XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 1 | // Copyright (c) Facebook, Inc. and its affiliates. |
| 2 | // All rights reserved. |
| 3 | // |
| 4 | // Copyright 2019 Google LLC |
| 5 | // |
| 6 | // This source code is licensed under the BSD-style license found in the |
| 7 | // LICENSE file in the root directory of this source tree. |
| 8 | |
| 9 | #pragma once |
| 10 | |
| 11 | #include <stddef.h> |
| 12 | #include <stdint.h> |
| 13 | |
| 14 | #include <xnnpack/params.h> |
| 15 | #include <xnnpack/common.h> |
| 16 | |
| 17 | #ifdef __cplusplus |
| 18 | extern "C" { |
| 19 | #endif |
| 20 | |
| 21 | |
Marat Dukhan | 08b7a97 | 2020-07-14 18:17:29 -0700 | [diff] [blame] | 22 | #define DECLARE_QU8_VADD_MINMAX_UKERNEL_FUNCTION(fn_name) \ |
| 23 | XNN_INTERNAL void fn_name( \ |
| 24 | size_t n, \ |
| 25 | const uint8_t* a, \ |
| 26 | const uint8_t* b, \ |
| 27 | uint8_t* y, \ |
| 28 | const union xnn_qu8_add_params* params); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 29 | |
Marat Dukhan | 08b7a97 | 2020-07-14 18:17:29 -0700 | [diff] [blame] | 30 | DECLARE_QU8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qu8_vadd_minmax_ukernel__neon) |
| 31 | DECLARE_QU8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qu8_vadd_minmax_ukernel__scalar) |
| 32 | DECLARE_QU8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qu8_vadd_minmax_ukernel__sse2) |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 33 | |
| 34 | |
Marat Dukhan | d9f3ad4 | 2020-08-10 12:30:58 -0700 | [diff] [blame] | 35 | #define DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(fn_name) \ |
| 36 | XNN_INTERNAL void fn_name( \ |
| 37 | size_t n, \ |
| 38 | const int8_t* input_x, \ |
| 39 | const int8_t* input_y, \ |
| 40 | int8_t* output, \ |
| 41 | const union xnn_qs8_add_params* params); |
| 42 | |
Marat Dukhan | ba7b279 | 2020-09-02 14:26:45 -0700 | [diff] [blame] | 43 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__neon_ld64_x8) |
| 44 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__neon_ld64_x16) |
| 45 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__neon_ld64_x24) |
| 46 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__neon_ld64_x32) |
| 47 | |
Marat Dukhan | d9f3ad4 | 2020-08-10 12:30:58 -0700 | [diff] [blame] | 48 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse2_mul16_ld64_x8) |
| 49 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse2_mul16_ld64_x16) |
| 50 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse2_mul16_ld64_x24) |
| 51 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse2_mul16_ld64_x32) |
| 52 | |
| 53 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul16_ld64_x8) |
| 54 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul16_ld64_x16) |
| 55 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul16_ld64_x24) |
| 56 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul16_ld64_x32) |
| 57 | |
Marat Dukhan | bb9225e | 2020-09-06 22:40:56 -0700 | [diff] [blame] | 58 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul32_ld32_x8) |
| 59 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul32_ld32_x16) |
| 60 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul32_ld32_x24) |
| 61 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__sse41_mul32_ld32_x32) |
| 62 | |
| 63 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__xop_mul32_ld32_x8) |
| 64 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__xop_mul32_ld32_x16) |
| 65 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__xop_mul32_ld32_x24) |
| 66 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__xop_mul32_ld32_x32) |
| 67 | |
Marat Dukhan | e6dc0b6 | 2020-09-08 23:57:14 -0700 | [diff] [blame] | 68 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__avx2_mul32_ld64_x8) |
| 69 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__avx2_mul32_ld64_x16) |
| 70 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__avx2_mul32_ld64_x24) |
| 71 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__avx2_mul32_ld64_x32) |
| 72 | |
Marat Dukhan | 5df27f8 | 2020-09-02 23:59:21 -0700 | [diff] [blame] | 73 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__wasmsimd_x8) |
| 74 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__wasmsimd_x16) |
| 75 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__wasmsimd_x24) |
| 76 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vadd_minmax_ukernel__wasmsimd_x32) |
| 77 | |
Marat Dukhan | ba7b279 | 2020-09-02 14:26:45 -0700 | [diff] [blame] | 78 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__neon_ld64_x8) |
| 79 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__neon_ld64_x16) |
| 80 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__neon_ld64_x24) |
| 81 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__neon_ld64_x32) |
| 82 | |
Marat Dukhan | 0270d9f | 2020-08-11 00:56:46 -0700 | [diff] [blame] | 83 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse2_mul16_ld64_x8) |
| 84 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse2_mul16_ld64_x16) |
| 85 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse2_mul16_ld64_x24) |
| 86 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse2_mul16_ld64_x32) |
| 87 | |
| 88 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul16_ld64_x8) |
| 89 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul16_ld64_x16) |
| 90 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul16_ld64_x24) |
| 91 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul16_ld64_x32) |
| 92 | |
Marat Dukhan | bb9225e | 2020-09-06 22:40:56 -0700 | [diff] [blame] | 93 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul32_ld32_x8) |
| 94 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul32_ld32_x16) |
| 95 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul32_ld32_x24) |
| 96 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__sse41_mul32_ld32_x32) |
| 97 | |
| 98 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__xop_mul32_ld32_x8) |
| 99 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__xop_mul32_ld32_x16) |
| 100 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__xop_mul32_ld32_x24) |
| 101 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__xop_mul32_ld32_x32) |
| 102 | |
Marat Dukhan | e6dc0b6 | 2020-09-08 23:57:14 -0700 | [diff] [blame] | 103 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__avx2_mul32_ld64_x8) |
| 104 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__avx2_mul32_ld64_x16) |
| 105 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__avx2_mul32_ld64_x24) |
| 106 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__avx2_mul32_ld64_x32) |
| 107 | |
Marat Dukhan | 5df27f8 | 2020-09-02 23:59:21 -0700 | [diff] [blame] | 108 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__wasmsimd_x8) |
| 109 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__wasmsimd_x16) |
| 110 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__wasmsimd_x24) |
| 111 | DECLARE_QS8_VADD_MINMAX_UKERNEL_FUNCTION(xnn_qs8_vaddc_minmax_ukernel__wasmsimd_x32) |
| 112 | |
Marat Dukhan | d9f3ad4 | 2020-08-10 12:30:58 -0700 | [diff] [blame] | 113 | |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 114 | #ifdef __cplusplus |
Marat Dukhan | 80fc932 | 2019-09-29 21:06:36 -0700 | [diff] [blame] | 115 | } // extern "C" |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 116 | #endif |