diff options
Diffstat (limited to 'intern/cycles/util/util_math.h')
| -rw-r--r-- | intern/cycles/util/util_math.h | 1031 |
1 files changed, 38 insertions, 993 deletions
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h index 77781ed4574..52b4fa859b7 100644 --- a/intern/cycles/util/util_math.h +++ b/intern/cycles/util/util_math.h @@ -28,12 +28,10 @@ #ifndef __KERNEL_OPENCL__ - -#include <float.h> -#include <math.h> -#include <stdio.h> - -#endif +# include <float.h> +# include <math.h> +# include <stdio.h> +#endif /* __KERNEL_OPENCL__ */ #include "util/util_types.h" @@ -43,49 +41,44 @@ CCL_NAMESPACE_BEGIN /* Division */ #ifndef M_PI_F -#define M_PI_F (3.1415926535897932f) /* pi */ +# define M_PI_F (3.1415926535897932f) /* pi */ #endif #ifndef M_PI_2_F -#define M_PI_2_F (1.5707963267948966f) /* pi/2 */ +# define M_PI_2_F (1.5707963267948966f) /* pi/2 */ #endif #ifndef M_PI_4_F -#define M_PI_4_F (0.7853981633974830f) /* pi/4 */ +# define M_PI_4_F (0.7853981633974830f) /* pi/4 */ #endif #ifndef M_1_PI_F -#define M_1_PI_F (0.3183098861837067f) /* 1/pi */ +# define M_1_PI_F (0.3183098861837067f) /* 1/pi */ #endif #ifndef M_2_PI_F -#define M_2_PI_F (0.6366197723675813f) /* 2/pi */ +# define M_2_PI_F (0.6366197723675813f) /* 2/pi */ #endif /* Multiplication */ #ifndef M_2PI_F -#define M_2PI_F (6.2831853071795864f) /* 2*pi */ +# define M_2PI_F (6.2831853071795864f) /* 2*pi */ #endif #ifndef M_4PI_F -#define M_4PI_F (12.566370614359172f) /* 4*pi */ +# define M_4PI_F (12.566370614359172f) /* 4*pi */ #endif /* Float sqrt variations */ - #ifndef M_SQRT2_F -#define M_SQRT2_F (1.4142135623730950f) /* sqrt(2) */ +# define M_SQRT2_F (1.4142135623730950f) /* sqrt(2) */ #endif - #ifndef M_LN2_F -#define M_LN2_F (0.6931471805599453f) /* ln(2) */ +# define M_LN2_F (0.6931471805599453f) /* ln(2) */ #endif - #ifndef M_LN10_F -#define M_LN10_F (2.3025850929940457f) /* ln(10) */ +# define M_LN10_F (2.3025850929940457f) /* ln(10) */ #endif /* Scalar */ #ifdef _WIN32 - -#ifndef __KERNEL_OPENCL__ - +# ifndef __KERNEL_OPENCL__ ccl_device_inline float fmaxf(float a, float b) { return (a > b)? a: b; @@ -95,13 +88,10 @@ ccl_device_inline float fminf(float a, float b) { return (a < b)? a: b; } - -#endif - -#endif +# endif /* !__KERNEL_OPENCL__ */ +#endif /* _WIN32 */ #ifndef __KERNEL_GPU__ - using std::isfinite; using std::isnan; @@ -157,8 +147,7 @@ ccl_device_inline T max4(const T& a, const T& b, const T& c, const T& d) { return max(max(a,b),max(c,d)); } - -#endif +#endif /* __KERNEL_GPU__ */ ccl_device_inline float min4(float a, float b, float c, float d) { @@ -170,13 +159,7 @@ ccl_device_inline float max4(float a, float b, float c, float d) return max(max(a, b), max(c, d)); } -ccl_device_inline float max3(float3 a) -{ - return max(max(a.x, a.y), a.z); -} - #ifndef __KERNEL_OPENCL__ - ccl_device_inline int clamp(int a, int mn, int mx) { return min(max(a, mn), mx); @@ -191,17 +174,14 @@ ccl_device_inline float mix(float a, float b, float t) { return a + t*(b - a); } - -#endif +#endif /* __KERNEL_OPENCL__ */ #ifndef __KERNEL_CUDA__ - ccl_device_inline float saturate(float a) { return clamp(a, 0.0f, 1.0f); } - -#endif +#endif /* __KERNEL_CUDA__ */ ccl_device_inline int float_to_int(float f) { @@ -242,453 +222,6 @@ ccl_device_inline int mod(int x, int m) return (x % m + m) % m; } -/* Float2 Vector */ - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline bool is_zero(const float2& a) -{ - return (a.x == 0.0f && a.y == 0.0f); -} - -#endif - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float average(const float2& a) -{ - return (a.x + a.y)*(1.0f/2.0f); -} - -#endif - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float2 operator-(const float2& a) -{ - return make_float2(-a.x, -a.y); -} - -ccl_device_inline float2 operator*(const float2& a, const float2& b) -{ - return make_float2(a.x*b.x, a.y*b.y); -} - -ccl_device_inline float2 operator*(const float2& a, float f) -{ - return make_float2(a.x*f, a.y*f); -} - -ccl_device_inline float2 operator*(float f, const float2& a) -{ - return make_float2(a.x*f, a.y*f); -} - -ccl_device_inline float2 operator/(float f, const float2& a) -{ - return make_float2(f/a.x, f/a.y); -} - -ccl_device_inline float2 operator/(const float2& a, float f) -{ - float invf = 1.0f/f; - return make_float2(a.x*invf, a.y*invf); -} - -ccl_device_inline float2 operator/(const float2& a, const float2& b) -{ - return make_float2(a.x/b.x, a.y/b.y); -} - -ccl_device_inline float2 operator+(const float2& a, const float2& b) -{ - return make_float2(a.x+b.x, a.y+b.y); -} - -ccl_device_inline float2 operator-(const float2& a, const float2& b) -{ - return make_float2(a.x-b.x, a.y-b.y); -} - -ccl_device_inline float2 operator+=(float2& a, const float2& b) -{ - return a = a + b; -} - -ccl_device_inline float2 operator*=(float2& a, const float2& b) -{ - return a = a * b; -} - -ccl_device_inline float2 operator*=(float2& a, float f) -{ - return a = a * f; -} - -ccl_device_inline float2 operator/=(float2& a, const float2& b) -{ - return a = a / b; -} - -ccl_device_inline float2 operator/=(float2& a, float f) -{ - float invf = 1.0f/f; - return a = a * invf; -} - - -ccl_device_inline float dot(const float2& a, const float2& b) -{ - return a.x*b.x + a.y*b.y; -} - -ccl_device_inline float cross(const float2& a, const float2& b) -{ - return (a.x*b.y - a.y*b.x); -} - -#endif - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline bool operator==(const int2 a, const int2 b) -{ - return (a.x == b.x && a.y == b.y); -} - -ccl_device_inline float len(const float2& a) -{ - return sqrtf(dot(a, a)); -} - -ccl_device_inline float2 normalize(const float2& a) -{ - return a/len(a); -} - -ccl_device_inline float2 normalize_len(const float2& a, float *t) -{ - *t = len(a); - return a/(*t); -} - -ccl_device_inline float2 safe_normalize(const float2& a) -{ - float t = len(a); - return (t != 0.0f)? a/t: a; -} - -ccl_device_inline bool operator==(const float2& a, const float2& b) -{ - return (a.x == b.x && a.y == b.y); -} - -ccl_device_inline bool operator!=(const float2& a, const float2& b) -{ - return !(a == b); -} - -ccl_device_inline float2 min(const float2& a, const float2& b) -{ - return make_float2(min(a.x, b.x), min(a.y, b.y)); -} - -ccl_device_inline float2 max(const float2& a, const float2& b) -{ - return make_float2(max(a.x, b.x), max(a.y, b.y)); -} - -ccl_device_inline float2 clamp(const float2& a, const float2& mn, const float2& mx) -{ - return min(max(a, mn), mx); -} - -ccl_device_inline float2 fabs(const float2& a) -{ - return make_float2(fabsf(a.x), fabsf(a.y)); -} - -ccl_device_inline float2 as_float2(const float4& a) -{ - return make_float2(a.x, a.y); -} - -#endif - -#ifndef __KERNEL_GPU__ - -ccl_device_inline void print_float2(const char *label, const float2& a) -{ - printf("%s: %.8f %.8f\n", label, (double)a.x, (double)a.y); -} - -#endif - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float2 interp(const float2& a, const float2& b, float t) -{ - return a + t*(b - a); -} - -#endif - -/* Float3 Vector */ - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float3 operator-(const float3& a) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_xor_ps(a.m128, _mm_castsi128_ps(_mm_set1_epi32(0x80000000)))); -#else - return make_float3(-a.x, -a.y, -a.z); -#endif -} - -ccl_device_inline float3 operator*(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_mul_ps(a.m128,b.m128)); -#else - return make_float3(a.x*b.x, a.y*b.y, a.z*b.z); -#endif -} - -ccl_device_inline float3 operator*(const float3& a, const float f) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_mul_ps(a.m128,_mm_set1_ps(f))); -#else - return make_float3(a.x*f, a.y*f, a.z*f); -#endif -} - -ccl_device_inline float3 operator*(const float f, const float3& a) -{ - /* TODO(sergey): Currently disabled, gives speedup but causes precision issues. */ -#if defined(__KERNEL_SSE__) && 0 - return float3(_mm_mul_ps(_mm_set1_ps(f), a.m128)); -#else - return make_float3(a.x*f, a.y*f, a.z*f); -#endif -} - -ccl_device_inline float3 operator/(const float f, const float3& a) -{ - /* TODO(sergey): Currently disabled, gives speedup but causes precision issues. */ -#if defined(__KERNEL_SSE__) && 0 - __m128 rc = _mm_rcp_ps(a.m128); - return float3(_mm_mul_ps(_mm_set1_ps(f),rc)); -#else - return make_float3(f / a.x, f / a.y, f / a.z); -#endif -} - -ccl_device_inline float3 operator/(const float3& a, const float f) -{ - float invf = 1.0f/f; - return a * invf; -} - -ccl_device_inline float3 operator/(const float3& a, const float3& b) -{ - /* TODO(sergey): Currently disabled, gives speedup but causes precision issues. */ -#if defined(__KERNEL_SSE__) && 0 - __m128 rc = _mm_rcp_ps(b.m128); - return float3(_mm_mul_ps(a, rc)); -#else - return make_float3(a.x / b.x, a.y / b.y, a.z / b.z); -#endif -} - -ccl_device_inline float3 operator+(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_add_ps(a.m128, b.m128)); -#else - return make_float3(a.x + b.x, a.y + b.y, a.z + b.z); -#endif -} - -ccl_device_inline float3 operator-(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_sub_ps(a.m128, b.m128)); -#else - return make_float3(a.x - b.x, a.y - b.y, a.z - b.z); -#endif -} - -ccl_device_inline float3 operator+=(float3& a, const float3& b) -{ - return a = a + b; -} - -ccl_device_inline float3 operator*=(float3& a, const float3& b) -{ - return a = a * b; -} - -ccl_device_inline float3 operator*=(float3& a, float f) -{ - return a = a * f; -} - -ccl_device_inline float3 operator/=(float3& a, const float3& b) -{ - return a = a / b; -} - -ccl_device_inline float3 operator/=(float3& a, float f) -{ - float invf = 1.0f/f; - return a = a * invf; -} - -ccl_device_inline float dot(const float3& a, const float3& b) -{ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) - return _mm_cvtss_f32(_mm_dp_ps(a, b, 0x7F)); -#else - return a.x*b.x + a.y*b.y + a.z*b.z; -#endif -} - -ccl_device_inline float dot_xy(const float3& a, const float3& b) -{ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) - return _mm_cvtss_f32(_mm_hadd_ps(_mm_mul_ps(a,b),b)); -#else - return a.x*b.x + a.y*b.y; -#endif -} - -ccl_device_inline float dot(const float4& a, const float4& b) -{ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) - return _mm_cvtss_f32(_mm_dp_ps(a, b, 0xFF)); -#else - return (a.x*b.x + a.y*b.y) + (a.z*b.z + a.w*b.w); -#endif -} - -ccl_device_inline float3 cross(const float3& a, const float3& b) -{ - float3 r = make_float3(a.y*b.z - a.z*b.y, a.z*b.x - a.x*b.z, a.x*b.y - a.y*b.x); - return r; -} - -#endif - -ccl_device_inline float len(const float3 a) -{ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) - return _mm_cvtss_f32(_mm_sqrt_ss(_mm_dp_ps(a.m128, a.m128, 0x7F))); -#else - return sqrtf(dot(a, a)); -#endif -} - -ccl_device_inline float len_squared(const float3 a) -{ - return dot(a, a); -} - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float len_squared(const float4& a) -{ - return dot(a, a); -} - -ccl_device_inline float3 normalize(const float3& a) -{ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) - __m128 norm = _mm_sqrt_ps(_mm_dp_ps(a.m128, a.m128, 0x7F)); - return float3(_mm_div_ps(a.m128, norm)); -#else - return a/len(a); -#endif -} - -#endif - -ccl_device_inline float3 saturate3(float3 a) -{ - return make_float3(saturate(a.x), saturate(a.y), saturate(a.z)); -} - -ccl_device_inline float3 normalize_len(const float3 a, float *t) -{ - *t = len(a); - float x = 1.0f / *t; - return a*x; -} - -ccl_device_inline float3 safe_normalize(const float3 a) -{ - float t = len(a); - return (t != 0.0f)? a * (1.0f/t) : a; -} - -ccl_device_inline float3 safe_normalize_len(const float3 a, float *t) -{ - *t = len(a); - return (*t != 0.0f)? a/(*t): a; -} - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline bool operator==(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return (_mm_movemask_ps(_mm_cmpeq_ps(a.m128, b.m128)) & 7) == 7; -#else - return (a.x == b.x && a.y == b.y && a.z == b.z); -#endif -} - -ccl_device_inline bool operator!=(const float3& a, const float3& b) -{ - return !(a == b); -} - -ccl_device_inline float3 min(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_min_ps(a.m128, b.m128)); -#else - return make_float3(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z)); -#endif -} - -ccl_device_inline float3 max(const float3& a, const float3& b) -{ -#ifdef __KERNEL_SSE__ - return float3(_mm_max_ps(a.m128, b.m128)); -#else - return make_float3(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z)); -#endif -} - -ccl_device_inline float3 clamp(const float3& a, const float3& mn, const float3& mx) -{ - return min(max(a, mn), mx); -} - -ccl_device_inline float3 fabs(const float3& a) -{ -#ifdef __KERNEL_SSE__ - __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff)); - return float3(_mm_and_ps(a.m128, mask)); -#else - return make_float3(fabsf(a.x), fabsf(a.y), fabsf(a.z)); -#endif -} - -#endif - ccl_device_inline float3 float2_to_float3(const float2 a) { return make_float3(a.x, a.y, 0.0f); @@ -704,509 +237,21 @@ ccl_device_inline float4 float3_to_float4(const float3 a) return make_float4(a.x, a.y, a.z, 1.0f); } -#ifndef __KERNEL_GPU__ - -ccl_device_inline void print_float3(const char *label, const float3& a) -{ - printf("%s: %.8f %.8f %.8f\n", label, (double)a.x, (double)a.y, (double)a.z); -} - -ccl_device_inline float3 rcp(const float3& a) -{ -#ifdef __KERNEL_SSE__ - const float4 r(_mm_rcp_ps(a.m128)); - return float3(_mm_sub_ps(_mm_add_ps(r, r), - _mm_mul_ps(_mm_mul_ps(r, r), a))); -#else - return make_float3(1.0f/a.x, 1.0f/a.y, 1.0f/a.z); -#endif -} - -#endif - -ccl_device_inline float3 interp(float3 a, float3 b, float t) -{ - return a + t*(b - a); -} - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float3 mix(const float3& a, const float3& b, float t) -{ - return a + t*(b - a); -} - -#endif - -ccl_device_inline bool is_zero(const float3 a) -{ -#ifdef __KERNEL_SSE__ - return a == make_float3(0.0f); -#else - return (a.x == 0.0f && a.y == 0.0f && a.z == 0.0f); -#endif -} - -ccl_device_inline float reduce_add(const float3 a) -{ - return (a.x + a.y + a.z); -} - -ccl_device_inline float average(const float3 a) -{ - return reduce_add(a)*(1.0f/3.0f); -} - -ccl_device_inline bool isequal_float3(const float3 a, const float3 b) -{ -#ifdef __KERNEL_OPENCL__ - return all(a == b); -#else - return a == b; -#endif -} - -/* Float4 Vector */ - -#ifdef __KERNEL_SSE__ - -template<size_t index_0, size_t index_1, size_t index_2, size_t index_3> -__forceinline const float4 shuffle(const float4& b) -{ - return float4(_mm_castsi128_ps( - _mm_shuffle_epi32(_mm_castps_si128(b), - _MM_SHUFFLE(index_3, index_2, index_1, index_0)))); -} - -#if defined(__KERNEL_SSE3__) -template<> __forceinline const float4 shuffle<0, 0, 2, 2>(const float4& b) -{ - return float4(_mm_moveldup_ps(b)); -} - -template<> __forceinline const float4 shuffle<1, 1, 3, 3>(const float4& b) -{ - return float4(_mm_movehdup_ps(b)); -} -#endif - -template<> __forceinline const float4 shuffle<0, 1, 0, 1>(const float4& b) -{ - return float4(_mm_castpd_ps(_mm_movedup_pd(_mm_castps_pd(b)))); -} - -#endif - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline float4 operator-(const float4& a) -{ -#ifdef __KERNEL_SSE__ - __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x80000000)); - return float4(_mm_xor_ps(a.m128, mask)); -#else - return make_float4(-a.x, -a.y, -a.z, -a.w); -#endif -} - -ccl_device_inline float4 operator*(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return float4(_mm_mul_ps(a.m128, b.m128)); -#else - return make_float4(a.x*b.x, a.y*b.y, a.z*b.z, a.w*b.w); -#endif -} - -ccl_device_inline float4 operator*(const float4& a, float f) -{ -#if defined(__KERNEL_SSE__) - return a * make_float4(f); -#else - return make_float4(a.x*f, a.y*f, a.z*f, a.w*f); -#endif -} - -ccl_device_inline float4 operator*(float f, const float4& a) -{ - return a * f; -} - -ccl_device_inline float4 rcp(const float4& a) -{ -#ifdef __KERNEL_SSE__ - float4 r(_mm_rcp_ps(a.m128)); - return float4(_mm_sub_ps(_mm_add_ps(r, r), - _mm_mul_ps(_mm_mul_ps(r, r), a))); -#else - return make_float4(1.0f/a.x, 1.0f/a.y, 1.0f/a.z, 1.0f/a.w); -#endif -} - -ccl_device_inline float4 operator/(const float4& a, float f) -{ - return a * (1.0f/f); -} - -ccl_device_inline float4 operator/(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return a * rcp(b); -#else - return make_float4(a.x/b.x, a.y/b.y, a.z/b.z, a.w/b.w); -#endif - -} - -ccl_device_inline float4 operator+(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return float4(_mm_add_ps(a.m128, b.m128)); -#else - return make_float4(a.x+b.x, a.y+b.y, a.z+b.z, a.w+b.w); -#endif -} - -ccl_device_inline float4 operator-(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return float4(_mm_sub_ps(a.m128, b.m128)); -#else - return make_float4(a.x-b.x, a.y-b.y, a.z-b.z, a.w-b.w); -#endif -} - -ccl_device_inline float4 operator+=(float4& a, const float4& b) -{ - return a = a + b; -} - -ccl_device_inline float4 operator*=(float4& a, const float4& b) -{ - return a = a * b; -} - -ccl_device_inline float4 operator/=(float4& a, float f) -{ - return a = a / f; -} - -ccl_device_inline int4 operator<(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - /* TODO(sergey): avoid cvt. */ - return int4(_mm_cvtps_epi32(_mm_cmplt_ps(a.m128, b.m128))); -#else - return make_int4(a.x < b.x, a.y < b.y, a.z < b.z, a.w < b.w); -#endif -} - -ccl_device_inline int4 operator>=(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - /* TODO(sergey): avoid cvt. */ - return int4(_mm_cvtps_epi32(_mm_cmpge_ps(a.m128, b.m128))); -#else - return make_int4(a.x >= b.x, a.y >= b.y, a.z >= b.z, a.w >= b.w); -#endif -} - -ccl_device_inline int4 operator<=(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - /* TODO(sergey): avoid cvt. */ - return int4(_mm_cvtps_epi32(_mm_cmple_ps(a.m128, b.m128))); -#else - return make_int4(a.x <= b.x, a.y <= b.y, a.z <= b.z, a.w <= b.w); -#endif -} - -ccl_device_inline bool operator==(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return (_mm_movemask_ps(_mm_cmpeq_ps(a.m128, b.m128)) & 15) == 15; -#else - return (a.x == b.x && a.y == b.y && a.z == b.z && a.w == b.w); -#endif -} - -ccl_device_inline float4 cross(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return (shuffle<1,2,0,0>(a)*shuffle<2,0,1,0>(b)) - (shuffle<2,0,1,0>(a)*shuffle<1,2,0,0>(b)); -#else - return make_float4(a.y*b.z - a.z*b.y, a.z*b.x - a.x*b.z, a.x*b.y - a.y*b.x, 0.0f); -#endif -} - -ccl_device_inline bool is_zero(const float4& a) -{ -#ifdef __KERNEL_SSE__ - return a == make_float4(0.0f); -#else - return (a.x == 0.0f && a.y == 0.0f && a.z == 0.0f && a.w == 0.0f); -#endif -} - -ccl_device_inline float reduce_add(const float4& a) -{ -#ifdef __KERNEL_SSE__ - float4 h(shuffle<1,0,3,2>(a) + a); - /* TODO(sergey): Investigate efficiency. */ - return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); -#else - return ((a.x + a.y) + (a.z + a.w)); -#endif -} - -ccl_device_inline float average(const float4& a) -{ - return reduce_add(a) * 0.25f; -} - -ccl_device_inline float len(const float4& a) -{ - return sqrtf(dot(a, a)); -} - -ccl_device_inline float4 normalize(const float4& a) -{ - return a/len(a); -} - -ccl_device_inline float4 safe_normalize(const float4& a) -{ - float t = len(a); - return (t != 0.0f)? a/t: a; -} - -ccl_device_inline float4 min(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return float4(_mm_min_ps(a.m128, b.m128)); -#else - return make_float4(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z), min(a.w, b.w)); -#endif -} - -ccl_device_inline float4 max(const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - return float4(_mm_max_ps(a.m128, b.m128)); -#else - return make_float4(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z), max(a.w, b.w)); -#endif -} - -#endif - -#ifndef __KERNEL_GPU__ - -ccl_device_inline float4 select(const int4& mask, const float4& a, const float4& b) -{ -#ifdef __KERNEL_SSE__ - /* TODO(sergey): avoid cvt. */ - return float4(_mm_or_ps(_mm_and_ps(_mm_cvtepi32_ps(mask), a), - _mm_andnot_ps(_mm_cvtepi32_ps(mask), b))); -#else - return make_float4((mask.x)? a.x: b.x, (mask.y)? a.y: b.y, (mask.z)? a.z: b.z, (mask.w)? a.w: b.w); -#endif -} - -ccl_device_inline float4 reduce_min(const float4& a) -{ -#ifdef __KERNEL_SSE__ - float4 h = min(shuffle<1,0,3,2>(a), a); - return min(shuffle<2,3,0,1>(h), h); -#else - return make_float4(min(min(a.x, a.y), min(a.z, a.w))); -#endif -} - -ccl_device_inline float4 reduce_max(const float4& a) -{ -#ifdef __KERNEL_SSE__ - float4 h = max(shuffle<1,0,3,2>(a), a); - return max(shuffle<2,3,0,1>(h), h); -#else - return make_float4(max(max(a.x, a.y), max(a.z, a.w))); -#endif -} - -#if 0 -ccl_device_inline float4 reduce_add(const float4& a) -{ -#ifdef __KERNEL_SSE__ - float4 h = shuffle<1,0,3,2>(a) + a; - return shuffle<2,3,0,1>(h) + h; -#else - return make_float4((a.x + a.y) + (a.z + a.w)); -#endif -} -#endif - -ccl_device_inline void print_float4(const char *label, const float4& a) -{ - printf("%s: %.8f %.8f %.8f %.8f\n", label, (double)a.x, (double)a.y, (double)a.z, (double)a.w); -} - -#endif - -/* Int2 */ - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline int2 operator+(const int2 &a, const int2 &b) -{ - return make_int2(a.x + b.x, a.y + b.y); -} - -ccl_device_inline int2 operator+=(int2 &a, const int2 &b) -{ - return a = a + b; -} - -ccl_device_inline int2 operator-(const int2 &a, const int2 &b) -{ - return make_int2(a.x - b.x, a.y - b.y); -} - -ccl_device_inline int2 operator*(const int2 &a, const int2 &b) -{ - return make_int2(a.x * b.x, a.y * b.y); -} +CCL_NAMESPACE_END -ccl_device_inline int2 operator/(const int2 &a, const int2 &b) -{ - return make_int2(a.x / b.x, a.y / b.y); -} +#include "util/util_math_int2.h" +#include "util/util_math_int3.h" +#include "util/util_math_int4.h" -#endif +#include "util/util_math_float2.h" +#include "util/util_math_float3.h" +#include "util/util_math_float4.h" -/* Int3 */ +CCL_NAMESPACE_BEGIN #ifndef __KERNEL_OPENCL__ - -ccl_device_inline int3 min(int3 a, int3 b) -{ -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE41__) - return int3(_mm_min_epi32(a.m128, b.m128)); -#else - return make_int3(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z)); -#endif -} - -ccl_device_inline int3 max(int3 a, int3 b) -{ -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE41__) - return int3(_mm_max_epi32(a.m128, b.m128)); -#else - return make_int3(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z)); -#endif -} - -ccl_device_inline int3 clamp(const int3& a, int mn, int mx) -{ -#ifdef __KERNEL_SSE__ - return min(max(a, make_int3(mn)), make_int3(mx)); -#else - return make_int3(clamp(a.x, mn, mx), clamp(a.y, mn, mx), clamp(a.z, mn, mx)); -#endif -} - -ccl_device_inline int3 clamp(const int3& a, int3& mn, int mx) -{ -#ifdef __KERNEL_SSE__ - return min(max(a, mn), make_int3(mx)); -#else - return make_int3(clamp(a.x, mn.x, mx), clamp(a.y, mn.y, mx), clamp(a.z, mn.z, mx)); -#endif -} - -#endif - -#ifndef __KERNEL_GPU__ - -ccl_device_inline void print_int3(const char *label, const int3& a) -{ - printf("%s: %d %d %d\n", label, a.x, a.y, a.z); -} - -#endif - -/* Int4 */ - -#ifndef __KERNEL_GPU__ - -ccl_device_inline int4 operator+(const int4& a, const int4& b) -{ -#ifdef __KERNEL_SSE__ - return int4(_mm_add_epi32(a.m128, b.m128)); -#else - return make_int4(a.x+b.x, a.y+b.y, a.z+b.z, a.w+b.w); -#endif -} - -ccl_device_inline int4 operator+=(int4& a, const int4& b) -{ - return a = a + b; -} - -ccl_device_inline int4 operator>>(const int4& a, int i) -{ -#ifdef __KERNEL_SSE__ - return int4(_mm_srai_epi32(a.m128, i)); -#else - return make_int4(a.x >> i, a.y >> i, a.z >> i, a.w >> i); -#endif -} - -ccl_device_inline int4 min(int4 a, int4 b) -{ -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE41__) - return int4(_mm_min_epi32(a.m128, b.m128)); -#else - return make_int4(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z), min(a.w, b.w)); -#endif -} - -ccl_device_inline int4 max(int4 a, int4 b) -{ -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE41__) - return int4(_mm_max_epi32(a.m128, b.m128)); -#else - return make_int4(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z), max(a.w, b.w)); -#endif -} - -ccl_device_inline int4 clamp(const int4& a, const int4& mn, const int4& mx) -{ - return min(max(a, mn), mx); -} - -ccl_device_inline int4 select(const int4& mask, const int4& a, const int4& b) -{ -#ifdef __KERNEL_SSE__ - const __m128 m = _mm_cvtepi32_ps(mask); - /* TODO(sergey): avoid cvt. */ - return int4(_mm_castps_si128(_mm_or_ps(_mm_and_ps(m, _mm_castsi128_ps(a)), - _mm_andnot_ps(m, _mm_castsi128_ps(b))))); -#else - return make_int4((mask.x)? a.x: b.x, (mask.y)? a.y: b.y, (mask.z)? a.z: b.z, (mask.w)? a.w: b.w); -#endif -} - -ccl_device_inline void print_int4(const char *label, const int4& a) -{ - printf("%s: %d %d %d %d\n", label, a.x, a.y, a.z, a.w); -} - -#endif - /* Int/Float conversion */ -#ifndef __KERNEL_OPENCL__ - ccl_device_inline int as_int(uint i) { union { uint ui; int i; } u; @@ -1256,7 +301,6 @@ ccl_device_inline float __uint_as_float(uint i) return u.f; } - /* Interpolation */ template<class A, class B> A lerp(const A& a, const A& b, const B& t) @@ -1266,12 +310,13 @@ template<class A, class B> A lerp(const A& a, const A& b, const B& t) /* Triangle */ -ccl_device_inline float triangle_area(const float3& v1, const float3& v2, const float3& v3) +ccl_device_inline float triangle_area(const float3& v1, + const float3& v2, + const float3& v3) { return len(cross(v3 - v2, v1 - v2))*0.5f; } - -#endif +#endif /* __KERNEL_OPENCL__ */ /* Versions of functions which are safe for fast math. */ ccl_device_inline bool isnan_safe(float f) @@ -1382,16 +427,16 @@ ccl_device_inline float3 rotate_around_axis(float3 p, float3 axis, float angle) float3 r; r.x = ((costheta + (1 - costheta) * axis.x * axis.x) * p.x) + - (((1 - costheta) * axis.x * axis.y - axis.z * sintheta) * p.y) + - (((1 - costheta) * axis.x * axis.z + axis.y * sintheta) * p.z); + (((1 - costheta) * axis.x * axis.y - axis.z * sintheta) * p.y) + + (((1 - costheta) * axis.x * axis.z + axis.y * sintheta) * p.z); r.y = (((1 - costheta) * axis.x * axis.y + axis.z * sintheta) * p.x) + - ((costheta + (1 - costheta) * axis.y * axis.y) * p.y) + - (((1 - costheta) * axis.y * axis.z - axis.x * sintheta) * p.z); + ((costheta + (1 - costheta) * axis.y * axis.y) * p.y) + + (((1 - costheta) * axis.y * axis.z - axis.x * sintheta) * p.z); r.z = (((1 - costheta) * axis.x * axis.z - axis.y * sintheta) * p.x) + - (((1 - costheta) * axis.y * axis.z + axis.x * sintheta) * p.y) + - ((costheta + (1 - costheta) * axis.z * axis.z) * p.z); + (((1 - costheta) * axis.y * axis.z + axis.x * sintheta) * p.y) + + ((costheta + (1 - costheta) * axis.z * axis.z) * p.z); return r; } |