archive/blender-archive
Archived
1
1
Fork 0
Code Pull Requests Packages Activity

Cycles: Add an experimental CUDA kernel.

Browse Source 
Now we build 2 .cubins per architecture (e.g. kernel_sm_21.cubin, kernel_experimental_sm_21.cubin).
The experimental kernel can be used by switching to the Experimental Feature Set: http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Experimental_Features

This enables Subsurface Scattering and Correlated Multi Jitter Sampling on GPU, while keeping the stability and performance of the regular kernel.

Differential Revision: https://developer.blender.org/D762
Patch by Sergey and myself.

Developer / Builder Note:
CUDA Toolkit 6.5 is highly recommended for this, also note that building the experimental kernel requires a lot of system memory (~7-8GB).
This commit is contained in:
Thomas Dinges
2014-08-26 17:02:03 +02:00
parent f6e049cd5a
commit fb3f32760d
7 changed files with 88 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
intern/cycles/kernel/SConscript
View File

@@ -83,30 +83,35 @@ if env['WITH_BF_CYCLES_CUDA_BINARIES']:
dependencies = ['kernel.cu'] + kernel.Glob('*.h') + kernel.Glob('../util/*.h') + kernel.Glob('svm/*.h') + kernel.Glob('geom/*.h') + kernel.Glob('closure/*.h')
last_cubin_file = None
configs = (("kernel_%s.cubin", ''),
("kernel_experimental_%s.cubin", ' -D__KERNEL_CUDA_EXPERIMENTAL__'))
# add command for each cuda architecture
for arch in cuda_archs:
if cuda_version < 60 and arch == "sm_50":
print("Can't build kernel for CUDA sm_50 architecture, skipping")
continue
for config in configs:
# TODO(sergey): Use dict instead ocouple in order to increase readability?
name = config[0]
extra_flags = config[1]
cubin_file = os.path.join(build_dir, "kernel_%s.cubin" % arch)
cubin_file = os.path.join(build_dir, name % arch)
current_flags = nvcc_flags + extra_flags
if env['BF_CYCLES_CUDA_ENV']:
MS_SDK = "C:\\Program Files\\Microsoft SDKs\\Windows\\v7.1\\Bin\\SetEnv.cmd"
command = "\"%s\" & \"%s\" -arch=%s %s \"%s\" -o \"%s\"" % (MS_SDK, nvcc, arch, nvcc_flags, kernel_file, cubin_file)
else:
command = "\"%s\" -arch=%s %s \"%s\" -o \"%s\"" % (nvcc, arch, nvcc_flags, kernel_file, cubin_file)
if env['BF_CYCLES_CUDA_ENV']:
MS_SDK = "C:\\Program Files\\Microsoft SDKs\\Windows\\v7.1\\Bin\\SetEnv.cmd"
command = "\"%s\" & \"%s\" -arch=%s %s \"%s\" -o \"%s\"" % (MS_SDK, nvcc, arch, current_flags, kernel_file, cubin_file)
else:
command = "\"%s\" -arch=%s %s \"%s\" -o \"%s\"" % (nvcc, arch, current_flags, kernel_file, cubin_file)
kernel.Command(cubin_file, 'kernel.cu', command)
kernel.Depends(cubin_file, dependencies)
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
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')