ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-20 21:09:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'c4b2da9cbd979eb9e32b4f20878d220b4f435a69' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-10-16 18:14:40 +00:00
parent 1b7c5502e2
commit 980f5036da
22 changed files with 1352 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/65_gemm_multiply_multiply/moe_gemm1_xdl_pk_i4.cpp
View File

@@ -289,7 +289,6 @@ int main(int argc, char* argv[])
case 0: break;
case 1:
a0_t_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
d0_t_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
d1_e_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
d2_e_n.GenerateTensorValue(GeneratorTensor_3<D2DataType>{0.0, 1.0});
@@ -303,7 +302,6 @@ int main(int argc, char* argv[])
break;
default:
a0_t_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
d0_t_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
d1_e_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
d2_e_n.GenerateTensorValue(GeneratorTensor_3<D2DataType>{0.0, 1.0});

4
example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp
View File

@@ -275,7 +275,7 @@ int main(int argc, char* argv[])
break;
case 3:
a0_t_k_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-1, 1});
d0_t_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
d1_e_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
d2_e_n.GenerateTensorValue(GeneratorTensor_3<D2DataType>{0.0, 1.0});
@@ -289,7 +289,7 @@ int main(int argc, char* argv[])
break;
default:
a0_t_k_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-1, 1});
d0_t_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
d1_e_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
d2_e_n.GenerateTensorValue(GeneratorTensor_3<D2DataType>{0.0, 1.0});