haraldwolff/CLBlast
1
0
Fork 0
mirror of https://github.com/CNugteren/CLBlast.git synced 2024-07-04 21:36:57 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'msvc2013_fixes'

Browse source
This commit is contained in:
Cedric Nugteren 2017-09-19 19:54:43 +02:00
parent a23cd8d13a ae1eeb4d1f
commit 44b59ec0cb
3 changed files with 29 additions and 19 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
src/database/database_structure.hpp
View file

@ -40,23 +40,24 @@ const std::string kDeviceTypeAll = "default";
const Name kDeviceNameDefault = {"default "};
struct DatabaseDevice {
const Name name;
const Params parameters; // parameter values
Name name;
Params parameters; // parameter values
};
struct DatabaseArchitecture {
const std::string name;
const std::vector<DatabaseDevice> devices;
std::string name;
std::vector<DatabaseDevice> devices;
};
struct DatabaseVendor {
const std::string type;
const std::string name;
const std::vector<DatabaseArchitecture> architectures;
std::string type;
std::string name;
std::vector<DatabaseArchitecture> architectures;
};
struct DatabaseEntry {
const std::string kernel;
const Precision precision;
const std::vector<std::string> parameter_names;
const std::vector<DatabaseVendor> vendors;
std::string kernel;
Precision precision;
std::vector<std::string> parameter_names;
std::vector<DatabaseVendor> vendors;
};
// =================================================================================================

8
src/routines/levelx/xaxpybatched.cpp
View file

@ -53,8 +53,12 @@ void XaxpyBatched<T>::DoAxpyBatched(const size_t n, const std::vector<T> &alphas
}
// Upload the arguments to the device
std::vector<int> x_offsets_int(x_offsets.begin(), x_offsets.end());
std::vector<int> y_offsets_int(y_offsets.begin(), y_offsets.end());
auto x_offsets_int = std::vector<int>(batch_count);
auto y_offsets_int = std::vector<int>(batch_count);
for (auto batch = size_t{ 0 }; batch < batch_count; ++batch) {
x_offsets_int[batch] = static_cast<int>(x_offsets[batch]);
y_offsets_int[batch] = static_cast<int>(y_offsets[batch]);
}
auto x_offsets_device = Buffer<int>(context_, BufferAccess::kReadOnly, batch_count);
auto y_offsets_device = Buffer<int>(context_, BufferAccess::kReadOnly, batch_count);
auto alphas_device = Buffer<T>(context_, BufferAccess::kReadOnly, batch_count);

17
src/routines/levelx/xgemmbatched.cpp
View file

@ -106,9 +106,14 @@ void XgemmBatched<T>::DoGemmBatched(const Layout layout, const Transpose a_trans
betas_device.Write(queue_, batch_count, betas);
// Converts the offset to integers
std::vector<int> a_offsets_int(a_offsets.begin(), a_offsets.end());
std::vector<int> b_offsets_int(b_offsets.begin(), b_offsets.end());
std::vector<int> c_offsets_int(c_offsets.begin(), c_offsets.end());
auto a_offsets_int = std::vector<int>(batch_count);
auto b_offsets_int = std::vector<int>(batch_count);
auto c_offsets_int = std::vector<int>(batch_count);
for (auto batch = size_t{ 0 }; batch < batch_count; ++batch) {
a_offsets_int[batch] = static_cast<int>(a_offsets[batch]);
b_offsets_int[batch] = static_cast<int>(b_offsets[batch]);
c_offsets_int[batch] = static_cast<int>(c_offsets[batch]);
}
// Selects which version of the batched GEMM to run
const auto do_gemm_direct = true;
@ -169,9 +174,9 @@ void XgemmBatched<T>::BatchedGemmIndirect(const size_t m, const size_t n, const
auto b_offsets_i = std::vector<int>(batch_count);
auto c_offsets_i = std::vector<int>(batch_count);
for (auto batch = size_t{0}; batch < batch_count; ++batch) {
a_offsets_i[batch] = batch * a_one_i * a_two_i;
b_offsets_i[batch] = batch * b_one_i * b_two_i;
c_offsets_i[batch] = batch * c_one_i * c_two_i;
a_offsets_i[batch] = static_cast<int>(batch * a_one_i * a_two_i);
b_offsets_i[batch] = static_cast<int>(batch * b_one_i * b_two_i);
c_offsets_i[batch] = static_cast<int>(batch * c_one_i * c_two_i);
}
// Determines whether or not temporary matrices are needed