Checkasm: assembly testing and benchmarking tool
[ffmpeg.git] / tests / checkasm / h264pred.c
1 /*
2  * Copyright (c) 2015 Henrik Gramner
3  *
4  * This file is part of Libav.
5  *
6  * Libav is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * Libav is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License along
17  * with Libav; if not, write to the Free Software Foundation, Inc.,
18  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19  */
20
21 #include <string.h>
22 #include "checkasm.h"
23 #include "libavcodec/avcodec.h"
24 #include "libavcodec/h264pred.h"
25 #include "libavutil/common.h"
26 #include "libavutil/intreadwrite.h"
27
28 static const int codec_ids[4] = { AV_CODEC_ID_H264, AV_CODEC_ID_VP8, AV_CODEC_ID_RV40, AV_CODEC_ID_SVQ3 };
29
30 static const char * const pred4x4_modes[4][15] = {
31     { /* H264 */
32         [VERT_PRED           ] = "vertical",
33         [HOR_PRED            ] = "horizontal",
34         [DC_PRED             ] = "dc",
35         [DIAG_DOWN_LEFT_PRED ] = "down_left",
36         [DIAG_DOWN_RIGHT_PRED] = "down_right",
37         [VERT_RIGHT_PRED     ] = "vertical_right",
38         [HOR_DOWN_PRED       ] = "horizontal_right",
39         [VERT_LEFT_PRED      ] = "vertical_left",
40         [HOR_UP_PRED         ] = "horizontal_up",
41         [LEFT_DC_PRED        ] = "left_dc",
42         [TOP_DC_PRED         ] = "top_dc",
43         [DC_128_PRED         ] = "dc_128",
44     },
45     { /* VP8 */
46         [VERT_PRED     ] = "vertical_vp8",
47         [HOR_PRED      ] = "horizontal_vp8",
48         [VERT_LEFT_PRED] = "vertical_left_vp8",
49         [TM_VP8_PRED   ] = "tm_vp8",
50         [DC_127_PRED   ] = "dc_127_vp8",
51         [DC_129_PRED   ] = "dc_129_vp8",
52     },
53     { /* RV40 */
54         [DIAG_DOWN_LEFT_PRED            ] = "down_left_rv40",
55         [VERT_LEFT_PRED                 ] = "vertical_left_rv40",
56         [HOR_UP_PRED                    ] = "horizontal_up_rv40",
57         [DIAG_DOWN_LEFT_PRED_RV40_NODOWN] = "down_left_nodown_rv40",
58         [HOR_UP_PRED_RV40_NODOWN        ] = "horizontal_up_nodown_rv40",
59         [VERT_LEFT_PRED_RV40_NODOWN     ] = "vertical_left_nodown_rv40",
60     },
61     { /* SVQ3 */
62         [DIAG_DOWN_LEFT_PRED] = "down_left_svq3",
63     },
64 };
65
66 static const char * const pred8x8_modes[4][11] = {
67     { /* H264 */
68         [DC_PRED8x8              ] = "dc",
69         [HOR_PRED8x8             ] = "horizontal",
70         [VERT_PRED8x8            ] = "vertical",
71         [PLANE_PRED8x8           ] = "plane",
72         [LEFT_DC_PRED8x8         ] = "left_dc",
73         [TOP_DC_PRED8x8          ] = "top_dc",
74         [DC_128_PRED8x8          ] = "dc_128",
75         [ALZHEIMER_DC_L0T_PRED8x8] = "mad_cow_dc_l0t",
76         [ALZHEIMER_DC_0LT_PRED8x8] = "mad_cow_dc_0lt",
77         [ALZHEIMER_DC_L00_PRED8x8] = "mad_cow_dc_l00",
78         [ALZHEIMER_DC_0L0_PRED8x8] = "mad_cow_dc_0l0",
79     },
80     { /* VP8 */
81         [PLANE_PRED8x8 ] = "tm_vp8",
82         [DC_127_PRED8x8] = "dc_127_vp8",
83         [DC_129_PRED8x8] = "dc_129_vp8",
84     },
85     { /* RV40 */
86         [DC_PRED8x8     ] = "dc_rv40",
87         [LEFT_DC_PRED8x8] = "left_dc_rv40",
88         [TOP_DC_PRED8x8 ] = "top_dc_rv40",
89     },
90     { /* SVQ3 */
91     },
92 };
93
94 static const char * const pred16x16_modes[4][9] = {
95     { /* H264 */
96         [DC_PRED8x8     ] = "dc",
97         [HOR_PRED8x8    ] = "horizontal",
98         [VERT_PRED8x8   ] = "vertical",
99         [PLANE_PRED8x8  ] = "plane",
100         [LEFT_DC_PRED8x8] = "left_dc",
101         [TOP_DC_PRED8x8 ] = "top_dc",
102         [DC_128_PRED8x8 ] = "dc_128",
103     },
104     { /* VP8 */
105         [PLANE_PRED8x8 ] = "tm_vp8",
106         [DC_127_PRED8x8] = "dc_127_vp8",
107         [DC_129_PRED8x8] = "dc_129_vp8",
108     },
109     { /* RV40 */
110         [PLANE_PRED8x8] = "plane_rv40",
111     },
112     { /* SVQ3 */
113         [PLANE_PRED8x8] = "plane_svq3",
114     },
115 };
116
117 static const uint32_t pixel_mask[3] = { 0xffffffff, 0x01ff01ff, 0x03ff03ff };
118
119 #define SIZEOF_PIXEL ((bit_depth + 7) / 8)
120 #define BUF_SIZE (3*16*17)
121
122 #define check_pred_func(func, name, mode_name)\
123     (mode_name && ((codec_ids[codec] == AV_CODEC_ID_H264) ?\
124     check_func(func, "pred%s_%s_%d", name, mode_name, bit_depth) :\
125     check_func(func, "pred%s_%s", name, mode_name)))
126
127 #define randomize_buffers()\
128     do {\
129         uint32_t mask = pixel_mask[bit_depth-8];\
130         int i;\
131         for (i = 0; i < BUF_SIZE; i += 4) {\
132             uint32_t r = rnd() & mask;\
133             AV_WN32A(buf0+i, r);\
134             AV_WN32A(buf1+i, r);\
135         }\
136     } while (0)
137
138 #define src0 (buf0 + 4*16) /* Offset to allow room for top and left */
139 #define src1 (buf1 + 4*16)
140
141 static void check_pred4x4(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
142                           int codec, int chroma_format, int bit_depth)
143 {
144     if (chroma_format == 1) {
145         uint8_t *topright = buf0 + 2*16;
146         int pred_mode;
147         for (pred_mode = 0; pred_mode < 15; pred_mode++) {
148             if (check_pred_func(h->pred4x4[pred_mode], "4x4", pred4x4_modes[codec][pred_mode])) {
149                 randomize_buffers();
150                 call_ref(src0, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
151                 call_new(src1, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
152                 if (memcmp(buf0, buf1, BUF_SIZE))
153                     fail();
154                 bench_new(src1, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
155             }
156         }
157     }
158 }
159
160 static void check_pred8x8(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
161                           int codec, int chroma_format, int bit_depth)
162 {
163     int pred_mode;
164     for (pred_mode = 0; pred_mode < 11; pred_mode++) {
165         if (check_pred_func(h->pred8x8[pred_mode], (chroma_format == 2) ? "8x16" : "8x8",
166                             pred8x8_modes[codec][pred_mode])) {
167             randomize_buffers();
168             call_ref(src0, (ptrdiff_t)24*SIZEOF_PIXEL);
169             call_new(src1, (ptrdiff_t)24*SIZEOF_PIXEL);
170             if (memcmp(buf0, buf1, BUF_SIZE))
171                 fail();
172             bench_new(src1, (ptrdiff_t)24*SIZEOF_PIXEL);
173         }
174     }
175 }
176
177 static void check_pred16x16(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
178                             int codec, int chroma_format, int bit_depth)
179 {
180     if (chroma_format == 1) {
181         int pred_mode;
182         for (pred_mode = 0; pred_mode < 9; pred_mode++) {
183             if (check_pred_func(h->pred16x16[pred_mode], "16x16", pred16x16_modes[codec][pred_mode])) {
184                 randomize_buffers();
185                 call_ref(src0, (ptrdiff_t)48);
186                 call_new(src1, (ptrdiff_t)48);
187                 if (memcmp(buf0, buf1, BUF_SIZE))
188                     fail();
189                 bench_new(src1, (ptrdiff_t)48);
190             }
191         }
192     }
193 }
194
195 static void check_pred8x8l(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
196                            int codec, int chroma_format, int bit_depth)
197 {
198     if (chroma_format == 1 && codec_ids[codec] == AV_CODEC_ID_H264) {
199         int pred_mode;
200         for (pred_mode = 0; pred_mode < 12; pred_mode++) {
201             if (check_pred_func(h->pred8x8l[pred_mode], "8x8l", pred4x4_modes[codec][pred_mode])) {
202                 int neighbors;
203                 for (neighbors = 0; neighbors <= 0xc000; neighbors += 0x4000) {
204                     int has_topleft  = neighbors & 0x8000;
205                     int has_topright = neighbors & 0x4000;
206
207                     if ((pred_mode == DIAG_DOWN_RIGHT_PRED || pred_mode == VERT_RIGHT_PRED) && !has_topleft)
208                         continue; /* Those aren't allowed according to the spec */
209
210                     randomize_buffers();
211                     call_ref(src0, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
212                     call_new(src1, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
213                     if (memcmp(buf0, buf1, BUF_SIZE))
214                         fail();
215                     bench_new(src1, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
216                 }
217             }
218         }
219     }
220 }
221
222 /* TODO: Add tests for H.264 lossless H/V prediction */
223
224 void checkasm_check_h264pred(void)
225 {
226     static const struct {
227         void (*func)(H264PredContext*, uint8_t*, uint8_t*, int, int, int);
228         const char *name;
229     } tests[] = {
230         { check_pred4x4,   "pred4x4"   },
231         { check_pred8x8,   "pred8x8"   },
232         { check_pred16x16, "pred16x16" },
233         { check_pred8x8l,  "pred8x8l"  },
234     };
235
236     DECLARE_ALIGNED(16, uint8_t, buf0)[BUF_SIZE];
237     DECLARE_ALIGNED(16, uint8_t, buf1)[BUF_SIZE];
238     H264PredContext h;
239     int test, codec, chroma_format, bit_depth;
240
241     for (test = 0; test < FF_ARRAY_ELEMS(tests); test++) {
242         for (codec = 0; codec < 4; codec++) {
243             int codec_id = codec_ids[codec];
244             for (bit_depth = 8; bit_depth <= (codec_id == AV_CODEC_ID_H264 ? 10 : 8); bit_depth++)
245                 for (chroma_format = 1; chroma_format <= (codec_id == AV_CODEC_ID_H264 ? 2 : 1); chroma_format++) {
246                     ff_h264_pred_init(&h, codec_id, bit_depth, chroma_format);
247                     tests[test].func(&h, buf0, buf1, codec, chroma_format, bit_depth);
248                 }
249         }
250         report("%s", tests[test].name);
251     }
252 }