ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-07 00:04:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

update V3 2GB check

Browse Source
This commit is contained in:
Graner, Johannes
2025-12-01 13:18:19 +00:00
parent 2695863e85
commit 765fcfc067
1 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp
View File

@@ -1503,9 +1503,17 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffleV3
}
constexpr long_index_t TwoGB = (long_index_t{1} << 31);
if(!(arg.a_grid_desc_k0_m_k1_.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
arg.b_grid_desc_k0_n_k1_.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB &&
arg.c_grid_desc_m_n_.GetElementSpaceSize() * sizeof(CDataType) <= TwoGB))
const bool a_small_enough = arg.a_grid_desc_k0_m_k1_.GetElementSpaceSize() /
(arg.split_k_offset_a_hack_ ? arg.k_batch_ : 1) *
sizeof(ADataType) <=
TwoGB;
const bool b_small_enough = arg.b_grid_desc_k0_n_k1_.GetElementSpaceSize() /
(arg.split_k_offset_b_hack_ ? arg.k_batch_ : 1) *
sizeof(BDataType) <=
TwoGB;
const bool c_small_enough =
arg.c_grid_desc_m_n_.GetElementSpaceSize() * sizeof(CDataType) <= TwoGB;
if(!(a_small_enough && b_small_enough && c_small_enough))
{
return false;
}