mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-09-22 18:09:06 +00:00
[feat] refactored extension module (#5298)
* [feat] refactored extension module * polish * polish * polish * polish * polish * polish * polish * polish * polish * polish
This commit is contained in:
24
extensions/layernorm/layernorm_cuda.py
Normal file
24
extensions/layernorm/layernorm_cuda.py
Normal file
@@ -0,0 +1,24 @@
|
||||
from ..cuda_extension import _CudaExtension
|
||||
from ..utils import append_nvcc_threads, get_cuda_cc_flag
|
||||
|
||||
|
||||
class LayerNormCudaExtension(_CudaExtension):
|
||||
def __init__(self):
|
||||
super().__init__(name="layernorm_cuda")
|
||||
|
||||
def sources_files(self):
|
||||
ret = [self.csrc_abs_path(fname) for fname in ["cuda/layer_norm_cuda.cpp", "cuda/layer_norm_cuda_kernel.cu"]]
|
||||
return ret
|
||||
|
||||
def include_dirs(self):
|
||||
ret = [self.get_cuda_home_include()]
|
||||
return ret
|
||||
|
||||
def cxx_flags(self):
|
||||
return ["-O3"] + self.version_dependent_macros
|
||||
|
||||
def nvcc_flags(self):
|
||||
extra_cuda_flags = ["-maxrregcount=50"]
|
||||
extra_cuda_flags.extend(get_cuda_cc_flag())
|
||||
ret = ["-O3", "--use_fast_math"] + extra_cuda_flags + self.version_dependent_macros
|
||||
return append_nvcc_threads(ret)
|
Reference in New Issue
Block a user