mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-02 04:31:25 +00:00
[CK_Tile] Fix gemm kernel for 4,64,16 and 64,4,16 warp tile sizes (#2262)
* debugging issue * debugging issue * debugging * debugging * reverting debugging code * clang formatted * updating default_config.json * fix ci failure * clang formatted
This commit is contained in:
@@ -149,7 +149,9 @@ struct DefaultGemm2DEpilogue : public Default2DEpilogue<Problem_, Policy_>
|
||||
else
|
||||
{
|
||||
// In this case each thread has just a single item in Ndim
|
||||
return WG::WarpGemmAttribute::Impl::kCNLane / WG::kN;
|
||||
return (WG::WarpGemmAttribute::Impl::kCNLane *
|
||||
WG::WarpGemmAttribute::Impl::kBNBlock) /
|
||||
WG::kN;
|
||||
}
|
||||
}
|
||||
// M is contiguous dimension
|
||||
@@ -158,7 +160,9 @@ struct DefaultGemm2DEpilogue : public Default2DEpilogue<Problem_, Policy_>
|
||||
if constexpr(isCTransposed)
|
||||
{
|
||||
// In this case each thread has just a single item in Mdim
|
||||
return WG::WarpGemmAttribute::Impl::kCNLane / WG::kN;
|
||||
return (WG::WarpGemmAttribute::Impl::kCNLane *
|
||||
WG::WarpGemmAttribute::Impl::kAMBlock) /
|
||||
WG::kN;
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user