diff options
| author | Sergey Sharybin <sergey.vfx@gmail.com> | 2015-04-05 20:13:10 +0300 |
|---|---|---|
| committer | Sergey Sharybin <sergey.vfx@gmail.com> | 2015-04-06 10:49:44 +0300 |
| commit | a9bb8d8a73cad2edd96297c868ae3a3ab4b04e27 (patch) | |
| tree | 42052afc9d53febf2e847b31492b210ecaee6b94 /intern/cycles/util/util_math_fast.h | |
| parent | 9f8ffd478e92ad69c334cc3281d8453ca9945782 (diff) | |
Cycles: de-duplicate fast/approximate erf function calculation
Our own implementation is in fact the same performance as in fast_math from
OpenShadingLanguage, but implementation from fast_math is using explicit madd
function, which increases chance of compiler deciding to use intrinsics.
Diffstat (limited to 'intern/cycles/util/util_math_fast.h')
| -rw-r--r-- | intern/cycles/util/util_math_fast.h | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/intern/cycles/util/util_math_fast.h b/intern/cycles/util/util_math_fast.h index 4ad81e9c015..1acceee22a5 100644 --- a/intern/cycles/util/util_math_fast.h +++ b/intern/cycles/util/util_math_fast.h @@ -539,7 +539,7 @@ ccl_device float fast_safe_powf(float x, float y) * bsdf_microfaset.h. */ -ccl_device float fast_erff(float x) +ccl_device_inline float fast_erff(float x) { /* Examined 1082130433 values of erff on [0,4]: 1.93715e-06 max error. */ /* Abramowitz and Stegun, 7.1.28. */ @@ -570,7 +570,7 @@ ccl_device_inline float fast_erfcf(float x) return 1.0f - fast_erff(x); } -ccl_device float fast_ierff(float x) +ccl_device_inline float fast_ierff(float x) { /* From: Approximating the erfinv function by Mike Giles. */ /* To avoid trouble at the limit, clamp input to 1-eps. */ |