Made picking of the number of compile threads more robust.

pull/1088/head
Davis King 7 years ago
parent d984a1f42d
commit a602126be3

@ -34,7 +34,7 @@ import platform
import subprocess
import multiprocessing
from distutils import log
from math import ceil
from math import ceil,floor
from setuptools import setup, Extension
from setuptools.command.build_ext import build_ext
@ -137,8 +137,7 @@ class CMakeBuild(build_ext):
else:
cmake_args += ['-DCMAKE_BUILD_TYPE=' + cfg]
# Do a parallel build
num_cores = int(ceil(multiprocessing.cpu_count()/2.0))
build_args += ['--', '-j'+str(num_cores)]
build_args += ['--', '-j'+str(num_available_cpu_cores(4))]
build_folder = os.path.abspath(self.build_temp)
@ -152,6 +151,20 @@ class CMakeBuild(build_ext):
print("Invoking CMake build: '{}'".format(['cmake', '--build', '.'] + build_args))
subprocess.check_call(['cmake', '--build', '.'] + build_args, cwd=build_folder)
def num_available_cpu_cores(ram_per_build_process_in_gb):
try:
mem_bytes = os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES')
mem_gib = mem_bytes/(1024.**3)
# Assume hyperthreading so divide by 2
num_cores = int(ceil(multiprocessing.cpu_count()/2.0))
# Require 4gb of ram per build thread.
mem_cores = int(floor(mem_gib/float(ram_per_build_process_in_gb)+0.5));
# We are limited either by RAM or CPU cores. So pick the limiting amount
# and return that.
return max(min(num_cores, mem_cores), 1)
except ValueError:
return 2 # just assume 2 if we can't get the os to tell us the right answer.
from setuptools.command.test import test as TestCommand
class PyTest(TestCommand):

Loading…
Cancel
Save