Frank Barchard | 9c7308f | 2020-08-31 17:03:01 -0700 | [diff] [blame] | 1 | // Auto-generated file. Do not edit! |
| 2 | // Template: src/f32-vbinary/vopc-scalar.c.in |
| 3 | // Generator: tools/xngen |
| 4 | // |
| 5 | // Copyright 2019 Google LLC |
| 6 | // |
| 7 | // This source code is licensed under the BSD-style license found in the |
| 8 | // LICENSE file in the root directory of this source tree. |
| 9 | |
| 10 | #include <assert.h> |
| 11 | |
| 12 | #include <xnnpack/common.h> |
| 13 | #include <xnnpack/math.h> |
| 14 | #include <xnnpack/vbinary.h> |
| 15 | |
| 16 | |
| 17 | void xnn_f32_vsubc_minmax_ukernel__scalar_x8( |
| 18 | size_t n, |
| 19 | const float* a, |
| 20 | const float* b, |
| 21 | float* y, |
| 22 | const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) |
| 23 | { |
| 24 | assert(n != 0); |
| 25 | assert(n % sizeof(float) == 0); |
| 26 | assert(a != NULL); |
| 27 | assert(b != NULL); |
| 28 | assert(y != NULL); |
| 29 | |
| 30 | const float vy_min = params->scalar.min; |
| 31 | const float vy_max = params->scalar.max; |
| 32 | |
| 33 | const float vb = *b; |
| 34 | for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) { |
| 35 | const float va0 = a[0]; |
| 36 | const float va1 = a[1]; |
| 37 | const float va2 = a[2]; |
| 38 | const float va3 = a[3]; |
| 39 | const float va4 = a[4]; |
| 40 | const float va5 = a[5]; |
| 41 | const float va6 = a[6]; |
| 42 | const float va7 = a[7]; |
| 43 | a += 8; |
| 44 | |
| 45 | float vy0 = va0 - vb; |
| 46 | float vy1 = va1 - vb; |
| 47 | float vy2 = va2 - vb; |
| 48 | float vy3 = va3 - vb; |
| 49 | float vy4 = va4 - vb; |
| 50 | float vy5 = va5 - vb; |
| 51 | float vy6 = va6 - vb; |
| 52 | float vy7 = va7 - vb; |
| 53 | |
| 54 | |
| 55 | vy0 = math_max_f32(vy0, vy_min); |
| 56 | vy1 = math_max_f32(vy1, vy_min); |
| 57 | vy2 = math_max_f32(vy2, vy_min); |
| 58 | vy3 = math_max_f32(vy3, vy_min); |
| 59 | vy4 = math_max_f32(vy4, vy_min); |
| 60 | vy5 = math_max_f32(vy5, vy_min); |
| 61 | vy6 = math_max_f32(vy6, vy_min); |
| 62 | vy7 = math_max_f32(vy7, vy_min); |
| 63 | |
| 64 | vy0 = math_min_f32(vy0, vy_max); |
| 65 | vy1 = math_min_f32(vy1, vy_max); |
| 66 | vy2 = math_min_f32(vy2, vy_max); |
| 67 | vy3 = math_min_f32(vy3, vy_max); |
| 68 | vy4 = math_min_f32(vy4, vy_max); |
| 69 | vy5 = math_min_f32(vy5, vy_max); |
| 70 | vy6 = math_min_f32(vy6, vy_max); |
| 71 | vy7 = math_min_f32(vy7, vy_max); |
| 72 | |
| 73 | y[0] = vy0; |
| 74 | y[1] = vy1; |
| 75 | y[2] = vy2; |
| 76 | y[3] = vy3; |
| 77 | y[4] = vy4; |
| 78 | y[5] = vy5; |
| 79 | y[6] = vy6; |
| 80 | y[7] = vy7; |
| 81 | y += 8; |
| 82 | } |
| 83 | if XNN_UNLIKELY(n != 0) { |
| 84 | do { |
| 85 | const float va = *a++; |
| 86 | float vy = va - vb; |
| 87 | vy = math_max_f32(vy, vy_min); |
| 88 | vy = math_min_f32(vy, vy_max); |
| 89 | *y++ = vy; |
| 90 | n -= sizeof(float); |
| 91 | } while (n != 0); |
| 92 | } |
| 93 | } |