Merge commit '9fcc1ee9fd9730efd865f530afde505f2556954d' into develop

This commit is contained in:
assistant-librarian[bot]
2025-08-18 17:12:50 +00:00
parent d436787ed0
commit 68b20e1d4f
113 changed files with 610 additions and 531 deletions

View File

@@ -65,7 +65,6 @@ float gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
ck_tile::tuple<>,
CLayout,
ck_tile::element_wise::PassThrough,
CodegenPipelineProblem::kBlockSize,
TilePartitioner::MPerBlock,
TilePartitioner::NPerBlock,
M_Warp,
@@ -81,8 +80,8 @@ float gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
using Kernel = ck_tile::GemmKernel<TilePartitioner, CodegenGemmPipeline, GemmEpilogue>;
auto kargs = Kernel::MakeKernelArgs(args);
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch);
constexpr dim3 blocks = Kernel::BlockSize();
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch);
const dim3 blocks = Kernel::BlockSize();
if(!Kernel::IsSupportedArgument(kargs))
{
@@ -100,10 +99,8 @@ float gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
<< std::endl;
}
float ave_time =
ck_tile::launch_kernel(s,
ck_tile::make_kernel<blocks.x, GemmConfig::kBlockPerCu>(
Kernel{}, grids, blocks, 0, kargs));
float ave_time = ck_tile::launch_kernel(
s, ck_tile::make_kernel<GemmConfig::kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
return ave_time;
};