs(void* in, void* out, size_t count);
void emit_broadcast(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisSet& broadcast_axes);
void emit_concat(codegen::CodeWriter& writer,
const std::string& element_type,
const std::vector<std::string>& args,
const std::string& out,
const std::vector<Shape>& in_shapes,
const Shape& out_shape,
const size_t concatenation_axis);
void emit_replace_slice(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& arg1, // replacement value
const std::string& out,
const Shape& arg1_shape,
const Shape& out_shape,
const Coordinate& lower_bounds,
const Coordinate& upper_bounds,
const Strides& strides);
void emit_slice(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const Coordinate& lower_bounds,
const Coordinate& upper_bounds,
const Strides& strides);
void emit_reshape(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisVector& arg0_axis_order);
void emit_sum(codegen::CodeWriter& writer,
const std::string& element_type,
const std::string& arg0, // replacement context
const std::string& out,
const Shape& arg0_shape,
const Shape& out_shape,
const AxisSet& reduction_axes);
}
}
}
......
src/ngraph/runtime/gpu/gpu_util.hpp
View file @ 59df2998
......@@ -22,6 +22,8 @@
#include <cudnn_v7.h>
#include <nvrtc.h>
//why use "do...while.."
//https://stackoverflow.com/questions/154136/why-use-apparently-meaningless-do-while-and-if-else-statements-in-macros
#define NVRTC_SAFE_CALL(x) \
do \
{ \
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment