liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Small improvements for sleef functions

Browse Source
This commit is contained in:
heckflosse
2018-01-14 22:34:38 +01:00
parent cbcb096339
commit 24a762c2c2
2 changed files with 14 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
rtengine/sleef.c
View File

@@ -12,10 +12,8 @@
#include <assert.h>
#include <stdint.h>
//#include <math.h>
#include "rt_math.h"
//#include <bits/nan.h>
//#include <bits/inf.h>
#include "opthelper.h"
#define PI4_A .7853981554508209228515625
#define PI4_B .794662735614792836713604629039764404296875e-8
@@ -1005,6 +1003,10 @@ __inline float xsinf(float d) {
}
__inline float xcosf(float d) {
#ifdef __SSE2__
// faster than scalar version
return xcosf(_mm_set_ss(d))[0];
#else
int q;
float u, s;
@@ -1027,9 +1029,15 @@ __inline float xcosf(float d) {
u = mlaf(s, u * d, d);
return u;
#endif
}
__inline float2 xsincosf(float d) {
#ifdef __SSE2__
// faster than scalar version
vfloat2 res = xsincosf(_mm_set_ss(d));
return {res.x[0], res.y[0]};
#else
int q;
float u, s, t;
float2 r;
@@ -1069,6 +1077,7 @@ __inline float2 xsincosf(float d) {
if (xisinff(d)) { r.x = r.y = rtengine::RT_NAN_F; }
return r;
#endif
}
__inline float xtanf(float d) {

16
rtengine/sleefsseavx.c
View File

@@ -12,10 +12,6 @@
#define SLEEFSSEAVX
#include <assert.h>
//#include <math.h>
//#include <bits/nan.h>
//#include <bits/inf.h>
//#include "sleefsseavx.h"
#include "rt_math.h"
#ifdef __SSE2__
#include "helpersse2.h"
@@ -30,8 +26,6 @@
#define INLINE inline
#endif
//
#define PI4_A .7853981554508209228515625
#define PI4_B .794662735614792836713604629039764404296875e-8
#define PI4_C .306161699786838294306516483068750264552437361480769e-16
@@ -41,8 +35,6 @@
#define L2L .28235290563031577122588448175013436025525412068e-12
#define R_LN2 1.442695040888963407359924681001892137426645954152985934135449406931
//
#define PI4_Af 0.78515625f
#define PI4_Bf 0.00024127960205078125f
#define PI4_Cf 6.3329935073852539062e-07f
@@ -55,8 +47,6 @@
#define INFINITYf ((float)rtengine::RT_INFINITY)
#define NANf ((float)rtengine::RT_NAN)
//
static INLINE vdouble vadd3(vdouble v0, vdouble v1, vdouble v2) {
return vadd(vadd(v0, v1), v2);
}
@@ -1323,10 +1313,8 @@ static INLINE vfloat xexpf(vfloat d) {
u = vldexpf(u, q);
u = vself(vmaskf_isminf(d), vcast_vf_f(0.0f), u);
// -104.0
u = vself(vmaskf_gt(vcast_vf_f(-104), d), vcast_vf_f(0), u);
return u;
// -104.0
return vselfnotzero(vmaskf_gt(vcast_vf_f(-104.f), d), u);
}
static INLINE vfloat xexpfNoCheck(vfloat d) { // this version does not check input values. Use it only when you know the input values are > -104.f e.g. when filling a lookup table