| // Auto-generated file. Do not edit! |
| // Template: src/f32-vunary/sse.c.in |
| // Generator: tools/xngen |
| // |
| // Copyright 2020 Google LLC |
| // |
| // This source code is licensed under the BSD-style license found in the |
| // LICENSE file in the root directory of this source tree. |
| |
| #include <assert.h> |
| |
| #include <xmmintrin.h> |
| |
| #include <xnnpack/common.h> |
| #include <xnnpack/math.h> |
| #include <xnnpack/vunary.h> |
| |
| |
| void xnn_f32_vneg_ukernel__sse_x8( |
| size_t n, |
| const float* x, |
| float* y, |
| const union xnn_f32_neg_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN |
| { |
| assert(n != 0); |
| assert(n % sizeof(float) == 0); |
| assert(x != NULL); |
| assert(y != NULL); |
| |
| const __m128 vsign_mask = _mm_load_ps(params->sse.sign_mask); |
| for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) { |
| const __m128 vx0123 = _mm_loadu_ps(x); |
| const __m128 vx4567 = _mm_loadu_ps(x + 4); |
| x += 8; |
| |
| const __m128 vy0123 = _mm_xor_ps(vx0123, vsign_mask); |
| const __m128 vy4567 = _mm_xor_ps(vx4567, vsign_mask); |
| |
| _mm_storeu_ps(y, vy0123); |
| _mm_storeu_ps(y + 4, vy4567); |
| y += 8; |
| } |
| for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) { |
| const __m128 vx = _mm_loadu_ps(x); |
| x += 4; |
| const __m128 vy = _mm_xor_ps(vx, vsign_mask); |
| _mm_storeu_ps(y, vy); |
| y += 4; |
| } |
| if XNN_UNLIKELY(n != 0) { |
| const __m128 vx = _mm_loadu_ps(x); |
| __m128 vy = _mm_xor_ps(vx, vsign_mask); |
| if (n & (2 * sizeof(float))) { |
| _mm_storel_pi((__m64*) y, vy); |
| vy = _mm_movehl_ps(vy, vy); |
| y += 2; |
| } |
| if (n & (1 * sizeof(float))) { |
| _mm_store_ss(y, vy); |
| } |
| } |
| } |