diff --git a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp index 1e82d59d51..6598e5eae8 100644 --- a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp +++ b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp @@ -37,7 +37,7 @@ using tile_grouped_conv_bwd_weight_bf16_instances_opt = std::tuple< GroupedConvolutionBackwardWeightInvoker, GroupedConvolutionBackwardWeightInvoker, - // Skinny kernels. + // Skinny GEMM-K kernels. GroupedConvolutionBackwardWeightInvoker, GroupedConvolutionBackwardWeightInvoker, GroupedConvolutionBackwardWeightInvoker, diff --git a/script/benchmark_ck_vs_ck_tile.py b/script/benchmark_ck_vs_ck_tile.py index dbd7b08ca6..7d5b71084a 100755 --- a/script/benchmark_ck_vs_ck_tile.py +++ b/script/benchmark_ck_vs_ck_tile.py @@ -234,6 +234,9 @@ def main(): profiler_commands = get_profiler_commands(args.input_file) print(f"Got {len(profiler_commands)} unique commands to run.") + if not os.path.exists(args.results_path): + os.makedirs(args.results_path) + results_file = os.path.join(args.results_path, f"ck_vs_ck_tile_results_{os.getpid()}.txt") for i, cmd in enumerate(profiler_commands):