Skip to content

Commit

Permalink
Merge pull request baidu-research#4 from ROCmSoftwarePlatform/amd
Browse files Browse the repository at this point in the history 
remove half.hpp dependency
  • Loading branch information
Mayank Daga authored Apr 4, 2018
2 parents bcfad48 + c7055ef commit 6395348
Show file tree
Hide file tree
Showing 2 changed files with 32 additions and 28 deletions.
51 changes: 32 additions & 19 deletions code/amd/conv_bench_rocm.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,11 +11,12 @@
#include "miopen_helper.h"
#include "conv_problems.h"

template<typename T>
class miopenCNN {
TensorDescriptor4d<float> x_desc_;
TensorDescriptor4d<float> h_desc_;
TensorDescriptor4d<T> x_desc_;
TensorDescriptor4d<T> h_desc_;

FilterDescriptor4d<float> w_desc_;
FilterDescriptor4d<T> w_desc_;

std::vector<int> output_dims_;
int num_repeats_;
Expand All @@ -28,7 +29,7 @@ class miopenCNN {
Tensor<float> bwd_inputs_workspace_;
Tensor<float> bwd_params_workspace_;

Tensor<float> h;
Tensor<T> h;

miopenConvFwdAlgorithm_t fwd_algo_;
miopenConvBwdDataAlgorithm_t bwd_inputs_algo_;
Expand All @@ -43,7 +44,7 @@ class miopenCNN {
public:

miopenCNN(int _w, int _h, int c, int n, int k, int r, int s,
int pad_w, int pad_h, int wstride, int hstride, Tensor<float> x, Tensor<float> w)
int pad_w, int pad_h, int wstride, int hstride, Tensor<T> x, Tensor<T> w)
:
miopen_handle_(),
x_desc_(n, c, _h, _w),
Expand All @@ -61,11 +62,11 @@ class miopenCNN {
&out_h,
&out_w));

h_desc_ = TensorDescriptor4d<float>(out_n, out_c, out_h, out_w);
h_desc_ = TensorDescriptor4d<T>(out_n, out_c, out_h, out_w);

output_dims_ = {out_w, out_h, out_c, out_n};

h = zeros<float>(output_dims_);
h = zeros<T>(output_dims_);


// Set fwd workspace size
Expand Down Expand Up @@ -166,7 +167,7 @@ class miopenCNN {

}

Tensor<float> getOutputTensor(){ return h; }
Tensor<T> getOutputTensor(){ return h; }

std::vector<int> get_output_dims() { return output_dims_; }

Expand All @@ -187,7 +188,7 @@ class miopenCNN {
}


void forward(Tensor<float> x, Tensor<float> filter, Tensor<float> h) {
void forward(Tensor<T> x, Tensor<T> filter, Tensor<T> h) {

// Convolution forward.
CHECK_MIOPEN_ERROR(miopenConvolutionForward(miopen_handle_.handle(),
Expand All @@ -207,7 +208,7 @@ class miopenCNN {

}

void backward_params(Tensor<float> x, Tensor<float> delta, Tensor<float> dW) {
void backward_params(Tensor<T> x, Tensor<T> delta, Tensor<T> dW) {

CHECK_MIOPEN_ERROR(miopenConvolutionBackwardWeights(miopen_handle_.handle(),
&alpha_,
Expand All @@ -227,7 +228,7 @@ class miopenCNN {

}

void backward_inputs(Tensor<float> filter, Tensor<float> delta, Tensor<float> dX) {
void backward_inputs(Tensor<T> filter, Tensor<T> delta, Tensor<T> dX) {

CHECK_MIOPEN_ERROR(miopenConvolutionBackwardData(miopen_handle_.handle(),
&alpha_,
Expand All @@ -247,6 +248,7 @@ class miopenCNN {
}
};

template<typename T>
std::tuple<int, int, int, std::string> time_cnn(
int k, int c, int r, int s,
int n, int h, int w,
Expand All @@ -257,11 +259,11 @@ std::tuple<int, int, int, std::string> time_cnn(


// Allocate memory for filter
auto filter = rand<float>(std::vector<int>{r, s, c, k});
auto filter = rand<T>(std::vector<int>{r, s, c, k});

// Allocate memory for input
auto input = rand<float>(std::vector<int>{w, h, c, n});
miopenCNN cnn(w, h, c, n, k, r, s, pad_w, pad_h, wstride, hstride, input, filter);
auto input = rand<T>(std::vector<int>{w, h, c, n});
miopenCNN<T> cnn(w, h, c, n, k, r, s, pad_w, pad_h, wstride, hstride, input, filter);

// Allocate memory for output tensor
auto output = cnn.getOutputTensor();
Expand All @@ -283,8 +285,8 @@ std::tuple<int, int, int, std::string> time_cnn(
int fwd_time = static_cast<int>(std::chrono::duration<double, std::micro>(end - start).count() / num_repeats);

// Allocate memory for backward pass wrt weights
auto delta = rand<float>(cnn.get_output_dims());
auto dW = zeros<float>(std::vector<int>{r, s, c, k});
auto delta = rand<T>(cnn.get_output_dims());
auto dW = zeros<T>(std::vector<int>{r, s, c, k});

// Warm up backward
cnn.backward_params(input, delta, dW);
Expand All @@ -303,7 +305,7 @@ std::tuple<int, int, int, std::string> time_cnn(
int bwd_params_time = static_cast<int>(std::chrono::duration<double, std::micro>(end - start).count() / num_repeats);

//Allocate memory for backward pass wrt inputs
auto dX = zeros<float>(std::vector<int>{w, h, c, n});
auto dX = zeros<T>(std::vector<int>{w, h, c, n});

//Warm up backward inputs
cnn.backward_inputs(filter, delta, dX);
Expand All @@ -330,11 +332,15 @@ std::tuple<int, int, int, std::string> time_cnn(
int main(int argc, char **argv) {

int num_repeats = 100;
std::string precision ="float";

hipFree(0);

if (argc > 1)
{
num_repeats = atoi(argv[1]);
precision = argv[2];
}

std::cout << std::setw(30) << "Times" << std::endl;
std::cout << std::setfill('-') << std::setw(190) << "-" << std::endl;
Expand Down Expand Up @@ -363,8 +369,15 @@ int main(int argc, char **argv) {
int fwd_time, bwd_inputs_time, bwd_params_time;
std::string fwd_algo_s;

std::tie(fwd_time, bwd_inputs_time, bwd_params_time, fwd_algo_s) =
time_cnn(k, c, r, s, n, h, w, pad_h, pad_w, hstride, wstride, num_repeats);
if( precision == "float" )
{
std::tie(fwd_time, bwd_inputs_time, bwd_params_time, fwd_algo_s) =
time_cnn<float>(k, c, r, s, n, h, w, pad_h, pad_w, hstride, wstride, num_repeats);
}
else
{
throw std::runtime_error("unknown precision");
}

std::cout << std::setw(5) << w;
std::cout << std::setw(7) << h;
Expand Down
9 changes: 0 additions & 9 deletions code/amd/miopen_helper.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,6 @@
#include <vector>

#include <miopen/miopen.h>
#include <half.hpp>

#include "hip_helper.h"

Expand Down Expand Up @@ -68,8 +67,6 @@ class TensorDescriptor {
miopenDataType_t type;
if (std::is_same<T, float>::value)
type = miopenFloat;
else if (std::is_same<T, half_float::half>::value)
type = miopenHalf;
else
throw std::runtime_error("Unknown type");

Expand Down Expand Up @@ -112,8 +109,6 @@ class TensorDescriptorArray
miopenDataType_t type;
if (std::is_same<T, float>::value)
type = miopenFloat;
else if (std::is_same<T, half_float::half>::value)
type = miopenHalf;
else
throw std::runtime_error("Unknown type");

Expand Down Expand Up @@ -150,8 +145,6 @@ class TensorDescriptor4d {
miopenDataType_t type;
if (std::is_same<T, float>::value)
type = miopenFloat;
else if (std::is_same<T, half_float::half>::value)
type = miopenHalf;
else
throw std::runtime_error("Unknown type");

Expand Down Expand Up @@ -187,8 +180,6 @@ class FilterDescriptor4d {
miopenDataType_t type;
if (std::is_same<T, float>::value)
type = miopenFloat;
else if (std::is_same<T, half_float::half>::value)
type = miopenHalf;
else
throw std::runtime_error("Unknown type");

Expand Down

0 comments on commit 6395348

Please sign in to comment.