/* Permission is hereby granted, free of charge, to any person * obtaining a copy of this software and associated documentation * files (the "Software"), to deal in the Software without * restriction, including without limitation the rights to use, copy, * modify, merge, publish, distribute, sublicense, and/or sell copies * of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be * included in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. * * Copyright: * 2017 Evan Nemerson * 2015-2017 John W. Ratcliff * 2015 Brandon Rowlett * 2015 Ken Fast */ #if !defined(SIMDE__SSE_H) #if !defined(SIMDE__SSE_H) #define SIMDE__SSE_H #endif #include "mmx.h" #if defined(SIMDE_SSE_NATIVE) #undef SIMDE_SSE_NATIVE #endif #if defined(SIMDE_SSE_FORCE_NATIVE) #define SIMDE_SSE_NATIVE #elif defined(__SSE__) && !defined(SIMDE_SSE_NO_NATIVE) && \ !defined(SIMDE_NO_NATIVE) #define SIMDE_SSE_NATIVE #elif defined(__ARM_NEON) && !defined(SIMDE_SSE_NO_NEON) && \ !defined(SIMDE_NO_NEON) #define SIMDE_SSE_NEON #endif #if defined(SIMDE_SSE_NATIVE) && !defined(SIMDE_MMX_NATIVE) #if defined(SIMDE_SSE_FORCE_NATIVE) #error Native SSE support requires native MMX support #else #warning Native SSE support requires native MMX support, disabling #undef SIMDE_SSE_NATIVE #endif #elif defined(SIMDE_SSE_NEON) && !defined(SIMDE_MMX_NEON) #warning SSE3 NEON support requires MMX NEON support, disabling #undef SIMDE_SSE3_NEON #endif #if defined(SIMDE_SSE_NATIVE) #include #else #if defined(SIMDE_SSE_NEON) #include #endif #if !defined(__INTEL_COMPILER) && defined(__STDC_VERSION__) && \ (__STDC_VERSION__ >= 201112L) && !defined(__STDC_NO_ATOMICS__) #include #elif defined(_WIN32) #include #endif #endif #include #include #define SIMDE_ALIGN(alignment) __attribute__((aligned(alignment))) SIMDE__BEGIN_DECLS typedef SIMDE_ALIGN(16) union { #if defined(SIMDE__ENABLE_GCC_VEC_EXT) int8_t i8 __attribute__((__vector_size__(16), __may_alias__)); int16_t i16 __attribute__((__vector_size__(16), __may_alias__)); int32_t i32 __attribute__((__vector_size__(16), __may_alias__)); int64_t i64 __attribute__((__vector_size__(16), __may_alias__)); uint8_t u8 __attribute__((__vector_size__(16), __may_alias__)); uint16_t u16 __attribute__((__vector_size__(16), __may_alias__)); uint32_t u32 __attribute__((__vector_size__(16), __may_alias__)); uint64_t u64 __attribute__((__vector_size__(16), __may_alias__)); #if defined(SIMDE__HAVE_INT128) simde_int128 i128 __attribute__((__vector_size__(16), __may_alias__)); simde_uint128 u128 __attribute__((__vector_size__(16), __may_alias__)); #endif simde_float32 f32 __attribute__((__vector_size__(16), __may_alias__)); #else int8_t i8[16]; int16_t i16[8]; int32_t i32[4]; int64_t i64[2]; uint8_t u8[16]; uint16_t u16[8]; uint32_t u32[4]; uint64_t u64[2]; #if defined(SIMDE__HAVE_INT128) simde_int128 i128[1]; simde_uint128 u128[1]; #endif simde_float32 f32[4]; #endif #if defined(SIMDE_SSE_NATIVE) __m128 n; #elif defined(SIMDE_SSE_NEON) int8x16_t neon_i8; int16x8_t neon_i16; int32x4_t neon_i32; int64x2_t neon_i64; uint8x16_t neon_u8; uint16x8_t neon_u16; uint32x4_t neon_u32; uint64x2_t neon_u64; float32x4_t neon_f32; #endif } simde__m128; #if defined(SIMDE_SSE_NATIVE) HEDLEY_STATIC_ASSERT(sizeof(__m128) == sizeof(simde__m128), "__m128 size doesn't match simde__m128 size"); SIMDE__FUNCTION_ATTRIBUTES simde__m128 SIMDE__M128_C(__m128 v) { simde__m128 r; r.n = v; return r; } #elif defined(SIMDE_SSE_NEON) #define SIMDE__M128_NEON_C(T, expr) \ (simde__m128) { .neon_##T = expr } #endif HEDLEY_STATIC_ASSERT(16 == sizeof(simde__m128), "simde__m128 size incorrect"); SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_add_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_add_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vaddq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i] + b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_add_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_add_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32_t b0 = vgetq_lane_f32(b.neon_f32, 0); float32x4_t value = vsetq_lane_f32(b0, vdupq_n_f32(0), 0); /* the upper values in the result must be the remnants of . */ r.neon_f32 = vaddq_f32(a.neon_f32, value); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_add_ps(a, b).f32, 4, 1, 2, 3); #else r.f32[0] = a.f32[0] + b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_and_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_and_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_i32 = vandq_s32(a.neon_i32, b.neon_i32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i32) / sizeof(r.i32[0])); i++) { r.i32[i] = a.i32[i] & b.i32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_andnot_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_andnot_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_i32 = vbicq_s32(b.neon_i32, a.neon_i32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i32) / sizeof(r.i32[0])); i++) { r.i32[i] = ~(a.i32[i]) & b.i32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_avg_pu16(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_avg_pu16(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u16 = vrhadd_u16(b.neon_u16, a.neon_u16); #else SIMDE__VECTORIZE for (size_t i = 0; i < 4; i++) { r.u16[i] = (a.u16[i] + b.u16[i] + 1) >> 1; } #endif return r; } #define simde_m_pavgw(a, b) simde_mm_avg_pu16(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_avg_pu8(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_avg_pu8(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u8 = vrhadd_u8(b.neon_u8, a.neon_u8); #else SIMDE__VECTORIZE for (size_t i = 0; i < 8; i++) { r.u8[i] = (a.u8[i] + b.u8[i] + 1) >> 1; } #endif return r; } #define simde_m_pavgb(a, b) simde_mm_avg_pu8(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpeq_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpeq_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vceqq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] == b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpeq_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpeq_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vceqq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmpeq_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] == b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpge_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpge_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcgeq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] >= b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpge_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) r.n = _mm_cmpge_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcgeq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmpge_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] >= b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpgt_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpgt_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcgtq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] > b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpgt_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) r.n = _mm_cmpgt_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcgtq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmpgt_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] > b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmple_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmple_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcleq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] <= b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmple_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmple_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcleq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmple_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] <= b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmplt_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmplt_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcltq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] < b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmplt_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmplt_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcltq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmplt_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] < b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpneq_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpneq_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vmvnq_u32(vceqq_f32(a.neon_f32, b.neon_f32)); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (a.f32[i] != b.f32[i]) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpneq_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpneq_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t e = vreinterpretq_f32_u32(vceqq_f32(a.neon_f32, b.neon_f32)); float32x4_t s = vreinterpretq_f32_u32(vmvnq_u32(vreinterpretq_u32_f32(e))); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmpneq_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (a.f32[0] != b.f32[0]) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = a.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnge_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpnge_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcltq_f32(a.neon_f32, b.neon_f32); #else r = simde_mm_cmplt_ps(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnge_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) r.n = _mm_cmpnge_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcltq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #else r = simde_mm_cmplt_ss(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpngt_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpngt_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcleq_f32(a.neon_f32, b.neon_f32); #else r = simde_mm_cmple_ps(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpngt_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) r.n = _mm_cmpngt_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcleq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #else r = simde_mm_cmple_ss(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnle_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpnle_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcgtq_f32(a.neon_f32, b.neon_f32); #else r = simde_mm_cmpgt_ps(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnle_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpnle_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t s = vreinterpretq_f32_u32(vcgtq_f32(a.neon_f32, b.neon_f32)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #else r = simde_mm_cmpgt_ss(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnlt_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpnlt_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_u32 = vcgeq_f32(a.neon_f32, b.neon_f32); #else r = simde_mm_cmpge_ps(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpnlt_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpnlt_ss(a.n, b.n); #else r = simde_mm_cmpge_ss(a, b); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpord_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpord_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) /* Note: NEON does not have ordered compare builtin Need to compare a eq a and b eq b to check for NaN Do AND of results to get final */ uint32x4_t ceqaa = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t ceqbb = vceqq_f32(b.neon_f32, b.neon_f32); r.neon_u32 = vandq_u32(ceqaa, ceqbb); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (isnan(a.f32[i]) || isnan(b.f32[i])) ? 0 : 0xffffffff; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpord_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpord_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t ceqaa = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t ceqbb = vceqq_f32(b.neon_f32, b.neon_f32); float32x4_t s = vreinterpretq_f32_u32(vandq_u32(ceqaa, ceqbb)); float32x4_t t = vextq_f32(a.neon_f32, s, 1); r.neon_f32 = vextq_f32(t, t, 3); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR(32, 16, a.f32, simde_mm_cmpord_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (isnan(a.f32[0]) || isnan(b.f32[0])) ? 0 : 0xffffffff; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpunord_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cmpunord_ps(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.u32[i] = (isnan(a.f32[i]) || isnan(b.f32[i])) ? 0xffffffff : 0; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cmpunord_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) r.n = _mm_cmpunord_ss(a.n, b.n); #elif defined(SIMDE__SHUFFLE_VECTOR) && defined(SIMDE_ASSUME_VECTORIZATION) r.f32 = SIMDE__SHUFFLE_VECTOR( 32, 16, a.f32, simde_mm_cmpunord_ps(a, b).f32, 4, 1, 2, 3); #else r.u32[0] = (isnan(a.f32[0]) || isnan(b.f32[0])) ? 0xffffffff : 0; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comieq_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comieq_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_or_b_nan = vmvnq_u32(vandq_u32(a_not_nan, b_not_nan)); uint32x4_t a_eq_b = vceqq_f32(a.neon_f32, b.neon_f32); return (vgetq_lane_u32(vorrq_u32(a_or_b_nan, a_eq_b), 0) != 0) ? 1 : 0; #else return a.f32[0] == b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comige_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comige_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_and_b_not_nan = vandq_u32(a_not_nan, b_not_nan); uint32x4_t a_ge_b = vcgeq_f32(a.neon_f32, b.neon_f32); return (vgetq_lane_u32(vandq_u32(a_and_b_not_nan, a_ge_b), 0) != 0) ? 1 : 0; #else return a.f32[0] >= b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comigt_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comigt_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_and_b_not_nan = vandq_u32(a_not_nan, b_not_nan); uint32x4_t a_gt_b = vcgtq_f32(a.neon_f32, b.neon_f32); return (vgetq_lane_u32(vandq_u32(a_and_b_not_nan, a_gt_b), 0) != 0) ? 1 : 0; #else return a.f32[0] > b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comile_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comile_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_or_b_nan = vmvnq_u32(vandq_u32(a_not_nan, b_not_nan)); uint32x4_t a_le_b = vcleq_f32(a.neon_f32, b.neon_f32); return (vgetq_lane_u32(vorrq_u32(a_or_b_nan, a_le_b), 0) != 0) ? 1 : 0; #else return a.f32[0] <= b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comilt_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comilt_ss(a.n, b.n); #elif defined(SIMDE_SSE_NATIVE) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_or_b_nan = vmvnq_u32(vandq_u32(a_not_nan, b_not_nan)); uint32x4_t a_lt_b = vcltq_f32(a.neon_f32, b.neon_f32); return (vgetq_lane_u32(vorrq_u32(a_or_b_nan, a_lt_b), 0) != 0) ? 1 : 0; #else return a.f32[0] < b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_comineq_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_comineq_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) uint32x4_t a_not_nan = vceqq_f32(a.neon_f32, a.neon_f32); uint32x4_t b_not_nan = vceqq_f32(b.neon_f32, b.neon_f32); uint32x4_t a_and_b_not_nan = vandq_u32(a_not_nan, b_not_nan); uint32x4_t a_neq_b = vmvnq_u32(vceqq_f32(a.neon_f32, b.neon_f32)); return (vgetq_lane_u32(vandq_u32(a_and_b_not_nan, a_neq_b), 0) != 0) ? 1 : 0; #else return a.f32[0] != b.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvt_pi2ps(simde__m128 a, simde__m64 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvt_pi2ps(a.n, b.n); #else r.f32[0] = (simde_float32)b.i32[0]; r.f32[1] = (simde_float32)b.i32[1]; r.i32[2] = a.i32[2]; r.i32[3] = a.i32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvt_ps2pi(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvt_ps2pi(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i32) / sizeof(r.i32[0])); i++) { r.i32[i] = (int32_t)a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvt_si2ss(simde__m128 a, int32_t b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvt_si2ss(a.n, b); #else r.f32[0] = (simde_float32)b; r.i32[1] = a.i32[1]; r.i32[2] = a.i32[2]; r.i32[3] = a.i32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int32_t simde_mm_cvt_ss2si(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_cvt_ss2si(a.n); #else return (int32_t)a.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpi16_ps(simde__m64 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpi16_ps(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = (simde_float32)a.i16[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpi32_ps(simde__m128 a, simde__m64 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpi32_ps(a.n, b.n); #else r.f32[0] = (simde_float32)b.i32[0]; r.f32[1] = (simde_float32)b.i32[1]; r.i32[2] = a.i32[2]; r.i32[3] = a.i32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpi32x2_ps(simde__m64 a, simde__m64 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpi32x2_ps(a.n, b.n); #else r.f32[0] = (simde_float32)a.i32[0]; r.f32[1] = (simde_float32)a.i32[1]; r.f32[2] = (simde_float32)b.i32[0]; r.f32[3] = (simde_float32)b.i32[1]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpi8_ps(simde__m64 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpi8_ps(a.n); #else r.f32[0] = (simde_float32)a.i8[0]; r.f32[1] = (simde_float32)a.i8[1]; r.f32[2] = (simde_float32)a.i8[2]; r.f32[3] = (simde_float32)a.i8[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvtps_pi16(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtps_pi16(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i16) / sizeof(r.i16[0])); i++) { r.i16[i] = (int16_t)a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvtps_pi32(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtps_pi32(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i32) / sizeof(r.i32[0])); i++) { r.i32[i] = (int32_t)a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvtps_pi8(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtps_pi8(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(a.f32) / sizeof(a.f32[0])); i++) { r.i8[i] = (int8_t)a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpu16_ps(simde__m64 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpu16_ps(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = (simde_float32)a.u16[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtpu8_ps(simde__m64 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtpu8_ps(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < 4; i++) { r.f32[i] = (simde_float32)a.u8[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtsi32_ss(simde__m128 a, int32_t b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtsi32_ss(a.n, b); #else r.f32[0] = (simde_float32)b; SIMDE__VECTORIZE for (size_t i = 1; i < 4; i++) { r.i32[i] = a.i32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_cvtsi64_ss(simde__m128 a, int64_t b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) && defined(SIMDE_ARCH_AMD64) #if !defined(__PGI) r.n = _mm_cvtsi64_ss(a.n, b); #else r.n = _mm_cvtsi64x_ss(a.n, b); #endif #else r.f32[0] = (simde_float32)b; SIMDE__VECTORIZE for (size_t i = 1; i < 4; i++) { r.i32[i] = a.i32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde_float32 simde_mm_cvtss_f32(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_cvtss_f32(a.n); #elif defined(SIMDE_SSE_NEON) return vgetq_lane_f32(a.neon_f32, 0); #else return a.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int32_t simde_mm_cvtss_si32(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_cvtss_si32(a.n); #else return (int32_t)a.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES int64_t simde_mm_cvtss_si64(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) && defined(SIMDE_ARCH_AMD64) #if !defined(__PGI) return _mm_cvtss_si64(a.n); #else return _mm_cvtss_si64x(a.n); #endif #else return (int64_t)a.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvtt_ps2pi(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvtt_ps2pi(a.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.i32[i] = (int32_t)truncf(a.f32[i]); } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int32_t simde_mm_cvtt_ss2si(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_cvtt_ss2si(a.n); #else return (int32_t)truncf(a.f32[0]); #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_cvttps_pi32(simde__m128 a) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_cvttps_pi32(a.n); #else r = simde_mm_cvtt_ps2pi(a); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int32_t simde_mm_cvttss_si32(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_cvttss_si32(a.n); #else return (int32_t)truncf(a.f32[0]); #endif } SIMDE__FUNCTION_ATTRIBUTES int64_t simde_mm_cvttss_si64(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) && defined(SIMDE_ARCH_AMD64) #if defined(__PGI) return _mm_cvttss_si64x(a.n); #else return _mm_cvttss_si64(a.n); #endif #else return (int64_t)truncf(a.f32[0]); #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_div_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_div_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x4_t recip0 = vrecpeq_f32(b.neon_f32); float32x4_t recip1 = vmulq_f32(recip0, vrecpsq_f32(recip0, b.neon_f32)); r.neon_f32 = vmulq_f32(a.neon_f32, recip1); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i] / b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_div_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_div_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32_t value = vgetq_lane_f32(simde_mm_div_ps(a, b).neon_f32, 0); r.neon_f32 = vsetq_lane_f32(value, a.neon_f32, 0); #else r.f32[0] = a.f32[0] / b.f32[0]; SIMDE__VECTORIZE for (size_t i = 1; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int32_t simde_mm_extract_pi16(simde__m64 a, const int imm8) { return a.u16[imm8]; } #if defined(SIMDE_SSE_NATIVE) #define simde_mm_extract_pi16(a, imm8) _mm_extract_pi16(a.n, imm8) #endif #define simde_m_pextrw(a, imm8) simde_mm_extract_pi16(a.n, imm8) enum { #if defined(SIMDE_SSE_NATIVE) simde_MM_ROUND_NEAREST = _MM_ROUND_NEAREST, simde_MM_ROUND_DOWN = _MM_ROUND_DOWN, simde_MM_ROUND_UP = _MM_ROUND_UP, simde_MM_ROUND_TOWARD_ZERO = _MM_ROUND_TOWARD_ZERO #else simde_MM_ROUND_NEAREST #if defined(FE_TONEAREST) = FE_TONEAREST #endif , simde_MM_ROUND_DOWN #if defined(FE_DOWNWARD) = FE_DOWNWARD #endif , simde_MM_ROUND_UP #if defined(FE_UPWARD) = FE_UPWARD #endif , simde_MM_ROUND_TOWARD_ZERO #if defined(FE_TOWARDZERO) = FE_TOWARDZERO #endif #endif }; SIMDE__FUNCTION_ATTRIBUTES unsigned int simde_MM_GET_ROUNDING_MODE(void) { #if defined(SIMDE_SSE_NATIVE) return _MM_GET_ROUNDING_MODE(); #else return fegetround(); #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_MM_SET_ROUNDING_MODE(unsigned int a) { #if defined(SIMDE_SSE_NATIVE) _MM_SET_ROUNDING_MODE(a); #else fesetround((int)a); #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_insert_pi16(simde__m64 a, int16_t i, const int imm8) { simde__m64 r; r.i64[0] = a.i64[0]; r.i16[imm8] = i; return r; } #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) #define simde_mm_insert_pi16(a, i, imm8) \ SIMDE__M64_C(_mm_insert_pi16((a).n, i, imm8)); #endif #define simde_m_pinsrw(a, i, imm8) \ SIMDE__M64_C(simde_mm_insert_pi16((a).n, i, imm8)); SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_load_ps(simde_float32 const mem_addr[HEDLEY_ARRAY_PARAM(4)]) { simde__m128 r; simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) r.n = _mm_load_ps(mem_addr); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vld1q_f32(mem_addr); #else memcpy(&r, mem_addr, sizeof(r.f32)); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_load_ps1(simde_float32 const *mem_addr) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_load_ps1(mem_addr); #else const simde_float32 v = *mem_addr; SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i32) / sizeof(r.i32[0])); i++) { r.f32[i] = v; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_load_ss(simde_float32 const *mem_addr) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_load_ss(mem_addr); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vsetq_lane_f32(*mem_addr, vdupq_n_f32(0), 0); #else r.f32[0] = *mem_addr; r.i32[1] = 0; r.i32[2] = 0; r.i32[3] = 0; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_load1_ps(simde_float32 const *mem_addr) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_load1_ps(mem_addr); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vld1q_dup_f32(mem_addr); #else r = simde_mm_load_ps1(mem_addr); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_loadh_pi(simde__m128 a, simde__m64 const *mem_addr) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_loadh_pi(a.n, (__m64 *)mem_addr); #else r.f32[0] = a.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = mem_addr->f32[0]; r.f32[3] = mem_addr->f32[1]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_loadl_pi(simde__m128 a, simde__m64 const *mem_addr) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_loadl_pi(a.n, (__m64 *)mem_addr); #else r.f32[0] = mem_addr->f32[0]; r.f32[1] = mem_addr->f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_loadr_ps(simde_float32 const mem_addr[HEDLEY_ARRAY_PARAM(4)]) { simde__m128 r; simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) r.n = _mm_loadr_ps(mem_addr); #else r.f32[0] = mem_addr[3]; r.f32[1] = mem_addr[2]; r.f32[2] = mem_addr[1]; r.f32[3] = mem_addr[0]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_loadu_ps(simde_float32 const mem_addr[HEDLEY_ARRAY_PARAM(4)]) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_loadu_ps(mem_addr); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vld1q_f32(mem_addr); #else r.f32[0] = mem_addr[0]; r.f32[1] = mem_addr[1]; r.f32[2] = mem_addr[2]; r.f32[3] = mem_addr[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_maskmove_si64(simde__m64 a, simde__m64 mask, char *mem_addr) { #if defined(SIMDE_SSE_NATIVE) _mm_maskmove_si64(a.n, mask.n, mem_addr); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(a.i8) / sizeof(a.i8[0])); i++) if (mask.i8[i] < 0) mem_addr[i] = a.i8[i]; #endif } #define simde_m_maskmovq(a, mask, mem_addr) \ simde_mm_maskmove_si64(a, mask, mem_addr) SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_max_pi16(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_max_pi16(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i16) / sizeof(r.i16[0])); i++) { r.i16[i] = (a.i16[i] > b.i16[i]) ? a.i16[i] : b.i16[i]; } #endif return r; } #define simde_m_pmaxsw(a, b) simde_mm_max_pi16(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_max_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_max_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vmaxq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = (a.f32[i] > b.f32[i]) ? a.f32[i] : b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_max_pu8(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_max_pu8(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.u8) / sizeof(r.u8[0])); i++) { r.u8[i] = (a.u8[i] > b.u8[i]) ? a.u8[i] : b.u8[i]; } #endif return r; } #define simde_m_pmaxub(a, b) simde_mm_max_pu8(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_max_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_max_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32_t value = vgetq_lane_f32(vmaxq_f32(a.neon_f32, b.neon_f32), 0); r.neon_f32 = vsetq_lane_f32(value, a.neon_f32, 0); #else r.f32[0] = (a.f32[0] > b.f32[0]) ? a.f32[0] : b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_min_pi16(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_min_pi16(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.i16) / sizeof(r.i16[0])); i++) { r.i16[i] = (a.i16[i] < b.i16[i]) ? a.i16[i] : b.i16[i]; } #endif return r; } #define simde_m_pminsw(a, b) simde_mm_min_pi16(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_min_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_min_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vminq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = (a.f32[i] < b.f32[i]) ? a.f32[i] : b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_min_pu8(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_min_pu8(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.u8) / sizeof(r.u8[0])); i++) { r.u8[i] = (a.u8[i] < b.u8[i]) ? a.u8[i] : b.u8[i]; } #endif return r; } #define simde_m_pminub(a, b) simde_mm_min_pu8(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_min_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_min_ss(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32_t value = vgetq_lane_f32(vminq_f32(a.neon_f32, b.neon_f32), 0); r.neon_f32 = vsetq_lane_f32(value, a.neon_f32, 0); #else r.f32[0] = (a.f32[0] < b.f32[0]) ? a.f32[0] : b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_move_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_move_ss(a.n, b.n); #else r.f32[0] = b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_movehl_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_movehl_ps(a.n, b.n); #else r.f32[0] = b.f32[2]; r.f32[1] = b.f32[3]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_movelh_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_movelh_ps(a.n, b.n); #else r.f32[0] = a.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = b.f32[0]; r.f32[3] = b.f32[1]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_movemask_pi8(simde__m64 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_movemask_pi8(a.n); #else int r = 0; const size_t nmemb = sizeof(a.i8) / sizeof(a.i8[0]); SIMDE__VECTORIZE_REDUCTION(| : r) for (size_t i = 0; i < nmemb; i++) { r |= (a.u8[nmemb - 1 - i] >> 7) << (nmemb - 1 - i); } return r; #endif } #define simde_m_pmovmskb(a, b) simde_mm_movemask_pi8(a, b) SIMDE__FUNCTION_ATTRIBUTES int simde_mm_movemask_ps(simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) return _mm_movemask_ps(a.n); #elif defined(SIMDE_SSE_NEON) /* TODO: check to see if NEON version is faster than the portable version */ static const uint32x4_t movemask = {1, 2, 4, 8}; static const uint32x4_t highbit = {0x80000000, 0x80000000, 0x80000000, 0x80000000}; uint32x4_t t0 = a.neon_u32; uint32x4_t t1 = vtstq_u32(t0, highbit); uint32x4_t t2 = vandq_u32(t1, movemask); uint32x2_t t3 = vorr_u32(vget_low_u32(t2), vget_high_u32(t2)); return vget_lane_u32(t3, 0) | vget_lane_u32(t3, 1); #else int r = 0; SIMDE__VECTORIZE_REDUCTION(| : r) for (size_t i = 0; i < sizeof(a.u32) / sizeof(a.u32[0]); i++) { r |= (a.u32[i] >> ((sizeof(a.u32[i]) * CHAR_BIT) - 1)) << i; } return r; #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_mul_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_mul_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vmulq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i] * b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_mul_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_mul_ss(a.n, b.n); #else r.f32[0] = a.f32[0] * b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_mulhi_pu16(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_mulhi_pu16(a.n, b.n); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.u16) / sizeof(r.u16[0])); i++) { r.u16[i] = (a.u16[i] * b.u16[i]) >> 16; } #endif return r; } #define simde_m_pmulhuw(a, b) simde_mm_mulhi_pu16(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_or_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_or_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_i32 = vorrq_s32(a.neon_i32, b.neon_i32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.u32) / sizeof(r.u32[0])); i++) { r.u32[i] = a.u32[i] | b.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_prefetch(char const *p, int i) { (void)p; (void)i; } #if defined(SIMDE_SSE_NATIVE) #define simde_mm_prefetch(p, i) _mm_prefetch(p, i) #endif SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_rcp_ps(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_rcp_ps(a.n); #elif defined(SIMDE_SSE_NEON) float32x4_t recip = vrecpeq_f32(a.neon_f32); #if !defined(SIMDE_MM_RCP_PS_ITERS) #define SIMDE_MM_RCP_PS_ITERS SIMDE_ACCURACY_ITERS #endif for (int i = 0; i < SIMDE_MM_RCP_PS_ITERS; ++i) { recip = vmulq_f32(recip, vrecpsq_f32(recip, a.neon_f32)); } r.neon_f32 = recip; #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = 1.0f / a.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_rcp_ss(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_rcp_ss(a.n); #else r.f32[0] = 1.0f / a.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_rsqrt_ps(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_rsqrt_ps(a.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vrsqrteq_f32(a.neon_f32); #elif defined(__STDC_IEC_559__) /* http://h14s.p5r.org/2012/09/0x5f3759df.html?mwh=1 */ SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.i32[i] = INT32_C(0x5f3759df) - (a.i32[i] >> 1); #if SIMDE_ACCURACY_ITERS > 2 const float half = SIMDE_FLOAT32_C(0.5) * a.f32[i]; for (int ai = 2; ai < SIMDE_ACCURACY_ITERS; ai++) r.f32[i] *= SIMDE_FLOAT32_C(1.5) - (half * r.f32[i] * r.f32[i]); #endif } #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = 1.0f / sqrtf(a.f32[i]); } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_rsqrt_ss(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_rsqrt_ss(a.n); #elif defined(__STDC_IEC_559__) { r.i32[0] = INT32_C(0x5f3759df) - (a.i32[0] >> 1); #if SIMDE_ACCURACY_ITERS > 2 float half = SIMDE_FLOAT32_C(0.5) * a.f32[0]; for (int ai = 2; ai < SIMDE_ACCURACY_ITERS; ai++) r.f32[0] *= SIMDE_FLOAT32_C(1.5) - (half * r.f32[0] * r.f32[0]); #endif } r.f32[0] = 1.0f / sqrtf(a.f32[0]); r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #else r.f32[0] = 1.0f / sqrtf(a.f32[0]); r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_sad_pu8(simde__m64 a, simde__m64 b) { simde__m64 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_sad_pu8(a.n, b.n); #else uint16_t sum = 0; SIMDE__VECTORIZE_REDUCTION(+ : sum) for (size_t i = 0; i < (sizeof(r.u8) / sizeof(r.u8[0])); i++) { sum += (uint8_t)abs(a.u8[i] - b.u8[i]); } r.i16[0] = sum; r.i16[1] = 0; r.i16[2] = 0; r.i16[3] = 0; #endif return r; } #define simde_m_psadbw(a, b) simde_mm_sad_pu8(a, b) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_set_ps(simde_float32 e3, simde_float32 e2, simde_float32 e1, simde_float32 e0) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_set_ps(e3, e2, e1, e0); #elif defined(SIMDE_SSE_NEON) SIMDE_ALIGN(16) simde_float32 data[4] = {e0, e1, e2, e3}; r.neon_f32 = vld1q_f32(data); #else r.f32[0] = e0; r.f32[1] = e1; r.f32[2] = e2; r.f32[3] = e3; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_set_ps1(simde_float32 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_set1_ps(a); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vdupq_n_f32(a); #else r = simde_mm_set_ps(a, a, a, a); #endif return r; } #define simde_mm_set1_ps(a) simde_mm_set_ps1(a) SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_set_ss(simde_float32 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_set_ss(a); #else r = simde_mm_set_ps(0, 0, 0, a); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_setr_ps(simde_float32 e3, simde_float32 e2, simde_float32 e1, simde_float32 e0) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_setr_ps(e3, e2, e1, e0); #elif defined(SIMDE_SSE_NEON) SIMDE_ALIGN(16) simde_float32 data[4] = {e3, e2, e1, e0}; r.neon_f32 = vld1q_f32(data); #else r = simde_mm_set_ps(e0, e1, e2, e3); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_setzero_ps(void) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_setzero_ps(); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vdupq_n_f32(0.0f); #else r = simde_mm_set_ps(0.0f, 0.0f, 0.0f, 0.0f); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_sfence(void) { /* TODO: Use Hedley. */ #if defined(SIMDE_SSE_NATIVE) _mm_sfence(); #elif defined(__GNUC__) && \ ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)) __atomic_thread_fence(__ATOMIC_SEQ_CST); #elif !defined(__INTEL_COMPILER) && defined(__STDC_VERSION__) && \ (__STDC_VERSION__ >= 201112L) && !defined(__STDC_NO_ATOMICS__) #if defined(__GNUC__) && (__GNUC__ == 4) && (__GNUC_MINOR__ < 9) __atomic_thread_fence(__ATOMIC_SEQ_CST); #else atomic_thread_fence(memory_order_seq_cst); #endif #elif defined(_MSC_VER) MemoryBarrier(); #elif defined(__GNUC__) && \ ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7)) __atomic_thread_fence(__ATOMIC_SEQ_CST); #elif HEDLEY_CLANG_HAS_FEATURE(c_atomic) __c11_atomic_thread_fence(__ATOMIC_SEQ_CST) #elif defined(__GNUC__) && \ ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)) __sync_synchronize(); #elif (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x5140)) || \ (defined(__SUNPRO_CC) && (__SUNPRO_CC >= 0x5140)) __atomic_thread_fence(__ATOMIC_SEQ_CST); #elif defined(_OPENMP) #pragma omp critical(simde_mm_sfence_) { } #endif } #define SIMDE_MM_SHUFFLE(z, y, x, w) \ (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) SIMDE__FUNCTION_ATTRIBUTES simde__m64 simde_mm_shuffle_pi16(simde__m64 a, const int imm8) { simde__m64 r; for (size_t i = 0; i < sizeof(r.u16) / sizeof(r.u16[0]); i++) { r.i16[i] = a.i16[(imm8 >> (i * 2)) & 3]; } return r; } #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) #define simde_mm_shuffle_pi16(a, imm8) SIMDE__M64_C(_mm_shuffle_pi16(a.n, imm8)) #elif defined(SIMDE__SHUFFLE_VECTOR) #define simde_mm_shuffle_pi16(a, imm8) \ ({ \ const simde__m64 simde__tmp_a_ = a; \ (simde__m64){.i16 = SIMDE__SHUFFLE_VECTOR( \ 16, 8, (simde__tmp_a_).i16, \ (simde__tmp_a_).i16, (((imm8)) & 3), \ (((imm8) >> 2) & 3), (((imm8) >> 4) & 3), \ (((imm8) >> 6) & 3))}; \ }) #endif #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) #define simde_m_pshufw(a, imm8) SIMDE__M64_C(_m_pshufw(a.n, imm8)) #else #define simde_m_pshufw(a, imm8) simde_mm_shuffle_pi16(a, imm8) #endif SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_shuffle_ps(simde__m128 a, simde__m128 b, const int imm8) { simde__m128 r; r.f32[0] = a.f32[(imm8 >> 0) & 3]; r.f32[1] = a.f32[(imm8 >> 2) & 3]; r.f32[2] = b.f32[(imm8 >> 4) & 3]; r.f32[3] = b.f32[(imm8 >> 6) & 3]; return r; } #if defined(SIMDE_SSE_NATIVE) && !defined(__PGI) #define simde_mm_shuffle_ps(a, b, imm8) \ SIMDE__M128_C(_mm_shuffle_ps(a.n, b.n, imm8)) #elif defined(SIMDE__SHUFFLE_VECTOR) #define simde_mm_shuffle_ps(a, b, imm8) \ ({ \ (simde__m128){.f32 = SIMDE__SHUFFLE_VECTOR( \ 32, 16, (a).f32, (b).f32, \ (((imm8)) & 3), (((imm8) >> 2) & 3), \ (((imm8) >> 4) & 3) + 4, \ (((imm8) >> 6) & 3) + 4)}; \ }) #endif SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_sqrt_ps(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_sqrt_ps(a.n); #elif defined(SIMDE_SSE_NEON) float32x4_t recipsq = vrsqrteq_f32(a.neon_f32); float32x4_t sq = vrecpeq_f32(recipsq); /* ??? use step versions of both sqrt and recip for better accuracy? */ r.neon_f32 = sq; #else SIMDE__VECTORIZE for (size_t i = 0; i < sizeof(r.f32) / sizeof(r.f32[0]); i++) { r.f32[i] = sqrtf(a.f32[i]); } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_sqrt_ss(simde__m128 a) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_sqrt_ss(a.n); #elif defined(SIMDE_SSE_NEON) float32_t value = vgetq_lane_f32(simde_mm_sqrt_ps(a).neon_f32, 0); r.neon_f32 = vsetq_lane_f32(value, a.neon_f32, 0); #else r.f32[0] = sqrtf(a.f32[0]); r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_store_ps(simde_float32 mem_addr[4], simde__m128 a) { simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) _mm_store_ps(mem_addr, a.n); #elif defined(SIMDE_SSE_NEON) vst1q_f32(mem_addr, a.neon_f32); #else SIMDE__VECTORIZE_ALIGNED(mem_addr : 16) for (size_t i = 0; i < sizeof(a.f32) / sizeof(a.f32[0]); i++) { mem_addr[i] = a.f32[i]; } #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_store_ps1(simde_float32 mem_addr[4], simde__m128 a) { simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) _mm_store_ps1(mem_addr, a.n); #else SIMDE__VECTORIZE_ALIGNED(mem_addr : 16) for (size_t i = 0; i < sizeof(a.f32) / sizeof(a.f32[0]); i++) { mem_addr[i] = a.f32[0]; } #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_store_ss(simde_float32 *mem_addr, simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) _mm_store_ss(mem_addr, a.n); #elif defined(SIMDE_SSE_NEON) vst1q_lane_f32(mem_addr, a.neon_f32, 0); #else *mem_addr = a.f32[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_store1_ps(simde_float32 mem_addr[4], simde__m128 a) { simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) _mm_store1_ps(mem_addr, a.n); #else simde_mm_store_ps1(mem_addr, a); #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_storeh_pi(simde__m64 *mem_addr, simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) _mm_storeh_pi(&(mem_addr->n), a.n); #else mem_addr->f32[0] = a.f32[2]; mem_addr->f32[1] = a.f32[3]; #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_storel_pi(simde__m64 *mem_addr, simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) _mm_storel_pi(&(mem_addr->n), a.n); #else mem_addr->f32[0] = a.f32[0]; mem_addr->f32[1] = a.f32[1]; #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_storer_ps(simde_float32 mem_addr[4], simde__m128 a) { simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) _mm_storer_ps(mem_addr, a.n); #else SIMDE__VECTORIZE_ALIGNED(mem_addr : 16) for (size_t i = 0; i < sizeof(a.f32) / sizeof(a.f32[0]); i++) { mem_addr[i] = a.f32[((sizeof(a.f32) / sizeof(a.f32[0])) - 1) - i]; } #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_storeu_ps(simde_float32 mem_addr[4], simde__m128 a) { #if defined(SIMDE_SSE_NATIVE) _mm_storeu_ps(mem_addr, a.n); #elif defined(SIMDE_SSE_NEON) vst1q_f32(mem_addr, a.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < sizeof(a.f32) / sizeof(a.f32[0]); i++) { mem_addr[i] = a.f32[i]; } #endif } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_sub_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_sub_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_f32 = vsubq_f32(a.neon_f32, b.neon_f32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.f32) / sizeof(r.f32[0])); i++) { r.f32[i] = a.f32[i] - b.f32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_sub_ss(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_sub_ss(a.n, b.n); #else r.f32[0] = a.f32[0] - b.f32[0]; r.f32[1] = a.f32[1]; r.f32[2] = a.f32[2]; r.f32[3] = a.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomieq_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomieq_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] == b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomige_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomige_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] >= b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomigt_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomigt_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] > b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomile_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomile_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] <= b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomilt_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomilt_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] < b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } SIMDE__FUNCTION_ATTRIBUTES int simde_mm_ucomineq_ss(simde__m128 a, simde__m128 b) { #if defined(SIMDE_SSE_NATIVE) return _mm_ucomineq_ss(a.n, b.n); #else fenv_t envp; int x = feholdexcept(&envp); int r = a.f32[0] != b.f32[0]; if (HEDLEY_LIKELY(x == 0)) fesetenv(&envp); return r; #endif } #if defined(SIMDE_SSE_NATIVE) #if defined(__has_builtin) #if __has_builtin(__builtin_ia32_undef128) #define SIMDE__HAVE_UNDEFINED128 #endif #elif !defined(__PGI) && !defined(SIMDE_BUG_GCC_REV_208793) #define SIMDE__HAVE_UNDEFINED128 #endif #endif SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_undefined_ps(void) { simde__m128 r; #if defined(SIMDE__HAVE_UNDEFINED128) r.n = _mm_undefined_ps(); #else r = simde_mm_setzero_ps(); #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_unpackhi_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_unpackhi_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x2_t a1 = vget_high_f32(a.neon_f32); float32x2_t b1 = vget_high_f32(b.neon_f32); float32x2x2_t result = vzip_f32(a1, b1); r.neon_f32 = vcombine_f32(result.val[0], result.val[1]); #else r.f32[0] = a.f32[2]; r.f32[1] = b.f32[2]; r.f32[2] = a.f32[3]; r.f32[3] = b.f32[3]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_unpacklo_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_unpacklo_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) float32x2_t a1 = vget_low_f32(a.neon_f32); float32x2_t b1 = vget_low_f32(b.neon_f32); float32x2x2_t result = vzip_f32(a1, b1); r.neon_f32 = vcombine_f32(result.val[0], result.val[1]); #else r.f32[0] = a.f32[0]; r.f32[1] = b.f32[0]; r.f32[2] = a.f32[1]; r.f32[3] = b.f32[1]; #endif return r; } SIMDE__FUNCTION_ATTRIBUTES simde__m128 simde_mm_xor_ps(simde__m128 a, simde__m128 b) { simde__m128 r; #if defined(SIMDE_SSE_NATIVE) r.n = _mm_xor_ps(a.n, b.n); #elif defined(SIMDE_SSE_NEON) r.neon_i32 = veorq_s32(a.neon_i32, b.neon_i32); #else SIMDE__VECTORIZE for (size_t i = 0; i < (sizeof(r.u32) / sizeof(r.u32[0])); i++) { r.u32[i] = a.u32[i] ^ b.u32[i]; } #endif return r; } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_stream_pi(simde__m64 *mem_addr, simde__m64 a) { #if defined(SIMDE_SSE_NATIVE) _mm_stream_pi(&(mem_addr->n), a.n); #else mem_addr->i64[0] = a.i64[0]; #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_stream_ps(simde_float32 mem_addr[4], simde__m128 a) { simde_assert_aligned(16, mem_addr); #if defined(SIMDE_SSE_NATIVE) _mm_stream_ps(mem_addr, a.n); #else SIMDE__ASSUME_ALIGNED(mem_addr, 16); memcpy(mem_addr, &a, sizeof(a)); #endif } SIMDE__FUNCTION_ATTRIBUTES uint32_t simde_mm_getcsr(void) { #if defined(SIMDE_SSE_NATIVE) return _mm_getcsr(); #else uint32_t r = 0; int rounding_mode = fegetround(); switch (rounding_mode) { case FE_TONEAREST: break; case FE_UPWARD: r |= 2 << 13; break; case FE_DOWNWARD: r |= 1 << 13; break; case FE_TOWARDZERO: r = 3 << 13; break; } return r; #endif } SIMDE__FUNCTION_ATTRIBUTES void simde_mm_setcsr(uint32_t a) { #if defined(SIMDE_SSE_NATIVE) _mm_setcsr(a); #else switch ((a >> 13) & 3) { case 0: fesetround(FE_TONEAREST); break; case 1: fesetround(FE_DOWNWARD); break; case 2: fesetround(FE_UPWARD); break; case 3: fesetround(FE_TOWARDZERO); break; } #endif } #define SIMDE_MM_TRANSPOSE4_PS(row0, row1, row2, row3) \ do { \ simde__m128 tmp3, tmp2, tmp1, tmp0; \ tmp0 = simde_mm_unpacklo_ps((row0), (row1)); \ tmp2 = simde_mm_unpacklo_ps((row2), (row3)); \ tmp1 = simde_mm_unpackhi_ps((row0), (row1)); \ tmp3 = simde_mm_unpackhi_ps((row2), (row3)); \ row0 = simde_mm_movelh_ps(tmp0, tmp2); \ row1 = simde_mm_movehl_ps(tmp2, tmp0); \ row2 = simde_mm_movelh_ps(tmp1, tmp3); \ row3 = simde_mm_movehl_ps(tmp3, tmp1); \ } while (0) SIMDE__END_DECLS #endif /* !defined(SIMDE__SSE_H) */