From 159bcf6750f529e35c5423442244f05b19f46ea3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ville=20Pietil=C3=A4?= <ville.pietila@amd.com>
Date: Tue, 21 Oct 2025 06:38:49 +0000
Subject: [PATCH] Small script improvements.

---
 .../gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp    | 2 +-
 script/benchmark_ck_vs_ck_tile.py                              | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)

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<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
     GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,    true,  CK_TILE_PIPELINE_COMPUTE_V4>,
 
-    // Skinny kernels.
+    // Skinny GEMM-K kernels.
     GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
     GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,    true,  CK_TILE_PIPELINE_COMPUTE_V4>,
     GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        1,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
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):