2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
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.
12 #include "./vpx_dsp_rtcd.h"
13 #include "vpx_ports/mem.h"
15 unsigned int vpx_avg_8x8_c(const uint8_t *s, int p) {
18 for (i = 0; i < 8; ++i, s += p)
19 for (j = 0; j < 8; sum += s[j], ++j) {
22 return (sum + 32) >> 6;
25 unsigned int vpx_avg_4x4_c(const uint8_t *s, int p) {
28 for (i = 0; i < 4; ++i, s += p)
29 for (j = 0; j < 4; sum += s[j], ++j) {
32 return (sum + 8) >> 4;
35 // src_diff: first pass, 9 bit, dynamic range [-255, 255]
36 // second pass, 12 bit, dynamic range [-2040, 2040]
37 static void hadamard_col8(const int16_t *src_diff, int src_stride,
39 int16_t b0 = src_diff[0 * src_stride] + src_diff[1 * src_stride];
40 int16_t b1 = src_diff[0 * src_stride] - src_diff[1 * src_stride];
41 int16_t b2 = src_diff[2 * src_stride] + src_diff[3 * src_stride];
42 int16_t b3 = src_diff[2 * src_stride] - src_diff[3 * src_stride];
43 int16_t b4 = src_diff[4 * src_stride] + src_diff[5 * src_stride];
44 int16_t b5 = src_diff[4 * src_stride] - src_diff[5 * src_stride];
45 int16_t b6 = src_diff[6 * src_stride] + src_diff[7 * src_stride];
46 int16_t b7 = src_diff[6 * src_stride] - src_diff[7 * src_stride];
67 // The order of the output coeff of the hadamard is not important. For
68 // optimization purposes the final transpose may be skipped.
69 void vpx_hadamard_8x8_c(const int16_t *src_diff, int src_stride,
74 int16_t *tmp_buf = &buffer[0];
75 for (idx = 0; idx < 8; ++idx) {
76 hadamard_col8(src_diff, src_stride, tmp_buf); // src_diff: 9 bit
77 // dynamic range [-255, 255]
83 for (idx = 0; idx < 8; ++idx) {
84 hadamard_col8(tmp_buf, 8, buffer2 + 8 * idx); // tmp_buf: 12 bit
85 // dynamic range [-2040, 2040]
87 // dynamic range [-16320, 16320]
91 for (idx = 0; idx < 64; ++idx) coeff[idx] = (tran_low_t)buffer2[idx];
94 // In place 16x16 2D Hadamard transform
95 void vpx_hadamard_16x16_c(const int16_t *src_diff, int src_stride,
98 for (idx = 0; idx < 4; ++idx) {
99 // src_diff: 9 bit, dynamic range [-255, 255]
100 const int16_t *src_ptr =
101 src_diff + (idx >> 1) * 8 * src_stride + (idx & 0x01) * 8;
102 vpx_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
105 // coeff: 15 bit, dynamic range [-16320, 16320]
106 for (idx = 0; idx < 64; ++idx) {
107 tran_low_t a0 = coeff[0];
108 tran_low_t a1 = coeff[64];
109 tran_low_t a2 = coeff[128];
110 tran_low_t a3 = coeff[192];
112 tran_low_t b0 = (a0 + a1) >> 1; // (a0 + a1): 16 bit, [-32640, 32640]
113 tran_low_t b1 = (a0 - a1) >> 1; // b0-b3: 15 bit, dynamic range
114 tran_low_t b2 = (a2 + a3) >> 1; // [-16320, 16320]
115 tran_low_t b3 = (a2 - a3) >> 1;
117 coeff[0] = b0 + b2; // 16 bit, [-32640, 32640]
119 coeff[128] = b0 - b2;
120 coeff[192] = b1 - b3;
126 // coeff: 16 bits, dynamic range [-32640, 32640].
127 // length: value range {16, 64, 256, 1024}.
128 int vpx_satd_c(const tran_low_t *coeff, int length) {
131 for (i = 0; i < length; ++i) satd += abs(coeff[i]);
133 // satd: 26 bits, dynamic range [-32640 * 1024, 32640 * 1024]
137 // Integer projection onto row vectors.
138 // height: value range {16, 32, 64}.
139 void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t *ref,
140 const int ref_stride, const int height) {
142 const int norm_factor = height >> 1;
143 for (idx = 0; idx < 16; ++idx) {
146 // hbuf[idx]: 14 bit, dynamic range [0, 16320].
147 for (i = 0; i < height; ++i) hbuf[idx] += ref[i * ref_stride];
148 // hbuf[idx]: 9 bit, dynamic range [0, 510].
149 hbuf[idx] /= norm_factor;
154 // width: value range {16, 32, 64}.
155 int16_t vpx_int_pro_col_c(const uint8_t *ref, const int width) {
158 // sum: 14 bit, dynamic range [0, 16320]
159 for (idx = 0; idx < width; ++idx) sum += ref[idx];
166 int vpx_vector_var_c(const int16_t *ref, const int16_t *src, const int bwl) {
168 int width = 4 << bwl;
169 int sse = 0, mean = 0, var;
171 for (i = 0; i < width; ++i) {
172 int diff = ref[i] - src[i]; // diff: dynamic range [-510, 510], 10 bits.
173 mean += diff; // mean: dynamic range 16 bits.
174 sse += diff * diff; // sse: dynamic range 26 bits.
177 // (mean * mean): dynamic range 31 bits.
178 var = sse - ((mean * mean) >> (bwl + 2));
182 void vpx_minmax_8x8_c(const uint8_t *s, int p, const uint8_t *d, int dp,
183 int *min, int *max) {
187 for (i = 0; i < 8; ++i, s += p, d += dp) {
188 for (j = 0; j < 8; ++j) {
189 int diff = abs(s[j] - d[j]);
190 *min = diff < *min ? diff : *min;
191 *max = diff > *max ? diff : *max;
196 #if CONFIG_VP9_HIGHBITDEPTH
197 unsigned int vpx_highbd_avg_8x8_c(const uint8_t *s8, int p) {
200 const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
201 for (i = 0; i < 8; ++i, s += p)
202 for (j = 0; j < 8; sum += s[j], ++j) {
205 return (sum + 32) >> 6;
208 unsigned int vpx_highbd_avg_4x4_c(const uint8_t *s8, int p) {
211 const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
212 for (i = 0; i < 4; ++i, s += p)
213 for (j = 0; j < 4; sum += s[j], ++j) {
216 return (sum + 8) >> 4;
219 void vpx_highbd_minmax_8x8_c(const uint8_t *s8, int p, const uint8_t *d8,
220 int dp, int *min, int *max) {
222 const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
223 const uint16_t *d = CONVERT_TO_SHORTPTR(d8);
226 for (i = 0; i < 8; ++i, s += p, d += dp) {
227 for (j = 0; j < 8; ++j) {
228 int diff = abs(s[j] - d[j]);
229 *min = diff < *min ? diff : *min;
230 *max = diff > *max ? diff : *max;
234 #endif // CONFIG_VP9_HIGHBITDEPTH