/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 | 97.3M | int h) \ |
36 | 97.3M | { \ |
37 | 97.3M | 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 | 97.3M | } \ qpeldsp.c:put_pixels8_l2_8 Line | Count | Source | 35 | 17.5M | int h) \ | 36 | 17.5M | { \ | 37 | 17.5M | int i; \ | 38 | 213M | for (i = 0; i < h; i++) { \ | 39 | 196M | pixel4 a, b; \ | 40 | 196M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 196M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 196M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 196M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 196M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 196M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 196M | rnd_avg_pixel4(a, b)); \ | 47 | 196M | } \ | 48 | 17.5M | } \ |
qpeldsp.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 463k | int h) \ | 36 | 463k | { \ | 37 | 463k | int i; \ | 38 | 4.46M | for (i = 0; i < h; i++) { \ | 39 | 3.99M | pixel4 a, b; \ | 40 | 3.99M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 3.99M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 3.99M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 3.99M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 3.99M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 3.99M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 3.99M | rnd_avg_pixel4(a, b)); \ | 47 | 3.99M | } \ | 48 | 463k | } \ |
hpeldsp.c:put_pixels8_l2_8 Line | Count | Source | 35 | 47.5M | int h) \ | 36 | 47.5M | { \ | 37 | 47.5M | int i; \ | 38 | 582M | for (i = 0; i < h; i++) { \ | 39 | 534M | pixel4 a, b; \ | 40 | 534M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 534M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 534M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 534M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 534M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 534M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 534M | rnd_avg_pixel4(a, b)); \ | 47 | 534M | } \ | 48 | 47.5M | } \ |
hpeldsp.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 578k | int h) \ | 36 | 578k | { \ | 37 | 578k | int i; \ | 38 | 7.36M | for (i = 0; i < h; i++) { \ | 39 | 6.78M | pixel4 a, b; \ | 40 | 6.78M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 6.78M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 6.78M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 6.78M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 6.78M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 6.78M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 6.78M | rnd_avg_pixel4(a, b)); \ | 47 | 6.78M | } \ | 48 | 578k | } \ |
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 | 9.06M | int h) \ | 36 | 9.06M | { \ | 37 | 9.06M | int i; \ | 38 | 119M | for (i = 0; i < h; i++) { \ | 39 | 110M | pixel4 a, b; \ | 40 | 110M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 110M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 110M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 110M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 110M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 110M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 110M | rnd_avg_pixel4(a, b)); \ | 47 | 110M | } \ | 48 | 9.06M | } \ |
h264qpel.c:avg_pixels8_l2_8 Line | Count | Source | 35 | 949k | int h) \ | 36 | 949k | { \ | 37 | 949k | int i; \ | 38 | 11.5M | for (i = 0; i < h; i++) { \ | 39 | 10.5M | pixel4 a, b; \ | 40 | 10.5M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 10.5M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 10.5M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 10.5M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 10.5M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 10.5M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 10.5M | rnd_avg_pixel4(a, b)); \ | 47 | 10.5M | } \ | 48 | 949k | } \ |
h264qpel.c:put_pixels8_l2_9 Line | Count | Source | 35 | 4.22M | int h) \ | 36 | 4.22M | { \ | 37 | 4.22M | int i; \ | 38 | 58.5M | for (i = 0; i < h; i++) { \ | 39 | 54.3M | pixel4 a, b; \ | 40 | 54.3M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 54.3M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 54.3M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 54.3M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 54.3M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 54.3M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 54.3M | rnd_avg_pixel4(a, b)); \ | 47 | 54.3M | } \ | 48 | 4.22M | } \ |
h264qpel.c:avg_pixels8_l2_9 Line | Count | Source | 35 | 767k | int h) \ | 36 | 767k | { \ | 37 | 767k | int i; \ | 38 | 10.2M | for (i = 0; i < h; i++) { \ | 39 | 9.50M | pixel4 a, b; \ | 40 | 9.50M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 9.50M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 9.50M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 9.50M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 9.50M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 9.50M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 9.50M | rnd_avg_pixel4(a, b)); \ | 47 | 9.50M | } \ | 48 | 767k | } \ |
h264qpel.c:put_pixels8_l2_10 Line | Count | Source | 35 | 5.66M | int h) \ | 36 | 5.66M | { \ | 37 | 5.66M | int i; \ | 38 | 81.0M | for (i = 0; i < h; i++) { \ | 39 | 75.3M | pixel4 a, b; \ | 40 | 75.3M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 75.3M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 75.3M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 75.3M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 75.3M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 75.3M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 75.3M | rnd_avg_pixel4(a, b)); \ | 47 | 75.3M | } \ | 48 | 5.66M | } \ |
h264qpel.c:avg_pixels8_l2_10 Line | Count | Source | 35 | 476k | int h) \ | 36 | 476k | { \ | 37 | 476k | int i; \ | 38 | 6.39M | for (i = 0; i < h; i++) { \ | 39 | 5.91M | pixel4 a, b; \ | 40 | 5.91M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 5.91M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 5.91M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 5.91M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 5.91M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 5.91M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 5.91M | rnd_avg_pixel4(a, b)); \ | 47 | 5.91M | } \ | 48 | 476k | } \ |
h264qpel.c:put_pixels8_l2_12 Line | Count | Source | 35 | 4.40M | int h) \ | 36 | 4.40M | { \ | 37 | 4.40M | int i; \ | 38 | 57.0M | 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 | 4.40M | } \ |
h264qpel.c:avg_pixels8_l2_12 Line | Count | Source | 35 | 533k | int h) \ | 36 | 533k | { \ | 37 | 533k | int i; \ | 38 | 6.49M | for (i = 0; i < h; i++) { \ | 39 | 5.96M | pixel4 a, b; \ | 40 | 5.96M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 5.96M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 5.96M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 5.96M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 5.96M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 5.96M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 5.96M | rnd_avg_pixel4(a, b)); \ | 47 | 5.96M | } \ | 48 | 533k | } \ |
h264qpel.c:put_pixels8_l2_14 Line | Count | Source | 35 | 3.91M | int h) \ | 36 | 3.91M | { \ | 37 | 3.91M | int i; \ | 38 | 56.1M | for (i = 0; i < h; i++) { \ | 39 | 52.2M | pixel4 a, b; \ | 40 | 52.2M | a = AV_RN4P(&src1[i * src_stride1]); \ | 41 | 52.2M | b = AV_RN4P(&src2[i * src_stride2]); \ | 42 | 52.2M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 43 | 52.2M | a = AV_RN4P(&src1[i * src_stride1 + 4 * (int)sizeof(pixel)]); \ | 44 | 52.2M | b = AV_RN4P(&src2[i * src_stride2 + 4 * (int)sizeof(pixel)]); \ | 45 | 52.2M | OP(*((pixel4 *) &dst[i * dst_stride + 4 * (int)sizeof(pixel)]), \ | 46 | 52.2M | rnd_avg_pixel4(a, b)); \ | 47 | 52.2M | } \ | 48 | 3.91M | } \ |
h264qpel.c:avg_pixels8_l2_14 Line | Count | Source | 35 | 1.16M | int h) \ | 36 | 1.16M | { \ | 37 | 1.16M | int i; \ | 38 | 17.1M | 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 | 1.16M | } \ |
|
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 | 11.0M | int h) \ |
57 | 11.0M | { \ |
58 | 11.0M | int i; \ |
59 | 55.4M | for (i = 0; i < h; i++) { \ |
60 | 44.3M | pixel4 a, b; \ |
61 | 44.3M | a = AV_RN4P(&src1[i * src_stride1]); \ |
62 | 44.3M | b = AV_RN4P(&src2[i * src_stride2]); \ |
63 | 44.3M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ |
64 | 44.3M | } \ |
65 | 11.0M | } \ 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 | 29.8k | int h) \ | 57 | 29.8k | { \ | 58 | 29.8k | int i; \ | 59 | 219k | for (i = 0; i < h; i++) { \ | 60 | 189k | pixel4 a, b; \ | 61 | 189k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 189k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 189k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 189k | } \ | 65 | 29.8k | } \ |
hpeldsp.c:avg_pixels4_l2_8 Line | Count | Source | 56 | 16.7k | int h) \ | 57 | 16.7k | { \ | 58 | 16.7k | int i; \ | 59 | 138k | for (i = 0; i < h; i++) { \ | 60 | 121k | pixel4 a, b; \ | 61 | 121k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 121k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 121k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 121k | } \ | 65 | 16.7k | } \ |
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 | 3.22M | int h) \ | 57 | 3.22M | { \ | 58 | 3.22M | int i; \ | 59 | 16.1M | for (i = 0; i < h; i++) { \ | 60 | 12.9M | pixel4 a, b; \ | 61 | 12.9M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 12.9M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 12.9M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 12.9M | } \ | 65 | 3.22M | } \ |
h264qpel.c:avg_pixels4_l2_8 Line | Count | Source | 56 | 197k | int h) \ | 57 | 197k | { \ | 58 | 197k | int i; \ | 59 | 989k | for (i = 0; i < h; i++) { \ | 60 | 791k | pixel4 a, b; \ | 61 | 791k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 791k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 791k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 791k | } \ | 65 | 197k | } \ |
h264qpel.c:put_pixels4_l2_9 Line | Count | Source | 56 | 1.48M | int h) \ | 57 | 1.48M | { \ | 58 | 1.48M | int i; \ | 59 | 7.41M | for (i = 0; i < h; i++) { \ | 60 | 5.93M | pixel4 a, b; \ | 61 | 5.93M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 5.93M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 5.93M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 5.93M | } \ | 65 | 1.48M | } \ |
h264qpel.c:avg_pixels4_l2_9 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_10 Line | Count | Source | 56 | 2.31M | int h) \ | 57 | 2.31M | { \ | 58 | 2.31M | int i; \ | 59 | 11.5M | for (i = 0; i < h; i++) { \ | 60 | 9.24M | pixel4 a, b; \ | 61 | 9.24M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 9.24M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 9.24M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 9.24M | } \ | 65 | 2.31M | } \ |
h264qpel.c:avg_pixels4_l2_10 Line | Count | Source | 56 | 151k | int h) \ | 57 | 151k | { \ | 58 | 151k | int i; \ | 59 | 758k | for (i = 0; i < h; i++) { \ | 60 | 606k | pixel4 a, b; \ | 61 | 606k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 606k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 606k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 606k | } \ | 65 | 151k | } \ |
h264qpel.c:put_pixels4_l2_12 Line | Count | Source | 56 | 2.15M | int h) \ | 57 | 2.15M | { \ | 58 | 2.15M | int i; \ | 59 | 10.7M | for (i = 0; i < h; i++) { \ | 60 | 8.60M | pixel4 a, b; \ | 61 | 8.60M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 8.60M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 8.60M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 8.60M | } \ | 65 | 2.15M | } \ |
h264qpel.c:avg_pixels4_l2_12 Line | Count | Source | 56 | 116k | int h) \ | 57 | 116k | { \ | 58 | 116k | int i; \ | 59 | 582k | for (i = 0; i < h; i++) { \ | 60 | 465k | pixel4 a, b; \ | 61 | 465k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 465k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 465k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 465k | } \ | 65 | 116k | } \ |
h264qpel.c:put_pixels4_l2_14 Line | Count | Source | 56 | 979k | int h) \ | 57 | 979k | { \ | 58 | 979k | int i; \ | 59 | 4.89M | for (i = 0; i < h; i++) { \ | 60 | 3.91M | pixel4 a, b; \ | 61 | 3.91M | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 3.91M | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 3.91M | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 3.91M | } \ | 65 | 979k | } \ |
h264qpel.c:avg_pixels4_l2_14 Line | Count | Source | 56 | 246k | int h) \ | 57 | 246k | { \ | 58 | 246k | int i; \ | 59 | 1.23M | for (i = 0; i < h; i++) { \ | 60 | 986k | pixel4 a, b; \ | 61 | 986k | a = AV_RN4P(&src1[i * src_stride1]); \ | 62 | 986k | b = AV_RN4P(&src2[i * src_stride2]); \ | 63 | 986k | OP(*((pixel4 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 64 | 986k | } \ | 65 | 246k | } \ |
|
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 | 851k | int h) \ |
74 | 851k | { \ |
75 | 851k | int i; \ |
76 | 2.65M | for (i = 0; i < h; i++) { \ |
77 | 1.80M | pixel4 a, b; \ |
78 | 1.80M | a = AV_RN2P(&src1[i * src_stride1]); \ |
79 | 1.80M | b = AV_RN2P(&src2[i * src_stride2]); \ |
80 | 1.80M | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ |
81 | 1.80M | } \ |
82 | 851k | } \ 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 | 28.5k | int h) \ | 74 | 28.5k | { \ | 75 | 28.5k | int i; \ | 76 | 135k | 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 | 28.5k | } \ |
hpeldsp.c:avg_pixels2_l2_8 Line | Count | Source | 73 | 27.5k | int h) \ | 74 | 27.5k | { \ | 75 | 27.5k | int i; \ | 76 | 133k | for (i = 0; i < h; i++) { \ | 77 | 106k | pixel4 a, b; \ | 78 | 106k | a = AV_RN2P(&src1[i * src_stride1]); \ | 79 | 106k | b = AV_RN2P(&src2[i * src_stride2]); \ | 80 | 106k | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 81 | 106k | } \ | 82 | 27.5k | } \ |
Line | Count | Source | 73 | 795k | int h) \ | 74 | 795k | { \ | 75 | 795k | int i; \ | 76 | 2.38M | for (i = 0; i < h; i++) { \ | 77 | 1.59M | pixel4 a, b; \ | 78 | 1.59M | a = AV_RN2P(&src1[i * src_stride1]); \ | 79 | 1.59M | b = AV_RN2P(&src2[i * src_stride2]); \ | 80 | 1.59M | OP(*((pixel2 *) &dst[i * dst_stride]), rnd_avg_pixel4(a, b)); \ | 81 | 1.59M | } \ | 82 | 795k | } \ |
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 | 12.6M | int h) \ |
91 | 12.6M | { \ |
92 | 12.6M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ |
93 | 12.6M | src_stride1, src_stride2, h); \ |
94 | 12.6M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ |
95 | 12.6M | src1 + 8 * sizeof(pixel), \ |
96 | 12.6M | src2 + 8 * sizeof(pixel), \ |
97 | 12.6M | dst_stride, src_stride1, \ |
98 | 12.6M | src_stride2, h); \ |
99 | 12.6M | } \ qpeldsp.c:put_pixels16_l2_8 Line | Count | Source | 90 | 3.65M | int h) \ | 91 | 3.65M | { \ | 92 | 3.65M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 3.65M | src_stride1, src_stride2, h); \ | 94 | 3.65M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 3.65M | src1 + 8 * sizeof(pixel), \ | 96 | 3.65M | src2 + 8 * sizeof(pixel), \ | 97 | 3.65M | dst_stride, src_stride1, \ | 98 | 3.65M | src_stride2, h); \ | 99 | 3.65M | } \ |
qpeldsp.c:avg_pixels16_l2_8 Line | Count | Source | 90 | 63.8k | int h) \ | 91 | 63.8k | { \ | 92 | 63.8k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 63.8k | src_stride1, src_stride2, h); \ | 94 | 63.8k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 63.8k | src1 + 8 * sizeof(pixel), \ | 96 | 63.8k | src2 + 8 * sizeof(pixel), \ | 97 | 63.8k | dst_stride, src_stride1, \ | 98 | 63.8k | src_stride2, h); \ | 99 | 63.8k | } \ |
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 | 2.34M | int h) \ | 91 | 2.34M | { \ | 92 | 2.34M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 2.34M | src_stride1, src_stride2, h); \ | 94 | 2.34M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 2.34M | src1 + 8 * sizeof(pixel), \ | 96 | 2.34M | src2 + 8 * sizeof(pixel), \ | 97 | 2.34M | dst_stride, src_stride1, \ | 98 | 2.34M | src_stride2, h); \ | 99 | 2.34M | } \ |
h264qpel.c:avg_pixels16_l2_8 Line | Count | Source | 90 | 185k | int h) \ | 91 | 185k | { \ | 92 | 185k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 185k | src_stride1, src_stride2, h); \ | 94 | 185k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 185k | src1 + 8 * sizeof(pixel), \ | 96 | 185k | src2 + 8 * sizeof(pixel), \ | 97 | 185k | dst_stride, src_stride1, \ | 98 | 185k | src_stride2, h); \ | 99 | 185k | } \ |
h264qpel.c:put_pixels16_l2_9 Line | Count | Source | 90 | 1.28M | int h) \ | 91 | 1.28M | { \ | 92 | 1.28M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.28M | src_stride1, src_stride2, h); \ | 94 | 1.28M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.28M | src1 + 8 * sizeof(pixel), \ | 96 | 1.28M | src2 + 8 * sizeof(pixel), \ | 97 | 1.28M | dst_stride, src_stride1, \ | 98 | 1.28M | src_stride2, h); \ | 99 | 1.28M | } \ |
h264qpel.c:avg_pixels16_l2_9 Line | Count | Source | 90 | 210k | int h) \ | 91 | 210k | { \ | 92 | 210k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 210k | src_stride1, src_stride2, h); \ | 94 | 210k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 210k | src1 + 8 * sizeof(pixel), \ | 96 | 210k | src2 + 8 * sizeof(pixel), \ | 97 | 210k | dst_stride, src_stride1, \ | 98 | 210k | src_stride2, h); \ | 99 | 210k | } \ |
h264qpel.c:put_pixels16_l2_10 Line | Count | Source | 90 | 1.87M | int h) \ | 91 | 1.87M | { \ | 92 | 1.87M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.87M | src_stride1, src_stride2, h); \ | 94 | 1.87M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.87M | src1 + 8 * sizeof(pixel), \ | 96 | 1.87M | src2 + 8 * sizeof(pixel), \ | 97 | 1.87M | dst_stride, src_stride1, \ | 98 | 1.87M | src_stride2, h); \ | 99 | 1.87M | } \ |
h264qpel.c:avg_pixels16_l2_10 Line | Count | Source | 90 | 131k | int h) \ | 91 | 131k | { \ | 92 | 131k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 131k | src_stride1, src_stride2, h); \ | 94 | 131k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 131k | src1 + 8 * sizeof(pixel), \ | 96 | 131k | src2 + 8 * sizeof(pixel), \ | 97 | 131k | dst_stride, src_stride1, \ | 98 | 131k | src_stride2, h); \ | 99 | 131k | } \ |
h264qpel.c:put_pixels16_l2_12 Line | Count | Source | 90 | 1.08M | int h) \ | 91 | 1.08M | { \ | 92 | 1.08M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.08M | src_stride1, src_stride2, h); \ | 94 | 1.08M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.08M | src1 + 8 * sizeof(pixel), \ | 96 | 1.08M | src2 + 8 * sizeof(pixel), \ | 97 | 1.08M | dst_stride, src_stride1, \ | 98 | 1.08M | src_stride2, h); \ | 99 | 1.08M | } \ |
h264qpel.c:avg_pixels16_l2_12 Line | Count | Source | 90 | 105k | int h) \ | 91 | 105k | { \ | 92 | 105k | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 105k | src_stride1, src_stride2, h); \ | 94 | 105k | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 105k | src1 + 8 * sizeof(pixel), \ | 96 | 105k | src2 + 8 * sizeof(pixel), \ | 97 | 105k | dst_stride, src_stride1, \ | 98 | 105k | src_stride2, h); \ | 99 | 105k | } \ |
h264qpel.c:put_pixels16_l2_14 Line | Count | Source | 90 | 1.30M | int h) \ | 91 | 1.30M | { \ | 92 | 1.30M | FUNC(OPNAME ## _pixels8_l2)(dst, src1, src2, dst_stride, \ | 93 | 1.30M | src_stride1, src_stride2, h); \ | 94 | 1.30M | FUNC(OPNAME ## _pixels8_l2)(dst + 8 * sizeof(pixel), \ | 95 | 1.30M | src1 + 8 * sizeof(pixel), \ | 96 | 1.30M | src2 + 8 * sizeof(pixel), \ | 97 | 1.30M | dst_stride, src_stride1, \ | 98 | 1.30M | src_stride2, h); \ | 99 | 1.30M | } \ |
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 | 121M | #define op_avg(a, b) a = rnd_avg_pixel4(a, b) |
102 | 2.19G | #define op_put(a, b) a = b |
103 | 121M | DEF_HPEL(avg, op_avg) |
104 | 2.19G | DEF_HPEL(put, op_put) |
105 | | #undef op_avg |
106 | | #undef op_put |