Cycles: Attempt to work around compilation of sm_20 and sm_21
authorSergey Sharybin <sergey.vfx@gmail.com>
Fri, 8 Sep 2017 16:37:54 +0000 (18:37 +0200)
committerSergey Sharybin <sergey.vfx@gmail.com>
Fri, 8 Sep 2017 16:37:54 +0000 (18:37 +0200)
commitae41a082889d45c696ab9426cfd921f61540c2ea
tree489042f04d331c501cc1d72f677920e64305418b
parent5d1a5001f4d039b421757419472287331ab03bd3
Cycles: Attempt to work around compilation of sm_20 and sm_21

Disabled forceinline for those architectures, which seems to be compiling
successfully more often.

There might be ~3% slowdown based on quick tests, but better be rendering
something rather than failing to compile kernels again and again.

Those architectures will be doomed for abandon once we'll switch to toolkit 9.
intern/cycles/kernel/kernel_compat_cuda.h