1 /*
2  *
3  * This file is part of FFmpeg.
4  *
5  * FFmpeg is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * FFmpeg is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License along
16  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18  */
19 
20 #include <string.h>
21 
22 #include "libavutil/common.h"
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/mem_internal.h"
25 
26 #include "libswscale/swscale.h"
27 #include "libswscale/swscale_internal.h"
28 
29 #include "checkasm.h"
30 
31 #define randomize_buffers(buf, size)      \
32     do {                                  \
33         int j;                            \
34         for (j = 0; j < size; j+=4)       \
35             AV_WN32(buf + j, rnd());      \
36     } while (0)
37 
38 // This reference function is the same approximate algorithm employed by the
39 // SIMD functions
ref_function(const int16_t *filter, int filterSize, const int16_t **src, uint8_t *dest, int dstW, const uint8_t *dither, int offset)40 static void ref_function(const int16_t *filter, int filterSize,
41                                                  const int16_t **src, uint8_t *dest, int dstW,
42                                                  const uint8_t *dither, int offset)
43 {
44     int i, d;
45     d = ((filterSize - 1) * 8 + dither[0]) >> 4;
46     for ( i = 0; i < dstW; i++) {
47         int16_t val = d;
48         int j;
49         union {
50             int val;
51             int16_t v[2];
52         } t;
53         for (j = 0; j < filterSize; j++){
54             t.val = (int)src[j][i + offset] * (int)filter[j];
55             val += t.v[1];
56         }
57         dest[i]= av_clip_uint8(val>>3);
58     }
59 }
60 
check_yuv2yuvX(void)61 static void check_yuv2yuvX(void)
62 {
63     struct SwsContext *ctx;
64     int fsi, osi, isi, i, j;
65     int dstW;
66 #define LARGEST_FILTER 16
67 #define FILTER_SIZES 4
68     static const int filter_sizes[FILTER_SIZES] = {1, 4, 8, 16};
69 #define LARGEST_INPUT_SIZE 512
70 #define INPUT_SIZES 6
71     static const int input_sizes[INPUT_SIZES] = {8, 24, 128, 144, 256, 512};
72 
73     declare_func_emms(AV_CPU_FLAG_MMX, void, const int16_t *filter,
74                       int filterSize, const int16_t **src, uint8_t *dest,
75                       int dstW, const uint8_t *dither, int offset);
76 
77     const int16_t **src;
78     LOCAL_ALIGNED_16(int16_t, src_pixels, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
79     LOCAL_ALIGNED_16(int16_t, filter_coeff, [LARGEST_FILTER]);
80     LOCAL_ALIGNED_16(uint8_t, dst0, [LARGEST_INPUT_SIZE]);
81     LOCAL_ALIGNED_16(uint8_t, dst1, [LARGEST_INPUT_SIZE]);
82     LOCAL_ALIGNED_16(uint8_t, dither, [LARGEST_INPUT_SIZE]);
83     union VFilterData{
84         const int16_t *src;
85         uint16_t coeff[8];
86     } *vFilterData;
87     uint8_t d_val = rnd();
88     memset(dither, d_val, LARGEST_INPUT_SIZE);
89     randomize_buffers((uint8_t*)src_pixels, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int16_t));
90     randomize_buffers((uint8_t*)filter_coeff, LARGEST_FILTER * sizeof(int16_t));
91     ctx = sws_alloc_context();
92     if (sws_init_context(ctx, NULL, NULL) < 0)
93         fail();
94 
95     ff_sws_init_scale(ctx);
96     for(isi = 0; isi < INPUT_SIZES; ++isi){
97         dstW = input_sizes[isi];
98         for(osi = 0; osi < 64; osi += 16){
99             for(fsi = 0; fsi < FILTER_SIZES; ++fsi){
100                 src = av_malloc(sizeof(int16_t*) * filter_sizes[fsi]);
101                 vFilterData = av_malloc((filter_sizes[fsi] + 2) * sizeof(union VFilterData));
102                 memset(vFilterData, 0, (filter_sizes[fsi] + 2) * sizeof(union VFilterData));
103                 for(i = 0; i < filter_sizes[fsi]; ++i){
104                     src[i] = &src_pixels[i * LARGEST_INPUT_SIZE];
105                     vFilterData[i].src = src[i];
106                     for(j = 0; j < 4; ++j)
107                         vFilterData[i].coeff[j + 4] = filter_coeff[i];
108                 }
109                 if (check_func(ctx->yuv2planeX, "yuv2yuvX_%d_%d_%d", filter_sizes[fsi], osi, dstW)){
110                     memset(dst0, 0, LARGEST_INPUT_SIZE * sizeof(dst0[0]));
111                     memset(dst1, 0, LARGEST_INPUT_SIZE * sizeof(dst1[0]));
112 
113                     // The reference function is not the scalar function selected when mmx
114                     // is deactivated as the SIMD functions do not give the same result as
115                     // the scalar ones due to rounding. The SIMD functions are activated by
116                     // the flag SWS_ACCURATE_RND
117                     ref_function(&filter_coeff[0], filter_sizes[fsi], src, dst0, dstW - osi, dither, osi);
118                     // There's no point in calling new for the reference function
119                     if(ctx->use_mmx_vfilter){
120                         call_new((const int16_t*)vFilterData, filter_sizes[fsi], src, dst1, dstW - osi, dither, osi);
121                         if (memcmp(dst0, dst1, LARGEST_INPUT_SIZE * sizeof(dst0[0])))
122                             fail();
123                         if(dstW == LARGEST_INPUT_SIZE)
124                             bench_new((const int16_t*)vFilterData, filter_sizes[fsi], src, dst1, dstW - osi, dither, osi);
125                     }
126                 }
127                 av_freep(&src);
128                 av_freep(&vFilterData);
129             }
130         }
131     }
132     sws_freeContext(ctx);
133 #undef FILTER_SIZES
134 }
135 
136 #undef SRC_PIXELS
137 #define SRC_PIXELS 512
138 
check_hscale(void)139 static void check_hscale(void)
140 {
141 #define MAX_FILTER_WIDTH 40
142 #define FILTER_SIZES 6
143     static const int filter_sizes[FILTER_SIZES] = { 4, 8, 12, 16, 32, 40 };
144 
145 #define HSCALE_PAIRS 2
146     static const int hscale_pairs[HSCALE_PAIRS][2] = {
147         { 8, 14 },
148         { 8, 18 },
149     };
150 
151 #define LARGEST_INPUT_SIZE 512
152 #define INPUT_SIZES 6
153     static const int input_sizes[INPUT_SIZES] = {8, 24, 128, 144, 256, 512};
154 
155     int i, j, fsi, hpi, width, dstWi;
156     struct SwsContext *ctx;
157 
158     // padded
159     LOCAL_ALIGNED_32(uint8_t, src, [FFALIGN(SRC_PIXELS + MAX_FILTER_WIDTH - 1, 4)]);
160     LOCAL_ALIGNED_32(uint32_t, dst0, [SRC_PIXELS]);
161     LOCAL_ALIGNED_32(uint32_t, dst1, [SRC_PIXELS]);
162 
163     // padded
164     LOCAL_ALIGNED_32(int16_t, filter, [SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH]);
165     LOCAL_ALIGNED_32(int32_t, filterPos, [SRC_PIXELS]);
166     LOCAL_ALIGNED_32(int16_t, filterAvx2, [SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH]);
167     LOCAL_ALIGNED_32(int32_t, filterPosAvx, [SRC_PIXELS]);
168 
169     // The dst parameter here is either int16_t or int32_t but we use void* to
170     // just cover both cases.
171     declare_func_emms(AV_CPU_FLAG_MMX, void, void *c, void *dst, int dstW,
172                       const uint8_t *src, const int16_t *filter,
173                       const int32_t *filterPos, int filterSize);
174 
175     int cpu_flags = av_get_cpu_flags();
176 
177     ctx = sws_alloc_context();
178     if (sws_init_context(ctx, NULL, NULL) < 0)
179         fail();
180 
181     randomize_buffers(src, SRC_PIXELS + MAX_FILTER_WIDTH - 1);
182 
183     for (hpi = 0; hpi < HSCALE_PAIRS; hpi++) {
184         for (fsi = 0; fsi < FILTER_SIZES; fsi++) {
185             for (dstWi = 0; dstWi < INPUT_SIZES; dstWi++) {
186                 width = filter_sizes[fsi];
187 
188                 ctx->srcBpc = hscale_pairs[hpi][0];
189                 ctx->dstBpc = hscale_pairs[hpi][1];
190                 ctx->hLumFilterSize = ctx->hChrFilterSize = width;
191 
192                 for (i = 0; i < SRC_PIXELS; i++) {
193                     filterPos[i] = i;
194                     filterPosAvx[i] = i;
195 
196                     // These filter cofficients are chosen to try break two corner
197                     // cases, namely:
198                     //
199                     // - Negative filter coefficients. The filters output signed
200                     //   values, and it should be possible to end up with negative
201                     //   output values.
202                     //
203                     // - Positive clipping. The hscale filter function has clipping
204                     //   at (1<<15) - 1
205                     //
206                     // The coefficients sum to the 1.0 point for the hscale
207                     // functions (1 << 14).
208 
209                     for (j = 0; j < width; j++) {
210                         filter[i * width + j] = -((1 << 14) / (width - 1));
211                     }
212                     filter[i * width + (rnd() % width)] = ((1 << 15) - 1);
213                 }
214 
215                 for (i = 0; i < MAX_FILTER_WIDTH; i++) {
216                     // These values should be unused in SIMD implementations but
217                     // may still be read, random coefficients here should help show
218                     // issues where they are used in error.
219 
220                     filter[SRC_PIXELS * width + i] = rnd();
221                 }
222                 ctx->dstW = ctx->chrDstW = input_sizes[dstWi];
223                 ff_sws_init_scale(ctx);
224                 memcpy(filterAvx2, filter, sizeof(uint16_t) * (SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH));
225                 if ((cpu_flags & AV_CPU_FLAG_AVX2) && !(cpu_flags & AV_CPU_FLAG_SLOW_GATHER))
226                     ff_shuffle_filter_coefficients(ctx, filterPosAvx, width, filterAvx2, SRC_PIXELS);
227 
228                 if (check_func(ctx->hcScale, "hscale_%d_to_%d__fs_%d_dstW_%d", ctx->srcBpc, ctx->dstBpc + 1, width, ctx->dstW)) {
229                     memset(dst0, 0, SRC_PIXELS * sizeof(dst0[0]));
230                     memset(dst1, 0, SRC_PIXELS * sizeof(dst1[0]));
231 
232                     call_ref(NULL, dst0, ctx->dstW, src, filter, filterPos, width);
233                     call_new(NULL, dst1, ctx->dstW, src, filterAvx2, filterPosAvx, width);
234                     if (memcmp(dst0, dst1, ctx->dstW * sizeof(dst0[0])))
235                         fail();
236                     bench_new(NULL, dst0, ctx->dstW, src, filter, filterPosAvx, width);
237                 }
238             }
239         }
240     }
241     sws_freeContext(ctx);
242 }
243 
checkasm_check_sw_scale(void)244 void checkasm_check_sw_scale(void)
245 {
246     check_hscale();
247     report("hscale");
248     check_yuv2yuvX();
249     report("yuv2yuvX");
250 }
251