Eric | 3afc1d1 | 2020-07-23 02:16:46 +0200 | [diff] [blame] | 1 | /*! \file conv_acc_neon.c |
| 2 | * Accelerated Viterbi decoder implementation |
| 3 | * for architectures with only NEON available. */ |
| 4 | /* |
| 5 | * (C) 2020 by sysmocom - s.f.m.c. GmbH |
| 6 | * Author: Eric Wild |
| 7 | * |
| 8 | * All Rights Reserved |
| 9 | * |
| 10 | * SPDX-License-Identifier: GPL-2.0+ |
| 11 | * |
| 12 | * This program is free software; you can redistribute it and/or modify |
| 13 | * it under the terms of the GNU General Public License as published by |
| 14 | * the Free Software Foundation; either version 2 of the License, or |
| 15 | * (at your option) any later version. |
| 16 | * |
| 17 | * This program is distributed in the hope that it will be useful, |
| 18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 20 | * GNU General Public License for more details. |
Eric | 3afc1d1 | 2020-07-23 02:16:46 +0200 | [diff] [blame] | 21 | */ |
| 22 | |
| 23 | #include <stdlib.h> |
| 24 | #include <stdint.h> |
| 25 | #include <malloc.h> |
| 26 | #include "config.h" |
| 27 | |
| 28 | #if defined(HAVE_NEON) |
| 29 | #include <arm_neon.h> |
| 30 | #endif |
| 31 | |
| 32 | /* align req is 16 on android because google was confused, 8 on sane platforms */ |
| 33 | #define NEON_ALIGN 8 |
| 34 | |
| 35 | #include <conv_acc_neon_impl.h> |
| 36 | |
| 37 | /* Aligned Memory Allocator |
| 38 | * NEON requires 8-byte memory alignment. We store relevant trellis values |
| 39 | * (accumulated sums, outputs, and path decisions) as 16 bit signed integers |
| 40 | * so the allocated memory is casted as such. |
| 41 | */ |
| 42 | __attribute__ ((visibility("hidden"))) |
| 43 | int16_t *osmo_conv_neon_vdec_malloc(size_t n) |
| 44 | { |
| 45 | return (int16_t *) memalign(NEON_ALIGN, sizeof(int16_t) * n); |
| 46 | } |
| 47 | |
| 48 | __attribute__ ((visibility("hidden"))) |
| 49 | void osmo_conv_neon_vdec_free(int16_t *ptr) |
| 50 | { |
| 51 | free(ptr); |
| 52 | } |
| 53 | |
| 54 | __attribute__ ((visibility("hidden"))) |
| 55 | void osmo_conv_neon_metrics_k5_n2(const int8_t *val, const int16_t *out, |
| 56 | int16_t *sums, int16_t *paths, int norm) |
| 57 | { |
| 58 | const int16_t _val[4] = { val[0], val[1], val[0], val[1] }; |
| 59 | |
| 60 | _neon_metrics_k5_n2(_val, out, sums, paths, norm); |
| 61 | } |
| 62 | |
| 63 | __attribute__ ((visibility("hidden"))) |
| 64 | void osmo_conv_neon_metrics_k5_n3(const int8_t *val, const int16_t *out, |
| 65 | int16_t *sums, int16_t *paths, int norm) |
| 66 | { |
| 67 | const int16_t _val[4] = { val[0], val[1], val[2], 0 }; |
| 68 | |
| 69 | _neon_metrics_k5_n4(_val, out, sums, paths, norm); |
| 70 | } |
| 71 | |
| 72 | __attribute__ ((visibility("hidden"))) |
| 73 | void osmo_conv_neon_metrics_k5_n4(const int8_t *val, const int16_t *out, |
| 74 | int16_t *sums, int16_t *paths, int norm) |
| 75 | { |
| 76 | const int16_t _val[4] = { val[0], val[1], val[2], val[3] }; |
| 77 | |
| 78 | _neon_metrics_k5_n4(_val, out, sums, paths, norm); |
| 79 | } |
| 80 | |
| 81 | __attribute__ ((visibility("hidden"))) |
| 82 | void osmo_conv_neon_metrics_k7_n2(const int8_t *val, const int16_t *out, |
| 83 | int16_t *sums, int16_t *paths, int norm) |
| 84 | { |
| 85 | const int16_t _val[4] = { val[0], val[1], val[0], val[1] }; |
| 86 | |
| 87 | _neon_metrics_k7_n2(_val, out, sums, paths, norm); |
| 88 | } |
| 89 | |
| 90 | __attribute__ ((visibility("hidden"))) |
| 91 | void osmo_conv_neon_metrics_k7_n3(const int8_t *val, const int16_t *out, |
| 92 | int16_t *sums, int16_t *paths, int norm) |
| 93 | { |
| 94 | const int16_t _val[4] = { val[0], val[1], val[2], 0 }; |
| 95 | |
| 96 | _neon_metrics_k7_n4(_val, out, sums, paths, norm); |
| 97 | } |
| 98 | |
| 99 | __attribute__ ((visibility("hidden"))) |
| 100 | void osmo_conv_neon_metrics_k7_n4(const int8_t *val, const int16_t *out, |
| 101 | int16_t *sums, int16_t *paths, int norm) |
| 102 | { |
| 103 | const int16_t _val[4] = { val[0], val[1], val[2], val[3] }; |
| 104 | |
| 105 | _neon_metrics_k7_n4(_val, out, sums, paths, norm); |
| 106 | } |