wehub-resource-sync
80 lines
3.0 KiB
Python
80 lines
3.0 KiB
Python
import argparse
|
|
import os
|
|
import subprocess
|
|
import shutil
|
|
|
|
def generate_for_one_arch(exePath, qnnSDKPath, socId, hexagonArch, srcMNNPath, outputDir):
|
|
"""
|
|
Calls the MNN2QNNModel tool for a single architecture.
|
|
"""
|
|
print(f"Generating for socId: {socId}, hexagonArch: {hexagonArch}...")
|
|
command = [
|
|
exePath,
|
|
qnnSDKPath,
|
|
str(socId),
|
|
str(hexagonArch),
|
|
srcMNNPath,
|
|
outputDir
|
|
]
|
|
print(f"Executing: {' '.join(command)}")
|
|
|
|
try:
|
|
result = subprocess.run(command, check=True, capture_output=True)
|
|
# print(result.stdout)
|
|
print(f"Successfully generated for socId: {socId}, hexagonArch: {hexagonArch}")
|
|
return True
|
|
except subprocess.CalledProcessError as e:
|
|
print(f"Error generating for socId: {socId}, hexagonArch: {hexagonArch}")
|
|
print(f"Return code: {e.returncode}")
|
|
print(f"Output:\n{e.stdout}")
|
|
print(f"Error output:\n{e.stderr}")
|
|
return False
|
|
|
|
def generate_for_all(exePath, qnnSDKPath, srcMNNPath, outputDir):
|
|
"""
|
|
Iterates through all combinations and calls generate_for_one_arch for each.
|
|
"""
|
|
combinations = [
|
|
[36, '69'],
|
|
[42, '69'],
|
|
[43, '73'],
|
|
[57, '75'],
|
|
[69, '79']
|
|
]
|
|
|
|
success_count = 0
|
|
for socId, hexagonArch in combinations:
|
|
if generate_for_one_arch(exePath, qnnSDKPath, socId, hexagonArch, srcMNNPath, outputDir):
|
|
success_count += 1
|
|
|
|
print(f"\nGeneration complete. {success_count}/{len(combinations)} architectures succeeded.")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
parser = argparse.ArgumentParser(
|
|
description="A tool to generate QNN offline caches for a specified model under different Qualcomm hardware architectures by calling 'MNN2QNNModel'.",
|
|
formatter_class=argparse.RawTextHelpFormatter
|
|
)
|
|
|
|
parser.add_argument("--MNN2QNNModel_path", required=True, help="(Required) Path to the executable file 'MNN2QNNModel'.")
|
|
parser.add_argument("--qnn_sdk_path", required=True, help="(Required) Path to the QNN SDK directory.")
|
|
parser.add_argument("--src_mnn_path", required=True, help="(Required) Path to the source MNN model file.")
|
|
parser.add_argument("--output_dir", default=".", help="(Optional) Directory to save the generated files. Default is the current working directory.")
|
|
|
|
args = parser.parse_args()
|
|
|
|
if not os.path.isfile(args.MNN2QNNModel_path):
|
|
parser.error(f"MNN2QNNModel_path does not exist or is not a file: {args.MNN2QNNModel_path}")
|
|
|
|
if not os.path.isdir(args.qnn_sdk_path):
|
|
parser.error(f"qnn_sdk_path does not exist or is not a directory: {args.qnn_sdk_path}")
|
|
|
|
if not os.path.isfile(args.src_mnn_path):
|
|
parser.error(f"src_mnn_path does not exist or is not a file: {args.src_mnn_path}")
|
|
|
|
if os.path.exists(args.output_dir):
|
|
shutil.rmtree(args.output_dir)
|
|
os.makedirs(args.output_dir)
|
|
|
|
generate_for_all(args.MNN2QNNModel_path, args.qnn_sdk_path, args.src_mnn_path, args.output_dir)
|