ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 04:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixes to "General 2D Reduction Kernel" (#2535) (#2656)

Browse Source 
* fix reduce2d

- revret the combine_partial_results() chnages
- remove auto from function def

* clang-format
This commit is contained in:
Yashvardhan Agarwal
2025-08-11 16:01:33 +03:00
committed by GitHub
parent 1e1ee758fa
commit 191c62967b
4 changed files with 7 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
include/ck_tile/ops/reduce/block/block_reduce2d.hpp
View File

@@ -183,16 +183,7 @@ struct BlockReduce2dSync
// pull data from remote lane
const auto v_remote = warp_shuffle(v_local, src_lane);
// For reduce, use combine_partial_results for operations that require it
if constexpr(ReduceFunc::requires_special_combine)
{
v_local = reduce_func.combine_partial_results(v_local, v_remote);
}
else
{
v_local = reduce_func(v_local, v_remote);
}
v_local = reduce_func(v_local, v_remote);
});
}
});
@@ -309,16 +300,7 @@ struct BlockReduce2dCrossWarpSync
static_for<0, num_reduce_warps - 1, 1>{}([&](auto i_1_n1) {
constexpr auto i_1 = number<i_1_n1 + 1>{};
const DataType v_remote = all_scratch[i_0 * num_reduce_warps + i_1];
// For reduce, use combine_partial_results for operations that require it
if constexpr(ReduceFunc::requires_special_combine)
{
v_local = reduce_func.combine_partial_results(v_local, v_remote);
}
else
{
v_local = reduce_func(v_local, v_remote);
}
v_local = reduce_func(v_local, v_remote);
});
y_tensor.get_thread_buffer()(i_0) = v_local;