/src/ffmpeg/libavcodec/hpel_template.c
Line | Count | Source |
1 | | /* |
2 | | * Copyright (c) 2000, 2001 Fabrice Bellard |
3 | | * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
4 | | * |
5 | | * This file is part of FFmpeg. |
6 | | * |
7 | | * FFmpeg is free software; you can redistribute it and/or |
8 | | * modify it under the terms of the GNU Lesser General Public |
9 | | * License as published by the Free Software Foundation; either |
10 | | * version 2.1 of the License, or (at your option) any later version. |
11 | | * |
12 | | * FFmpeg is distributed in the hope that it will be useful, |
13 | | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | | * Lesser General Public License for more details. |
16 | | * |
17 | | * You should have received a copy of the GNU Lesser General Public |
18 | | * License along with FFmpeg; if not, write to the Free Software |
19 | | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 | | */ |
21 | | |
22 | | #include "libavutil/intreadwrite.h" |
23 | | |
24 | | #include "pixels.h" |
25 | | |
26 | | #include "bit_depth_template.c" |
27 | | |
28 | | #define DEF_HPEL(OPNAME, OP) \ |
29 | | static inline void FUNC(OPNAME ## _pixels8_l2)(uint8_t *dst, \ |
30 | | const uint8_t *src1, \ |
31 | | const uint8_t *src2, \ |
32 | | ptrdiff_t dst_stride, \ |
33 | | ptrdiff_t src_stride1, \ |
34 | | ptrdiff_t src_stride2, \ |
35 | 96.3M | int h) \ |
36 | 96.3M | { \ |
37 | 96.3M | int i; \ |
38 | 1.23G | for (i = 0; i < h; i++) { \ |
39 | 1.13G | pixel4 a, b; \ |
40 | 1.13G | a = AV_RN4P(&src1[i * src_stride1]); \ |
41 | 1.13G | b = AV_RN4P(&src2[i * src_stride2]); \ |
42 | 1.13G | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ |
43 | 1.13G | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ |
44 | 1.13G | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ |
45 | 1.13G | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ |
46 | 1.13G | rnd_avg_pixel4(a, b)); \ |
47 | 1.13G | } \ |
48 | 96.3M | } \ qpeldsp.c:put_pixels8_l2_8 Line | Count | Source | 35 | 19.2M | int h) \ | 36 | 19.2M | { \ | 37 | 19.2M | int i; \ | 38 | 243M | for (i = 0; i < h; i++) { \ | 39 | 223M | pixel4 a, b; \ | 40 | 223M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 223M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 223M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 223M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 223M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 223M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 223M | rnd_avg_pixel4(a, b)); \ | 47 | 223M | } \ | 48 | 19.2M | } \ |
qpeldsp.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 363k | int h) \ | 36 | 363k | { \ | 37 | 363k | int i; \ | 38 | 3.73M | for (i = 0; i < h; i++) { \ | 39 | 3.37M | pixel4 a, b; \ | 40 | 3.37M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 3.37M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 3.37M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 3.37M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 3.37M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 3.37M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 3.37M | rnd_avg_pixel4(a, b)); \ | 47 | 3.37M | } \ | 48 | 363k | } \ |
hpeldsp.c:put_pixels8_l2_8 Line | Count | Source | 35 | 44.4M | int h) \ | 36 | 44.4M | { \ | 37 | 44.4M | int i; \ | 38 | 550M | for (i = 0; i < h; i++) { \ | 39 | 506M | pixel4 a, b; \ | 40 | 506M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 506M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 506M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 506M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 506M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 506M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 506M | rnd_avg_pixel4(a, b)); \ | 47 | 506M | } \ | 48 | 44.4M | } \ |
hpeldsp.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 657k | int h) \ | 36 | 657k | { \ | 37 | 657k | int i; \ | 38 | 8.53M | for (i = 0; i < h; i++) { \ | 39 | 7.87M | pixel4 a, b; \ | 40 | 7.87M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 7.87M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 7.87M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 7.87M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 7.87M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 7.87M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 7.87M | rnd_avg_pixel4(a, b)); \ | 47 | 7.87M | } \ | 48 | 657k | } \ |
Unexecuted instantiation: snow.c:avg_pixels8_l2_8 Unexecuted instantiation: snow.c:put_pixels8_l2_8 h264qpel.c:put_pixels8_l2_8 Line | Count | Source | 35 | 8.99M | int h) \ | 36 | 8.99M | { \ | 37 | 8.99M | int i; \ | 38 | 117M | for (i = 0; i < h; i++) { \ | 39 | 108M | pixel4 a, b; \ | 40 | 108M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 108M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 108M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 108M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 108M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 108M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 108M | rnd_avg_pixel4(a, b)); \ | 47 | 108M | } \ | 48 | 8.99M | } \ |
h264qpel.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 989k | int h) \ | 36 | 989k | { \ | 37 | 989k | int i; \ | 38 | 11.9M | for (i = 0; i < h; i++) { \ | 39 | 11.0M | pixel4 a, b; \ | 40 | 11.0M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 11.0M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 11.0M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 11.0M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 11.0M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 11.0M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 11.0M | rnd_avg_pixel4(a, b)); \ | 47 | 11.0M | } \ | 48 | 989k | } \ |
h264qpel.c:put_pixels8_l2_9 Line | Count | Source | 35 | 4.39M | int h) \ | 36 | 4.39M | { \ | 37 | 4.39M | int i; \ | 38 | 60.7M | for (i = 0; i < h; i++) { \ | 39 | 56.3M | pixel4 a, b; \ | 40 | 56.3M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 56.3M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 56.3M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 56.3M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 56.3M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 56.3M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 56.3M | rnd_avg_pixel4(a, b)); \ | 47 | 56.3M | } \ | 48 | 4.39M | } \ |
h264qpel.c:avg_pixels8_l2_9 Line | Count | Source | 35 | 768k | int h) \ | 36 | 768k | { \ | 37 | 768k | int i; \ | 38 | 10.2M | for (i = 0; i < h; i++) { \ | 39 | 9.52M | pixel4 a, b; \ | 40 | 9.52M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 9.52M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 9.52M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 9.52M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 9.52M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 9.52M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 9.52M | rnd_avg_pixel4(a, b)); \ | 47 | 9.52M | } \ | 48 | 768k | } \ |
h264qpel.c:put_pixels8_l2_10 Line | Count | Source | 35 | 5.85M | int h) \ | 36 | 5.85M | { \ | 37 | 5.85M | int i; \ | 38 | 82.9M | for (i = 0; i < h; i++) { \ | 39 | 77.1M | pixel4 a, b; \ | 40 | 77.1M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 77.1M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 77.1M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 77.1M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 77.1M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 77.1M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 77.1M | rnd_avg_pixel4(a, b)); \ | 47 | 77.1M | } \ | 48 | 5.85M | } \ |
h264qpel.c:avg_pixels8_l2_10 Line | Count | Source | 35 | 479k | int h) \ | 36 | 479k | { \ | 37 | 479k | int i; \ | 38 | 6.38M | for (i = 0; i < h; i++) { \ | 39 | 5.90M | pixel4 a, b; \ | 40 | 5.90M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 5.90M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 5.90M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 5.90M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 5.90M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 5.90M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 5.90M | rnd_avg_pixel4(a, b)); \ | 47 | 5.90M | } \ | 48 | 479k | } \ |
h264qpel.c:put_pixels8_l2_12 Line | Count | Source | 35 | 4.48M | int h) \ | 36 | 4.48M | { \ | 37 | 4.48M | int i; \ | 38 | 57.9M | for (i = 0; i < h; i++) { \ | 39 | 53.4M | pixel4 a, b; \ | 40 | 53.4M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 53.4M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 53.4M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 53.4M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 53.4M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 53.4M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 53.4M | rnd_avg_pixel4(a, b)); \ | 47 | 53.4M | } \ | 48 | 4.48M | } \ |
h264qpel.c:avg_pixels8_l2_12 Line | Count | Source | 35 | 550k | int h) \ | 36 | 550k | { \ | 37 | 550k | int i; \ | 38 | 6.67M | for (i = 0; i < h; i++) { \ | 39 | 6.12M | pixel4 a, b; \ | 40 | 6.12M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 6.12M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 6.12M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 6.12M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 6.12M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 6.12M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 6.12M | rnd_avg_pixel4(a, b)); \ | 47 | 6.12M | } \ | 48 | 550k | } \ |
h264qpel.c:put_pixels8_l2_14 Line | Count | Source | 35 | 3.96M | int h) \ | 36 | 3.96M | { \ | 37 | 3.96M | int i; \ | 38 | 56.6M | for (i = 0; i < h; i++) { \ | 39 | 52.6M | pixel4 a, b; \ | 40 | 52.6M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 52.6M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 52.6M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 52.6M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 52.6M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 52.6M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 52.6M | rnd_avg_pixel4(a, b)); \ | 47 | 52.6M | } \ | 48 | 3.96M | } \ |
h264qpel.c:avg_pixels8_l2_14 Line | Count | Source | 35 | 1.15M | int h) \ | 36 | 1.15M | { \ | 37 | 1.15M | int i; \ | 38 | 17.0M | for (i = 0; i < h; i++) { \ | 39 | 15.9M | pixel4 a, b; \ | 40 | 15.9M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 15.9M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 15.9M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 15.9M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 15.9M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 15.9M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 15.9M | rnd_avg_pixel4(a, b)); \ | 47 | 15.9M | } \ | 48 | 1.15M | } \ |
|
49 | | \ |
50 | | static inline void FUNC(OPNAME ## _pixels4_l2)(uint8_t *dst, \ |
51 | | const uint8_t *src1, \ |
52 | | const uint8_t *src2, \ |
53 | | ptrdiff_t dst_stride, \ |
54 | | ptrdiff_t src_stride1, \ |
55 | | ptrdiff_t src_stride2, \ |
56 | 11.2M | int h) \ |
57 | 11.2M | { \ |
58 | 11.2M | int i; \ |
59 | 56.2M | for (i = 0; i < h; i++) { \ |
60 | 45.0M | pixel4 a, b; \ |
61 | 45.0M | a = AV_RN4P(&src1[i * src_stride1]); \ |
62 | 45.0M | b = AV_RN4P(&src2[i * src_stride2]); \ |
63 | 45.0M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ |
64 | 45.0M | } \ |
65 | 11.2M | } \ Unexecuted instantiation: qpeldsp.c:avg_pixels4_l2_8 Unexecuted instantiation: qpeldsp.c:put_pixels4_l2_8 hpeldsp.c:put_pixels4_l2_8 Line | Count | Source | 56 | 29.2k | int h) \ | 57 | 29.2k | { \ | 58 | 29.2k | int i; \ | 59 | 217k | for (i = 0; i < h; i++) { \ | 60 | 188k | pixel4 a, b; \ | 61 | 188k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 188k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 188k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 188k | } \ | 65 | 29.2k | } \ |
hpeldsp.c:avg_pixels4_l2_8 Line | Count | Source | 56 | 16.3k | int h) \ | 57 | 16.3k | { \ | 58 | 16.3k | int i; \ | 59 | 133k | for (i = 0; i < h; i++) { \ | 60 | 116k | pixel4 a, b; \ | 61 | 116k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 116k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 116k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 116k | } \ | 65 | 16.3k | } \ |
Unexecuted instantiation: snow.c:avg_pixels4_l2_8 Unexecuted instantiation: snow.c:put_pixels4_l2_8 h264qpel.c:put_pixels4_l2_8 Line | Count | Source | 56 | 3.14M | int h) \ | 57 | 3.14M | { \ | 58 | 3.14M | int i; \ | 59 | 15.7M | for (i = 0; i < h; i++) { \ | 60 | 12.5M | pixel4 a, b; \ | 61 | 12.5M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 12.5M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 12.5M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 12.5M | } \ | 65 | 3.14M | } \ |
h264qpel.c:avg_pixels4_l2_8 Line | Count | Source | 56 | 195k | int h) \ | 57 | 195k | { \ | 58 | 195k | int i; \ | 59 | 976k | for (i = 0; i < h; i++) { \ | 60 | 781k | pixel4 a, b; \ | 61 | 781k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 781k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 781k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 781k | } \ | 65 | 195k | } \ |
h264qpel.c:put_pixels4_l2_9 Line | Count | Source | 56 | 1.48M | int h) \ | 57 | 1.48M | { \ | 58 | 1.48M | int i; \ | 59 | 7.40M | for (i = 0; i < h; i++) { \ | 60 | 5.92M | pixel4 a, b; \ | 61 | 5.92M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 5.92M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 5.92M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 5.92M | } \ | 65 | 1.48M | } \ |
h264qpel.c:avg_pixels4_l2_9 Line | Count | Source | 56 | 152k | int h) \ | 57 | 152k | { \ | 58 | 152k | int i; \ | 59 | 762k | for (i = 0; i < h; i++) { \ | 60 | 609k | pixel4 a, b; \ | 61 | 609k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 609k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 609k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 609k | } \ | 65 | 152k | } \ |
h264qpel.c:put_pixels4_l2_10 Line | Count | Source | 56 | 2.41M | int h) \ | 57 | 2.41M | { \ | 58 | 2.41M | int i; \ | 59 | 12.0M | for (i = 0; i < h; i++) { \ | 60 | 9.64M | pixel4 a, b; \ | 61 | 9.64M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 9.64M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 9.64M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 9.64M | } \ | 65 | 2.41M | } \ |
h264qpel.c:avg_pixels4_l2_10 Line | Count | Source | 56 | 155k | int h) \ | 57 | 155k | { \ | 58 | 155k | int i; \ | 59 | 775k | for (i = 0; i < h; i++) { \ | 60 | 620k | pixel4 a, b; \ | 61 | 620k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 620k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 620k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 620k | } \ | 65 | 155k | } \ |
h264qpel.c:put_pixels4_l2_12 Line | Count | Source | 56 | 2.23M | int h) \ | 57 | 2.23M | { \ | 58 | 2.23M | int i; \ | 59 | 11.1M | for (i = 0; i < h; i++) { \ | 60 | 8.94M | pixel4 a, b; \ | 61 | 8.94M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 8.94M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 8.94M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 8.94M | } \ | 65 | 2.23M | } \ |
h264qpel.c:avg_pixels4_l2_12 Line | Count | Source | 56 | 128k | int h) \ | 57 | 128k | { \ | 58 | 128k | int i; \ | 59 | 641k | for (i = 0; i < h; i++) { \ | 60 | 513k | pixel4 a, b; \ | 61 | 513k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 513k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 513k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 513k | } \ | 65 | 128k | } \ |
h264qpel.c:put_pixels4_l2_14 Line | Count | Source | 56 | 1.03M | int h) \ | 57 | 1.03M | { \ | 58 | 1.03M | int i; \ | 59 | 5.16M | for (i = 0; i < h; i++) { \ | 60 | 4.13M | pixel4 a, b; \ | 61 | 4.13M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 4.13M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 4.13M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 4.13M | } \ | 65 | 1.03M | } \ |
h264qpel.c:avg_pixels4_l2_14 Line | Count | Source | 56 | 245k | int h) \ | 57 | 245k | { \ | 58 | 245k | int i; \ | 59 | 1.22M | for (i = 0; i < h; i++) { \ | 60 | 981k | pixel4 a, b; \ | 61 | 981k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 981k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 981k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 981k | } \ | 65 | 245k | } \ |
|
66 | | \ |
67 | | static inline void FUNC(OPNAME ## _pixels2_l2)(uint8_t *dst, \ |
68 | | const uint8_t *src1, \ |
69 | | const uint8_t *src2, \ |
70 | | ptrdiff_t dst_stride, \ |
71 | | ptrdiff_t src_stride1, \ |
72 | | ptrdiff_t src_stride2, \ |
73 | 769k | int h) \ |
74 | 769k | { \ |
75 | 769k | int i; \ |
76 | 2.40M | for (i = 0; i < h; i++) { \ |
77 | 1.63M | pixel4 a, b; \ |
78 | 1.63M | a = AV_RN2P(&src1[i * src_stride1]); \ |
79 | 1.63M | b = AV_RN2P(&src2[i * src_stride2]); \ |
80 | 1.63M | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ |
81 | 1.63M | } \ |
82 | 769k | } \ Unexecuted instantiation: qpeldsp.c:avg_pixels2_l2_8 Unexecuted instantiation: qpeldsp.c:put_pixels2_l2_8 hpeldsp.c:put_pixels2_l2_8 Line | Count | Source | 73 | 28.6k | int h) \ | 74 | 28.6k | { \ | 75 | 28.6k | int i; \ | 76 | 136k | for (i = 0; i < h; i++) { \ | 77 | 107k | pixel4 a, b; \ | 78 | 107k | a = AV_RN2P(&src1[i * src_stride1]); \ | 79 | 107k | b = AV_RN2P(&src2[i * src_stride2]); \ | 80 | 107k | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 81 | 107k | } \ | 82 | 28.6k | } \ |
hpeldsp.c:avg_pixels2_l2_8 Line | Count | Source | 73 | 26.8k | int h) \ | 74 | 26.8k | { \ | 75 | 26.8k | int i; \ | 76 | 129k | for (i = 0; i < h; i++) { \ | 77 | 102k | pixel4 a, b; \ | 78 | 102k | a = AV_RN2P(&src1[i * src_stride1]); \ | 79 | 102k | b = AV_RN2P(&src2[i * src_stride2]); \ | 80 | 102k | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 81 | 102k | } \ | 82 | 26.8k | } \ |
Line | Count | Source | 73 | 713k | int h) \ | 74 | 713k | { \ | 75 | 713k | int i; \ | 76 | 2.14M | for (i = 0; i < h; i++) { \ | 77 | 1.42M | pixel4 a, b; \ | 78 | 1.42M | a = AV_RN2P(&src1[i * src_stride1]); \ | 79 | 1.42M | b = AV_RN2P(&src2[i * src_stride2]); \ | 80 | 1.42M | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 81 | 1.42M | } \ | 82 | 713k | } \ |
Unexecuted instantiation: snow.c:avg_pixels2_l2_8 Unexecuted instantiation: h264qpel.c:avg_pixels2_l2_8 Unexecuted instantiation: h264qpel.c:put_pixels2_l2_8 Unexecuted instantiation: h264qpel.c:avg_pixels2_l2_9 Unexecuted instantiation: h264qpel.c:put_pixels2_l2_9 Unexecuted instantiation: h264qpel.c:avg_pixels2_l2_10 Unexecuted instantiation: h264qpel.c:put_pixels2_l2_10 Unexecuted instantiation: h264qpel.c:avg_pixels2_l2_12 Unexecuted instantiation: h264qpel.c:put_pixels2_l2_12 Unexecuted instantiation: h264qpel.c:avg_pixels2_l2_14 Unexecuted instantiation: h264qpel.c:put_pixels2_l2_14 |
83 | | \ |
84 | | static inline void FUNC(OPNAME ## _pixels16_l2)(uint8_t *dst, \ |
85 | | const uint8_t *src1, \ |
86 | | const uint8_t *src2, \ |
87 | | ptrdiff_t dst_stride, \ |
88 | | ptrdiff_t src_stride1, \ |
89 | | ptrdiff_t src_stride2, \ |
90 | 13.5M | int h) \ |
91 | 13.5M | { \ |
92 | 13.5M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ |
93 | 13.5M | src_stride1, src_stride2, h); \ |
94 | 13.5M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ |
95 | 13.5M | src1 + 8 * sizeof(pixel), \ |
96 | 13.5M | src2 + 8 * sizeof(pixel), \ |
97 | 13.5M | dst_stride, src_stride1, \ |
98 | 13.5M | src_stride2, h); \ |
99 | 13.5M | } \ qpeldsp.c:put_pixels16_l2_8 Line | Count | Source | 90 | 4.53M | int h) \ | 91 | 4.53M | { \ | 92 | 4.53M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 4.53M | src_stride1, src_stride2, h); \ | 94 | 4.53M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 4.53M | src1 + 8 * sizeof(pixel), \ | 96 | 4.53M | src2 + 8 * sizeof(pixel), \ | 97 | 4.53M | dst_stride, src_stride1, \ | 98 | 4.53M | src_stride2, h); \ | 99 | 4.53M | } \ |
qpeldsp.c:avg_pixels16_l2_8 Line | Count | Source | 90 | 57.1k | int h) \ | 91 | 57.1k | { \ | 92 | 57.1k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 57.1k | src_stride1, src_stride2, h); \ | 94 | 57.1k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 57.1k | src1 + 8 * sizeof(pixel), \ | 96 | 57.1k | src2 + 8 * sizeof(pixel), \ | 97 | 57.1k | dst_stride, src_stride1, \ | 98 | 57.1k | src_stride2, h); \ | 99 | 57.1k | } \ |
Unexecuted instantiation: hpeldsp.c:avg_pixels16_l2_8 Unexecuted instantiation: hpeldsp.c:put_pixels16_l2_8 Unexecuted instantiation: snow.c:avg_pixels16_l2_8 Unexecuted instantiation: snow.c:put_pixels16_l2_8 h264qpel.c:put_pixels16_l2_8 Line | Count | Source | 90 | 2.30M | int h) \ | 91 | 2.30M | { \ | 92 | 2.30M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 2.30M | src_stride1, src_stride2, h); \ | 94 | 2.30M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 2.30M | src1 + 8 * sizeof(pixel), \ | 96 | 2.30M | src2 + 8 * sizeof(pixel), \ | 97 | 2.30M | dst_stride, src_stride1, \ | 98 | 2.30M | src_stride2, h); \ | 99 | 2.30M | } \ |
h264qpel.c:avg_pixels16_l2_8 Line | Count | Source | 90 | 193k | int h) \ | 91 | 193k | { \ | 92 | 193k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 193k | src_stride1, src_stride2, h); \ | 94 | 193k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 193k | src1 + 8 * sizeof(pixel), \ | 96 | 193k | src2 + 8 * sizeof(pixel), \ | 97 | 193k | dst_stride, src_stride1, \ | 98 | 193k | src_stride2, h); \ | 99 | 193k | } \ |
h264qpel.c:put_pixels16_l2_9 Line | Count | Source | 90 | 1.32M | int h) \ | 91 | 1.32M | { \ | 92 | 1.32M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.32M | src_stride1, src_stride2, h); \ | 94 | 1.32M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.32M | src1 + 8 * sizeof(pixel), \ | 96 | 1.32M | src2 + 8 * sizeof(pixel), \ | 97 | 1.32M | dst_stride, src_stride1, \ | 98 | 1.32M | src_stride2, h); \ | 99 | 1.32M | } \ |
h264qpel.c:avg_pixels16_l2_9 Line | Count | Source | 90 | 211k | int h) \ | 91 | 211k | { \ | 92 | 211k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 211k | src_stride1, src_stride2, h); \ | 94 | 211k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 211k | src1 + 8 * sizeof(pixel), \ | 96 | 211k | src2 + 8 * sizeof(pixel), \ | 97 | 211k | dst_stride, src_stride1, \ | 98 | 211k | src_stride2, h); \ | 99 | 211k | } \ |
h264qpel.c:put_pixels16_l2_10 Line | Count | Source | 90 | 1.89M | int h) \ | 91 | 1.89M | { \ | 92 | 1.89M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.89M | src_stride1, src_stride2, h); \ | 94 | 1.89M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.89M | src1 + 8 * sizeof(pixel), \ | 96 | 1.89M | src2 + 8 * sizeof(pixel), \ | 97 | 1.89M | dst_stride, src_stride1, \ | 98 | 1.89M | src_stride2, h); \ | 99 | 1.89M | } \ |
h264qpel.c:avg_pixels16_l2_10 Line | Count | Source | 90 | 129k | int h) \ | 91 | 129k | { \ | 92 | 129k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 129k | src_stride1, src_stride2, h); \ | 94 | 129k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 129k | src1 + 8 * sizeof(pixel), \ | 96 | 129k | src2 + 8 * sizeof(pixel), \ | 97 | 129k | dst_stride, src_stride1, \ | 98 | 129k | src_stride2, h); \ | 99 | 129k | } \ |
h264qpel.c:put_pixels16_l2_12 Line | Count | Source | 90 | 1.09M | int h) \ | 91 | 1.09M | { \ | 92 | 1.09M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.09M | src_stride1, src_stride2, h); \ | 94 | 1.09M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.09M | src1 + 8 * sizeof(pixel), \ | 96 | 1.09M | src2 + 8 * sizeof(pixel), \ | 97 | 1.09M | dst_stride, src_stride1, \ | 98 | 1.09M | src_stride2, h); \ | 99 | 1.09M | } \ |
h264qpel.c:avg_pixels16_l2_12 Line | Count | Source | 90 | 107k | int h) \ | 91 | 107k | { \ | 92 | 107k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 107k | src_stride1, src_stride2, h); \ | 94 | 107k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 107k | src1 + 8 * sizeof(pixel), \ | 96 | 107k | src2 + 8 * sizeof(pixel), \ | 97 | 107k | dst_stride, src_stride1, \ | 98 | 107k | src_stride2, h); \ | 99 | 107k | } \ |
h264qpel.c:put_pixels16_l2_14 Line | Count | Source | 90 | 1.31M | int h) \ | 91 | 1.31M | { \ | 92 | 1.31M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.31M | src_stride1, src_stride2, h); \ | 94 | 1.31M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.31M | src1 + 8 * sizeof(pixel), \ | 96 | 1.31M | src2 + 8 * sizeof(pixel), \ | 97 | 1.31M | dst_stride, src_stride1, \ | 98 | 1.31M | src_stride2, h); \ | 99 | 1.31M | } \ |
h264qpel.c:avg_pixels16_l2_14 Line | Count | Source | 90 | 415k | int h) \ | 91 | 415k | { \ | 92 | 415k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 415k | src_stride1, src_stride2, h); \ | 94 | 415k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 415k | src1 + 8 * sizeof(pixel), \ | 96 | 415k | src2 + 8 * sizeof(pixel), \ | 97 | 415k | dst_stride, src_stride1, \ | 98 | 415k | src_stride2, h); \ | 99 | 415k | } \ |
|
100 | | |
101 | 123M | #define op_avg(a, b) a = rnd_avg_pixel4(a, b) |
102 | 2.20G | #define op_put(a, b) a = b |
103 | 123M | DEF_HPEL(avg, op_avg) |
104 | 2.20G | DEF_HPEL(put, op_put) |
105 | | #undef op_avg |
106 | | #undef op_put |