ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Prevent instantiation of undefined FP8 operators. (#1639)

Browse Source
This commit is contained in:
Andriy Roshchenko
2024-11-05 14:58:29 -07:00
committed by GitHub
parent 54440cf562
commit 365f39aed0
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
example/44_elementwise_permute/elementwise_scale_permute_amax_2D_fp16_fp8.cpp
View File

@@ -68,7 +68,7 @@ using DeviceElementwisePermuteInstance = ck::tensor_operation::device::DeviceEle
using DeviceReduceInstance =
ck::tensor_operation::device::DeviceReduceMultiBlock<OutputDataType,
OutputDataType,
ScaleDataType,
OutputDataType,
NumDim,
NumDim,
@@ -108,7 +108,8 @@ void reference_scale_permute_amax(Tensor<InputDataType>& input,
host_output_scaled_casted_transposed(m, k) = y1;
const OutputDataType y_fabs =
ck::type_convert<OutputDataType>(ck::math::abs(ck::type_convert<float>(y0)));
host_output_amax(0) = ck::math::max(y_fabs, host_output_amax(0));
host_output_amax(0) = ck::type_convert<OutputDataType>(ck::math::max(
ck::type_convert<float>(y_fabs), ck::type_convert<float>(host_output_amax(0))));
}
}
}