1 /*
2 * Copyright (c) 2016 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <math.h>
12 #include <assert.h>
13 #include "./vpx_dsp_rtcd.h"
14 #include "vpx_dsp/psnr.h"
15 #include "vpx_scale/yv12config.h"
16
vpx_sse_to_psnr(double samples,double peak,double sse)17 double vpx_sse_to_psnr(double samples, double peak, double sse) {
18 if (sse > 0.0) {
19 const double psnr = 10.0 * log10(samples * peak * peak / sse);
20 return psnr > MAX_PSNR ? MAX_PSNR : psnr;
21 } else {
22 return MAX_PSNR;
23 }
24 }
25
26 /* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
27 * and highbd_8_variance(). It should not.
28 */
encoder_sse(const uint8_t * a,int a_stride,const uint8_t * b,int b_stride,int w,int h)29 static int64_t encoder_sse(const uint8_t *a, int a_stride, const uint8_t *b,
30 int b_stride, int w, int h) {
31 int i, j;
32 int64_t sse = 0;
33
34 for (i = 0; i < h; i++) {
35 for (j = 0; j < w; j++) {
36 const int diff = a[j] - b[j];
37 sse += diff * diff;
38 }
39
40 a += a_stride;
41 b += b_stride;
42 }
43
44 return sse;
45 }
46
47 #if CONFIG_VP9_HIGHBITDEPTH
encoder_highbd_sse(const uint8_t * a8,int a_stride,const uint8_t * b8,int b_stride,int w,int h)48 static int64_t encoder_highbd_sse(const uint8_t *a8, int a_stride,
49 const uint8_t *b8, int b_stride, int w,
50 int h) {
51 int i, j;
52 int64_t sse = 0;
53
54 const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
55 const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
56
57 for (i = 0; i < h; i++) {
58 for (j = 0; j < w; j++) {
59 const int diff = a[j] - b[j];
60 sse += diff * diff;
61 }
62 a += a_stride;
63 b += b_stride;
64 }
65
66 return sse;
67 }
68 #endif // CONFIG_VP9_HIGHBITDEPTH
69
get_sse(const uint8_t * a,int a_stride,const uint8_t * b,int b_stride,int width,int height)70 static int64_t get_sse(const uint8_t *a, int a_stride, const uint8_t *b,
71 int b_stride, int width, int height) {
72 const int dw = width % 16;
73 const int dh = height % 16;
74 int64_t total_sse = 0;
75 int x, y;
76
77 if (dw > 0) {
78 total_sse += encoder_sse(&a[width - dw], a_stride, &b[width - dw], b_stride,
79 dw, height);
80 }
81
82 if (dh > 0) {
83 total_sse +=
84 encoder_sse(&a[(height - dh) * a_stride], a_stride,
85 &b[(height - dh) * b_stride], b_stride, width - dw, dh);
86 }
87
88 for (y = 0; y < height / 16; ++y) {
89 const uint8_t *pa = a;
90 const uint8_t *pb = b;
91 for (x = 0; x < width / 16; ++x) {
92 total_sse += vpx_sse(pa, a_stride, pb, b_stride, 16, 16);
93
94 pa += 16;
95 pb += 16;
96 }
97
98 a += 16 * a_stride;
99 b += 16 * b_stride;
100 }
101
102 return total_sse;
103 }
104
105 #if CONFIG_VP9_HIGHBITDEPTH
highbd_get_sse_shift(const uint8_t * a8,int a_stride,const uint8_t * b8,int b_stride,int width,int height,unsigned int input_shift)106 static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
107 const uint8_t *b8, int b_stride, int width,
108 int height, unsigned int input_shift) {
109 const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
110 const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
111 int64_t total_sse = 0;
112 int x, y;
113 for (y = 0; y < height; ++y) {
114 for (x = 0; x < width; ++x) {
115 int64_t diff;
116 diff = (a[x] >> input_shift) - (b[x] >> input_shift);
117 total_sse += diff * diff;
118 }
119 a += a_stride;
120 b += b_stride;
121 }
122 return total_sse;
123 }
124
highbd_get_sse(const uint8_t * a,int a_stride,const uint8_t * b,int b_stride,int width,int height)125 static int64_t highbd_get_sse(const uint8_t *a, int a_stride, const uint8_t *b,
126 int b_stride, int width, int height) {
127 int64_t total_sse = 0;
128 int x, y;
129 const int dw = width % 16;
130 const int dh = height % 16;
131 if (dw > 0) {
132 total_sse += encoder_highbd_sse(&a[width - dw], a_stride, &b[width - dw],
133 b_stride, dw, height);
134 }
135 if (dh > 0) {
136 total_sse += encoder_highbd_sse(&a[(height - dh) * a_stride], a_stride,
137 &b[(height - dh) * b_stride], b_stride,
138 width - dw, dh);
139 }
140 for (y = 0; y < height / 16; ++y) {
141 const uint8_t *pa = a;
142 const uint8_t *pb = b;
143 for (x = 0; x < width / 16; ++x) {
144 total_sse += vpx_highbd_sse(pa, a_stride, pb, b_stride, 16, 16);
145 pa += 16;
146 pb += 16;
147 }
148 a += 16 * a_stride;
149 b += 16 * b_stride;
150 }
151 return total_sse;
152 }
153 #endif // CONFIG_VP9_HIGHBITDEPTH
154
vpx_get_y_sse(const YV12_BUFFER_CONFIG * a,const YV12_BUFFER_CONFIG * b)155 int64_t vpx_get_y_sse(const YV12_BUFFER_CONFIG *a,
156 const YV12_BUFFER_CONFIG *b) {
157 assert(a->y_crop_width == b->y_crop_width);
158 assert(a->y_crop_height == b->y_crop_height);
159
160 return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
161 a->y_crop_width, a->y_crop_height);
162 }
163
164 #if CONFIG_VP9_HIGHBITDEPTH
vpx_highbd_get_y_sse(const YV12_BUFFER_CONFIG * a,const YV12_BUFFER_CONFIG * b)165 int64_t vpx_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
166 const YV12_BUFFER_CONFIG *b) {
167 assert(a->y_crop_width == b->y_crop_width);
168 assert(a->y_crop_height == b->y_crop_height);
169 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
170 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
171
172 return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
173 a->y_crop_width, a->y_crop_height);
174 }
175 #endif // CONFIG_VP9_HIGHBITDEPTH
176
177 #if CONFIG_VP9_HIGHBITDEPTH
vpx_calc_highbd_psnr(const YV12_BUFFER_CONFIG * a,const YV12_BUFFER_CONFIG * b,PSNR_STATS * psnr,uint32_t bit_depth,uint32_t in_bit_depth)178 void vpx_calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
179 const YV12_BUFFER_CONFIG *b, PSNR_STATS *psnr,
180 uint32_t bit_depth, uint32_t in_bit_depth) {
181 const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
182 const int heights[3] = { a->y_crop_height, a->uv_crop_height,
183 a->uv_crop_height };
184 const uint8_t *a_planes[3] = { a->y_buffer, a->u_buffer, a->v_buffer };
185 const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
186 const uint8_t *b_planes[3] = { b->y_buffer, b->u_buffer, b->v_buffer };
187 const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
188 int i;
189 uint64_t total_sse = 0;
190 uint32_t total_samples = 0;
191 const double peak = (double)((1 << in_bit_depth) - 1);
192 const unsigned int input_shift = bit_depth - in_bit_depth;
193
194 for (i = 0; i < 3; ++i) {
195 const int w = widths[i];
196 const int h = heights[i];
197 const uint32_t samples = w * h;
198 uint64_t sse;
199 if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
200 if (input_shift) {
201 sse = highbd_get_sse_shift(a_planes[i], a_strides[i], b_planes[i],
202 b_strides[i], w, h, input_shift);
203 } else {
204 sse = highbd_get_sse(a_planes[i], a_strides[i], b_planes[i],
205 b_strides[i], w, h);
206 }
207 } else {
208 sse = get_sse(a_planes[i], a_strides[i], b_planes[i], b_strides[i], w, h);
209 }
210 psnr->sse[1 + i] = sse;
211 psnr->samples[1 + i] = samples;
212 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
213
214 total_sse += sse;
215 total_samples += samples;
216 }
217
218 psnr->sse[0] = total_sse;
219 psnr->samples[0] = total_samples;
220 psnr->psnr[0] =
221 vpx_sse_to_psnr((double)total_samples, peak, (double)total_sse);
222 }
223
224 #endif // !CONFIG_VP9_HIGHBITDEPTH
225
vpx_calc_psnr(const YV12_BUFFER_CONFIG * a,const YV12_BUFFER_CONFIG * b,PSNR_STATS * psnr)226 void vpx_calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
227 PSNR_STATS *psnr) {
228 static const double peak = 255.0;
229 const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
230 const int heights[3] = { a->y_crop_height, a->uv_crop_height,
231 a->uv_crop_height };
232 const uint8_t *a_planes[3] = { a->y_buffer, a->u_buffer, a->v_buffer };
233 const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
234 const uint8_t *b_planes[3] = { b->y_buffer, b->u_buffer, b->v_buffer };
235 const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
236 int i;
237 uint64_t total_sse = 0;
238 uint32_t total_samples = 0;
239
240 for (i = 0; i < 3; ++i) {
241 const int w = widths[i];
242 const int h = heights[i];
243 const uint32_t samples = w * h;
244 const uint64_t sse =
245 get_sse(a_planes[i], a_strides[i], b_planes[i], b_strides[i], w, h);
246 psnr->sse[1 + i] = sse;
247 psnr->samples[1 + i] = samples;
248 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
249
250 total_sse += sse;
251 total_samples += samples;
252 }
253
254 psnr->sse[0] = total_sse;
255 psnr->samples[0] = total_samples;
256 psnr->psnr[0] =
257 vpx_sse_to_psnr((double)total_samples, peak, (double)total_sse);
258 }
259