diff options
| author | Brecht Van Lommel <brechtvanlommel@pandora.be> | 2013-06-19 21:54:23 +0400 |
|---|---|---|
| committer | Brecht Van Lommel <brechtvanlommel@pandora.be> | 2013-06-19 21:54:23 +0400 |
| commit | 16204bd64759fddc940800f39fc91461ee340424 (patch) | |
| tree | e2f9f61775df316001e0bb328b15bf97fdf6555d /intern/cycles/kernel/kernel_jitter.h | |
| parent | a7416641e6cb634e4ac5fd3ad14be57e308b52fa (diff) | |
Cycles: prepare to make CUDA 5.0 the official version we use
* 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.
Diffstat (limited to 'intern/cycles/kernel/kernel_jitter.h')
| -rw-r--r-- | intern/cycles/kernel/kernel_jitter.h | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/intern/cycles/kernel/kernel_jitter.h b/intern/cycles/kernel/kernel_jitter.h index 17704b48cee..3e1a18ab469 100644 --- a/intern/cycles/kernel/kernel_jitter.h +++ b/intern/cycles/kernel/kernel_jitter.h @@ -137,7 +137,7 @@ __device_inline float cmj_randfloat(uint i, uint p) } #ifdef __CMJ__ -__device_noinline float cmj_sample_1D(int s, int N, int p) +__device float cmj_sample_1D(int s, int N, int p) { uint x = cmj_permute(s, N, p * 0x68bc21eb); float jx = cmj_randfloat(s, p * 0x967a889b); @@ -146,7 +146,7 @@ __device_noinline float cmj_sample_1D(int s, int N, int p) return (x + jx)*invN; } -__device_noinline void cmj_sample_2D(int s, int N, int p, float *fx, float *fy) +__device void cmj_sample_2D(int s, int N, int p, float *fx, float *fy) { int m = float_to_int(sqrtf(N)); int n = (N + m - 1)/m; |