p83571920
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!21131 sponge ops 0730 

Merge pull request !21131 from jiahongQian/master1
This commit is contained in:
i-robot 2021-07-31 08:07:07 +00:00 committed by Gitee
parent dbe8016596 16796c7241
commit 707307cb32
12 changed files with 368 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/common_sponge.cuh
View File

@ -359,4 +359,8 @@ __global__ static void Print(const size_t size, const int *input_x) {
return;
}
__device__ static VECTOR Make_Vector_Not_Exceed_Value(VECTOR vector, const float value) {
return fminf(1.0, value * rnorm3df(vector.x, vector.y, vector.z)) * vector;
}
#endif // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_COMMON_SPONGE_H_

41
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_gradient_descent_impl.cu Normal file
View File

@ -0,0 +1,41 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_gradient_descent_impl.cuh"
#include "backend/kernel_compiler/gpu/cuda_impl/util.cuh"
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/common_sponge.cuh"
__global__ void MD_Iteration_Gradient_Descent(const int atom_numbers, VECTOR *crd, VECTOR *frc,
const float learning_rate) {
int i = blockDim.x * blockIdx.x + threadIdx.x;
if (i < atom_numbers) {
crd[i].x = crd[i].x + learning_rate * frc[i].x;
crd[i].y = crd[i].y + learning_rate * frc[i].y;
crd[i].z = crd[i].z + learning_rate * frc[i].z;
frc[i].x = 0.;
frc[i].y = 0.;
frc[i].z = 0.;
}
}
void MDIterationGradientDescent(const int atom_numbers, float *crd, float *frc, const float learning_rate,
cudaStream_t stream) {
VECTOR *d_crd = reinterpret_cast<VECTOR *>(crd);
VECTOR *d_frc = reinterpret_cast<VECTOR *>(frc);
MD_Iteration_Gradient_Descent<<<ceilf(static_cast<float>(atom_numbers) / 128), 128, 0, stream>>>(
atom_numbers, d_crd, d_frc, learning_rate);
}

25
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_gradient_descent_impl.cuh Normal file
View File

@ -0,0 +1,25 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_MD_ITERATION_GRADIENT_DESCENT_IMPL_H_
#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_MD_ITERATION_GRADIENT_DESCENT_IMPL_H_
#include <curand_kernel.h>
#include "runtime/device/gpu/cuda_common.h"
void MDIterationGradientDescent(const int atom_numbers, float *crd, float *frc, const float learning_rate,
cudaStream_t stream);
#endif

2
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_leap_frog_liujian_gpu_impl.cu
View File

@ -54,7 +54,7 @@ void MD_Iteration_Leap_Frog_With_LiuJian(const int atom_numbers, const float hal
curandStatePhilox4_32_10_t *rand_state, float *rand_frc, float *output,
cudaStream_t stream) {
Rand_Normal<<<ceilf(static_cast<float>(float4_numbers) / 32.), 32, 0, stream>>>(float4_numbers, rand_state,
reinterpret_cast<float4 *>(rand_frc));
reinterpret_cast<float4 *>(rand_frc));
VECTOR *d_vel = reinterpret_cast<VECTOR *>(vel);
VECTOR *d_crd = reinterpret_cast<VECTOR *>(crd);
VECTOR *d_frc = reinterpret_cast<VECTOR *>(frc);

44
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_leap_frog_with_max_vel_impl.cu Normal file
View File

@ -0,0 +1,44 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_leap_frog_with_max_vel_impl.cuh"
#include "backend/kernel_compiler/gpu/cuda_impl/util.cuh"
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/common_sponge.cuh"
__global__ void MD_Iteration_Leap_Frog_With_Max_Velocity(const int atom_numbers, VECTOR *vel, VECTOR *crd, VECTOR *frc,
VECTOR *acc, const float *inverse_mass, const float dt,
const float max_velocity) {
int i = blockDim.x * blockIdx.x + threadIdx.x;
if (i < atom_numbers) {
VECTOR acc_i = inverse_mass[i] * frc[i];
VECTOR vel_i = vel[i] + dt * acc_i;
vel_i = Make_Vector_Not_Exceed_Value(vel_i, max_velocity);
vel[i] = vel_i;
crd[i] = crd[i] + dt * vel_i;
frc[i] = {0.0f, 0.0f, 0.0f};
}
}
void MDIterationLeapFrogWithMaxVelocity(const int atom_numbers, float *vel, float *crd, float *frc, float *acc,
const float *inverse_mass, const float dt, const float max_velocity,
cudaStream_t stream) {
VECTOR *d_vel = reinterpret_cast<VECTOR *>(vel);
VECTOR *d_crd = reinterpret_cast<VECTOR *>(crd);
VECTOR *d_frc = reinterpret_cast<VECTOR *>(frc);
VECTOR *d_acc = reinterpret_cast<VECTOR *>(acc);
MD_Iteration_Leap_Frog_With_Max_Velocity<<<ceilf(static_cast<float>(atom_numbers) / 128), 128, 0, stream>>>(
atom_numbers, d_vel, d_crd, d_frc, d_acc, inverse_mass, dt, max_velocity);
}

26
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_leap_frog_with_max_vel_impl.cuh Normal file
View File

@ -0,0 +1,26 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_MD_ITERATION_LEAP_FROG_WITH_MAX_VEL_IMPL_H_
#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_MD_ITERATION_LEAP_FROG_WITH_MAX_VEL_IMPL_H_
#include <curand_kernel.h>
#include "runtime/device/gpu/cuda_common.h"
void MDIterationLeapFrogWithMaxVelocity(const int atom_numbers, float *vel, float *crd, float *frc, float *acc,
const float *inverse_mass, const float dt, const float max_velocity,
cudaStream_t stream);
#endif // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_SPONGE_MD_ITERATION_LEAP_FROG_WITH_MAX_VEL_IMPL_H_

2
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_setup_random_state_gpu_impl.cu
View File

@ -21,7 +21,7 @@
void MD_Iteration_Setup_Random_State(int float4_numbers, curandStatePhilox4_32_10_t *rand_state, int seed,
cudaStream_t stream) {
Setup_Rand_Normal_Kernel<<<ceilf(static_cast<float>(float4_numbers) / 32.), 32, 0, stream>>>(float4_numbers,
rand_state, seed);
rand_state, seed);
}
void MD_Iteration_Setup_Random_State(int float4_numbers, curandStatePhilox4_32_10_t *rand_state, int seed,

12
mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sponge/pme/pme_common.cuh
View File

@ -108,16 +108,16 @@ __global__ static void PME_Atom_Near(const UNSIGNED_INT_VECTOR *uint_crd, int *P
UNSIGNED_INT_VECTOR *temp_uxyz = &PME_uxyz[atom];
int k, tempux, tempuy, tempuz;
float tempf;
tempf = static_cast<float> (uint_crd[atom].uint_x) * periodic_factor_inverse_x;
tempux = static_cast<int> (tempf);
tempf = static_cast<float>(uint_crd[atom].uint_x) * periodic_factor_inverse_x;
tempux = static_cast<int>(tempf);
PME_frxyz[atom].x = tempf - tempux;
tempf = static_cast<float> (uint_crd[atom].uint_y) * periodic_factor_inverse_y;
tempuy = static_cast<int> (tempf);
tempf = static_cast<float>(uint_crd[atom].uint_y) * periodic_factor_inverse_y;
tempuy = static_cast<int>(tempf);
PME_frxyz[atom].y = tempf - tempuy;
tempf = static_cast<float> (uint_crd[atom].uint_z) * periodic_factor_inverse_z;
tempuz = static_cast<int> (tempf);
tempf = static_cast<float>(uint_crd[atom].uint_z) * periodic_factor_inverse_z;
tempuz = static_cast<int>(tempf);
PME_frxyz[atom].z = tempf - tempuz;
if (tempux != (*temp_uxyz).uint_x || tempuy != (*temp_uxyz).uint_y || tempuz != (*temp_uxyz).uint_z) {

26
mindspore/ccsrc/backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_gradient_descent_kernel.cc Normal file
View File

@ -0,0 +1,26 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#include "backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_gradient_descent_kernel.h"
namespace mindspore {
namespace kernel {
MS_REG_GPU_KERNEL_ONE(
MDIterationGradientDescent,
KernelAttr().AddInputAttr(kNumberTypeFloat32).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
MDIterationGradientDescentGpuKernel, float)
} // namespace kernel
} // namespace mindspore

77
mindspore/ccsrc/backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_gradient_descent_kernel.h Normal file
View File

@ -0,0 +1,77 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_MD_ITERATION_GRADIENT_DESCENT_KERNEL_H_
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_MD_ITERATION_GRADIENT_DESCENT_KERNEL_H_
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_gradient_descent_impl.cuh"
#include <cuda_runtime_api.h>
#include <map>
#include <string>
#include <vector>
#include "backend/kernel_compiler/gpu/gpu_kernel.h"
#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
#include "runtime/device/gpu/cuda_common.h"
namespace mindspore {
namespace kernel {
template <typename T>
class MDIterationGradientDescentGpuKernel : public GpuKernel {
public:
MDIterationGradientDescentGpuKernel() {}
~MDIterationGradientDescentGpuKernel() override = default;
bool Init(const CNodePtr &kernel_node) override {
// get bond_numbers
kernel_node_ = kernel_node;
atom_numbers = static_cast<int>(GetAttr<int64_t>(kernel_node, "atom_numbers"));
learning_rate = static_cast<float>(GetAttr<float>(kernel_node, "learning_rate"));
InitSizeLists();
return true;
}
const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
auto crd = GetDeviceAddress<float>(inputs, 0);
auto frc = GetDeviceAddress<float>(inputs, 1);
MDIterationGradientDescent(atom_numbers, crd, frc, learning_rate, reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
protected:
void InitSizeLists() override {
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
output_size_list_.push_back(sizeof(T));
}
private:
std::vector<size_t> input_size_list_;
std::vector<size_t> output_size_list_;
std::vector<size_t> workspace_size_list_;
int atom_numbers;
float learning_rate;
};
} // namespace kernel
} // namespace mindspore
#endif

31
mindspore/ccsrc/backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_leap_frog_with_max_vel_kernel.cc Normal file
View File

@ -0,0 +1,31 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#include "backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_leap_frog_with_max_vel_kernel.h"
namespace mindspore {
namespace kernel {
MS_REG_GPU_KERNEL_ONE(MDIterationLeapFrogWithMaxVel,
KernelAttr()
.AddInputAttr(kNumberTypeFloat32)
.AddInputAttr(kNumberTypeFloat32)
.AddInputAttr(kNumberTypeFloat32)
.AddInputAttr(kNumberTypeFloat32)
.AddInputAttr(kNumberTypeFloat32)
.AddOutputAttr(kNumberTypeFloat32),
MDIterationLeapFrogWithMaxVelGpuKernel, float)
} // namespace kernel
} // namespace mindspore

86
mindspore/ccsrc/backend/kernel_compiler/gpu/sponge/nvtit/md_iteration_leap_frog_with_max_vel_kernel.h Normal file
View File

@ -0,0 +1,86 @@
/**
* Copyright 2021 Huawei Technologies Co., Ltd
*
* 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.
*/
#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_MD_ITERATION_LEAP_FROG_WITH_MAX_VEL_KERNEL_H_
#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_MD_ITERATION_LEAP_FROG_WITH_MAX_VEL_KERNEL_H_
#include "backend/kernel_compiler/gpu/cuda_impl/sponge/nvtit/md_iteration_leap_frog_with_max_vel_impl.cuh"
#include <cuda_runtime_api.h>
#include <map>
#include <string>
#include <vector>
#include "backend/kernel_compiler/gpu/gpu_kernel.h"
#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
#include "runtime/device/gpu/cuda_common.h"
namespace mindspore {
namespace kernel {
template <typename T>
class MDIterationLeapFrogWithMaxVelGpuKernel : public GpuKernel {
public:
MDIterationLeapFrogWithMaxVelGpuKernel() {}
~MDIterationLeapFrogWithMaxVelGpuKernel() override = default;
bool Init(const CNodePtr &kernel_node) override {
// get bond_numbers
kernel_node_ = kernel_node;
atom_numbers = static_cast<int>(GetAttr<int64_t>(kernel_node, "atom_numbers"));
dt = static_cast<float>(GetAttr<float>(kernel_node, "dt"));
max_velocity = static_cast<float>(GetAttr<float>(kernel_node, "max_velocity"));
InitSizeLists();
return true;
}
const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
auto vel = GetDeviceAddress<float>(inputs, 0);
auto crd = GetDeviceAddress<float>(inputs, 1);
auto frc = GetDeviceAddress<float>(inputs, 2);
auto acc = GetDeviceAddress<float>(inputs, 3);
auto inverse_mass = GetDeviceAddress<float>(inputs, 4);
MDIterationLeapFrogWithMaxVelocity(atom_numbers, vel, crd, frc, acc, inverse_mass, dt, max_velocity,
reinterpret_cast<cudaStream_t>(stream_ptr));
return true;
}
protected:
void InitSizeLists() override {
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
input_size_list_.push_back(atom_numbers * 3 * sizeof(T));
input_size_list_.push_back(atom_numbers * sizeof(T));
output_size_list_.push_back(sizeof(T));
}
private:
std::vector<size_t> input_size_list_;
std::vector<size_t> output_size_list_;
std::vector<size_t> workspace_size_list_;
int atom_numbers;
float dt;
float max_velocity;
};
} // namespace kernel
} // namespace mindspore
#endif