forked from bartvdbraak/blender
16204bd647
* Add CUDA compiler version detection to cmake/scons/runtime * Remove noinline in kernel_shader.h and reenable --use_fast_math if CUDA 5.x is used, these were workarounds for CUDA 4.2 bugs * Change max number of registers to 32 for sm 2.x (based on performance tests from Martijn Berger and confirmed here), and also for NVidia OpenCL. Overall it seems that with these changes and the latest CUDA 5.0 download, that performance is as good as or better than the 2.67b release with the scenes and graphics cards I tested.
128 lines
4.4 KiB
Python
128 lines
4.4 KiB
Python
#!/usr/bin/env python
|
|
#
|
|
# ***** BEGIN GPL LICENSE BLOCK *****
|
|
#
|
|
# This program is free software; you can redistribute it and/or
|
|
# modify it under the terms of the GNU General Public License
|
|
# as published by the Free Software Foundation; either version 2
|
|
# of the License, or (at your option) any later version.
|
|
#
|
|
# This program is distributed in the hope that it will be useful,
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
# GNU General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License
|
|
# along with this program; if not, write to the Free Software Foundation,
|
|
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
#
|
|
# The Original Code is Copyright (C) 2011, Blender Foundation
|
|
# All rights reserved.
|
|
#
|
|
# The Original Code is: all of this file.
|
|
#
|
|
# Contributor(s): Nathan Letwory.
|
|
#
|
|
# ***** END GPL LICENSE BLOCK *****
|
|
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import os
|
|
import Blender as B
|
|
|
|
def normpath(path):
|
|
return os.path.abspath(os.path.normpath(path))
|
|
|
|
Import ('env')
|
|
|
|
kernel_binaries = []
|
|
|
|
#Bitness
|
|
if B.bitness == 32:
|
|
bits = 32
|
|
else:
|
|
bits = 64
|
|
|
|
if env['WITH_BF_CYCLES_CUDA_BINARIES']:
|
|
kernel = env.Clone()
|
|
|
|
# cuda info
|
|
nvcc = env['BF_CYCLES_CUDA_NVCC']
|
|
cuda_archs = env['BF_CYCLES_CUDA_BINARIES_ARCH']
|
|
|
|
# build directory
|
|
root_build_dir = normpath(env['BF_BUILDDIR'])
|
|
build_dir = os.path.join(root_build_dir, 'intern/cycles/kernel')
|
|
|
|
# source directories and files
|
|
source_dir = Dir('.').srcnode().path
|
|
kernel_file = os.path.join(source_dir, "kernel.cu")
|
|
util_dir = os.path.join(source_dir, "../util")
|
|
svm_dir = os.path.join(source_dir, "../svm")
|
|
closure_dir = os.path.join(source_dir, "../closure")
|
|
|
|
# get CUDA version
|
|
nvcc_pipe = subprocess.Popen([nvcc, "--version"],stdout=subprocess.PIPE,stderr=subprocess.PIPE)
|
|
output, erroroutput = nvcc_pipe.communicate()
|
|
cuda_major_minor = re.findall(r'release (\d+).(\d+)', output)[0]
|
|
cuda_version = int(cuda_major_minor[0])*10 + int(cuda_major_minor[1])
|
|
|
|
if cuda_version != 50:
|
|
print("CUDA version %d.%d detected, build may succeed but only CUDA 5.0 is officially supported." % (cuda_version/10, cuda_version%10))
|
|
|
|
# nvcc flags
|
|
nvcc_flags = "-m%s" % (bits)
|
|
nvcc_flags += " --cubin --ptxas-options=\"-v\""
|
|
nvcc_flags += " -D__KERNEL_CUDA_VERSION__=%d" % (cuda_version)
|
|
nvcc_flags += " -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -DNVCC"
|
|
nvcc_flags += " -I \"%s\" -I \"%s\" -I \"%s\"" % (util_dir, svm_dir, closure_dir)
|
|
|
|
# dependencies
|
|
dependencies = ['kernel.cu'] + kernel.Glob('*.h') + kernel.Glob('../util/*.h') + kernel.Glob('svm/*.h') + kernel.Glob('closure/*.h')
|
|
last_cubin_file = None
|
|
|
|
# add command for each cuda architecture
|
|
for arch in cuda_archs:
|
|
cubin_file = os.path.join(build_dir, "kernel_%s.cubin" % arch)
|
|
|
|
# build flags depending on CUDA version and arch
|
|
if cuda_version < 50:
|
|
# CUDA 4.x
|
|
if arch.startswith("sm_1"):
|
|
# sm_1x
|
|
cuda_arch_flags = "--maxrregcount=24 --opencc-options -OPT:Olimit=0"
|
|
elif arch.startswith("sm_2"):
|
|
# sm_2x
|
|
cuda_arch_flags = "--maxrregcount=24"
|
|
else:
|
|
# sm_3x
|
|
cuda_arch_flags = "--maxrregcount=32"
|
|
else:
|
|
# CUDA 5.x
|
|
if arch.startswith("sm_1"):
|
|
# sm_1x
|
|
cuda_arch_flags = "--maxrregcount=24 --opencc-options -OPT:Olimit=0 --use_fast_math"
|
|
elif arch.startswith("sm_2"):
|
|
# sm_2x
|
|
cuda_arch_flags = "--maxrregcount=32 --use_fast_math"
|
|
else:
|
|
# sm_3x
|
|
cuda_arch_flags = "--maxrregcount=32 --use_fast_math"
|
|
|
|
command = "\"%s\" -arch=%s %s %s \"%s\" -o \"%s\"" % (nvcc, arch, nvcc_flags, cuda_arch_flags, kernel_file, cubin_file)
|
|
|
|
kernel.Command(cubin_file, 'kernel.cu', command)
|
|
kernel.Depends(cubin_file, dependencies)
|
|
|
|
kernel_binaries.append(cubin_file)
|
|
|
|
if not env['WITH_BF_CYCLES_CUDA_THREADED_COMPILE']:
|
|
# trick to compile one kernel at a time to reduce memory usage
|
|
if last_cubin_file:
|
|
kernel.Depends(cubin_file, last_cubin_file)
|
|
last_cubin_file = cubin_file
|
|
|
|
Return('kernel_binaries')
|
|
|