/src/aom/aom_dsp/psnr.c

Source (jump to first uncovered line)
/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <assert.h>
#include <math.h>

#include "config/aom_dsp_rtcd.h"

#include "aom_dsp/psnr.h"
#include "aom_scale/yv12config.h"

double aom_sse_to_psnr(double samples, double peak, double sse) {
  if (sse > 0.0) {
    const double psnr = 10.0 * log10(samples * peak * peak / sse);
    return psnr > MAX_PSNR ? MAX_PSNR : psnr;
  } else {
    return MAX_PSNR;
  }
}

static void encoder_variance(const uint8_t *a, int a_stride, const uint8_t *b,
                             int b_stride, int w, int h, unsigned int *sse,
                             int *sum) {
  int i, j;

  *sum = 0;
  *sse = 0;

  for (i = 0; i < h; i++) {
    for (j = 0; j < w; j++) {
      const int diff = a[j] - b[j];
      *sum += diff;
      *sse += diff * diff;
    }

    a += a_stride;
    b += b_stride;
  }
}

#if CONFIG_AV1_HIGHBITDEPTH
static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
                                      const uint8_t *b8, int b_stride, int w,
                                      int h, uint64_t *sse, int64_t *sum) {
  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
  int64_t tsum = 0;
  uint64_t tsse = 0;
  for (int i = 0; i < h; ++i) {
    int32_t lsum = 0;
    for (int j = 0; j < w; ++j) {
      const int diff = a[j] - b[j];
      lsum += diff;
      tsse += (uint32_t)(diff * diff);
    }
    tsum += lsum;
    a += a_stride;
    b += b_stride;
  }
  *sum = tsum;
  *sse = tsse;
}

static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
                                      const uint8_t *b8, int b_stride, int w,
                                      int h, unsigned int *sse, int *sum) {
  uint64_t sse_long = 0;
  int64_t sum_long = 0;
  encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long,
                            &sum_long);
  *sse = (unsigned int)sse_long;
  *sum = (int)sum_long;
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

static int64_t get_sse(const uint8_t *a, int a_stride, const uint8_t *b,
                       int b_stride, int width, int height) {
  const int dw = width % 16;
  const int dh = height % 16;
  int64_t total_sse = 0;
  unsigned int sse = 0;
  int sum = 0;
  int x, y;

  if (dw > 0) {
    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride, dw,
                     height, &sse, &sum);
    total_sse += sse;
  }

  if (dh > 0) {
    encoder_variance(&a[(height - dh) * a_stride], a_stride,
                     &b[(height - dh) * b_stride], b_stride, width - dw, dh,
                     &sse, &sum);
    total_sse += sse;
  }

  for (y = 0; y < height / 16; ++y) {
    const uint8_t *pa = a;
    const uint8_t *pb = b;
    for (x = 0; x < width / 16; ++x) {
      aom_mse16x16(pa, a_stride, pb, b_stride, &sse);
      total_sse += sse;

      pa += 16;
      pb += 16;
    }

    a += 16 * a_stride;
    b += 16 * b_stride;
  }

  return total_sse;
}

#if CONFIG_AV1_HIGHBITDEPTH
static int64_t 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) {
  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
  int64_t total_sse = 0;
  int x, y;
  for (y = 0; y < height; ++y) {
    for (x = 0; x < width; ++x) {
      int64_t diff;
      diff = (a[x] >> input_shift) - (b[x] >> input_shift);
      total_sse += diff * diff;
    }
    a += a_stride;
    b += b_stride;
  }
  return total_sse;
}

static int64_t highbd_get_sse(const uint8_t *a, int a_stride, const uint8_t *b,
                              int b_stride, int width, int height) {
  int64_t total_sse = 0;
  int x, y;
  const int dw = width % 16;
  const int dh = height % 16;
  unsigned int sse = 0;
  int sum = 0;
  if (dw > 0) {
    encoder_highbd_8_variance(&a[width - dw], a_stride, &b[width - dw],
                              b_stride, dw, height, &sse, &sum);
    total_sse += sse;
  }
  if (dh > 0) {
    encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
                              &b[(height - dh) * b_stride], b_stride,
                              width - dw, dh, &sse, &sum);
    total_sse += sse;
  }
  for (y = 0; y < height / 16; ++y) {
    const uint8_t *pa = a;
    const uint8_t *pb = b;
    for (x = 0; x < width / 16; ++x) {
      aom_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
      total_sse += sse;
      pa += 16;
      pb += 16;
    }
    a += 16 * a_stride;
    b += 16 * b_stride;
  }
  return total_sse;
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

uint64_t aom_get_y_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
                       int vstart, int height) {
  return aom_var_2d_u8(a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
                       width, height) /
         (width * height);
}

uint64_t aom_get_u_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
                       int vstart, int height) {
  return aom_var_2d_u8(a->u_buffer + vstart * a->uv_stride + hstart,
                       a->uv_stride, width, height) /
         (width * height);
}

uint64_t aom_get_v_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
                       int vstart, int height) {
  return aom_var_2d_u8(a->v_buffer + vstart * a->uv_stride + hstart,
                       a->uv_stride, width, height) /
         (width * height);
}

int64_t aom_get_y_sse_part(const YV12_BUFFER_CONFIG *a,
                           const YV12_BUFFER_CONFIG *b, int hstart, int width,
                           int vstart, int height) {
  return get_sse(a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
                 b->y_buffer + vstart * b->y_stride + hstart, b->y_stride,
                 width, height);
}

int64_t aom_get_y_sse(const YV12_BUFFER_CONFIG *a,
                      const YV12_BUFFER_CONFIG *b) {
  assert(a->y_crop_width == b->y_crop_width);
  assert(a->y_crop_height == b->y_crop_height);

  return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
                 a->y_crop_width, a->y_crop_height);
}

int64_t aom_get_u_sse_part(const YV12_BUFFER_CONFIG *a,
                           const YV12_BUFFER_CONFIG *b, int hstart, int width,
                           int vstart, int height) {
  return get_sse(a->u_buffer + vstart * a->uv_stride + hstart, a->uv_stride,
                 b->u_buffer + vstart * b->uv_stride + hstart, b->uv_stride,
                 width, height);
}

int64_t aom_get_u_sse(const YV12_BUFFER_CONFIG *a,
                      const YV12_BUFFER_CONFIG *b) {
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);

  return get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride,
                 a->uv_crop_width, a->uv_crop_height);
}

int64_t aom_get_v_sse_part(const YV12_BUFFER_CONFIG *a,
                           const YV12_BUFFER_CONFIG *b, int hstart, int width,
                           int vstart, int height) {
  return get_sse(a->v_buffer + vstart * a->uv_stride + hstart, a->uv_stride,
                 b->v_buffer + vstart * b->uv_stride + hstart, b->uv_stride,
                 width, height);
}

int64_t aom_get_v_sse(const YV12_BUFFER_CONFIG *a,
                      const YV12_BUFFER_CONFIG *b) {
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);

  return get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride,
                 a->uv_crop_width, a->uv_crop_height);
}

#if CONFIG_AV1_HIGHBITDEPTH
uint64_t aom_highbd_get_y_var(const YV12_BUFFER_CONFIG *a, int hstart,
                              int width, int vstart, int height) {
  return aom_var_2d_u16(a->y_buffer + vstart * a->y_stride + hstart,
                        a->y_stride, width, height) /
         (width * height);
}

uint64_t aom_highbd_get_u_var(const YV12_BUFFER_CONFIG *a, int hstart,
                              int width, int vstart, int height) {
  return aom_var_2d_u16(a->u_buffer + vstart * a->uv_stride + hstart,
                        a->uv_stride, width, height) /
         (width * height);
}

uint64_t aom_highbd_get_v_var(const YV12_BUFFER_CONFIG *a, int hstart,
                              int width, int vstart, int height) {
  return aom_var_2d_u16(a->v_buffer + vstart * a->uv_stride + hstart,
                        a->uv_stride, width, height) /
         (width * height);
}

int64_t aom_highbd_get_y_sse_part(const YV12_BUFFER_CONFIG *a,
                                  const YV12_BUFFER_CONFIG *b, int hstart,
                                  int width, int vstart, int height) {
  return highbd_get_sse(
      a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
      b->y_buffer + vstart * b->y_stride + hstart, b->y_stride, width, height);
}

int64_t aom_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
                             const YV12_BUFFER_CONFIG *b) {
  assert(a->y_crop_width == b->y_crop_width);
  assert(a->y_crop_height == b->y_crop_height);
  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);

  return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
                        a->y_crop_width, a->y_crop_height);
}

int64_t aom_highbd_get_u_sse_part(const YV12_BUFFER_CONFIG *a,
                                  const YV12_BUFFER_CONFIG *b, int hstart,
                                  int width, int vstart, int height) {
  return highbd_get_sse(a->u_buffer + vstart * a->uv_stride + hstart,
                        a->uv_stride,
                        b->u_buffer + vstart * b->uv_stride + hstart,
                        b->uv_stride, width, height);
}

int64_t aom_highbd_get_u_sse(const YV12_BUFFER_CONFIG *a,
                             const YV12_BUFFER_CONFIG *b) {
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);
  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);

  return highbd_get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride,
                        a->uv_crop_width, a->uv_crop_height);
}

int64_t aom_highbd_get_v_sse_part(const YV12_BUFFER_CONFIG *a,
                                  const YV12_BUFFER_CONFIG *b, int hstart,
                                  int width, int vstart, int height) {
  return highbd_get_sse(a->v_buffer + vstart * a->uv_stride + hstart,
                        a->uv_stride,
                        b->v_buffer + vstart * b->uv_stride + hstart,
                        b->uv_stride, width, height);
}

int64_t aom_highbd_get_v_sse(const YV12_BUFFER_CONFIG *a,
                             const YV12_BUFFER_CONFIG *b) {
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);
  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);

  return highbd_get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride,
                        a->uv_crop_width, a->uv_crop_height);
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

int64_t aom_get_sse_plane(const YV12_BUFFER_CONFIG *a,
                          const YV12_BUFFER_CONFIG *b, int plane, int highbd) {
#if CONFIG_AV1_HIGHBITDEPTH
  if (highbd) {
    switch (plane) {
      case 0: return aom_highbd_get_y_sse(a, b);
      case 1: return aom_highbd_get_u_sse(a, b);
      case 2: return aom_highbd_get_v_sse(a, b);
      default: assert(plane >= 0 && plane <= 2); return 0;
    }
  } else {
    switch (plane) {
      case 0: return aom_get_y_sse(a, b);
      case 1: return aom_get_u_sse(a, b);
      case 2: return aom_get_v_sse(a, b);
      default: assert(plane >= 0 && plane <= 2); return 0;
    }
  }
#else
  (void)highbd;
  switch (plane) {
    case 0: return aom_get_y_sse(a, b);
    case 1: return aom_get_u_sse(a, b);
    case 2: return aom_get_v_sse(a, b);
    default: assert(plane >= 0 && plane <= 2); return 0;
  }
#endif
}

#if CONFIG_AV1_HIGHBITDEPTH
void aom_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) {
  assert(a->y_crop_width == b->y_crop_width);
  assert(a->y_crop_height == b->y_crop_height);
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);
  const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
  const int heights[3] = { a->y_crop_height, a->uv_crop_height,
                           a->uv_crop_height };
  const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
  const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
  int i;
  uint64_t total_sse = 0;
  uint32_t total_samples = 0;
  double peak = (double)((1 << in_bit_depth) - 1);
  const unsigned int input_shift = bit_depth - in_bit_depth;

  for (i = 0; i < 3; ++i) {
    const int w = widths[i];
    const int h = heights[i];
    const uint32_t samples = w * h;
    uint64_t sse;
    if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
      if (input_shift) {
        sse = highbd_get_sse_shift(a->buffers[i], a_strides[i], b->buffers[i],
                                   b_strides[i], w, h, input_shift);
      } else {
        sse = highbd_get_sse(a->buffers[i], a_strides[i], b->buffers[i],
                             b_strides[i], w, h);
      }
    } else {
      sse = get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w,
                    h);
    }
    psnr->sse[1 + i] = sse;
    psnr->samples[1 + i] = samples;
    psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);

    total_sse += sse;
    total_samples += samples;
  }

  psnr->sse[0] = total_sse;
  psnr->samples[0] = total_samples;
  psnr->psnr[0] =
      aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);

  // Compute PSNR based on stream bit depth
  if ((a->flags & YV12_FLAG_HIGHBITDEPTH) && (in_bit_depth < bit_depth)) {
    peak = (double)((1 << bit_depth) - 1);
    total_sse = 0;
    total_samples = 0;
    for (i = 0; i < 3; ++i) {
      const int w = widths[i];
      const int h = heights[i];
      const uint32_t samples = w * h;
      uint64_t sse;
      sse = highbd_get_sse(a->buffers[i], a_strides[i], b->buffers[i],
                           b_strides[i], w, h);
      psnr->sse_hbd[1 + i] = sse;
      psnr->samples_hbd[1 + i] = samples;
      psnr->psnr_hbd[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);
      total_sse += sse;
      total_samples += samples;
    }

    psnr->sse_hbd[0] = total_sse;
    psnr->samples_hbd[0] = total_samples;
    psnr->psnr_hbd[0] =
        aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);
  }
}
#endif

void aom_calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
                   PSNR_STATS *psnr) {
  assert(a->y_crop_width == b->y_crop_width);
  assert(a->y_crop_height == b->y_crop_height);
  assert(a->uv_crop_width == b->uv_crop_width);
  assert(a->uv_crop_height == b->uv_crop_height);
  static const double peak = 255.0;
  const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
  const int heights[3] = { a->y_crop_height, a->uv_crop_height,
                           a->uv_crop_height };
  const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
  const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
  int i;
  uint64_t total_sse = 0;
  uint32_t total_samples = 0;

  for (i = 0; i < 3; ++i) {
    const int w = widths[i];
    const int h = heights[i];
    const uint32_t samples = w * h;
    const uint64_t sse =
        get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w, h);
    psnr->sse[1 + i] = sse;
    psnr->samples[1 + i] = samples;
    psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);

    total_sse += sse;
    total_samples += samples;
  }

  psnr->sse[0] = total_sse;
  psnr->samples[0] = total_samples;
  psnr->psnr[0] =
      aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);
}

Coverage Report

Created: 2022-08-24 06:17

Line	Count	Source (jump to first uncovered line)
1		/*
2		* Copyright (c) 2016, Alliance for Open Media. All rights reserved
3		*
4		* This source code is subject to the terms of the BSD 2 Clause License and
5		* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6		* was not distributed with this source code in the LICENSE file, you can
7		* obtain it at www.aomedia.org/license/software. If the Alliance for Open
8		* Media Patent License 1.0 was not distributed with this source code in the
9		* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10		*/
11
12		#include <assert.h>
13		#include <math.h>
14
15		#include "config/aom_dsp_rtcd.h"
16
17		#include "aom_dsp/psnr.h"
18		#include "aom_scale/yv12config.h"
19
20	0	double aom_sse_to_psnr(double samples, double peak, double sse) {
21	0	if (sse > 0.0) {
22	0	const double psnr = 10.0 * log10(samples * peak * peak / sse);
23	0	return psnr > MAX_PSNR ? MAX_PSNR : psnr;
24	0	} else {
25	0	return MAX_PSNR;
26	0	}
27	0	}
28
29		static void encoder_variance(const uint8_t a, int a_stride, const uint8_t b,
30		int b_stride, int w, int h, unsigned int *sse,
31	23.5k	int *sum) {
32	23.5k	int i, j;
33
34	23.5k	*sum = 0;
35	23.5k	*sse = 0;
36
37	1.18M	for (i = 0; i < h; i++) {
38	16.4M	for (j = 0; j < w; j++) {
39	15.3M	const int diff = a[j] - b[j];
40	15.3M	*sum += diff;
41	15.3M	sse += diff diff;
42	15.3M	}
43
44	1.16M	a += a_stride;
45	1.16M	b += b_stride;
46	1.16M	}
47	23.5k	}
48
49		#if CONFIG_AV1_HIGHBITDEPTH
50		static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
51		const uint8_t *b8, int b_stride, int w,
52	0	int h, uint64_t sse, int64_t sum) {
53	0	const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
54	0	const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
55	0	int64_t tsum = 0;
56	0	uint64_t tsse = 0;
57	0	for (int i = 0; i < h; ++i) {
58	0	int32_t lsum = 0;
59	0	for (int j = 0; j < w; ++j) {
60	0	const int diff = a[j] - b[j];
61	0	lsum += diff;
62	0	tsse += (uint32_t)(diff * diff);
63	0	}
64	0	tsum += lsum;
65	0	a += a_stride;
66	0	b += b_stride;
67	0	}
68	0	*sum = tsum;
69	0	*sse = tsse;
70	0	}
71
72		static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
73		const uint8_t *b8, int b_stride, int w,
74	0	int h, unsigned int sse, int sum) {
75	0	uint64_t sse_long = 0;
76	0	int64_t sum_long = 0;
77	0	encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long,
78	0	&sum_long);
79	0	*sse = (unsigned int)sse_long;
80	0	*sum = (int)sum_long;
81	0	}
82		#endif // CONFIG_AV1_HIGHBITDEPTH
83
84		static int64_t get_sse(const uint8_t a, int a_stride, const uint8_t b,
85	13.1k	int b_stride, int width, int height) {
86	13.1k	const int dw = width % 16;
87	13.1k	const int dh = height % 16;
88	13.1k	int64_t total_sse = 0;
89	13.1k	unsigned int sse = 0;
90	13.1k	int sum = 0;
91	13.1k	int x, y;
92
93	13.1k	if (dw > 0) {
94	11.9k	encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride, dw,
95	11.9k	height, &sse, &sum);
96	11.9k	total_sse += sse;
97	11.9k	}
98
99	13.1k	if (dh > 0) {
100	11.6k	encoder_variance(&a[(height - dh) * a_stride], a_stride,
101	11.6k	&b[(height - dh) * b_stride], b_stride, width - dw, dh,
102	11.6k	&sse, &sum);
103	11.6k	total_sse += sse;
104	11.6k	}
105
106	81.8k	for (y = 0; y < height / 16; ++y) {
107	68.6k	const uint8_t *pa = a;
108	68.6k	const uint8_t *pb = b;
109	567k	for (x = 0; x < width / 16; ++x) {
110	498k	aom_mse16x16(pa, a_stride, pb, b_stride, &sse);
111	498k	total_sse += sse;
112
113	498k	pa += 16;
114	498k	pb += 16;
115	498k	}
116
117	68.6k	a += 16 * a_stride;
118	68.6k	b += 16 * b_stride;
119	68.6k	}
120
121	13.1k	return total_sse;
122	13.1k	}
123
124		#if CONFIG_AV1_HIGHBITDEPTH
125		static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
126		const uint8_t *b8, int b_stride, int width,
127	0	int height, unsigned int input_shift) {
128	0	const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
129	0	const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
130	0	int64_t total_sse = 0;
131	0	int x, y;
132	0	for (y = 0; y < height; ++y) {
133	0	for (x = 0; x < width; ++x) {
134	0	int64_t diff;
135	0	diff = (a[x] >> input_shift) - (b[x] >> input_shift);
136	0	total_sse += diff * diff;
137	0	}
138	0	a += a_stride;
139	0	b += b_stride;
140	0	}
141	0	return total_sse;
142	0	}
143
144		static int64_t highbd_get_sse(const uint8_t a, int a_stride, const uint8_t b,
145	0	int b_stride, int width, int height) {
146	0	int64_t total_sse = 0;
147	0	int x, y;
148	0	const int dw = width % 16;
149	0	const int dh = height % 16;
150	0	unsigned int sse = 0;
151	0	int sum = 0;
152	0	if (dw > 0) {
153	0	encoder_highbd_8_variance(&a[width - dw], a_stride, &b[width - dw],
154	0	b_stride, dw, height, &sse, &sum);
155	0	total_sse += sse;
156	0	}
157	0	if (dh > 0) {
158	0	encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
159	0	&b[(height - dh) * b_stride], b_stride,
160	0	width - dw, dh, &sse, &sum);
161	0	total_sse += sse;
162	0	}
163	0	for (y = 0; y < height / 16; ++y) {
164	0	const uint8_t *pa = a;
165	0	const uint8_t *pb = b;
166	0	for (x = 0; x < width / 16; ++x) {
167	0	aom_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
168	0	total_sse += sse;
169	0	pa += 16;
170	0	pb += 16;
171	0	}
172	0	a += 16 * a_stride;
173	0	b += 16 * b_stride;
174	0	}
175	0	return total_sse;
176	0	}
177		#endif // CONFIG_AV1_HIGHBITDEPTH
178
179		uint64_t aom_get_y_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
180	0	int vstart, int height) {
181	0	return aom_var_2d_u8(a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
182	0	width, height) /
183	0	(width * height);
184	0	}
185
186		uint64_t aom_get_u_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
187	0	int vstart, int height) {
188	0	return aom_var_2d_u8(a->u_buffer + vstart * a->uv_stride + hstart,
189	0	a->uv_stride, width, height) /
190	0	(width * height);
191	0	}
192
193		uint64_t aom_get_v_var(const YV12_BUFFER_CONFIG *a, int hstart, int width,
194	0	int vstart, int height) {
195	0	return aom_var_2d_u8(a->v_buffer + vstart * a->uv_stride + hstart,
196	0	a->uv_stride, width, height) /
197	0	(width * height);
198	0	}
199
200		int64_t aom_get_y_sse_part(const YV12_BUFFER_CONFIG *a,
201		const YV12_BUFFER_CONFIG *b, int hstart, int width,
202	0	int vstart, int height) {
203	0	return get_sse(a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
204	0	b->y_buffer + vstart * b->y_stride + hstart, b->y_stride,
205	0	width, height);
206	0	}
207
208		int64_t aom_get_y_sse(const YV12_BUFFER_CONFIG *a,
209	4.23k	const YV12_BUFFER_CONFIG *b) {
210	4.23k	assert(a->y_crop_width == b->y_crop_width);
211	4.23k	assert(a->y_crop_height == b->y_crop_height);
212
213	4.23k	return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
214	4.23k	a->y_crop_width, a->y_crop_height);
215	4.23k	}
216
217		int64_t aom_get_u_sse_part(const YV12_BUFFER_CONFIG *a,
218		const YV12_BUFFER_CONFIG *b, int hstart, int width,
219	0	int vstart, int height) {
220	0	return get_sse(a->u_buffer + vstart * a->uv_stride + hstart, a->uv_stride,
221	0	b->u_buffer + vstart * b->uv_stride + hstart, b->uv_stride,
222	0	width, height);
223	0	}
224
225		int64_t aom_get_u_sse(const YV12_BUFFER_CONFIG *a,
226	4.48k	const YV12_BUFFER_CONFIG *b) {
227	4.48k	assert(a->uv_crop_width == b->uv_crop_width);
228	4.48k	assert(a->uv_crop_height == b->uv_crop_height);
229
230	4.48k	return get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride,
231	4.48k	a->uv_crop_width, a->uv_crop_height);
232	4.48k	}
233
234		int64_t aom_get_v_sse_part(const YV12_BUFFER_CONFIG *a,
235		const YV12_BUFFER_CONFIG *b, int hstart, int width,
236	0	int vstart, int height) {
237	0	return get_sse(a->v_buffer + vstart * a->uv_stride + hstart, a->uv_stride,
238	0	b->v_buffer + vstart * b->uv_stride + hstart, b->uv_stride,
239	0	width, height);
240	0	}
241
242		int64_t aom_get_v_sse(const YV12_BUFFER_CONFIG *a,
243	4.48k	const YV12_BUFFER_CONFIG *b) {
244	4.48k	assert(a->uv_crop_width == b->uv_crop_width);
245	4.48k	assert(a->uv_crop_height == b->uv_crop_height);
246
247	4.48k	return get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride,
248	4.48k	a->uv_crop_width, a->uv_crop_height);
249	4.48k	}
250
251		#if CONFIG_AV1_HIGHBITDEPTH
252		uint64_t aom_highbd_get_y_var(const YV12_BUFFER_CONFIG *a, int hstart,
253	0	int width, int vstart, int height) {
254	0	return aom_var_2d_u16(a->y_buffer + vstart * a->y_stride + hstart,
255	0	a->y_stride, width, height) /
256	0	(width * height);
257	0	}
258
259		uint64_t aom_highbd_get_u_var(const YV12_BUFFER_CONFIG *a, int hstart,
260	0	int width, int vstart, int height) {
261	0	return aom_var_2d_u16(a->u_buffer + vstart * a->uv_stride + hstart,
262	0	a->uv_stride, width, height) /
263	0	(width * height);
264	0	}
265
266		uint64_t aom_highbd_get_v_var(const YV12_BUFFER_CONFIG *a, int hstart,
267	0	int width, int vstart, int height) {
268	0	return aom_var_2d_u16(a->v_buffer + vstart * a->uv_stride + hstart,
269	0	a->uv_stride, width, height) /
270	0	(width * height);
271	0	}
272
273		int64_t aom_highbd_get_y_sse_part(const YV12_BUFFER_CONFIG *a,
274		const YV12_BUFFER_CONFIG *b, int hstart,
275	0	int width, int vstart, int height) {
276	0	return highbd_get_sse(
277	0	a->y_buffer + vstart * a->y_stride + hstart, a->y_stride,
278	0	b->y_buffer + vstart * b->y_stride + hstart, b->y_stride, width, height);
279	0	}
280
281		int64_t aom_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
282	0	const YV12_BUFFER_CONFIG *b) {
283	0	assert(a->y_crop_width == b->y_crop_width);
284	0	assert(a->y_crop_height == b->y_crop_height);
285	0	assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
286	0	assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
287
288	0	return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
289	0	a->y_crop_width, a->y_crop_height);
290	0	}
291
292		int64_t aom_highbd_get_u_sse_part(const YV12_BUFFER_CONFIG *a,
293		const YV12_BUFFER_CONFIG *b, int hstart,
294	0	int width, int vstart, int height) {
295	0	return highbd_get_sse(a->u_buffer + vstart * a->uv_stride + hstart,
296	0	a->uv_stride,
297	0	b->u_buffer + vstart * b->uv_stride + hstart,
298	0	b->uv_stride, width, height);
299	0	}
300
301		int64_t aom_highbd_get_u_sse(const YV12_BUFFER_CONFIG *a,
302	0	const YV12_BUFFER_CONFIG *b) {
303	0	assert(a->uv_crop_width == b->uv_crop_width);
304	0	assert(a->uv_crop_height == b->uv_crop_height);
305	0	assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
306	0	assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
307
308	0	return highbd_get_sse(a->u_buffer, a->uv_stride, b->u_buffer, b->uv_stride,
309	0	a->uv_crop_width, a->uv_crop_height);
310	0	}
311
312		int64_t aom_highbd_get_v_sse_part(const YV12_BUFFER_CONFIG *a,
313		const YV12_BUFFER_CONFIG *b, int hstart,
314	0	int width, int vstart, int height) {
315	0	return highbd_get_sse(a->v_buffer + vstart * a->uv_stride + hstart,
316	0	a->uv_stride,
317	0	b->v_buffer + vstart * b->uv_stride + hstart,
318	0	b->uv_stride, width, height);
319	0	}
320
321		int64_t aom_highbd_get_v_sse(const YV12_BUFFER_CONFIG *a,
322	0	const YV12_BUFFER_CONFIG *b) {
323	0	assert(a->uv_crop_width == b->uv_crop_width);
324	0	assert(a->uv_crop_height == b->uv_crop_height);
325	0	assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
326	0	assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
327
328	0	return highbd_get_sse(a->v_buffer, a->uv_stride, b->v_buffer, b->uv_stride,
329	0	a->uv_crop_width, a->uv_crop_height);
330	0	}
331		#endif // CONFIG_AV1_HIGHBITDEPTH
332
333		int64_t aom_get_sse_plane(const YV12_BUFFER_CONFIG *a,
334	13.1k	const YV12_BUFFER_CONFIG *b, int plane, int highbd) {
335	13.1k	#if CONFIG_AV1_HIGHBITDEPTH
336	13.1k	if (highbd) {
337	0	switch (plane) {
338	0	case 0: return aom_highbd_get_y_sse(a, b);
339	0	case 1: return aom_highbd_get_u_sse(a, b);
340	0	case 2: return aom_highbd_get_v_sse(a, b);
341	0	default: assert(plane >= 0 && plane <= 2); return 0;
342	0	}
343	13.1k	} else {
344	13.1k	switch (plane) {
345	4.23k	case 0: return aom_get_y_sse(a, b);
346	4.48k	case 1: return aom_get_u_sse(a, b);
347	4.48k	case 2: return aom_get_v_sse(a, b);
348	0	default: assert(plane >= 0 && plane <= 2); return 0;
349	13.1k	}
350	13.1k	}
351		#else
352		(void)highbd;
353		switch (plane) {
354		case 0: return aom_get_y_sse(a, b);
355		case 1: return aom_get_u_sse(a, b);
356		case 2: return aom_get_v_sse(a, b);
357		default: assert(plane >= 0 && plane <= 2); return 0;
358		}
359		#endif
360	13.1k	}
361
362		#if CONFIG_AV1_HIGHBITDEPTH
363		void aom_calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
364		const YV12_BUFFER_CONFIG b, PSNR_STATS psnr,
365	0	uint32_t bit_depth, uint32_t in_bit_depth) {
366	0	assert(a->y_crop_width == b->y_crop_width);
367	0	assert(a->y_crop_height == b->y_crop_height);
368	0	assert(a->uv_crop_width == b->uv_crop_width);
369	0	assert(a->uv_crop_height == b->uv_crop_height);
370	0	const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
371	0	const int heights[3] = { a->y_crop_height, a->uv_crop_height,
372	0	a->uv_crop_height };
373	0	const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
374	0	const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
375	0	int i;
376	0	uint64_t total_sse = 0;
377	0	uint32_t total_samples = 0;
378	0	double peak = (double)((1 << in_bit_depth) - 1);
379	0	const unsigned int input_shift = bit_depth - in_bit_depth;
380
381	0	for (i = 0; i < 3; ++i) {
382	0	const int w = widths[i];
383	0	const int h = heights[i];
384	0	const uint32_t samples = w * h;
385	0	uint64_t sse;
386	0	if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
387	0	if (input_shift) {
388	0	sse = highbd_get_sse_shift(a->buffers[i], a_strides[i], b->buffers[i],
389	0	b_strides[i], w, h, input_shift);
390	0	} else {
391	0	sse = highbd_get_sse(a->buffers[i], a_strides[i], b->buffers[i],
392	0	b_strides[i], w, h);
393	0	}
394	0	} else {
395	0	sse = get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w,
396	0	h);
397	0	}
398	0	psnr->sse[1 + i] = sse;
399	0	psnr->samples[1 + i] = samples;
400	0	psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);
401
402	0	total_sse += sse;
403	0	total_samples += samples;
404	0	}
405
406	0	psnr->sse[0] = total_sse;
407	0	psnr->samples[0] = total_samples;
408	0	psnr->psnr[0] =
409	0	aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);
410
411		// Compute PSNR based on stream bit depth
412	0	if ((a->flags & YV12_FLAG_HIGHBITDEPTH) && (in_bit_depth < bit_depth)) {
413	0	peak = (double)((1 << bit_depth) - 1);
414	0	total_sse = 0;
415	0	total_samples = 0;
416	0	for (i = 0; i < 3; ++i) {
417	0	const int w = widths[i];
418	0	const int h = heights[i];
419	0	const uint32_t samples = w * h;
420	0	uint64_t sse;
421	0	sse = highbd_get_sse(a->buffers[i], a_strides[i], b->buffers[i],
422	0	b_strides[i], w, h);
423	0	psnr->sse_hbd[1 + i] = sse;
424	0	psnr->samples_hbd[1 + i] = samples;
425	0	psnr->psnr_hbd[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);
426	0	total_sse += sse;
427	0	total_samples += samples;
428	0	}
429
430	0	psnr->sse_hbd[0] = total_sse;
431	0	psnr->samples_hbd[0] = total_samples;
432	0	psnr->psnr_hbd[0] =
433	0	aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);
434	0	}
435	0	}
436		#endif
437
438		void aom_calc_psnr(const YV12_BUFFER_CONFIG a, const YV12_BUFFER_CONFIG b,
439	0	PSNR_STATS *psnr) {
440	0	assert(a->y_crop_width == b->y_crop_width);
441	0	assert(a->y_crop_height == b->y_crop_height);
442	0	assert(a->uv_crop_width == b->uv_crop_width);
443	0	assert(a->uv_crop_height == b->uv_crop_height);
444	0	static const double peak = 255.0;
445	0	const int widths[3] = { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
446	0	const int heights[3] = { a->y_crop_height, a->uv_crop_height,
447	0	a->uv_crop_height };
448	0	const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
449	0	const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
450	0	int i;
451	0	uint64_t total_sse = 0;
452	0	uint32_t total_samples = 0;
453
454	0	for (i = 0; i < 3; ++i) {
455	0	const int w = widths[i];
456	0	const int h = heights[i];
457	0	const uint32_t samples = w * h;
458	0	const uint64_t sse =
459	0	get_sse(a->buffers[i], a_strides[i], b->buffers[i], b_strides[i], w, h);
460	0	psnr->sse[1 + i] = sse;
461	0	psnr->samples[1 + i] = samples;
462	0	psnr->psnr[1 + i] = aom_sse_to_psnr(samples, peak, (double)sse);
463
464	0	total_sse += sse;
465	0	total_samples += samples;
466	0	}
467
468	0	psnr->sse[0] = total_sse;
469	0	psnr->samples[0] = total_samples;
470	0	psnr->psnr[0] =
471	0	aom_sse_to_psnr((double)total_samples, peak, (double)total_sse);
472	0	}