Fix the fp8 gemm for large tensors on MI300. (#1011)

* Fix the fp8 conversion

* Try clipping value before conversion

* Fix return

* Simplify with a const

* reduce the gemm input tensor values to reduce round-off error

* replace if-else with lambda

* fix syntax

---------

Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
This commit is contained in:
Illia Silin
2023-10-27 21:10:47 -07:00
committed by GitHub
parent 6fe0bc7e72
commit f46a6ffad8
4 changed files with 10 additions and 8 deletions

View File

@@ -75,8 +75,8 @@ int profile_gemm_impl(int do_verification,
b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
break;
default:
a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 0.1});
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.05, 0.05});
}
using AElementOp = ck::tensor_operation::element_wise::PassThrough;