mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-01 12:11:19 +00:00
[CK_TILE] Add Flatmm MX FP8 (#3208)
* Use async for flatmm mxfp4 * Fix preshuffle * Add flatmm mxfp8 * Thanks, Copilot * Thanks Copilot again~
This commit is contained in:
@@ -432,6 +432,12 @@ CK_TILE_HOST void reference_mx_gemm(const HostTensor<ADataType>& a_m_k,
|
||||
a_m_k_scaled(m, k) = a_f4_lo * a_scale;
|
||||
a_m_k_scaled(m, k + 1) = a_f4_hi * a_scale;
|
||||
}
|
||||
else
|
||||
{
|
||||
a_m_k_scaled(m, k) =
|
||||
ck_tile::type_convert<AccDataType>((a_m_k(m, k))) *
|
||||
ck_tile::type_convert<AccDataType>(scale_a(m, k / ScaleBlockSize));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user