ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-19 20:40:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support for dtypes (fp8, bf8, bf16 and fp16) for the ck_tile/03_gemm example. (#1845)

Browse Source 
* Support bf16/fb8/bf8 datatypes for ck_tile/gemm

* remove commented out code.

* Addressing code review comments and enabling universal_gemm for all the supported data types.

* Merge conflict resolution.

* Solve the memory pipeline compilation error. Merge with the new change of CShuffle

* finish the feature, pass the tests

* Fix the pipeline and add the benchmark script for other data types

---------

Co-authored-by: ThomasNing <thomas.ning@amd.com>

[ROCm/composable_kernel commit: ab5d027866]
This commit is contained in:
kylasa
2025-02-06 14:07:38 -08:00
committed by Sam Wu
parent f5d3690565
commit 0aee5c2d16
21 changed files with 598 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
include/ck_tile/host/reference/reference_gemm.hpp
View File

@@ -80,13 +80,14 @@ __global__ void naive_gemm_kernel(ADataType* A,
int b_index = (std::is_same_v<LayoutB, tensor_layout::gemm::ColumnMajor>)
? col * strideB + k
: k * strideB + col;
acc += static_cast<AccDataType>(A[a_index]) * static_cast<AccDataType>(B[b_index]);
acc += ck_tile::type_convert<AccDataType>(A[a_index]) *
ck_tile::type_convert<AccDataType>(B[b_index]);
}
int c_index = (std::is_same_v<LayoutC, tensor_layout::gemm::RowMajor>)
? row * strideC + col
: col * strideC + row;
C[c_index] = acc;
C[c_index] = ck_tile::type_convert<CDataType>(acc);
}
}