aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #736 from egaburov/native_trigonometry

Browse Source 
Native trigonometry
This commit is contained in:
Dmitry Babokin
2014-02-20 19:18:35 +03:00
parent c3b041f7ae 193bba77b0
commit f280b32fa4
43 changed files with 1037 additions and 188 deletions
Download Patch File Download Diff File Expand all files Collapse all files

270
stdlib.ispc
View File

@@ -2335,8 +2335,12 @@ static inline uniform float frexp(uniform float x, uniform int * uniform pw2) {
__declspec(safe)
static inline float sin(float x_full) {
if (__math_lib == __math_lib_svml) {
return __svml_sinf(x_full);
if (__have_native_trigonometry)
{
return __sin_varying_float(x_full);
}
else if (__math_lib == __math_lib_svml) {
return __svml_sinf(x_full);
}
else if (__math_lib == __math_lib_system) {
float ret;
@@ -2397,7 +2401,11 @@ static inline float sin(float x_full) {
__declspec(safe)
static inline uniform float sin(uniform float x_full) {
if (__math_lib == __math_lib_system ||
if (__have_native_trigonometry)
{
return __sin_uniform_float(x_full);
}
else if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
return __stdlib_sinf(x_full);
}
@@ -2462,25 +2470,29 @@ static inline uniform float sin(uniform float x_full) {
__declspec(safe)
static inline float asin(float x) {
bool isneg = x < 0;
x = abs(x);
static inline float asin(float x0) {
bool isneg = x0< 0;
float x = abs(x0);
bool isnan = (x > 1);
float v;
if (__math_lib == __math_lib_svml) {
return __svml_asinf(x);
if (__have_native_trigonometry)
{
return __asin_varying_float(x0);
}
else if (__math_lib == __math_lib_svml) {
return __svml_asinf(x0);
}
else if (__math_lib == __math_lib_system) {
float ret;
foreach_active (i) {
uniform float r = __stdlib_asinf(extract(x, i));
uniform float r = __stdlib_asinf(extract(x0, i));
ret = insert(ret, i, r);
}
return ret;
}
else if (__math_lib == __math_lib_ispc)
{
// sollya
// fpminimax(((asin(x)-pi/2)/-sqrt(1-x)), [|0,1,2,3,4,5,6,7,8,9,10|],
// [|single...|], [1e-20;.9999999999999999]);
@@ -2496,7 +2508,9 @@ static inline float asin(float x) {
x * (3.05023305118083953857421875e-2f +
x * (-1.2897425331175327301025390625e-2f +
x * 2.38926825113594532012939453125e-3f)))))))));
}
else if (__math_lib == __math_lib_ispc_fast)
{
// sollya
// fpminimax(((asin(x)-pi/2)/-sqrt(1-x)), [|0,1,2,3,4,5|],[|single...|],
// [1e-20;.9999999999999999]);
@@ -2507,6 +2521,7 @@ static inline float asin(float x) {
x * (-4.489909112453460693359375e-2f +
x * (1.928029954433441162109375e-2f +
x * (-4.3095736764371395111083984375e-3f)))));
}
v *= -sqrt(1.f - x);
v = v + 1.57079637050628662109375;
@@ -2521,18 +2536,21 @@ static inline float asin(float x) {
__declspec(safe)
static inline uniform float asin(uniform float x) {
uniform bool isneg = x < 0;
x = abs(x);
static inline uniform float asin(uniform float x0) {
uniform bool isneg = x0 < 0;
uniform float x = abs(x0);
uniform bool isnan = (x > 1);
uniform float v;
if (__math_lib == __math_lib_svml ||
if (__have_native_trigonometry)
{
return __asin_uniform_float(x0);
}
else if (__math_lib == __math_lib_svml ||
__math_lib == __math_lib_system) {
return __stdlib_asinf(x);
return __stdlib_asinf(x0);
}
else if (__math_lib == __math_lib_ispc)
{
// sollya
// fpminimax(((asin(x)-pi/2)/-sqrt(1-x)), [|0,1,2,3,4,5,6,7,8,9,10|],
// [|single...|], [1e-20;.9999999999999999]);
@@ -2548,7 +2566,9 @@ static inline uniform float asin(uniform float x) {
x * (3.05023305118083953857421875e-2f +
x * (-1.2897425331175327301025390625e-2f +
x * 2.38926825113594532012939453125e-3f)))))))));
}
else if (__math_lib == __math_lib_ispc_fast)
{
// sollya
// fpminimax(((asin(x)-pi/2)/-sqrt(1-x)), [|0,1,2,3,4,5|],[|single...|],
// [1e-20;.9999999999999999]);
@@ -2559,6 +2579,7 @@ static inline uniform float asin(uniform float x) {
x * (-4.489909112453460693359375e-2f +
x * (1.928029954433441162109375e-2f +
x * (-4.3095736764371395111083984375e-3f)))));
}
v *= -sqrt(1.f - x);
v = v + 1.57079637050628662109375;
@@ -2574,6 +2595,10 @@ static inline uniform float asin(uniform float x) {
__declspec(safe)
static inline float cos(float x_full) {
if (__have_native_trigonometry)
{
return __cos_varying_float(x_full);
}
if (__math_lib == __math_lib_svml) {
return __svml_cosf(x_full);
}
@@ -2635,7 +2660,11 @@ static inline float cos(float x_full) {
__declspec(safe)
static inline uniform float cos(uniform float x_full) {
if (__math_lib == __math_lib_system ||
if (__have_native_trigonometry)
{
return __cos_uniform_float(x_full);
}
else if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
return __stdlib_cosf(x_full);
}
@@ -2700,22 +2729,34 @@ static inline uniform float cos(uniform float x_full) {
__declspec(safe)
static inline float acos(float v) {
if (__have_native_trigonometry)
return __acos_varying_float(v);
else
return 1.57079637050628662109375 - asin(v);
}
__declspec(safe)
static inline double acos(const double v) {
if (__have_native_trigonometry)
return __acos_varying_double(v);
else
return 1.57079637050628662109375d0 - asin(v);
}
__declspec(safe)
static inline uniform float acos(uniform float v) {
if (__have_native_trigonometry)
return __acos_uniform_float(v);
else
return 1.57079637050628662109375 - asin(v);
}
__declspec(safe)
static inline uniform double acos(const uniform double v) {
if (__have_native_trigonometry)
return __acos_uniform_double(v);
else
return 1.57079637050628662109375d0 - asin(v);
}
@@ -2723,6 +2764,10 @@ static inline uniform double acos(const uniform double v) {
__declspec(safe)
static inline void sincos(float x_full, varying float * uniform sin_result,
varying float * uniform cos_result) {
if (__have_native_trigonometry)
{
__sincos_varying_float(x_full,sin_result,cos_result);
}
if (__math_lib == __math_lib_svml) {
__svml_sincosf(x_full, sin_result, cos_result);
}
@@ -2793,6 +2838,10 @@ static inline void sincos(float x_full, varying float * uniform sin_result,
__declspec(safe)
static inline void sincos(uniform float x_full, uniform float * uniform sin_result,
uniform float * uniform cos_result) {
if (__have_native_trigonometry)
{
__sincos_uniform_float(x_full, sin_result, cos_result);
}
if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
__stdlib_sincosf(x_full, sin_result, cos_result);
@@ -2855,7 +2904,11 @@ static inline void sincos(uniform float x_full, uniform float * uniform sin_resu
__declspec(safe)
static inline float tan(float x_full) {
if (__math_lib == __math_lib_svml) {
if (__have_native_trigonometry)
{
return __tan_varying_float(x_full);
}
else if (__math_lib == __math_lib_svml) {
return __svml_tanf(x_full);
}
else if (__math_lib == __math_lib_system) {
@@ -2934,7 +2987,11 @@ static inline float tan(float x_full) {
__declspec(safe)
static inline uniform float tan(uniform float x_full) {
if (__math_lib == __math_lib_system ||
if (__have_native_trigonometry)
{
return __tan_uniform_float(x_full);
}
else if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
return __stdlib_tanf(x_full);
}
@@ -3006,7 +3063,11 @@ static inline uniform float tan(uniform float x_full) {
__declspec(safe)
static inline float atan(float x_full) {
if (__math_lib == __math_lib_svml) {
if (__have_native_trigonometry)
{
return __atan_varying_float(x_full);
}
else if (__math_lib == __math_lib_svml) {
return __svml_atanf(x_full);
}
else if (__math_lib == __math_lib_system) {
@@ -3057,7 +3118,11 @@ static inline float atan(float x_full) {
__declspec(safe)
static inline uniform float atan(uniform float x_full) {
if (__math_lib == __math_lib_system ||
if (__have_native_trigonometry)
{
return __atan_uniform_float(x_full);
}
else if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
return __stdlib_atanf(x_full);
}
@@ -3101,7 +3166,11 @@ static inline uniform float atan(uniform float x_full) {
__declspec(safe)
static inline float atan2(float y, float x) {
if (__math_lib == __math_lib_svml) {
if (__have_native_trigonometry)
{
return __atan2_varying_float(y,x);
}
else if (__math_lib == __math_lib_svml) {
return __svml_atan2f(y, x);
}
else if (__math_lib == __math_lib_system) {
@@ -3140,7 +3209,11 @@ static inline float atan2(float y, float x) {
__declspec(safe)
static inline uniform float atan2(uniform float y, uniform float x) {
if (__math_lib == __math_lib_system ||
if (__have_native_trigonometry)
{
return __atan2_uniform_float(y,x);
}
else if (__math_lib == __math_lib_system ||
__math_lib == __math_lib_svml) {
return __stdlib_atan2f(y, x);
}
@@ -3675,12 +3748,14 @@ static inline uniform double frexp(uniform double x, uniform int * uniform pw2)
__declspec(safe)
static inline double sin(double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_trigonometry)
{
return __sin_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_sind(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return sin((float)x);
else {
double ret;
foreach_active (i) {
@@ -3690,23 +3765,46 @@ static inline double sin(double x) {
return ret;
}
}
__declspec(safe)
static inline double asin(double x) {
if (__have_native_trigonometry)
{
return __asin_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_asind(x);
}
else {
double ret;
foreach_active (i) {
uniform double r = __stdlib_asin(extract(x, i));
ret = insert(ret, i, r);
}
return ret;
}
}
__declspec(safe)
static inline uniform double sin(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return sin((float)x);
if (__have_native_trigonometry)
{
return __sin_uniform_double(x);
}
else
return __stdlib_sin(x);
}
__declspec(safe)
static inline double asin(const double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_trigonometry)
{
return __asin_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_asind(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return asin((float)x);
else {
double ret;
foreach_active (i) {
@@ -3719,12 +3817,14 @@ static inline double asin(const double x) {
__declspec(safe)
static inline double cos(const double x) {
if (__have_native_trigonometry)
{
return __cos_varying_double(x);
}
if (__math_lib == __math_lib_svml)
{
return __svml_cosd(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return cos((float)x);
else {
double ret;
foreach_active (i) {
@@ -3737,8 +3837,10 @@ static inline double cos(const double x) {
__declspec(safe)
static inline uniform double cos(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return cos((float)x);
if (__have_native_trigonometry)
{
return __cos_uniform_double(x);
}
else
return __stdlib_cos(x);
}
@@ -3746,16 +3848,14 @@ static inline uniform double cos(uniform double x) {
__declspec(safe)
static inline void sincos(double x, varying double * uniform sin_result,
varying double * uniform cos_result) {
if (__have_native_trigonometry)
{
__sincos_varying_double(x,sin_result,cos_result);
}
if (__math_lib == __math_lib_svml)
{
__svml_sincosd(x, sin_result, cos_result);
}
else if (__math_lib == __math_lib_ispc_fast) {
float sr, cr;
sincos((float)x, &sr, &cr);
*sin_result = sr;
*cos_result = cr;
}
else {
foreach_active (i) {
uniform double sr, cr;
@@ -3769,11 +3869,9 @@ static inline void sincos(double x, varying double * uniform sin_result,
__declspec(safe)
static inline void sincos(uniform double x, uniform double * uniform sin_result,
uniform double * uniform cos_result) {
if (__math_lib == __math_lib_ispc_fast) {
uniform float sr, cr;
sincos((uniform float)x, &sr, &cr);
*sin_result = sr;
*cos_result = cr;
if (__have_native_trigonometry)
{
__sincos_uniform_double(x,sin_result, cos_result);
}
else
__stdlib_sincos(x, sin_result, cos_result);
@@ -3781,12 +3879,14 @@ static inline void sincos(uniform double x, uniform double * uniform sin_result,
__declspec(safe)
static inline double tan(double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_trigonometry)
{
return __tan_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_tand(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return tan((float)x);
else {
double ret;
foreach_active (i) {
@@ -3799,16 +3899,20 @@ static inline double tan(double x) {
__declspec(safe)
static inline uniform double tan(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return tan((float)x);
if (__have_native_trigonometry)
{
return __tan_uniform_double(x);
}
else
return __stdlib_tan(x);
}
__declspec(safe)
static inline double atan(double x) {
if (__math_lib == __math_lib_ispc_fast)
return atan((float)x);
if (__have_native_trigonometry)
{
return __atan_varying_double(x);
}
else {
double ret;
foreach_active (i) {
@@ -3821,20 +3925,24 @@ static inline double atan(double x) {
__declspec(safe)
static inline uniform double atan(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return atan((float)x);
if (__have_native_trigonometry)
{
return __atan_uniform_double(x);
}
else
return __stdlib_atan(x);
}
__declspec(safe)
static inline double atan2(double y, double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_trigonometry)
{
return __atan2_varying_double(y,x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_atan2d(y,x);
}
else if (__math_lib == __math_lib_ispc_fast)
return atan2((float)y, (float)x);
else {
double ret;
foreach_active (i) {
@@ -3847,20 +3955,23 @@ static inline double atan2(double y, double x) {
__declspec(safe)
static inline uniform double atan2(uniform double y, uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return atan2((float)y, (float)x);
if (__have_native_trigonometry)
{
return __atan2_uniform_double(y,x);
}
else
return __stdlib_atan2(y, x);
}
__declspec(safe)
static inline double exp(double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_transcendentals) {
return __exp_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_expd(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return exp((float)x);
else {
double ret;
foreach_active (i) {
@@ -3873,20 +3984,22 @@ static inline double exp(double x) {
__declspec(safe)
static inline uniform double exp(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return exp((float)x);
if (__have_native_transcendentals) {
return __exp_uniform_double(x);
}
else
return __stdlib_exp(x);
}
__declspec(safe)
static inline double log(double x) {
if (__math_lib == __math_lib_svml)
if (__have_native_transcendentals) {
return __log_varying_double(x);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_logd(x);
}
else if (__math_lib == __math_lib_ispc_fast)
return log((float)x);
else {
double ret;
foreach_active (i) {
@@ -3899,20 +4012,22 @@ static inline double log(double x) {
__declspec(safe)
static inline uniform double log(uniform double x) {
if (__math_lib == __math_lib_ispc_fast)
return log((float)x);
if (__have_native_transcendentals) {
return __log_uniform_double(x);
}
else
return __stdlib_log(x);
}
__declspec(safe)
static inline double pow(double a, double b) {
if (__math_lib == __math_lib_svml)
if (__have_native_transcendentals) {
return __pow_varying_double(a,b);
}
else if (__math_lib == __math_lib_svml)
{
return __svml_powd(a,b);
}
else if (__math_lib == __math_lib_ispc_fast)
return pow((float)a, (float)b);
else {
double ret;
foreach_active (i) {
@@ -3925,8 +4040,9 @@ static inline double pow(double a, double b) {
__declspec(safe)
static inline uniform double pow(uniform double a, uniform double b) {
if (__math_lib == __math_lib_ispc_fast)
return pow((float)a, (float)b);
if (__have_native_transcendentals) {
return __pow_uniform_double(a,b);
}
else
return __stdlib_pow(a, b);
}