haraldwolff
/
CLBlast
mirror of https://github.com/CNugteren/CLBlast.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fixed a failing TRSV test using a CPU with Apple OpenCL

pull/269/head
Cedric Nugteren 2018-03-15 20:58:42 +01:00
parent f4d96e80c3
commit 7a756cbce7
3 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/database/apple_cpu_fallback.hpp
View File

@ -41,7 +41,7 @@ const DatabaseEntry XgerApple = {
"Xger", Precision::kAny, {"WGS1", "WGS2", "WPT"}, { { kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 64, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } } } } } } }
};
const DatabaseEntry XtrsvApple = {
"Xtrsv", Precision::kAny, {"TRSV_BLOCK_SIZE"}, { { kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 32, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } } } } } } }
"Xtrsv", Precision::kAny, {"TRSV_BLOCK_SIZE"}, { { kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 16, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } } } } } } }
};
const DatabaseEntry XgemmApple = {
"Xgemm", Precision::kAny, {"KWG", "KWI", "MDIMA", "MDIMC", "MWG", "NDIMB", "NDIMC", "NWG", "SA", "SB", "STRM", "STRN", "VWM", "VWN"}, { { kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1 } } } } } } }

2
src/kernels/level2/xtrsv.opencl
View File

@ -18,7 +18,7 @@ R"(
// =================================================================================================
#if defined(ROUTINE_TRSV)
__kernel __attribute__((reqd_work_group_size(64, 1, 1)))
__kernel __attribute__((reqd_work_group_size(16, 1, 1)))
void FillVector(const int n, const int inc, const int offset,
__global real* restrict dest, const real_arg arg_value) {
const real value = GetRealArg(arg_value);

4
src/routines/common.cpp
View File

@ -125,8 +125,8 @@ void FillVector(Queue &queue, const Device &device,
kernel.SetArgument(2, static_cast<int>(offset));
kernel.SetArgument(3, dest());
kernel.SetArgument(4, GetRealArg(constant_value));
auto local = std::vector<size_t>{64};
auto global = std::vector<size_t>{Ceil(n, 64)};
auto local = std::vector<size_t>{16};
auto global = std::vector<size_t>{Ceil(n, 16)};
RunKernel(kernel, queue, device, global, local, event, waitForEvents);
}