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

Fixed half-precision tests for im2col and col2im

pull/331/head
Cedric Nugteren 2018-11-01 21:44:21 +01:00
parent 4215bbe62a
commit 469c346a8e
2 changed files with 40 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
test/routines/levelx/xcol2im.hpp
View File

@ -197,6 +197,26 @@ StatusCode RunReference(const Arguments<T> &args, BuffersHost<T> &buffers_host)
return StatusCode::kSuccess;
}
// Half-precision version calling the above reference implementation after conversions
template <>
StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &buffers_host) {
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.channels = args.channels; args2.height = args.height; args2.width = args.width;
args2.kernel_h = args.kernel_h; args2.kernel_w = args.kernel_w;
args2.pad_h = args.pad_h; args2.pad_w = args.pad_w;
args2.stride_h = args.stride_h; args2.stride_w = args.stride_w;
args2.dilation_h = args.dilation_h; args2.dilation_w = args.dilation_w;
args2.a_offset = args.a_offset; args2.b_offset = args.b_offset;
auto status = RunReference(args2, buffers2);
FloatToHalfBuffer(buffers_host.a_mat, buffers2.a_mat);
return status;
}
// =================================================================================================
} // namespace clblast

20
test/routines/levelx/xim2col.hpp
View File

@ -188,6 +188,26 @@ StatusCode RunReference(const Arguments<T> &args, BuffersHost<T> &buffers_host)
return StatusCode::kSuccess;
}
// Half-precision version calling the above reference implementation after conversions
template <>
StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &buffers_host) {
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.channels = args.channels; args2.height = args.height; args2.width = args.width;
args2.kernel_h = args.kernel_h; args2.kernel_w = args.kernel_w;
args2.pad_h = args.pad_h; args2.pad_w = args.pad_w;
args2.stride_h = args.stride_h; args2.stride_w = args.stride_w;
args2.dilation_h = args.dilation_h; args2.dilation_w = args.dilation_w;
args2.a_offset = args.a_offset; args2.b_offset = args.b_offset;
auto status = RunReference(args2, buffers2);
FloatToHalfBuffer(buffers_host.b_mat, buffers2.b_mat);
return status;
}
// =================================================================================================
} // namespace clblast