MCPcopy Create free account
hub / github.com/PaddlePaddle/FastDeploy / get_gencode_flags

Function get_gencode_flags

custom_ops/setup_ops.py:155–179  ·  view source on GitHub ↗

Get gencode flags for current device or input.

(archs)

Source from the content-addressed store, hash-verified

153
154
155def get_gencode_flags(archs):
156 """
157 Get gencode flags for current device or input.
158 """
159 cc_s = get_sm_version(archs)
160 flags = []
161 for cc_val in cc_s:
162 if cc_val == 90:
163 arch_code = "90a"
164 flags += [
165 "-gencode",
166 f"arch=compute_{arch_code},code=sm_{arch_code}",
167 ]
168 elif cc_val == 100: # Assuming 100 is the code for Blackwell SM10.x
169 # Per NVIDIA dev blog, for CUTLASS and architecture-specific features on CC 10.0, use '100a'
170 # https://developer.nvidia.com/blog/nvidia-blackwell-and-nvidia-cuda-12-9-introduce-family-specific-architecture-features/
171 # "The CUTLASS build instructions specify using the a flag when building for devices of CC 9.0 and 10.0"
172 arch_code = "100a"
173 flags += [
174 "-gencode",
175 f"arch=compute_{arch_code},code=sm_{arch_code}",
176 ]
177 else:
178 flags += ["-gencode", f"arch=compute_{cc_val},code=sm_{cc_val}"]
179 return flags
180
181
182def find_end_files(directory, end_str):

Callers 1

setup_ops.pyFile · 0.85

Calls 1

get_sm_versionFunction · 0.70

Tested by

no test coverage detected