forked from bartvdbraak/blender
Attempted fix for T40363: CUDA 30% slowdown in testbuilds compared to 2.70.
CMake had this --fast-math flag but scons not, makes a big difference on some files. Slightly slower rendering might still happen though, but it should not be this much.
This commit is contained in:
parent
033b4ffddd
commit
b33d83bf51
@ -74,7 +74,7 @@ if env['WITH_BF_CYCLES_CUDA_BINARIES']:
|
||||
|
||||
# nvcc flags
|
||||
nvcc_flags = "-m%s" % (bits)
|
||||
nvcc_flags += " --cubin --ptxas-options=\"-v\""
|
||||
nvcc_flags += " --cubin --ptxas-options=\"-v\" --use_fast_math"
|
||||
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\" -I \"%s\"" % (util_dir, svm_dir, geom_dir, closure_dir)
|
||||
|
Loading…
Reference in New Issue
Block a user