Marat Dukhan | 2dbb944 | 2020-05-12 20:43:43 -0700 | [diff] [blame^] | 1 | // Copyright 2020 Google LLC |
| 2 | // |
| 3 | // This source code is licensed under the BSD-style license found in the |
| 4 | // LICENSE file in the root directory of this source tree. |
| 5 | |
| 6 | #include <assert.h> |
| 7 | #include <stddef.h> |
| 8 | |
| 9 | #include <smmintrin.h> |
| 10 | |
| 11 | #include <xnnpack/math-stubs.h> |
| 12 | |
| 13 | |
| 14 | void xnn_math_f32_roundz__sse41( |
| 15 | size_t n, |
| 16 | const float* input, |
| 17 | float* output) |
| 18 | { |
| 19 | assert(n % (4 * sizeof(float)) == 0); |
| 20 | |
| 21 | for (; n != 0; n -= 4 * sizeof(float)) { |
| 22 | const __m128 vx = _mm_load_ps(input); |
| 23 | input += 4; |
| 24 | |
| 25 | const __m128 vy = _mm_round_ps(vx, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC); |
| 26 | |
| 27 | _mm_store_ps(output, vy); |
| 28 | output += 4; |
| 29 | } |
| 30 | } |