4a4f043bc4235c046d2b58e00f2b80665ded11bf introduced a compile-time increase of cycles kernel.cpp with msvc from 1:07.86 on this box to 00:03:41.00
3c85e1ca1a916fe2ded9ab508f4cd55a2ee22549 increased this to 00:08:37.29
at ~9 minutes per kernel, cycles_kernel now takes about 50 minutes to compile.
this patch brings this back down to 00:01:12.92 (for kernel.cpp)
Given the method is marked ccl_device_noinline i expected a perf hit, but to my surprise I got about a 1% speedup out of it on pavillon_barcelone_cpu.
I tested the patch with the cpu render, it broke compiling with osl and i haven't tested gpu rendering at all. Posting this incomplete patch cause i'm out of time for the night hoping dingto will pickup where i left of tomorrow morning, if not ... no big deal, i'll finish the patch over the weekend.