ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

CGEMM examples bf16, fp32, int8 (#332)

Browse Source 
* Add int8 specialization for elementwise Add and Subtract.

* CGEMM examples bf16, fp32, int8

* Add convert reference output to CDataType.

* Skip BF16 data type during testing.

* Lower K value to get rid of accumulation error.

* Fix merge artifact.

* Fix changed function name: GetElementSpaceSize()

* Fix merge artifact.

Co-authored-by: Adam Osewski <aosewski@amd.com>
This commit is contained in:
Adam Osewski
2022-08-02 21:52:27 +02:00
committed by GitHub
parent 984b3722bf
commit fb0dc35861
9 changed files with 649 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
View File

@@ -51,6 +51,13 @@ struct Add
const float y_tmp = x1_tmp + x2_tmp;
y = ck::type_convert<bhalf_t>(y_tmp);
}
template <>
__host__ __device__ constexpr void
operator()<int8_t>(int8_t& y, const int8_t& x0, const int8_t& x1) const
{
y = x0 + x1;
};
};
struct Subtract
@@ -88,6 +95,13 @@ struct Subtract
const float y_tmp = x1_tmp - x2_tmp;
y = ck::type_convert<bhalf_t>(y_tmp);
}
template <>
__host__ __device__ constexpr void
operator()<int8_t>(int8_t& y, const int8_t& x0, const int8_t& x1) const
{
y = x0 - x1;
};
};
struct Bilinear