XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 1 | // Copyright 2019 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 | #pragma once |
| 7 | |
| 8 | #include <stddef.h> |
| 9 | #include <stdint.h> |
| 10 | |
| 11 | #include <xnnpack/params.h> |
| 12 | #include <xnnpack/common.h> |
| 13 | |
| 14 | #ifdef __cplusplus |
| 15 | extern "C" { |
| 16 | #endif |
| 17 | |
| 18 | |
Marat Dukhan | 355ab43 | 2020-04-09 19:01:52 -0700 | [diff] [blame] | 19 | #define DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(fn_name) \ |
| 20 | XNN_INTERNAL void fn_name( \ |
| 21 | uint32_t m, \ |
| 22 | uint32_t n, \ |
| 23 | const float* a, \ |
| 24 | const float* w, \ |
| 25 | const int32_t* dmap, \ |
| 26 | const uint32_t* nmap, \ |
| 27 | float* c, \ |
Marat Dukhan | eb09a6b | 2020-04-08 17:34:32 -0700 | [diff] [blame] | 28 | const union xnn_f32_minmax_params* params); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 29 | |
Marat Dukhan | 355ab43 | 2020-04-09 19:01:52 -0700 | [diff] [blame] | 30 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_1x1__scalar) |
| 31 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_1x1__scalar_pipelined) |
| 32 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_2x1__scalar) |
| 33 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_2x1__scalar_pipelined) |
| 34 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__neonfma) |
Marat Dukhan | 355ab43 | 2020-04-09 19:01:52 -0700 | [diff] [blame] | 35 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__neonfma_pipelined) |
| 36 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__neonfma_unroll2) |
| 37 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__scalar) |
| 38 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__scalar_pipelined) |
| 39 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__sse) |
Frank Barchard | 9e05340 | 2020-10-19 15:29:08 -0700 | [diff] [blame] | 40 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_arm) |
Frank Barchard | c451e8a | 2020-10-21 17:13:12 -0700 | [diff] [blame] | 41 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_arm_unroll2) |
| 42 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_arm_unroll4) |
Frank Barchard | 9e05340 | 2020-10-19 15:29:08 -0700 | [diff] [blame] | 43 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_x86) |
Frank Barchard | c451e8a | 2020-10-21 17:13:12 -0700 | [diff] [blame] | 44 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_x86_unroll2) |
| 45 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x1__wasmsimd_x86_unroll4) |
Frank Barchard | 846c0c6 | 2020-10-26 15:01:39 -0700 | [diff] [blame] | 46 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x2__neonfma) |
| 47 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_4x4__neonfma) |
| 48 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__neonfma) |
| 49 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__neonfma_pipelined) |
| 50 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__neonfma_unroll2) |
| 51 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__scalar) |
| 52 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__scalar_pipelined) |
| 53 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__sse) |
| 54 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_arm) |
| 55 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_arm_unroll2) |
| 56 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_arm_unroll4) |
Frank Barchard | 9e05340 | 2020-10-19 15:29:08 -0700 | [diff] [blame] | 57 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_x86) |
Frank Barchard | c451e8a | 2020-10-21 17:13:12 -0700 | [diff] [blame] | 58 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_x86_unroll2) |
| 59 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x1__wasmsimd_x86_unroll4) |
Frank Barchard | 846c0c6 | 2020-10-26 15:01:39 -0700 | [diff] [blame] | 60 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x2__neonfma) |
| 61 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x2__scalar) |
| 62 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x4__neonfma) |
| 63 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_8x4__scalar) |
| 64 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_12x1__neonfma) |
| 65 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_12x2__neonfma) |
| 66 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_12x4__neonfma) |
| 67 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__neonfma) |
| 68 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__neonfma_pipelined) |
| 69 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__neonfma_unroll2) |
| 70 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__sse) |
| 71 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_arm) |
| 72 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_arm_unroll2) |
| 73 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_arm_unroll4) |
Frank Barchard | 9e05340 | 2020-10-19 15:29:08 -0700 | [diff] [blame] | 74 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_x86) |
Frank Barchard | c451e8a | 2020-10-21 17:13:12 -0700 | [diff] [blame] | 75 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_x86_unroll2) |
| 76 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x1__wasmsimd_x86_unroll4) |
Frank Barchard | 846c0c6 | 2020-10-26 15:01:39 -0700 | [diff] [blame] | 77 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x2__neonfma) |
| 78 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_16x4__neonfma) |
| 79 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__neonfma) |
| 80 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__neonfma_pipelined) |
| 81 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__neonfma_unroll2) |
| 82 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__sse) |
| 83 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_arm) |
| 84 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_arm_unroll2) |
| 85 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_arm_unroll4) |
| 86 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_x86) |
| 87 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_x86_unroll2) |
| 88 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x1__wasmsimd_x86_unroll4) |
| 89 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x2__neonfma) |
| 90 | DECLARE_F32_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f32_spmm_minmax_ukernel_32x4__neonfma) |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 91 | |
Marat Dukhan | 355ab43 | 2020-04-09 19:01:52 -0700 | [diff] [blame] | 92 | #define DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(fn_name) \ |
| 93 | XNN_INTERNAL void fn_name( \ |
| 94 | uint32_t m, \ |
| 95 | uint32_t n, \ |
| 96 | const void* a, \ |
| 97 | const void* w, \ |
| 98 | const int32_t* dmap, \ |
| 99 | const uint32_t* nmap, \ |
| 100 | void* c, \ |
Frank Barchard | 875be77 | 2020-04-16 09:50:09 -0700 | [diff] [blame] | 101 | const struct xnn_f16_scaleminmax_params* params); |
Marat Dukhan | bdb56f5 | 2020-02-05 21:42:49 -0800 | [diff] [blame] | 102 | |
Marat Dukhan | 355ab43 | 2020-04-09 19:01:52 -0700 | [diff] [blame] | 103 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_8x1__neonfp16arith) |
| 104 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_8x1__neonfp16arith_unroll2) |
| 105 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_16x1__neonfp16arith) |
| 106 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_16x1__neonfp16arith_unroll2) |
| 107 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_24x1__neonfp16arith) |
| 108 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_24x1__neonfp16arith_unroll2) |
| 109 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_32x1__neonfp16arith) |
| 110 | DECLARE_F16_SPMM_MINMAX_UKERNEL_FUNCTION(xnn_f16_spmm_minmax_ukernel_32x1__neonfp16arith_unroll2) |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 111 | |
| 112 | #ifdef __cplusplus |
Marat Dukhan | 80fc932 | 2019-09-29 21:06:36 -0700 | [diff] [blame] | 113 | } // extern "C" |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 114 | #endif |