From 4cf6524731c0856bfc6d6c0b7bcb2df0ad28b97a Mon Sep 17 00:00:00 2001 From: Brecht Van Lommel Date: Mon, 25 Jul 2022 21:16:34 +0200 Subject: Fix Cycles Metal build errors after recent changes float8 is a reserved type in Metal, but is not implemented. So rename to float8_t for now. Also move back intersection handlers to kernel.metal, they can't be in the class that encapsulates the other Metal kernel functions. --- intern/cycles/util/math_float8.h | 384 ++++++++++++++++----------------- intern/cycles/util/types_float8.h | 21 +- intern/cycles/util/types_float8_impl.h | 30 +-- 3 files changed, 219 insertions(+), 216 deletions(-) (limited to 'intern/cycles/util') diff --git a/intern/cycles/util/math_float8.h b/intern/cycles/util/math_float8.h index 8ed8d56a034..b538cfbe70b 100644 --- a/intern/cycles/util/math_float8.h +++ b/intern/cycles/util/math_float8.h @@ -14,187 +14,187 @@ CCL_NAMESPACE_BEGIN * Declaration. */ -ccl_device_inline float8 operator+(const float8 &a, const float8 &b); -ccl_device_inline float8 operator+(const float8 &a, const float f); -ccl_device_inline float8 operator+(const float f, const float8 &a); +ccl_device_inline float8_t operator+(const float8_t a, const float8_t b); +ccl_device_inline float8_t operator+(const float8_t a, const float f); +ccl_device_inline float8_t operator+(const float f, const float8_t a); -ccl_device_inline float8 operator-(const float8 &a); -ccl_device_inline float8 operator-(const float8 &a, const float8 &b); -ccl_device_inline float8 operator-(const float8 &a, const float f); -ccl_device_inline float8 operator-(const float f, const float8 &a); +ccl_device_inline float8_t operator-(const float8_t a); +ccl_device_inline float8_t operator-(const float8_t a, const float8_t b); +ccl_device_inline float8_t operator-(const float8_t a, const float f); +ccl_device_inline float8_t operator-(const float f, const float8_t a); -ccl_device_inline float8 operator*(const float8 &a, const float8 &b); -ccl_device_inline float8 operator*(const float8 &a, const float f); -ccl_device_inline float8 operator*(const float f, const float8 &a); +ccl_device_inline float8_t operator*(const float8_t a, const float8_t b); +ccl_device_inline float8_t operator*(const float8_t a, const float f); +ccl_device_inline float8_t operator*(const float f, const float8_t a); -ccl_device_inline float8 operator/(const float8 &a, const float8 &b); -ccl_device_inline float8 operator/(const float8 &a, float f); -ccl_device_inline float8 operator/(const float f, const float8 &a); +ccl_device_inline float8_t operator/(const float8_t a, const float8_t b); +ccl_device_inline float8_t operator/(const float8_t a, float f); +ccl_device_inline float8_t operator/(const float f, const float8_t a); -ccl_device_inline float8 operator+=(float8 &a, const float8 &b); +ccl_device_inline float8_t operator+=(float8_t a, const float8_t b); -ccl_device_inline float8 operator*=(float8 &a, const float8 &b); -ccl_device_inline float8 operator*=(float8 &a, float f); +ccl_device_inline float8_t operator*=(float8_t a, const float8_t b); +ccl_device_inline float8_t operator*=(float8_t a, float f); -ccl_device_inline float8 operator/=(float8 &a, float f); +ccl_device_inline float8_t operator/=(float8_t a, float f); -ccl_device_inline bool operator==(const float8 &a, const float8 &b); +ccl_device_inline bool operator==(const float8_t a, const float8_t b); -ccl_device_inline float8 rcp(const float8 &a); -ccl_device_inline float8 sqrt(const float8 &a); -ccl_device_inline float8 sqr(const float8 &a); -ccl_device_inline bool is_zero(const float8 &a); -ccl_device_inline float average(const float8 &a); -ccl_device_inline float8 min(const float8 &a, const float8 &b); -ccl_device_inline float8 max(const float8 &a, const float8 &b); -ccl_device_inline float8 clamp(const float8 &a, const float8 &mn, const float8 &mx); -ccl_device_inline float8 fabs(const float8 &a); -ccl_device_inline float8 mix(const float8 &a, const float8 &b, float t); +ccl_device_inline float8_t rcp(const float8_t a); +ccl_device_inline float8_t sqrt(const float8_t a); +ccl_device_inline float8_t sqr(const float8_t a); +ccl_device_inline bool is_zero(const float8_t a); +ccl_device_inline float average(const float8_t a); +ccl_device_inline float8_t min(const float8_t a, const float8_t b); +ccl_device_inline float8_t max(const float8_t a, const float8_t b); +ccl_device_inline float8_t clamp(const float8_t a, const float8_t mn, const float8_t mx); +ccl_device_inline float8_t fabs(const float8_t a); +ccl_device_inline float8_t mix(const float8_t a, const float8_t b, float t); +ccl_device_inline float8_t saturate(const float8_t a); -ccl_device_inline float8 safe_divide(const float8 a, const float b); -ccl_device_inline float8 safe_divide(const float8 a, const float8 b); +ccl_device_inline float8_t safe_divide(const float8_t a, const float b); +ccl_device_inline float8_t safe_divide(const float8_t a, const float8_t b); -ccl_device_inline float reduce_min(const float8 &a); -ccl_device_inline float reduce_max(const float8 &a); -ccl_device_inline float reduce_add(const float8 &a); +ccl_device_inline float reduce_min(const float8_t a); +ccl_device_inline float reduce_max(const float8_t a); +ccl_device_inline float reduce_add(const float8_t a); -ccl_device_inline float8 saturate(const float8 &a); -ccl_device_inline bool isequal(const float8 a, const float8 b); +ccl_device_inline bool isequal(const float8_t a, const float8_t b); /******************************************************************************* * Definition. */ -ccl_device_inline float8 zero_float8() +ccl_device_inline float8_t zero_float8_t() { #ifdef __KERNEL_AVX2__ - return float8(_mm256_setzero_ps()); + return float8_t(_mm256_setzero_ps()); #else - return make_float8(0.0f); + return make_float8_t(0.0f); #endif } -ccl_device_inline float8 one_float8() +ccl_device_inline float8_t one_float8_t() { - return make_float8(1.0f); + return make_float8_t(1.0f); } -ccl_device_inline float8 operator+(const float8 &a, const float8 &b) +ccl_device_inline float8_t operator+(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_add_ps(a.m256, b.m256)); + return float8_t(_mm256_add_ps(a.m256, b.m256)); #else - return make_float8( + return make_float8_t( a.a + b.a, a.b + b.b, a.c + b.c, a.d + b.d, a.e + b.e, a.f + b.f, a.g + b.g, a.h + b.h); #endif } -ccl_device_inline float8 operator+(const float8 &a, const float f) +ccl_device_inline float8_t operator+(const float8_t a, const float f) { - return a + make_float8(f); + return a + make_float8_t(f); } -ccl_device_inline float8 operator+(const float f, const float8 &a) +ccl_device_inline float8_t operator+(const float f, const float8_t a) { - return make_float8(f) + a; + return make_float8_t(f) + a; } -ccl_device_inline float8 operator-(const float8 &a) +ccl_device_inline float8_t operator-(const float8_t a) { #ifdef __KERNEL_AVX2__ __m256 mask = _mm256_castsi256_ps(_mm256_set1_epi32(0x80000000)); - return float8(_mm256_xor_ps(a.m256, mask)); + return float8_t(_mm256_xor_ps(a.m256, mask)); #else - return make_float8(-a.a, -a.b, -a.c, -a.d, -a.e, -a.f, -a.g, -a.h); + return make_float8_t(-a.a, -a.b, -a.c, -a.d, -a.e, -a.f, -a.g, -a.h); #endif } -ccl_device_inline float8 operator-(const float8 &a, const float8 &b) +ccl_device_inline float8_t operator-(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_sub_ps(a.m256, b.m256)); + return float8_t(_mm256_sub_ps(a.m256, b.m256)); #else - return make_float8( + return make_float8_t( a.a - b.a, a.b - b.b, a.c - b.c, a.d - b.d, a.e - b.e, a.f - b.f, a.g - b.g, a.h - b.h); #endif } -ccl_device_inline float8 operator-(const float8 &a, const float f) +ccl_device_inline float8_t operator-(const float8_t a, const float f) { - return a - make_float8(f); + return a - make_float8_t(f); } -ccl_device_inline float8 operator-(const float f, const float8 &a) +ccl_device_inline float8_t operator-(const float f, const float8_t a) { - return make_float8(f) - a; + return make_float8_t(f) - a; } -ccl_device_inline float8 operator*(const float8 &a, const float8 &b) +ccl_device_inline float8_t operator*(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_mul_ps(a.m256, b.m256)); + return float8_t(_mm256_mul_ps(a.m256, b.m256)); #else - return make_float8( + return make_float8_t( a.a * b.a, a.b * b.b, a.c * b.c, a.d * b.d, a.e * b.e, a.f * b.f, a.g * b.g, a.h * b.h); #endif } -ccl_device_inline float8 operator*(const float8 &a, const float f) +ccl_device_inline float8_t operator*(const float8_t a, const float f) { - return a * make_float8(f); + return a * make_float8_t(f); } -ccl_device_inline float8 operator*(const float f, const float8 &a) +ccl_device_inline float8_t operator*(const float f, const float8_t a) { - return make_float8(f) * a; + return make_float8_t(f) * a; } -ccl_device_inline float8 operator/(const float8 &a, const float8 &b) +ccl_device_inline float8_t operator/(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_div_ps(a.m256, b.m256)); + return float8_t(_mm256_div_ps(a.m256, b.m256)); #else - return make_float8( + return make_float8_t( a.a / b.a, a.b / b.b, a.c / b.c, a.d / b.d, a.e / b.e, a.f / b.f, a.g / b.g, a.h / b.h); #endif } -ccl_device_inline float8 operator/(const float8 &a, const float f) +ccl_device_inline float8_t operator/(const float8_t a, const float f) { - return a / make_float8(f); + return a / make_float8_t(f); } -ccl_device_inline float8 operator/(const float f, const float8 &a) +ccl_device_inline float8_t operator/(const float f, const float8_t a) { - return make_float8(f) / a; + return make_float8_t(f) / a; } -ccl_device_inline float8 operator+=(float8 &a, const float8 &b) +ccl_device_inline float8_t operator+=(float8_t a, const float8_t b) { return a = a + b; } -ccl_device_inline float8 operator-=(float8 &a, const float8 &b) +ccl_device_inline float8_t operator-=(float8_t a, const float8_t b) { return a = a - b; } -ccl_device_inline float8 operator*=(float8 &a, const float8 &b) +ccl_device_inline float8_t operator*=(float8_t a, const float8_t b) { return a = a * b; } -ccl_device_inline float8 operator*=(float8 &a, float f) +ccl_device_inline float8_t operator*=(float8_t a, float f) { return a = a * f; } -ccl_device_inline float8 operator/=(float8 &a, float f) +ccl_device_inline float8_t operator/=(float8_t a, float f) { return a = a / f; } -ccl_device_inline bool operator==(const float8 &a, const float8 &b) +ccl_device_inline bool operator==(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ return (_mm256_movemask_ps(_mm256_castsi256_ps( @@ -206,160 +206,195 @@ ccl_device_inline bool operator==(const float8 &a, const float8 &b) #endif } -ccl_device_inline float8 rcp(const float8 &a) +ccl_device_inline float8_t rcp(const float8_t a) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_rcp_ps(a.m256)); + return float8_t(_mm256_rcp_ps(a.m256)); #else - return make_float8(1.0f / a.a, - 1.0f / a.b, - 1.0f / a.c, - 1.0f / a.d, - 1.0f / a.e, - 1.0f / a.f, - 1.0f / a.g, - 1.0f / a.h); + return make_float8_t(1.0f / a.a, + 1.0f / a.b, + 1.0f / a.c, + 1.0f / a.d, + 1.0f / a.e, + 1.0f / a.f, + 1.0f / a.g, + 1.0f / a.h); #endif } -ccl_device_inline float8 sqrt(const float8 &a) +ccl_device_inline float8_t sqrt(const float8_t a) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_sqrt_ps(a.m256)); + return float8_t(_mm256_sqrt_ps(a.m256)); #else - return make_float8(sqrtf(a.a), - sqrtf(a.b), - sqrtf(a.c), - sqrtf(a.d), - sqrtf(a.e), - sqrtf(a.f), - sqrtf(a.g), - sqrtf(a.h)); + return make_float8_t(sqrtf(a.a), + sqrtf(a.b), + sqrtf(a.c), + sqrtf(a.d), + sqrtf(a.e), + sqrtf(a.f), + sqrtf(a.g), + sqrtf(a.h)); #endif } -ccl_device_inline float8 sqr(const float8 &a) +ccl_device_inline float8_t sqr(const float8_t a) { return a * a; } -ccl_device_inline bool is_zero(const float8 &a) +ccl_device_inline bool is_zero(const float8_t a) { - return a == make_float8(0.0f); + return a == make_float8_t(0.0f); } -ccl_device_inline float average(const float8 &a) +ccl_device_inline float average(const float8_t a) { return reduce_add(a) / 8.0f; } -ccl_device_inline float8 min(const float8 &a, const float8 &b) +ccl_device_inline float8_t min(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_min_ps(a.m256, b.m256)); + return float8_t(_mm256_min_ps(a.m256, b.m256)); #else - return make_float8(min(a.a, b.a), - min(a.b, b.b), - min(a.c, b.c), - min(a.d, b.d), - min(a.e, b.e), - min(a.f, b.f), - min(a.g, b.g), - min(a.h, b.h)); + return make_float8_t(min(a.a, b.a), + min(a.b, b.b), + min(a.c, b.c), + min(a.d, b.d), + min(a.e, b.e), + min(a.f, b.f), + min(a.g, b.g), + min(a.h, b.h)); #endif } -ccl_device_inline float8 max(const float8 &a, const float8 &b) +ccl_device_inline float8_t max(const float8_t a, const float8_t b) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_max_ps(a.m256, b.m256)); + return float8_t(_mm256_max_ps(a.m256, b.m256)); #else - return make_float8(max(a.a, b.a), - max(a.b, b.b), - max(a.c, b.c), - max(a.d, b.d), - max(a.e, b.e), - max(a.f, b.f), - max(a.g, b.g), - max(a.h, b.h)); + return make_float8_t(max(a.a, b.a), + max(a.b, b.b), + max(a.c, b.c), + max(a.d, b.d), + max(a.e, b.e), + max(a.f, b.f), + max(a.g, b.g), + max(a.h, b.h)); #endif } -ccl_device_inline float8 clamp(const float8 &a, const float8 &mn, const float8 &mx) +ccl_device_inline float8_t clamp(const float8_t a, const float8_t mn, const float8_t mx) { return min(max(a, mn), mx); } -ccl_device_inline float8 fabs(const float8 &a) +ccl_device_inline float8_t fabs(const float8_t a) { #ifdef __KERNEL_AVX2__ - return float8(_mm256_and_ps(a.m256, _mm256_castsi256_ps(_mm256_set1_epi32(0x7fffffff)))); + return float8_t(_mm256_and_ps(a.m256, _mm256_castsi256_ps(_mm256_set1_epi32(0x7fffffff)))); #else - return make_float8(fabsf(a.a), - fabsf(a.b), - fabsf(a.c), - fabsf(a.d), - fabsf(a.e), - fabsf(a.f), - fabsf(a.g), - fabsf(a.h)); + return make_float8_t(fabsf(a.a), + fabsf(a.b), + fabsf(a.c), + fabsf(a.d), + fabsf(a.e), + fabsf(a.f), + fabsf(a.g), + fabsf(a.h)); #endif } -ccl_device_inline float8 mix(const float8 &a, const float8 &b, float t) +ccl_device_inline float8_t mix(const float8_t a, const float8_t b, float t) { return a + t * (b - a); } -ccl_device_inline float reduce_min(const float8 &a) +ccl_device_inline float8_t saturate(const float8_t a) +{ + return clamp(a, make_float8_t(0.0f), make_float8_t(1.0f)); +} + +ccl_device_inline float8_t exp(float8_t v) +{ + return make_float8_t( + expf(v.a), expf(v.b), expf(v.c), expf(v.d), expf(v.e), expf(v.f), expf(v.g), expf(v.h)); +} + +ccl_device_inline float8_t log(float8_t v) +{ + return make_float8_t( + logf(v.a), logf(v.b), logf(v.c), logf(v.d), logf(v.e), logf(v.f), logf(v.g), logf(v.h)); +} + +ccl_device_inline float dot(const float8_t a, const float8_t b) +{ +#ifdef __KERNEL_AVX2__ + float8_t t(_mm256_dp_ps(a.m256, b.m256, 0xFF)); + return t[0] + t[4]; +#else + return (a.a * b.a) + (a.b * b.b) + (a.c * b.c) + (a.d * b.d) + (a.e * b.e) + (a.f * b.f) + + (a.g * b.g) + (a.h * b.h); +#endif +} + +ccl_device_inline float8_t pow(float8_t v, float e) +{ + return make_float8_t(powf(v.a, e), + powf(v.b, e), + powf(v.c, e), + powf(v.d, e), + powf(v.e, e), + powf(v.f, e), + powf(v.g, e), + powf(v.h, e)); +} + +ccl_device_inline float reduce_min(const float8_t a) { return min(min(min(a.a, a.b), min(a.c, a.d)), min(min(a.e, a.f), min(a.g, a.h))); } -ccl_device_inline float reduce_max(const float8 &a) +ccl_device_inline float reduce_max(const float8_t a) { return max(max(max(a.a, a.b), max(a.c, a.d)), max(max(a.e, a.f), max(a.g, a.h))); } -ccl_device_inline float reduce_add(const float8 &a) +ccl_device_inline float reduce_add(const float8_t a) { #ifdef __KERNEL_AVX2__ - float8 b(_mm256_hadd_ps(a.m256, a.m256)); - float8 h(_mm256_hadd_ps(b.m256, b.m256)); + float8_t b(_mm256_hadd_ps(a.m256, a.m256)); + float8_t h(_mm256_hadd_ps(b.m256, b.m256)); return h[0] + h[4]; #else return a.a + a.b + a.c + a.d + a.e + a.f + a.g + a.h; #endif } -ccl_device_inline float8 saturate(const float8 &a) -{ - return clamp(a, make_float8(0.0f), make_float8(1.0f)); -} - -ccl_device_inline bool isequal(const float8 a, const float8 b) +ccl_device_inline bool isequal(const float8_t a, const float8_t b) { return a == b; } -ccl_device_inline float8 safe_divide(const float8 a, const float b) +ccl_device_inline float8_t safe_divide(const float8_t a, const float b) { - return (b != 0.0f) ? a / b : make_float8(0.0f); + return (b != 0.0f) ? a / b : make_float8_t(0.0f); } -ccl_device_inline float8 safe_divide(const float8 a, const float8 b) +ccl_device_inline float8_t safe_divide(const float8_t a, const float8_t b) { - return make_float8((b.a != 0.0f) ? a.a / b.a : 0.0f, - (b.b != 0.0f) ? a.b / b.b : 0.0f, - (b.c != 0.0f) ? a.c / b.c : 0.0f, - (b.d != 0.0f) ? a.d / b.d : 0.0f, - (b.e != 0.0f) ? a.e / b.e : 0.0f, - (b.f != 0.0f) ? a.f / b.f : 0.0f, - (b.g != 0.0f) ? a.g / b.g : 0.0f, - (b.h != 0.0f) ? a.h / b.h : 0.0f); + return make_float8_t((b.a != 0.0f) ? a.a / b.a : 0.0f, + (b.b != 0.0f) ? a.b / b.b : 0.0f, + (b.c != 0.0f) ? a.c / b.c : 0.0f, + (b.d != 0.0f) ? a.d / b.d : 0.0f, + (b.e != 0.0f) ? a.e / b.e : 0.0f, + (b.f != 0.0f) ? a.f / b.f : 0.0f, + (b.g != 0.0f) ? a.g / b.g : 0.0f, + (b.h != 0.0f) ? a.h / b.h : 0.0f); } -ccl_device_inline float8 ensure_finite(float8 v) +ccl_device_inline float8_t ensure_finite(float8_t v) { v.a = ensure_finite(v.a); v.b = ensure_finite(v.b); @@ -373,47 +408,12 @@ ccl_device_inline float8 ensure_finite(float8 v) return v; } -ccl_device_inline bool isfinite_safe(float8 v) +ccl_device_inline bool isfinite_safe(float8_t v) { return isfinite_safe(v.a) && isfinite_safe(v.b) && isfinite_safe(v.c) && isfinite_safe(v.d) && isfinite_safe(v.e) && isfinite_safe(v.f) && isfinite_safe(v.g) && isfinite_safe(v.h); } -ccl_device_inline float8 pow(float8 v, float e) -{ - return make_float8(powf(v.a, e), - powf(v.b, e), - powf(v.c, e), - powf(v.d, e), - powf(v.e, e), - powf(v.f, e), - powf(v.g, e), - powf(v.h, e)); -} - -ccl_device_inline float8 exp(float8 v) -{ - return make_float8( - expf(v.a), expf(v.b), expf(v.c), expf(v.d), expf(v.e), expf(v.f), expf(v.g), expf(v.h)); -} - -ccl_device_inline float8 log(float8 v) -{ - return make_float8( - logf(v.a), logf(v.b), logf(v.c), logf(v.d), logf(v.e), logf(v.f), logf(v.g), logf(v.h)); -} - -ccl_device_inline float dot(const float8 &a, const float8 &b) -{ -#ifdef __KERNEL_AVX2__ - float8 t(_mm256_dp_ps(a.m256, b.m256, 0xFF)); - return t[0] + t[4]; -#else - return (a.a * b.a) + (a.b * b.b) + (a.c * b.c) + (a.d * b.d) + (a.e * b.e) + (a.f * b.f) + - (a.g * b.g) + (a.h * b.h); -#endif -} - CCL_NAMESPACE_END #endif /* __UTIL_MATH_FLOAT8_H__ */ diff --git a/intern/cycles/util/types_float8.h b/intern/cycles/util/types_float8.h index f04dc675c84..bb9798932ac 100644 --- a/intern/cycles/util/types_float8.h +++ b/intern/cycles/util/types_float8.h @@ -11,10 +11,13 @@ CCL_NAMESPACE_BEGIN +/* float8 is a reserved type in Metal that has not been implemented. For + * that reason this is named float8_t. */ + #ifdef __KERNEL_GPU__ -struct float8 +struct float8_t #else -struct ccl_try_align(32) float8 +struct ccl_try_align(32) float8_t #endif { #ifdef __KERNEL_AVX2__ @@ -25,14 +28,14 @@ struct ccl_try_align(32) float8 }; }; - __forceinline float8(); - __forceinline float8(const float8 &a); - __forceinline explicit float8(const __m256 &a); + __forceinline float8_t(); + __forceinline float8_t(const float8_t &a); + __forceinline explicit float8_t(const __m256 &a); __forceinline operator const __m256 &() const; __forceinline operator __m256 &(); - __forceinline float8 &operator=(const float8 &a); + __forceinline float8_t &operator=(const float8_t &a); #else /* __KERNEL_AVX2__ */ float a, b, c, d, e, f, g, h; @@ -44,9 +47,9 @@ struct ccl_try_align(32) float8 #endif }; -ccl_device_inline float8 make_float8(float f); -ccl_device_inline float8 -make_float8(float a, float b, float c, float d, float e, float f, float g, float h); +ccl_device_inline float8_t make_float8_t(float f); +ccl_device_inline float8_t +make_float8_t(float a, float b, float c, float d, float e, float f, float g, float h); CCL_NAMESPACE_END diff --git a/intern/cycles/util/types_float8_impl.h b/intern/cycles/util/types_float8_impl.h index 21931c55071..2ab464a791b 100644 --- a/intern/cycles/util/types_float8_impl.h +++ b/intern/cycles/util/types_float8_impl.h @@ -16,29 +16,29 @@ CCL_NAMESPACE_BEGIN #ifdef __KERNEL_AVX2__ -__forceinline float8::float8() +__forceinline float8_t::float8_t() { } -__forceinline float8::float8(const float8 &f) : m256(f.m256) +__forceinline float8_t::float8_t(const float8_t &f) : m256(f.m256) { } -__forceinline float8::float8(const __m256 &f) : m256(f) +__forceinline float8_t::float8_t(const __m256 &f) : m256(f) { } -__forceinline float8::operator const __m256 &() const +__forceinline float8_t::operator const __m256 &() const { return m256; } -__forceinline float8::operator __m256 &() +__forceinline float8_t::operator __m256 &() { return m256; } -__forceinline float8 &float8::operator=(const float8 &f) +__forceinline float8_t &float8_t::operator=(const float8_t &f) { m256 = f.m256; return *this; @@ -46,14 +46,14 @@ __forceinline float8 &float8::operator=(const float8 &f) #endif /* __KERNEL_AVX2__ */ #ifndef __KERNEL_GPU__ -__forceinline float float8::operator[](int i) const +__forceinline float float8_t::operator[](int i) const { util_assert(i >= 0); util_assert(i < 8); return *(&a + i); } -__forceinline float &float8::operator[](int i) +__forceinline float &float8_t::operator[](int i) { util_assert(i >= 0); util_assert(i < 8); @@ -61,23 +61,23 @@ __forceinline float &float8::operator[](int i) } #endif -ccl_device_inline float8 make_float8(float f) +ccl_device_inline float8_t make_float8_t(float f) { #ifdef __KERNEL_AVX2__ - float8 r(_mm256_set1_ps(f)); + float8_t r(_mm256_set1_ps(f)); #else - float8 r = {f, f, f, f, f, f, f, f}; + float8_t r = {f, f, f, f, f, f, f, f}; #endif return r; } -ccl_device_inline float8 -make_float8(float a, float b, float c, float d, float e, float f, float g, float h) +ccl_device_inline float8_t +make_float8_t(float a, float b, float c, float d, float e, float f, float g, float h) { #ifdef __KERNEL_AVX2__ - float8 r(_mm256_setr_ps(a, b, c, d, e, f, g, h)); + float8_t r(_mm256_setr_ps(a, b, c, d, e, f, g, h)); #else - float8 r = {a, b, c, d, e, f, g, h}; + float8_t r = {a, b, c, d, e, f, g, h}; #endif return r; } -- cgit v1.2.3