ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-11 17:51:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix profiler command conversion and execution scripts.

Browse Source
This commit is contained in:
Ville Pietilä
2026-01-19 07:35:41 -05:00
parent 5f2f7b5ee2
commit a5c81472d0
2 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
script/RetinaNet/convert_pytorch_to_ck.py
View File

@@ -20,7 +20,7 @@ class Colors:
# Configuration constants
DATA_TYPE = 1 # FP16
LAYOUT = 3 # NGCHW_GKCYX_NGKHW
INDEX_TYPE = 0 # 32-bit (only for forward)
INDEX_TYPE = 1 # 64-bit (only for forward)
VERIFY = 0 # No verification
INIT = 1 # Integer initialization
LOG = 0 # No log printing
@@ -97,7 +97,7 @@ class PyTorchToCKConverter:
dilation = [dilation] * num_dim_spatial
return {
'operation_type': 'profile_grouped_conv_fwd',
'operation_type': 'grouped_conv_fwd',
'profiler_args': {
'data_type': DATA_TYPE,
'layout': LAYOUT,
@@ -176,7 +176,7 @@ class PyTorchToCKConverter:
# Backward data (computing gradient w.r.t. input)
if output_mask[0]: # grad_input
results.append({
'operation_type': 'profile_grouped_conv_bwd_data',
'operation_type': 'grouped_conv_bwd_data',
'profiler_args': {
'data_type': DATA_TYPE,
'layout': LAYOUT,
@@ -207,7 +207,7 @@ class PyTorchToCKConverter:
# Backward weight (computing gradient w.r.t. weight)
if output_mask[1]: # grad_weight
results.append({
'operation_type': 'profile_grouped_conv_bwd_weight',
'operation_type': 'grouped_conv_bwd_weight',
'profiler_args': {
'data_type': DATA_TYPE,
'layout': LAYOUT,
@@ -255,9 +255,9 @@ class PyTorchToCKConverter:
results = self.extract_backward_conv_params(op)
if results:
for result in results:
if result['operation_type'] == 'profile_grouped_conv_bwd_data':
if result['operation_type'] == 'grouped_conv_bwd_data':
self.stats['backward_data'] += 1
elif result['operation_type'] == 'profile_grouped_conv_bwd_weight':
elif result['operation_type'] == 'grouped_conv_bwd_weight':
self.stats['backward_weight'] += 1
return results
else:
@@ -291,7 +291,7 @@ class PyTorchToCKConverter:
if results:
self.converted_ops.extend(results)
for result in results:
op_type = result['operation_type'].replace('profile_grouped_conv_', '')
op_type = result['operation_type'].replace('grouped_conv_', '')
print(f"{Colors.GREEN}✓{Colors.RESET} [{i}/{len(pytorch_ops)}] Converted: {op_name} (rank {priority}) → {op_type}")
else:
print(f"{Colors.YELLOW}⚠{Colors.RESET} [{i}/{len(pytorch_ops)}] Skipped: {op_name} (rank {priority})")
@@ -327,13 +327,13 @@ def main():
description='Convert PyTorch convolution JSON to CK Profiler configuration JSON'
)
parser.add_argument(
'input',
'--input',
nargs='?',
default='build/test-data/conv_repros_ir.json',
help='Input PyTorch JSON file (default: build/test-data/conv_repros_ir.json)'
)
parser.add_argument(
'output',
'--output',
nargs='?',
default='ck_profiler_configs.json',
help='Output CK Profiler JSON file (default: ck_profiler_configs.json)'

4
script/RetinaNet/run_ck_profiler.py
View File

@@ -43,10 +43,10 @@ class CKProfilerExecutor:
args = config['profiler_args']
# Build argument list
cmd = [str(self.profiler_path / op_type)]
cmd = [str(self.profiler_path), op_type]
# Add arguments based on operation type
if op_type == 'profile_grouped_conv_fwd':
if op_type == 'grouped_conv_fwd':
# Forward convolution arguments
cmd.extend([
str(args['data_type']),