Changeset View
Changeset View
Standalone View
Standalone View
intern/cycles/util/util_math_float3.h
| Show All 21 Lines | |||||
| #endif | #endif | ||||
| CCL_NAMESPACE_BEGIN | CCL_NAMESPACE_BEGIN | ||||
| /******************************************************************************* | /******************************************************************************* | ||||
| * Declaration. | * Declaration. | ||||
| */ | */ | ||||
| #ifndef __KERNEL_OPENCL__ | |||||
| ccl_device_inline float3 operator-(const float3 &a); | ccl_device_inline float3 operator-(const float3 &a); | ||||
| ccl_device_inline float3 operator*(const float3 &a, const float3 &b); | ccl_device_inline float3 operator*(const float3 &a, const float3 &b); | ||||
| ccl_device_inline float3 operator*(const float3 &a, const float f); | ccl_device_inline float3 operator*(const float3 &a, const float f); | ||||
| ccl_device_inline float3 operator*(const float f, const float3 &a); | ccl_device_inline float3 operator*(const float f, const float3 &a); | ||||
| ccl_device_inline float3 operator/(const float f, const float3 &a); | ccl_device_inline float3 operator/(const float f, const float3 &a); | ||||
| ccl_device_inline float3 operator/(const float3 &a, const float f); | ccl_device_inline float3 operator/(const float3 &a, const float f); | ||||
| ccl_device_inline float3 operator/(const float3 &a, const float3 &b); | ccl_device_inline float3 operator/(const float3 &a, const float3 &b); | ||||
| ccl_device_inline float3 operator+(const float3 &a, const float f); | ccl_device_inline float3 operator+(const float3 &a, const float f); | ||||
| Show All 19 Lines | |||||
| ccl_device_inline float3 max(const float3 &a, const float3 &b); | ccl_device_inline float3 max(const float3 &a, const float3 &b); | ||||
| ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx); | ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx); | ||||
| ccl_device_inline float3 fabs(const float3 &a); | ccl_device_inline float3 fabs(const float3 &a); | ||||
| ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t); | ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t); | ||||
| ccl_device_inline float3 rcp(const float3 &a); | ccl_device_inline float3 rcp(const float3 &a); | ||||
| ccl_device_inline float3 sqrt(const float3 &a); | ccl_device_inline float3 sqrt(const float3 &a); | ||||
| ccl_device_inline float3 floor(const float3 &a); | ccl_device_inline float3 floor(const float3 &a); | ||||
| ccl_device_inline float3 ceil(const float3 &a); | ccl_device_inline float3 ceil(const float3 &a); | ||||
| #endif /* !__KERNEL_OPENCL__ */ | |||||
| ccl_device_inline float min3(float3 a); | ccl_device_inline float min3(float3 a); | ||||
| ccl_device_inline float max3(float3 a); | ccl_device_inline float max3(float3 a); | ||||
| ccl_device_inline float len(const float3 a); | ccl_device_inline float len(const float3 a); | ||||
| ccl_device_inline float len_squared(const float3 a); | ccl_device_inline float len_squared(const float3 a); | ||||
| ccl_device_inline float3 reflect(const float3 incident, const float3 normal); | ccl_device_inline float3 reflect(const float3 incident, const float3 normal); | ||||
| ccl_device_inline float3 project(const float3 v, const float3 v_proj); | ccl_device_inline float3 project(const float3 v, const float3 v_proj); | ||||
| Show All 25 Lines | |||||
| #endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 one_float3() | ccl_device_inline float3 one_float3() | ||||
| { | { | ||||
| return make_float3(1.0f, 1.0f, 1.0f); | return make_float3(1.0f, 1.0f, 1.0f); | ||||
| } | } | ||||
| #ifndef __KERNEL_OPENCL__ | |||||
| ccl_device_inline float3 operator-(const float3 &a) | ccl_device_inline float3 operator-(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_xor_ps(a.m128, _mm_castsi128_ps(_mm_set1_epi32(0x80000000)))); | return float3(_mm_xor_ps(a.m128, _mm_castsi128_ps(_mm_set1_epi32(0x80000000)))); | ||||
| # else | #else | ||||
| return make_float3(-a.x, -a.y, -a.z); | return make_float3(-a.x, -a.y, -a.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator*(const float3 &a, const float3 &b) | ccl_device_inline float3 operator*(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_mul_ps(a.m128, b.m128)); | return float3(_mm_mul_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(a.x * b.x, a.y * b.y, a.z * b.z); | return make_float3(a.x * b.x, a.y * b.y, a.z * b.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator*(const float3 &a, const float f) | ccl_device_inline float3 operator*(const float3 &a, const float f) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_mul_ps(a.m128, _mm_set1_ps(f))); | return float3(_mm_mul_ps(a.m128, _mm_set1_ps(f))); | ||||
| # else | #else | ||||
| return make_float3(a.x * f, a.y * f, a.z * f); | return make_float3(a.x * f, a.y * f, a.z * f); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator*(const float f, const float3 &a) | ccl_device_inline float3 operator*(const float f, const float3 &a) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE__) | ||||
| return float3(_mm_mul_ps(_mm_set1_ps(f), a.m128)); | return float3(_mm_mul_ps(_mm_set1_ps(f), a.m128)); | ||||
| # else | #else | ||||
| return make_float3(a.x * f, a.y * f, a.z * f); | return make_float3(a.x * f, a.y * f, a.z * f); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator/(const float f, const float3 &a) | ccl_device_inline float3 operator/(const float f, const float3 &a) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE__) | ||||
| return float3(_mm_div_ps(_mm_set1_ps(f), a.m128)); | return float3(_mm_div_ps(_mm_set1_ps(f), a.m128)); | ||||
| # else | #else | ||||
| return make_float3(f / a.x, f / a.y, f / a.z); | return make_float3(f / a.x, f / a.y, f / a.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator/(const float3 &a, const float f) | ccl_device_inline float3 operator/(const float3 &a, const float f) | ||||
| { | { | ||||
| float invf = 1.0f / f; | float invf = 1.0f / f; | ||||
| return a * invf; | return a * invf; | ||||
| } | } | ||||
| ccl_device_inline float3 operator/(const float3 &a, const float3 &b) | ccl_device_inline float3 operator/(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE__) | ||||
| return float3(_mm_div_ps(a.m128, b.m128)); | return float3(_mm_div_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(a.x / b.x, a.y / b.y, a.z / b.z); | return make_float3(a.x / b.x, a.y / b.y, a.z / b.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator+(const float3 &a, const float f) | ccl_device_inline float3 operator+(const float3 &a, const float f) | ||||
| { | { | ||||
| return a + make_float3(f, f, f); | return a + make_float3(f, f, f); | ||||
| } | } | ||||
| ccl_device_inline float3 operator+(const float3 &a, const float3 &b) | ccl_device_inline float3 operator+(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_add_ps(a.m128, b.m128)); | return float3(_mm_add_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(a.x + b.x, a.y + b.y, a.z + b.z); | return make_float3(a.x + b.x, a.y + b.y, a.z + b.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator-(const float3 &a, const float f) | ccl_device_inline float3 operator-(const float3 &a, const float f) | ||||
| { | { | ||||
| return a - make_float3(f, f, f); | return a - make_float3(f, f, f); | ||||
| } | } | ||||
| ccl_device_inline float3 operator-(const float3 &a, const float3 &b) | ccl_device_inline float3 operator-(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_sub_ps(a.m128, b.m128)); | return float3(_mm_sub_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(a.x - b.x, a.y - b.y, a.z - b.z); | return make_float3(a.x - b.x, a.y - b.y, a.z - b.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 operator+=(float3 &a, const float3 &b) | ccl_device_inline float3 operator+=(float3 &a, const float3 &b) | ||||
| { | { | ||||
| return a = a + b; | return a = a + b; | ||||
| } | } | ||||
| ccl_device_inline float3 operator-=(float3 &a, const float3 &b) | ccl_device_inline float3 operator-=(float3 &a, const float3 &b) | ||||
| Show All 19 Lines | |||||
| ccl_device_inline float3 operator/=(float3 &a, float f) | ccl_device_inline float3 operator/=(float3 &a, float f) | ||||
| { | { | ||||
| float invf = 1.0f / f; | float invf = 1.0f / f; | ||||
| return a = a * invf; | return a = a * invf; | ||||
| } | } | ||||
| ccl_device_inline bool operator==(const float3 &a, const float3 &b) | ccl_device_inline bool operator==(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return (_mm_movemask_ps(_mm_cmpeq_ps(a.m128, b.m128)) & 7) == 7; | return (_mm_movemask_ps(_mm_cmpeq_ps(a.m128, b.m128)) & 7) == 7; | ||||
| # else | #else | ||||
| return (a.x == b.x && a.y == b.y && a.z == b.z); | return (a.x == b.x && a.y == b.y && a.z == b.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline bool operator!=(const float3 &a, const float3 &b) | ccl_device_inline bool operator!=(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| return !(a == b); | return !(a == b); | ||||
| } | } | ||||
| ccl_device_inline float distance(const float3 &a, const float3 &b) | ccl_device_inline float distance(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| return len(a - b); | return len(a - b); | ||||
| } | } | ||||
| ccl_device_inline float dot(const float3 &a, const float3 &b) | ccl_device_inline float dot(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | ||||
| return _mm_cvtss_f32(_mm_dp_ps(a, b, 0x7F)); | return _mm_cvtss_f32(_mm_dp_ps(a, b, 0x7F)); | ||||
| # else | #else | ||||
| return a.x * b.x + a.y * b.y + a.z * b.z; | return a.x * b.x + a.y * b.y + a.z * b.z; | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float dot_xy(const float3 &a, const float3 &b) | ccl_device_inline float dot_xy(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | ||||
| return _mm_cvtss_f32(_mm_hadd_ps(_mm_mul_ps(a, b), b)); | return _mm_cvtss_f32(_mm_hadd_ps(_mm_mul_ps(a, b), b)); | ||||
| # else | #else | ||||
| return a.x * b.x + a.y * b.y; | return a.x * b.x + a.y * b.y; | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 cross(const float3 &a, const float3 &b) | 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); | 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; | return r; | ||||
| } | } | ||||
| ccl_device_inline float3 normalize(const float3 &a) | ccl_device_inline float3 normalize(const float3 &a) | ||||
| { | { | ||||
| # if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | #if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__) | ||||
| __m128 norm = _mm_sqrt_ps(_mm_dp_ps(a.m128, a.m128, 0x7F)); | __m128 norm = _mm_sqrt_ps(_mm_dp_ps(a.m128, a.m128, 0x7F)); | ||||
| return float3(_mm_div_ps(a.m128, norm)); | return float3(_mm_div_ps(a.m128, norm)); | ||||
| # else | #else | ||||
| return a / len(a); | return a / len(a); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 min(const float3 &a, const float3 &b) | ccl_device_inline float3 min(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_min_ps(a.m128, b.m128)); | return float3(_mm_min_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z)); | return make_float3(min(a.x, b.x), min(a.y, b.y), min(a.z, b.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 max(const float3 &a, const float3 &b) | ccl_device_inline float3 max(const float3 &a, const float3 &b) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_max_ps(a.m128, b.m128)); | return float3(_mm_max_ps(a.m128, b.m128)); | ||||
| # else | #else | ||||
| return make_float3(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z)); | return make_float3(max(a.x, b.x), max(a.y, b.y), max(a.z, b.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx) | ccl_device_inline float3 clamp(const float3 &a, const float3 &mn, const float3 &mx) | ||||
| { | { | ||||
| return min(max(a, mn), mx); | return min(max(a, mn), mx); | ||||
| } | } | ||||
| ccl_device_inline float3 fabs(const float3 &a) | ccl_device_inline float3 fabs(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| # ifdef __KERNEL_NEON__ | # ifdef __KERNEL_NEON__ | ||||
| return float3(vabsq_f32(a.m128)); | return float3(vabsq_f32(a.m128)); | ||||
| # else | # else | ||||
| __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff)); | __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff)); | ||||
| return float3(_mm_and_ps(a.m128, mask)); | return float3(_mm_and_ps(a.m128, mask)); | ||||
| # endif | # endif | ||||
| # else | #else | ||||
| return make_float3(fabsf(a.x), fabsf(a.y), fabsf(a.z)); | return make_float3(fabsf(a.x), fabsf(a.y), fabsf(a.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 sqrt(const float3 &a) | ccl_device_inline float3 sqrt(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_sqrt_ps(a)); | return float3(_mm_sqrt_ps(a)); | ||||
| # else | #else | ||||
| return make_float3(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z)); | return make_float3(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 floor(const float3 &a) | ccl_device_inline float3 floor(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_floor_ps(a)); | return float3(_mm_floor_ps(a)); | ||||
| # else | #else | ||||
| return make_float3(floorf(a.x), floorf(a.y), floorf(a.z)); | return make_float3(floorf(a.x), floorf(a.y), floorf(a.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 ceil(const float3 &a) | ccl_device_inline float3 ceil(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| return float3(_mm_ceil_ps(a)); | return float3(_mm_ceil_ps(a)); | ||||
| # else | #else | ||||
| return make_float3(ceilf(a.x), ceilf(a.y), ceilf(a.z)); | return make_float3(ceilf(a.x), ceilf(a.y), ceilf(a.z)); | ||||
| # endif | #endif | ||||
| } | } | ||||
| ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t) | ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t) | ||||
| { | { | ||||
| return a + t * (b - a); | return a + t * (b - a); | ||||
| } | } | ||||
| ccl_device_inline float3 rcp(const float3 &a) | ccl_device_inline float3 rcp(const float3 &a) | ||||
| { | { | ||||
| # ifdef __KERNEL_SSE__ | #ifdef __KERNEL_SSE__ | ||||
| /* Don't use _mm_rcp_ps due to poor precision. */ | /* Don't use _mm_rcp_ps due to poor precision. */ | ||||
| return float3(_mm_div_ps(_mm_set_ps1(1.0f), a.m128)); | return float3(_mm_div_ps(_mm_set_ps1(1.0f), a.m128)); | ||||
| # else | #else | ||||
| return make_float3(1.0f / a.x, 1.0f / a.y, 1.0f / a.z); | return make_float3(1.0f / a.x, 1.0f / a.y, 1.0f / a.z); | ||||
| # endif | #endif | ||||
| } | } | ||||
| #endif /* !__KERNEL_OPENCL__ */ | |||||
| ccl_device_inline float min3(float3 a) | ccl_device_inline float min3(float3 a) | ||||
| { | { | ||||
| return min(min(a.x, a.y), a.z); | return min(min(a.x, a.y), a.z); | ||||
| } | } | ||||
| ccl_device_inline float max3(float3 a) | ccl_device_inline float max3(float3 a) | ||||
| { | { | ||||
| ▲ Show 20 Lines • Show All 110 Lines • ▼ Show 20 Lines | |||||
| ccl_device_inline float average(const float3 a) | ccl_device_inline float average(const float3 a) | ||||
| { | { | ||||
| return reduce_add(a) * (1.0f / 3.0f); | return reduce_add(a) * (1.0f / 3.0f); | ||||
| } | } | ||||
| ccl_device_inline bool isequal_float3(const float3 a, const float3 b) | ccl_device_inline bool isequal_float3(const float3 a, const float3 b) | ||||
| { | { | ||||
| #ifdef __KERNEL_OPENCL__ | |||||
| return all(a == b); | |||||
| #else | |||||
| return a == b; | return a == b; | ||||
| #endif | |||||
| } | } | ||||
| ccl_device_inline float3 pow3(float3 v, float e) | ccl_device_inline float3 pow3(float3 v, float e) | ||||
| { | { | ||||
| return make_float3(powf(v.x, e), powf(v.y, e), powf(v.z, e)); | return make_float3(powf(v.x, e), powf(v.y, e), powf(v.z, e)); | ||||
| } | } | ||||
| ccl_device_inline float3 exp3(float3 v) | ccl_device_inline float3 exp3(float3 v) | ||||
| Show All 40 Lines | |||||