diff options
Diffstat (limited to 'intern/cycles/util/util_math.h')
-rw-r--r-- | intern/cycles/util/util_math.h | 1361 |
1 files changed, 129 insertions, 1232 deletions
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h index 3f4d3e06c0b..fb04d49bcd9 100644 --- a/intern/cycles/util/util_math.h +++ b/intern/cycles/util/util_math.h @@ -22,15 +22,18 @@ * Basic math functions on scalar and vector types. This header is used by * both the kernel code when compiled as C++, and other C++ non-kernel code. */ -#ifndef __KERNEL_OPENCL__ +#ifndef __KERNEL_GPU__ +# include <cmath> +#endif -#include <float.h> -#include <math.h> -#include <stdio.h> -#endif +#ifndef __KERNEL_OPENCL__ +# include <float.h> +# include <math.h> +# include <stdio.h> +#endif /* __KERNEL_OPENCL__ */ -#include "util_types.h" +#include "util/util_types.h" CCL_NAMESPACE_BEGIN @@ -38,49 +41,44 @@ CCL_NAMESPACE_BEGIN /* Division */ #ifndef M_PI_F -#define M_PI_F ((float)3.14159265358979323846264338327950288) /* pi */ +# define M_PI_F (3.1415926535897932f) /* pi */ #endif #ifndef M_PI_2_F -#define M_PI_2_F ((float)1.57079632679489661923132169163975144) /* pi/2 */ +# define M_PI_2_F (1.5707963267948966f) /* pi/2 */ #endif #ifndef M_PI_4_F -#define M_PI_4_F ((float)0.785398163397448309615660845819875721) /* pi/4 */ +# define M_PI_4_F (0.7853981633974830f) /* pi/4 */ #endif #ifndef M_1_PI_F -#define M_1_PI_F ((float)0.318309886183790671537767526745028724) /* 1/pi */ +# define M_1_PI_F (0.3183098861837067f) /* 1/pi */ #endif #ifndef M_2_PI_F -#define M_2_PI_F ((float)0.636619772367581343075535053490057448) /* 2/pi */ +# define M_2_PI_F (0.6366197723675813f) /* 2/pi */ #endif /* Multiplication */ #ifndef M_2PI_F -#define M_2PI_F ((float)6.283185307179586476925286766559005768) /* 2*pi */ +# define M_2PI_F (6.2831853071795864f) /* 2*pi */ #endif #ifndef M_4PI_F -#define M_4PI_F ((float)12.56637061435917295385057353311801153) /* 4*pi */ +# define M_4PI_F (12.566370614359172f) /* 4*pi */ #endif /* Float sqrt variations */ - #ifndef M_SQRT2_F -#define M_SQRT2_F ((float)1.41421356237309504880) /* sqrt(2) */ +# define M_SQRT2_F (1.4142135623730950f) /* sqrt(2) */ #endif - #ifndef M_LN2_F -#define M_LN2_F ((float)0.6931471805599453) /* ln(2) */ +# define M_LN2_F (0.6931471805599453f) /* ln(2) */ #endif - #ifndef M_LN10_F -#define M_LN10_F ((float)2.3025850929940457) /* 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; @@ -90,12 +88,13 @@ 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; +using std::sqrt; ccl_device_inline int abs(int x) { @@ -149,8 +148,7 @@ 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) { @@ -162,526 +160,141 @@ 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); -} - -ccl_device_inline float clamp(float a, float mn, float mx) -{ - return min(max(a, mn), mx); -} - -ccl_device_inline float mix(float a, float b, float t) -{ - return a + t*(b - a); -} - -#endif - -#ifndef __KERNEL_CUDA__ - -ccl_device_inline float saturate(float a) -{ - return clamp(a, 0.0f, 1.0f); -} - -#endif - -ccl_device_inline int float_to_int(float f) -{ - return (int)f; -} - -ccl_device_inline int floor_to_int(float f) -{ - return float_to_int(floorf(f)); -} - -ccl_device_inline int ceil_to_int(float f) -{ - return float_to_int(ceilf(f)); -} - -ccl_device_inline float signf(float f) -{ - return (f < 0.0f)? -1.0f: 1.0f; -} - -ccl_device_inline float nonzerof(float f, float eps) -{ - if(fabsf(f) < eps) - return signf(f)*eps; - else - return f; -} - -ccl_device_inline float smoothstepf(float f) -{ - float ff = f*f; - return (3.0f*ff - 2.0f*ff*f); -} - -ccl_device_inline int mod(int x, int m) -{ - return (x % m + m) % m; -} - -/* Float2 Vector */ - #ifndef __KERNEL_OPENCL__ +/* Int/Float conversion */ -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) +ccl_device_inline int as_int(uint i) { - return make_float2(a.x/b.x, a.y/b.y); + union { uint ui; int i; } u; + u.ui = i; + return u.i; } -ccl_device_inline float2 operator+(const float2& a, const float2& b) +ccl_device_inline uint as_uint(int i) { - return make_float2(a.x+b.x, a.y+b.y); + union { uint ui; int i; } u; + u.i = i; + return u.ui; } -ccl_device_inline float2 operator-(const float2& a, const float2& b) +ccl_device_inline uint as_uint(float f) { - return make_float2(a.x-b.x, a.y-b.y); + union { uint i; float f; } u; + u.f = f; + return u.i; } -ccl_device_inline float2 operator+=(float2& a, const float2& b) +ccl_device_inline int __float_as_int(float f) { - return a = a + b; + union { int i; float f; } u; + u.f = f; + return u.i; } -ccl_device_inline float2 operator*=(float2& a, const float2& b) +ccl_device_inline float __int_as_float(int i) { - return a = a * b; + union { int i; float f; } u; + u.i = i; + return u.f; } -ccl_device_inline float2 operator*=(float2& a, float f) +ccl_device_inline uint __float_as_uint(float f) { - return a = a * f; + union { uint i; float f; } u; + u.f = f; + return u.i; } -ccl_device_inline float2 operator/=(float2& a, const float2& b) +ccl_device_inline float __uint_as_float(uint i) { - return a = a / b; + union { uint i; float f; } u; + u.i = i; + return u.f; } +#endif /* __KERNEL_OPENCL__ */ -ccl_device_inline float2 operator/=(float2& a, float f) +/* Versions of functions which are safe for fast math. */ +ccl_device_inline bool isnan_safe(float f) { - float invf = 1.0f/f; - return a = a * invf; + unsigned int x = __float_as_uint(f); + return (x << 1) > 0xff000000u; } - -ccl_device_inline float dot(const float2& a, const float2& b) +ccl_device_inline bool isfinite_safe(float f) { - return a.x*b.x + a.y*b.y; + /* By IEEE 754 rule, 2*Inf equals Inf */ + unsigned int x = __float_as_uint(f); + return (f == f) && (x == 0 || x == (1u << 31) || (f != 2.0f*f)) && !((x << 1) > 0xff000000u); } -ccl_device_inline float cross(const float2& a, const float2& b) +ccl_device_inline float ensure_finite(float v) { - return (a.x*b.y - a.y*b.x); + return isfinite_safe(v)? v : 0.0f; } -#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) +ccl_device_inline int clamp(int a, int mn, int 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) -{ - /* TODO(sergey): Disabled for now, causes crashes in certain cases. */ -#if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) && 0 - __m128 norm = _mm_sqrt_ps(_mm_dp_ps(a.m128, a.m128, 0x7F)); - return _mm_div_ps(a.m128, norm); -#else - return a/len(a); -#endif -} - -#endif - -ccl_device_inline float3 saturate3(float3 a) +ccl_device_inline float clamp(float a, float mn, float mx) { - return make_float3(saturate(a.x), saturate(a.y), saturate(a.z)); + return min(max(a, mn), mx); } -ccl_device_inline float3 normalize_len(const float3 a, float *t) +ccl_device_inline float mix(float a, float b, float t) { - *t = len(a); - float x = 1.0f / *t; - return a*x; + return a + t*(b - a); } +#endif /* __KERNEL_OPENCL__ */ -ccl_device_inline float3 safe_normalize(const float3 a) +#ifndef __KERNEL_CUDA__ +ccl_device_inline float saturate(float a) { - float t = len(a); - return (t != 0.0f)? a * (1.0f/t) : a; + return clamp(a, 0.0f, 1.0f); } +#endif /* __KERNEL_CUDA__ */ -ccl_device_inline float3 safe_normalize_len(const float3 a, float *t) +ccl_device_inline int float_to_int(float f) { - *t = len(a); - return (*t != 0.0f)? a/(*t): a; + return (int)f; } -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline bool operator==(const float3& a, const float3& b) +ccl_device_inline int floor_to_int(float f) { -#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 + return float_to_int(floorf(f)); } -ccl_device_inline bool operator!=(const float3& a, const float3& b) +ccl_device_inline int ceil_to_int(float f) { - return !(a == b); + return float_to_int(ceilf(f)); } -ccl_device_inline float3 min(const float3& a, const float3& b) +ccl_device_inline float signf(float f) { -#ifdef __KERNEL_SSE__ - return _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 + return (f < 0.0f)? -1.0f: 1.0f; } -ccl_device_inline float3 max(const float3& a, const float3& b) +ccl_device_inline float nonzerof(float f, float eps) { -#ifdef __KERNEL_SSE__ - return _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 + if(fabsf(f) < eps) + return signf(f)*eps; + else + return f; } -ccl_device_inline float3 clamp(const float3& a, const float3& mn, const float3& mx) +ccl_device_inline float smoothstepf(float f) { - return min(max(a, mn), mx); + float ff = f*f; + return (3.0f*ff - 2.0f*ff*f); } -ccl_device_inline float3 fabs(const float3& a) +ccl_device_inline int mod(int x, int m) { -#ifdef __KERNEL_SSE__ - __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff)); - return _mm_and_ps(a.m128, mask); -#else - return make_float3(fabsf(a.x), fabsf(a.y), fabsf(a.z)); -#endif + return (x % m + m) % m; } -#endif - ccl_device_inline float3 float2_to_float3(const float2 a) { return make_float3(a.x, a.y, 0.0f); @@ -697,545 +310,19 @@ 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__ - float4 r = _mm_rcp_ps(a.m128); - return _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 _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(b), _MM_SHUFFLE(index_3, index_2, index_1, index_0))); -} - -template<> __forceinline const float4 shuffle<0, 0, 2, 2>(const float4& b) -{ - return _mm_moveldup_ps(b); -} - -template<> __forceinline const float4 shuffle<1, 1, 3, 3>(const float4& b) -{ - return _mm_movehdup_ps(b); -} - -template<> __forceinline const float4 shuffle<0, 1, 0, 1>(const float4& b) -{ - return _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 _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) -{ - /* TODO(sergey): Disabled for now, causes crashes in certain cases. */ -#if defined(__KERNEL_SSE__) && 0 - return _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 _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) -{ - /* TODO(sergey): Disabled for now, causes crashes in certain cases. */ -#if defined(__KERNEL_SSE__) && 0 - return _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 _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__ - return _mm_cvtps_epi32(_mm_cmplt_ps(a.m128, b.m128)); /* todo: avoid cvt */ -#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__ - return _mm_cvtps_epi32(_mm_cmpge_ps(a.m128, b.m128)); /* todo: avoid cvt */ -#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__ - return _mm_cvtps_epi32(_mm_cmple_ps(a.m128, b.m128)); /* todo: avoid cvt */ -#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; - return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); /* todo: efficiency? */ -#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 _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 _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__ - return _mm_or_ps(_mm_and_ps(_mm_cvtepi32_ps(mask), a), _mm_andnot_ps(_mm_cvtepi32_ps(mask), b)); /* todo: avoid cvt */ -#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_device_inline int2 operator/(const int2 &a, const int2 &b) -{ - return make_int2(a.x / b.x, a.y / b.y); -} - -#endif - -/* Int3 */ - -#ifndef __KERNEL_OPENCL__ - -ccl_device_inline int3 min(int3 a, int3 b) -{ -#if defined(__KERNEL_SSE__) && defined(__KERNEL_SSE41__) - return _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 _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 _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 _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 _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 _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__ - __m128 m = _mm_cvtepi32_ps(mask); - return _mm_castps_si128(_mm_or_ps(_mm_and_ps(m, _mm_castsi128_ps(a)), _mm_andnot_ps(m, _mm_castsi128_ps(b)))); /* todo: avoid cvt */ -#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_NAMESPACE_END -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); -} +#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" -/* Int/Float conversion */ +CCL_NAMESPACE_BEGIN #ifndef __KERNEL_OPENCL__ - -ccl_device_inline int as_int(uint i) -{ - union { uint ui; int i; } u; - u.ui = i; - return u.i; -} - -ccl_device_inline uint as_uint(int i) -{ - union { uint ui; int i; } u; - u.i = i; - return u.ui; -} - -ccl_device_inline uint as_uint(float f) -{ - union { uint i; float f; } u; - u.f = f; - return u.i; -} - -ccl_device_inline int __float_as_int(float f) -{ - union { int i; float f; } u; - u.f = f; - return u.i; -} - -ccl_device_inline float __int_as_float(int i) -{ - union { int i; float f; } u; - u.i = i; - return u.f; -} - -ccl_device_inline uint __float_as_uint(float f) -{ - union { uint i; float f; } u; - u.f = f; - return u.i; -} - -ccl_device_inline float __uint_as_float(uint i) -{ - union { uint i; float f; } u; - u.i = i; - return u.f; -} - /* Interpolation */ template<class A, class B> A lerp(const A& a, const A& b, const B& t) @@ -1243,15 +330,23 @@ template<class A, class B> A lerp(const A& a, const A& b, const B& t) return (A)(a * ((B)1 - t) + b * t); } +#endif /* __KERNEL_OPENCL__ */ + /* Triangle */ -ccl_device_inline float triangle_area(const float3& v1, const float3& v2, const float3& v3) +#ifndef __KERNEL_OPENCL__ +ccl_device_inline float triangle_area(const float3& v1, + const float3& v2, + const float3& v3) +#else +ccl_device_inline float triangle_area(const float3 v1, + const float3 v2, + const float3 v3) +#endif { return len(cross(v3 - v2, v1 - v2))*0.5f; } -#endif - /* Orthonormal vectors */ ccl_device_inline void make_orthonormals(const float3 N, float3 *a, float3 *b) @@ -1310,7 +405,7 @@ ccl_device_inline float3 safe_divide_even_color(float3 a, float3 b) y = (b.y != 0.0f)? a.y/b.y: 0.0f; z = (b.z != 0.0f)? a.z/b.z: 0.0f; - /* try to get grey even if b is zero */ + /* try to get gray even if b is zero */ if(b.x == 0.0f) { if(b.y == 0.0f) { x = z; @@ -1347,16 +442,16 @@ 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; } @@ -1405,17 +500,17 @@ ccl_device float safe_powf(float a, float b) return compatible_powf(a, b); } -ccl_device float safe_logf(float a, float b) +ccl_device float safe_divide(float a, float b) { - if(UNLIKELY(a < 0.0f || b < 0.0f)) - return 0.0f; - - return logf(a)/logf(b); + return (b != 0.0f)? a/b: 0.0f; } -ccl_device float safe_divide(float a, float b) +ccl_device float safe_logf(float a, float b) { - return (b != 0.0f)? a/b: 0.0f; + if(UNLIKELY(a <= 0.0f || b <= 0.0f)) + return 0.0f; + + return safe_divide(logf(a),logf(b)); } ccl_device float safe_modulo(float a, float b) @@ -1432,181 +527,9 @@ ccl_device_inline float beta(float x, float y) #endif } -/* Ray Intersection */ - -ccl_device bool ray_sphere_intersect( - float3 ray_P, float3 ray_D, float ray_t, - float3 sphere_P, float sphere_radius, - float3 *isect_P, float *isect_t) -{ - float3 d = sphere_P - ray_P; - float radiussq = sphere_radius*sphere_radius; - float tsq = dot(d, d); - - if(tsq > radiussq) { /* ray origin outside sphere */ - float tp = dot(d, ray_D); - - if(tp < 0.0f) /* dir points away from sphere */ - return false; - - float dsq = tsq - tp*tp; /* pythagoras */ - - if(dsq > radiussq) /* closest point on ray outside sphere */ - return false; - - float t = tp - sqrtf(radiussq - dsq); /* pythagoras */ - - if(t < ray_t) { - *isect_t = t; - *isect_P = ray_P + ray_D*t; - return true; - } - } - - return false; -} - -ccl_device bool ray_aligned_disk_intersect( - float3 ray_P, float3 ray_D, float ray_t, - float3 disk_P, float disk_radius, - float3 *isect_P, float *isect_t) -{ - /* aligned disk normal */ - float disk_t; - float3 disk_N = normalize_len(ray_P - disk_P, &disk_t); - float div = dot(ray_D, disk_N); - - if(UNLIKELY(div == 0.0f)) - return false; - - /* compute t to intersection point */ - float t = -disk_t/div; - if(t < 0.0f || t > ray_t) - return false; - - /* test if within radius */ - float3 P = ray_P + ray_D*t; - if(len_squared(P - disk_P) > disk_radius*disk_radius) - return false; - - *isect_P = P; - *isect_t = t; - - return true; -} - -ccl_device bool ray_triangle_intersect( - float3 ray_P, float3 ray_D, float ray_t, - float3 v0, float3 v1, float3 v2, - float3 *isect_P, float *isect_t) +ccl_device_inline float xor_signmask(float x, int y) { - /* Calculate intersection */ - float3 e1 = v1 - v0; - float3 e2 = v2 - v0; - float3 s1 = cross(ray_D, e2); - - const float divisor = dot(s1, e1); - if(UNLIKELY(divisor == 0.0f)) - return false; - - const float invdivisor = 1.0f/divisor; - - /* compute first barycentric coordinate */ - const float3 d = ray_P - v0; - const float u = dot(d, s1)*invdivisor; - if(u < 0.0f) - return false; - - /* Compute second barycentric coordinate */ - const float3 s2 = cross(d, e1); - const float v = dot(ray_D, s2)*invdivisor; - if(v < 0.0f) - return false; - - const float b0 = 1.0f - u - v; - if(b0 < 0.0f) - return false; - - /* compute t to intersection point */ - const float t = dot(e2, s2)*invdivisor; - if(t < 0.0f || t > ray_t) - return false; - - *isect_t = t; - *isect_P = ray_P + ray_D*t; - - return true; -} - -ccl_device_inline bool ray_triangle_intersect_uv( - float3 ray_P, float3 ray_D, float ray_t, - float3 v0, float3 v1, float3 v2, - float *isect_u, float *isect_v, float *isect_t) -{ - /* Calculate intersection */ - float3 e1 = v1 - v0; - float3 e2 = v2 - v0; - float3 s1 = cross(ray_D, e2); - - const float divisor = dot(s1, e1); - if(UNLIKELY(divisor == 0.0f)) - return false; - - const float invdivisor = 1.0f/divisor; - - /* compute first barycentric coordinate */ - const float3 d = ray_P - v0; - const float u = dot(d, s1)*invdivisor; - if(u < 0.0f) - return false; - - /* Compute second barycentric coordinate */ - const float3 s2 = cross(d, e1); - const float v = dot(ray_D, s2)*invdivisor; - if(v < 0.0f) - return false; - - const float b0 = 1.0f - u - v; - if(b0 < 0.0f) - return false; - - /* compute t to intersection point */ - const float t = dot(e2, s2)*invdivisor; - if(t < 0.0f || t > ray_t) - return false; - - *isect_u = u; - *isect_v = v; - *isect_t = t; - - return true; -} - -ccl_device bool ray_quad_intersect(float3 ray_P, float3 ray_D, float ray_mint, float ray_maxt, - float3 quad_P, float3 quad_u, float3 quad_v, float3 quad_n, - float3 *isect_P, float *isect_t, float *isect_u, float *isect_v) -{ - float t = -(dot(ray_P, quad_n) - dot(quad_P, quad_n)) / dot(ray_D, quad_n); - if(t < ray_mint || t > ray_maxt) - return false; - - float3 hit = ray_P + t*ray_D; - float3 inplane = hit - quad_P; - - float u = dot(inplane, quad_u) / dot(quad_u, quad_u) + 0.5f; - if(u < 0.0f || u > 1.0f) - return false; - - float v = dot(inplane, quad_v) / dot(quad_v, quad_v) + 0.5f; - if(v < 0.0f || v > 1.0f) - return false; - - if(isect_P) *isect_P = hit; - if(isect_t) *isect_t = t; - if(isect_u) *isect_u = u; - if(isect_v) *isect_v = v; - - return true; + return __int_as_float(__float_as_int(x) ^ y); } /* projections */ @@ -1643,32 +566,6 @@ ccl_device_inline float2 map_to_sphere(const float3 co) return make_float2(u, v); } -ccl_device_inline int util_max_axis(float3 vec) -{ -#ifdef __KERNEL_SSE__ - __m128 a = shuffle<0,0,1,1>(vec.m128); - __m128 b = shuffle<1,2,2,1>(vec.m128); - __m128 c = _mm_cmpgt_ps(a, b); - int mask = _mm_movemask_ps(c) & 0x7; - static const char tab[8] = {2, 2, 2, 0, 1, 2, 1, 0}; - return tab[mask]; -#else - if(vec.x > vec.y) { - if(vec.x > vec.z) - return 0; - else - return 2; - } - else { - if(vec.y > vec.z) - return 1; - else - return 2; - } -#endif -} - CCL_NAMESPACE_END #endif /* __UTIL_MATH_H__ */ - |