github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-04-29 04:05:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

updated c++17 compiler flags (#4983)

Browse Source
This commit is contained in:
アマデウス 2023-10-27 18:19:56 +08:00 committed by GitHub
parent 1db6727678
commit 4e4a10c97d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 15 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
examples/community/roberta/preprocessing/Makefile
View File

@ -1,4 +1,4 @@
CXXFLAGS += -O3 -Wall -shared -std=c++14 -fPIC -fdiagnostics-color CXXFLAGS += -O3 -Wall -shared -std=c++14 -std=c++17 -fPIC -fdiagnostics-color
CPPFLAGS += $(shell python3 -m pybind11 --includes) CPPFLAGS += $(shell python3 -m pybind11 --includes)
LIBNAME = mask LIBNAME = mask
LIBEXT = $(shell python3-config --extension-suffix) LIBEXT = $(shell python3-config --extension-suffix)

12
op_builder/cpu_adam.py
View File

@ -21,12 +21,22 @@ class CPUAdamBuilder(Builder):
return [self.csrc_abs_path("includes"), self.get_cuda_home_include()] return [self.csrc_abs_path("includes"), self.get_cuda_home_include()]
def cxx_flags(self): def cxx_flags(self):
extra_cxx_flags = ["-std=c++14", "-lcudart", "-lcublas", "-g", "-Wno-reorder", "-fopenmp", "-march=native"] extra_cxx_flags = [
"-std=c++14",
"-std=c++17",
"-lcudart",
"-lcublas",
"-g",
"-Wno-reorder",
"-fopenmp",
"-march=native",
]
return ["-O3"] + self.version_dependent_macros + extra_cxx_flags return ["-O3"] + self.version_dependent_macros + extra_cxx_flags
def nvcc_flags(self): def nvcc_flags(self):
extra_cuda_flags = [ extra_cuda_flags = [
"-std=c++14", "-std=c++14",
"-std=c++17",
"-U__CUDA_NO_HALF_OPERATORS__", "-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__", "-U__CUDA_NO_HALF_CONVERSIONS__",
"-U__CUDA_NO_HALF2_OPERATORS__", "-U__CUDA_NO_HALF2_OPERATORS__",

2
op_builder/gptq.py
View File

@ -37,12 +37,12 @@ class GPTQBuilder(Builder):
extra_cuda_flags = [ extra_cuda_flags = [
"-v", "-v",
"-std=c++14", "-std=c++14",
"-std=c++17",
"-U__CUDA_NO_HALF_OPERATORS__", "-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__", "-U__CUDA_NO_HALF_CONVERSIONS__",
"-U__CUDA_NO_HALF2_OPERATORS__", "-U__CUDA_NO_HALF2_OPERATORS__",
"-DTHRUST_IGNORE_CUB_VERSION_CHECK", "-DTHRUST_IGNORE_CUB_VERSION_CHECK",
"-lcublas", "-lcublas",
"-std=c++17",
] ]
for arch in torch.cuda.get_arch_list(): for arch in torch.cuda.get_arch_list():

1
op_builder/multi_head_attn.py
View File

@ -35,6 +35,7 @@ class MultiHeadAttnBuilder(Builder):
def nvcc_flags(self): def nvcc_flags(self):
extra_cuda_flags = [ extra_cuda_flags = [
"-std=c++14", "-std=c++14",
"-std=c++17",
"-U__CUDA_NO_HALF_OPERATORS__", "-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__", "-U__CUDA_NO_HALF_CONVERSIONS__",
"-U__CUDA_NO_HALF2_OPERATORS__", "-U__CUDA_NO_HALF2_OPERATORS__",

1
op_builder/scaled_masked_softmax.py
View File

@ -25,6 +25,7 @@ class ScaledMaskedSoftmaxBuilder(Builder):
def nvcc_flags(self): def nvcc_flags(self):
extra_cuda_flags = [ extra_cuda_flags = [
"-std=c++14", "-std=c++14",
"-std=c++17",
"-U__CUDA_NO_HALF_OPERATORS__", "-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__", "-U__CUDA_NO_HALF_CONVERSIONS__",
"-U__CUDA_NO_HALF2_OPERATORS__", "-U__CUDA_NO_HALF2_OPERATORS__",