Commit | Line | Data |
---|---|---|
72b9787e JB |
1 | /***************************************************************************** |
2 | * Copyright (C) 2013 x265 project | |
3 | * | |
4 | * Authors: Min Chen <chenm003@163.com> | |
5 | * | |
6 | * This program 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 | * This program 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 | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. | |
19 | * | |
20 | * This program is also available under a commercial proprietary license. | |
21 | * For more information, contact us at license @ x265.com. | |
22 | *****************************************************************************/ | |
23 | ||
24 | #include "common.h" | |
25 | #include "predict.h" | |
26 | #include "intrapredharness.h" | |
27 | ||
28 | using namespace x265; | |
29 | ||
30 | IntraPredHarness::IntraPredHarness() | |
31 | { | |
32 | for (int i = 0; i < INPUT_SIZE; i++) | |
33 | pixel_buff[i] = rand() % PIXEL_MAX; | |
34 | ||
35 | initROM(); | |
36 | } | |
37 | ||
38 | bool IntraPredHarness::check_dc_primitive(intra_pred_t ref, intra_pred_t opt, int width) | |
39 | { | |
40 | int j = Predict::ADI_BUF_STRIDE; | |
41 | intptr_t stride = FENC_STRIDE; | |
42 | ||
43 | #if _DEBUG | |
44 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); | |
45 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); | |
46 | #endif | |
47 | ||
48 | for (int i = 0; i <= 100; i++) | |
49 | { | |
50 | int rand_filter = rand() & 1; | |
51 | if (width > 16) | |
52 | rand_filter = 0; | |
53 | ||
54 | pixel left[MAX_CU_SIZE * 2 + 1]; | |
55 | for (int k = 0; k < width * 2 + 1; k++) | |
56 | { | |
57 | left[k] = pixel_buff[j - 1 + k * Predict::ADI_BUF_STRIDE]; | |
58 | } | |
59 | ||
60 | ref(pixel_out_c, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, rand_filter); | |
61 | checked(opt, pixel_out_vec, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, rand_filter); | |
62 | ||
63 | for (int k = 0; k < width; k++) | |
64 | { | |
65 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) | |
66 | return false; | |
67 | } | |
68 | ||
69 | reportfail(); | |
70 | j += FENC_STRIDE; | |
71 | } | |
72 | ||
73 | return true; | |
74 | } | |
75 | ||
76 | bool IntraPredHarness::check_planar_primitive(intra_pred_t ref, intra_pred_t opt, int width) | |
77 | { | |
78 | int j = Predict::ADI_BUF_STRIDE; | |
79 | intptr_t stride = FENC_STRIDE; | |
80 | ||
81 | #if _DEBUG | |
82 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); | |
83 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); | |
84 | #endif | |
85 | ||
86 | for (int i = 0; i <= 100; i++) | |
87 | { | |
88 | pixel left[MAX_CU_SIZE * 2 + 1]; | |
89 | for (int k = 0; k < width * 2 + 1; k++) | |
90 | { | |
91 | left[k] = pixel_buff[j - 1 + k * Predict::ADI_BUF_STRIDE]; | |
92 | } | |
93 | ||
94 | ref(pixel_out_c, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, 0); | |
95 | checked(opt, pixel_out_vec, stride, pixel_buff + j - Predict::ADI_BUF_STRIDE, left + 1, 0, 0); | |
96 | ||
97 | for (int k = 0; k < width; k++) | |
98 | { | |
99 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) | |
100 | return false; | |
101 | } | |
102 | ||
103 | reportfail(); | |
104 | j += FENC_STRIDE; | |
105 | } | |
106 | ||
107 | return true; | |
108 | } | |
109 | ||
110 | bool IntraPredHarness::check_angular_primitive(const intra_pred_t ref[][NUM_TR_SIZE], const intra_pred_t opt[][NUM_TR_SIZE]) | |
111 | { | |
112 | int j = Predict::ADI_BUF_STRIDE; | |
113 | intptr_t stride = FENC_STRIDE; | |
114 | ||
115 | #if _DEBUG | |
116 | memset(pixel_out_vec, 0xCD, OUTPUT_SIZE); | |
117 | memset(pixel_out_c, 0xCD, OUTPUT_SIZE); | |
118 | #endif | |
119 | ||
120 | for (int size = 2; size <= 5; size++) | |
121 | { | |
122 | int width = (1 << size); | |
123 | for (int i = 0; i <= 100; i++) | |
124 | { | |
125 | int bFilter = (width <= 16) && (rand() % 2); | |
126 | for (int pmode = 2; pmode <= 34; pmode++) | |
127 | { | |
128 | if (!opt[pmode][size - 2]) | |
129 | continue; | |
130 | ||
131 | pixel * refAbove = pixel_buff + j; | |
132 | pixel * refLeft = refAbove + 3 * width; | |
133 | refLeft[0] = refAbove[0]; | |
134 | ||
135 | checked(opt[pmode][size - 2], pixel_out_vec, stride, refLeft, refAbove, pmode, bFilter); | |
136 | ref[pmode][size - 2](pixel_out_c, stride, refLeft, refAbove, pmode, bFilter); | |
137 | ||
138 | for (int k = 0; k < width; k++) | |
139 | { | |
140 | if (memcmp(pixel_out_vec + k * FENC_STRIDE, pixel_out_c + k * FENC_STRIDE, width * sizeof(pixel))) | |
141 | return false; | |
142 | } | |
143 | ||
144 | reportfail(); | |
145 | } | |
146 | ||
147 | j += FENC_STRIDE; | |
148 | } | |
149 | } | |
150 | ||
151 | return true; | |
152 | } | |
153 | ||
154 | bool IntraPredHarness::check_allangs_primitive(const intra_allangs_t ref[], const intra_allangs_t opt[]) | |
155 | { | |
156 | int j = Predict::ADI_BUF_STRIDE; | |
157 | int isLuma; | |
158 | ||
159 | #if _DEBUG | |
160 | memset(pixel_out_33_vec, 0xCD, OUTPUT_SIZE_33); | |
161 | memset(pixel_out_33_c, 0xCD, OUTPUT_SIZE_33); | |
162 | #endif | |
163 | ||
164 | for (int size = 2; size <= 5; size++) | |
165 | { | |
166 | if (opt[size - 2] == NULL) continue; | |
167 | ||
168 | const int width = (1 << size); | |
169 | ||
170 | for (int i = 0; i <= 100; i++) | |
171 | { | |
172 | isLuma = (width <= 16) ? true : false; // bFilter is true for 4x4, 8x8, 16x16 and false for 32x32 | |
173 | ||
174 | pixel * refAbove0 = pixel_buff + j; | |
175 | pixel * refLeft0 = refAbove0 + 3 * width; | |
176 | ||
177 | pixel * refAbove1 = pixel_buff + j + 3 * FENC_STRIDE; // keep this offset, since vector code may broken input buffer range [-(width-1), 0] | |
178 | pixel * refLeft1 = refAbove1 + 3 * width + FENC_STRIDE; | |
179 | refLeft0[0] = refAbove0[0] = refLeft1[0] = refAbove1[0]; | |
180 | ||
181 | ref[size - 2](pixel_out_33_c, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); | |
182 | checked(opt[size - 2], pixel_out_33_vec, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); | |
183 | ||
184 | for (int p = 2 - 2; p <= 34 - 2; p++) | |
185 | { | |
186 | for (int k = 0; k < width; k++) | |
187 | { | |
188 | if (memcmp(pixel_out_33_c + p * (width * width) + k * width, pixel_out_33_vec + p * (width * width) + k * width, width * sizeof(pixel))) | |
189 | { | |
190 | printf("\nFailed: (%dx%d) Mode(%2d), Line[%2d], bfilter=%d\n", width, width, p + 2, k, isLuma); | |
191 | opt[size - 2](pixel_out_33_vec, refAbove0, refLeft0, refAbove1, refLeft1, isLuma); | |
192 | return false; | |
193 | } | |
194 | } | |
195 | } | |
196 | ||
197 | reportfail(); | |
198 | j += FENC_STRIDE; | |
199 | } | |
200 | } | |
201 | ||
202 | return true; | |
203 | } | |
204 | ||
205 | bool IntraPredHarness::testCorrectness(const EncoderPrimitives& ref, const EncoderPrimitives& opt) | |
206 | { | |
207 | for (int i = BLOCK_4x4; i <= BLOCK_32x32; i++) | |
208 | { | |
209 | if (opt.intra_pred[1][i]) | |
210 | { | |
211 | const int size = (1 << (i + 2)); | |
212 | if (!check_dc_primitive(ref.intra_pred[1][i], opt.intra_pred[1][i], size)) | |
213 | { | |
214 | printf("intra_dc %dx%d failed\n", size, size); | |
215 | return false; | |
216 | } | |
217 | } | |
218 | if (opt.intra_pred[0][i]) | |
219 | { | |
220 | const int size = (1 << (i + 2)); | |
221 | if (!check_planar_primitive(ref.intra_pred[0][i], opt.intra_pred[0][i], size)) | |
222 | { | |
223 | printf("intra_planar %dx%d failed\n", size, size); | |
224 | return false; | |
225 | } | |
226 | } | |
227 | } | |
228 | ||
229 | // NOTE: always call since this function have check pointer in loop | |
230 | if (!check_angular_primitive(ref.intra_pred, opt.intra_pred)) | |
231 | { | |
232 | printf("intra_angular failed\n"); | |
233 | return false; | |
234 | } | |
235 | ||
236 | if (opt.intra_pred_allangs[0]) | |
237 | { | |
238 | if (!check_allangs_primitive(ref.intra_pred_allangs, opt.intra_pred_allangs)) | |
239 | { | |
240 | printf("intra_allangs failed\n"); | |
241 | return false; | |
242 | } | |
243 | } | |
244 | ||
245 | return true; | |
246 | } | |
247 | ||
248 | void IntraPredHarness::measureSpeed(const EncoderPrimitives& ref, const EncoderPrimitives& opt) | |
249 | { | |
250 | int width = 64; | |
251 | uint16_t srcStride = 96; | |
252 | ||
253 | for (int i = BLOCK_4x4; i <= BLOCK_32x32; i++) | |
254 | { | |
255 | const int size = (1 << (i + 2)); | |
256 | if (opt.intra_pred[1][i]) | |
257 | { | |
258 | printf("intra_dc_%dx%d[f=0]", size, size); | |
259 | REPORT_SPEEDUP(opt.intra_pred[1][i], ref.intra_pred[1][i], | |
260 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 0); | |
261 | if (size <= 16) | |
262 | { | |
263 | printf("intra_dc_%dx%d[f=1]", size, size); | |
264 | REPORT_SPEEDUP(opt.intra_pred[1][i], ref.intra_pred[1][i], | |
265 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 1); | |
266 | } | |
267 | } | |
268 | if (opt.intra_pred[0][i]) | |
269 | { | |
270 | printf("intra_planar %2dx%d", size, size); | |
271 | REPORT_SPEEDUP(opt.intra_pred[0][i], ref.intra_pred[0][i], | |
272 | pixel_out_vec, FENC_STRIDE, pixel_buff + srcStride, pixel_buff, 0, 0); | |
273 | } | |
274 | if (opt.intra_pred_allangs[i]) | |
275 | { | |
276 | bool bFilter = (size <= 16); | |
277 | pixel * refAbove = pixel_buff + srcStride; | |
278 | pixel * refLeft = refAbove + 3 * size; | |
279 | refLeft[0] = refAbove[0]; | |
280 | printf("intra_allangs%dx%d", size, size); | |
281 | REPORT_SPEEDUP(opt.intra_pred_allangs[i], ref.intra_pred_allangs[i], | |
282 | pixel_out_33_vec, refAbove, refLeft, refAbove, refLeft, bFilter); | |
283 | } | |
284 | } | |
285 | ||
286 | for (int ii = 2; ii <= 5; ii++) | |
287 | { | |
288 | for (int p = 2; p <= 34; p += 1) | |
289 | { | |
290 | int pmode = p; //(rand()%33)+2; | |
291 | if (opt.intra_pred[pmode][ii - 2]) | |
292 | { | |
293 | width = (1 << ii); | |
294 | bool bFilter = (width <= 16); | |
295 | pixel * refAbove = pixel_buff + srcStride; | |
296 | pixel * refLeft = refAbove + 3 * width; | |
297 | refLeft[0] = refAbove[0]; | |
298 | printf("intra_ang%dx%d[%2d]", width, width, pmode); | |
299 | REPORT_SPEEDUP(opt.intra_pred[pmode][ii - 2], ref.intra_pred[pmode][ii - 2], | |
300 | pixel_out_vec, FENC_STRIDE, refAbove, refLeft, pmode, bFilter); | |
301 | } | |
302 | } | |
303 | } | |
304 | } |