Mercurial > libavcodec.hg
annotate ppc/int_altivec.c @ 5445:5581a40c673a libavcodec
exchange the values of MV_DIR_FORWARD and MV_DIR_BACKWARD (this is more sane,
matches the order of some other stuff and allows some simplifications)
author | michael |
---|---|
date | Wed, 01 Aug 2007 22:12:52 +0000 |
parents | 669a97223dc7 |
children | f7cbb7733146 |
rev | line source |
---|---|
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
1 /* |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
2 * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org> |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
3 * |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
4 * This file is part of FFmpeg. |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
5 * |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
7 * modify it under the terms of the GNU Lesser General Public |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
8 * License as published by the Free Software Foundation; either |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
10 * |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
14 * Lesser General Public License for more details. |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
15 * |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
16 * You should have received a copy of the GNU Lesser General Public |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
17 * License along with FFmpeg; if not, write to the Free Software |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
19 */ |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
20 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
21 /** |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
22 ** @file int_altivec.c |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
23 ** integer misc ops. |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
24 **/ |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
25 |
5010
d5ba514e3f4a
Add libavcodec to compiler include flags in order to simplify header
diego
parents:
4838
diff
changeset
|
26 #include "dsputil.h" |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
27 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
28 #include "gcc_fixes.h" |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
29 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
30 #include "dsputil_altivec.h" |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
31 |
5255 | 32 static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2, |
33 int size) { | |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
34 int i, size16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
35 vector signed char vpix1; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
36 vector signed short vpix2, vdiff, vpix1l,vpix1h; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
37 union { vector signed int vscore; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
38 int32_t score[4]; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
39 } u; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
40 u.vscore = vec_splat_s32(0); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
41 // |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
42 //XXX lazy way, fix it later |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
43 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
44 #define vec_unaligned_load(b) \ |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
45 vec_perm(vec_ld(0,b),vec_ld(15,b),vec_lvsl(0, b)); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
46 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
47 size16 = size >> 4; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
48 while(size16) { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
49 // score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
50 //load pix1 and the first batch of pix2 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
51 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
52 vpix1 = vec_unaligned_load(pix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
53 vpix2 = vec_unaligned_load(pix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
54 pix2 += 8; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
55 //unpack |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
56 vpix1h = vec_unpackh(vpix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
57 vdiff = vec_sub(vpix1h, vpix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
58 vpix1l = vec_unpackl(vpix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
59 // load another batch from pix2 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
60 vpix2 = vec_unaligned_load(pix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
61 u.vscore = vec_msum(vdiff, vdiff, u.vscore); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
62 vdiff = vec_sub(vpix1l, vpix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
63 u.vscore = vec_msum(vdiff, vdiff, u.vscore); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
64 pix1 += 16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
65 pix2 += 8; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
66 size16--; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
67 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
68 u.vscore = vec_sums(u.vscore, vec_splat_s32(0)); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
69 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
70 size %= 16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
71 for (i = 0; i < size; i++) { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
72 u.score[3] += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
73 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
74 return u.score[3]; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
75 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
76 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
77 void int_init_altivec(DSPContext* c, AVCodecContext *avctx) |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
78 { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
79 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
80 } |