Commit 7a3b49e5 authored by Chao Liu's avatar Chao Liu
Browse files

Merge remote-tracking branch 'origin/develop' into contraction

parents e07b3d8e d3051d75
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
#include <memory>
#include <functional>
#include <thread>
#include <chrono>
#include <hip/hip_runtime.h> #include <hip/hip_runtime.h>
#include <hip/hip_fp16.h>
#include "stream_config.hpp"
#include "ck/options.hpp"
template <typename T>
__global__ void set_buffer_value(T* p, T x, uint64_t buffer_element_size)
{
for(uint64_t i = threadIdx.x; i < buffer_element_size; i += blockDim.x)
{
p[i] = x;
}
}
inline void hip_check_error(hipError_t x) #include "ck/ck.hpp"
{ #include "ck/stream_config.hpp"
if(x != hipSuccess) #include "ck/device_utility/hip_check_error.hpp"
{
std::ostringstream ss;
ss << "HIP runtime error: " << hipGetErrorString(x) << ". " << __FILE__ << ": " << __LINE__
<< "in function: " << __func__;
throw std::runtime_error(ss.str());
}
}
struct DeviceMem
{
DeviceMem() = delete;
DeviceMem(std::size_t mem_size);
void* GetDeviceBuffer();
std::size_t GetBufferSize();
void ToDevice(const void* p);
void FromDevice(void* p);
void SetZero();
template <typename T>
void SetValue(T x)
{
if(mMemSize % sizeof(T) != 0)
{
throw std::runtime_error("wrong! not entire DeviceMem will be set");
}
set_buffer_value<T><<<1, 1024>>>(static_cast<T*>(mpDeviceBuf), x, mMemSize / sizeof(T));
}
~DeviceMem();
void* mpDeviceBuf;
std::size_t mMemSize;
};
struct KernelTimerImpl;
struct KernelTimer
{
KernelTimer();
~KernelTimer();
void Start();
void End();
float GetElapsedTime() const;
std::unique_ptr<KernelTimerImpl> impl;
};
template <typename... Args, typename F> template <typename... Args, typename F>
float launch_and_time_kernel(const StreamConfig& stream_config, float launch_and_time_kernel(const StreamConfig& stream_config,
...@@ -97,17 +38,27 @@ float launch_and_time_kernel(const StreamConfig& stream_config, ...@@ -97,17 +38,27 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
printf("Start running %d times...\n", nrepeat); printf("Start running %d times...\n", nrepeat);
KernelTimer timer; hipEvent_t start, stop;
timer.Start();
hip_check_error(hipEventCreate(&start));
hip_check_error(hipEventCreate(&stop));
hip_check_error(hipDeviceSynchronize());
hip_check_error(hipEventRecord(start, stream_config.stream_id_));
for(int i = 0; i < nrepeat; ++i) for(int i = 0; i < nrepeat; ++i)
{ {
kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...); kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
} }
timer.End(); hip_check_error(hipEventRecord(stop, stream_config.stream_id_));
hip_check_error(hipEventSynchronize(stop));
float total_time = 0;
hip_check_error(hipEventElapsedTime(&total_time, start, stop));
return timer.GetElapsedTime() / nrepeat; return total_time / nrepeat;
} }
else else
{ {
......
#pragma once
#define CK_TIME_KERNEL 1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP #ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP #define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP #ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP #define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP #define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
#include <hip/hip_runtime.h> #include <hip/hip_runtime.h>
......
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_STATIC_TENSOR_HPP #ifndef CK_STATIC_TENSOR_HPP
#define CK_STATIC_TENSOR_HPP #define CK_STATIC_TENSOR_HPP
......
#ifndef CK_CLUSTER_DESCRIPTOR_HPP // SPDX-License-Identifier: MIT
#define CK_CLUSTER_DESCRIPTOR_HPP // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "common_header.hpp" #pragma once
#include "tensor_adaptor.hpp"
#include "ck/utility/common_header.hpp"
#include "ck/tensor_description/tensor_adaptor.hpp"
namespace ck { namespace ck {
...@@ -30,4 +32,3 @@ __host__ __device__ constexpr auto make_cluster_descriptor( ...@@ -30,4 +32,3 @@ __host__ __device__ constexpr auto make_cluster_descriptor(
} }
} // namespace ck } // namespace ck
#endif
#ifndef CK_MULTI_INDEX_TRANSFORM_HPP // SPDX-License-Identifier: MIT
#define CK_MULTI_INDEX_TRANSFORM_HPP // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "common_header.hpp" #pragma once
#include "multi_index.hpp"
#include "ck/utility/common_header.hpp"
#include "ck/utility/multi_index.hpp"
namespace ck { namespace ck {
...@@ -1950,4 +1952,3 @@ struct Modulo ...@@ -1950,4 +1952,3 @@ struct Modulo
} }
}; };
} // namespace ck } // namespace ck
#endif
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