Skip to content

GitLab

"tests/git@developer.sourcefind.cn:OpenDAS/mmcv.git" did not exist on "a47451b47a1c88668c8284c885be92b5965e5447"
Unverified Commit 0a72e4df authored by Chao Liu's avatar Chao Liu Committed by GitHub
Browse files

Change initialization method of tensor for iGEMM (#49) 

* change init method
parent 58a80570
Hide whitespace changes
Inline Side-by-side
Showing with 103 additions and 81 deletions
driver/conv_bwd_data_driver_v2.cpp
View file @ 0a72e4df
...@@ -179,26 +179,38 @@ int main(int argc, char* argv[]) ...@@ -179,26 +179,38 @@ int main(int argc, char* argv[])
std::size_t num_thread = std::thread::hardware_concurrency(); std::size_t num_thread = std::thread::hardware_concurrency();
if(do_verification) switch(init_method)
{ {
switch(init_method) case 0:
{ // no initialization
case 0: break;
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 1:
out.GenerateTensorValue(GeneratorTensor_1{}, num_thread); out.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
case 1: break;
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 2:
out.GenerateTensorValue(GeneratorTensor_1{}, num_thread); out.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
case 2: break;
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 3:
out.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); out.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
default: break;
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 4:
out.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); out.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
} wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break;
case 5:
out.GenerateTensorValue(GeneratorTensor_3<float>{0.0, 1.0}, num_thread);
wei.GenerateTensorValue(GeneratorTensor_3<float>{-0.5, 0.5}, num_thread);
break;
default:
out.GenerateTensorValue(GeneratorTensor_2{1, 5}, num_thread);
auto gen_wei = [](auto... is) {
return GeneratorTensor_2{1, 5}(is...) * GeneratorTensor_Checkboard{}(is...);
};
wei.GenerateTensorValue(gen_wei, num_thread);
} }
auto f_make_for_device_nchw = [&]() { auto f_make_for_device_nchw = [&]() {
......
driver/conv_driver_v2.cpp
View file @ 0a72e4df
...@@ -205,34 +205,38 @@ int main(int argc, char* argv[]) ...@@ -205,34 +205,38 @@ int main(int argc, char* argv[])
std::size_t num_thread = std::thread::hardware_concurrency(); std::size_t num_thread = std::thread::hardware_concurrency();
if(do_verification) switch(init_method)
{ {
switch(init_method) case 0:
{ // no initialization
case 0: break;
in.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 1:
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); in.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
case 1: break;
in.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 2:
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); in.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
case 2: break;
in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 3:
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
case 3: break;
in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 4:
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
default: break;
in.GenerateTensorValue(GeneratorTensor_2{1, 5}, num_thread); case 5:
in.GenerateTensorValue(GeneratorTensor_3<float>{0.0, 1.0}, num_thread);
auto gen_wei = [](auto... is) { wei.GenerateTensorValue(GeneratorTensor_3<float>{-0.5, 0.5}, num_thread);
return GeneratorTensor_2{1, 5}(is...) * GeneratorTensor_Checkboard{}(is...); break;
}; default:
wei.GenerateTensorValue(gen_wei, num_thread); in.GenerateTensorValue(GeneratorTensor_2{1, 5}, num_thread);
}
auto gen_wei = [](auto... is) {
return GeneratorTensor_2{1, 5}(is...) * GeneratorTensor_Checkboard{}(is...);
};
wei.GenerateTensorValue(gen_wei, num_thread);
} }
auto f_make_for_device_nchw = [&]() { auto f_make_for_device_nchw = [&]() {
......
driver/conv_driver_v2_olc.cpp
View file @ 0a72e4df
...@@ -152,34 +152,38 @@ int main(int argc, char* argv[]) ...@@ -152,34 +152,38 @@ int main(int argc, char* argv[])
std::size_t num_thread = std::thread::hardware_concurrency(); std::size_t num_thread = std::thread::hardware_concurrency();
if(do_verification) switch(init_method)
{ {
switch(init_method) case 0:
{ // no initialization
case 0: break;
in.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 1:
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); in.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
case 1: break;
in.GenerateTensorValue(GeneratorTensor_1{}, num_thread); case 2:
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); in.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
case 2: break;
in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 3:
wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread); in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_1{}, num_thread);
case 3: break;
in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); case 4:
wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread); in.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
break; wei.GenerateTensorValue(GeneratorTensor_2{-5, 5}, num_thread);
default: break;
in.GenerateTensorValue(GeneratorTensor_2{1, 5}, num_thread); case 5:
in.GenerateTensorValue(GeneratorTensor_3<float>{0.0, 1.0}, num_thread);
auto gen_wei = [](auto... is) { wei.GenerateTensorValue(GeneratorTensor_3<float>{-0.5, 0.5}, num_thread);
return GeneratorTensor_2{1, 5}(is...) * GeneratorTensor_Checkboard{}(is...); break;
}; default:
wei.GenerateTensorValue(gen_wei, num_thread); in.GenerateTensorValue(GeneratorTensor_2{1, 5}, num_thread);
}
auto gen_wei = [](auto... is) {
return GeneratorTensor_2{1, 5}(is...) * GeneratorTensor_Checkboard{}(is...);
};
wei.GenerateTensorValue(gen_wei, num_thread);
} }
auto f_make_for_device_nchw = [&]() { auto f_make_for_device_nchw = [&]() {
......
driver/include/host_tensor_generator.hpp
View file @ 0a72e4df
...@@ -9,7 +9,7 @@ struct GeneratorTensor_1 ...@@ -9,7 +9,7 @@ struct GeneratorTensor_1
int value = 1; int value = 1;
template <typename... Is> template <typename... Is>
double operator()(Is... is) float operator()(Is... is)
{ {
return value; return value;
} }
...@@ -21,29 +21,31 @@ struct GeneratorTensor_2 ...@@ -21,29 +21,31 @@ struct GeneratorTensor_2
int max_value = 1; int max_value = 1;
template <typename... Is> template <typename... Is>
double operator()(Is...) float operator()(Is...)
{ {
return (std::rand() % (max_value - min_value)) + min_value; return (std::rand() % (max_value - min_value)) + min_value;
} }
}; };
template <typename T>
struct GeneratorTensor_3 struct GeneratorTensor_3
{ {
T min_value = 0;
T max_value = 1;
template <typename... Is> template <typename... Is>
double operator()(Is... is) float operator()(Is...)
{ {
std::array<ck::index_t, sizeof...(Is)> dims = {{static_cast<ck::index_t>(is)...}}; float tmp = float(std::rand()) / float(RAND_MAX);
auto f_acc = [](auto a, auto b) { return 10 * a + b; };
return std::accumulate(dims.begin(), dims.end(), ck::index_t(0), f_acc); return min_value + tmp * (max_value - min_value);
} }
}; };
struct GeneratorTensor_Checkboard struct GeneratorTensor_Checkboard
{ {
template <typename... Ts> template <typename... Ts>
double operator()(Ts... Xs) const float operator()(Ts... Xs) const
{ {
std::array<ck::index_t, sizeof...(Ts)> dims = {{static_cast<ck::index_t>(Xs)...}}; std::array<ck::index_t, sizeof...(Ts)> dims = {{static_cast<ck::index_t>(Xs)...}};
return std::accumulate(dims.begin(), return std::accumulate(dims.begin(),
......
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