Mercurial > libavcodec.hg
annotate ppc/int_altivec.c @ 10366:55f4c141dbb0 libavcodec
cosmetics: reindent and line wrap after last commit
author | jbr |
---|---|
date | Mon, 05 Oct 2009 21:33:50 +0000 |
parents | 9f4b529bd5c0 |
children | 5da7180afadf |
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 /** |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8494
diff
changeset
|
22 ** @file libavcodec/ppc/int_altivec.c |
4838
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 |
9364 | 26 #include "config.h" |
27 #if HAVE_ALTIVEC_H | |
28 #include <altivec.h> | |
29 #endif | |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
30 |
9364 | 31 #include "libavcodec/dsputil.h" |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
32 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
33 #include "dsputil_altivec.h" |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
34 |
7204 | 35 #include "types_altivec.h" |
36 | |
5255 | 37 static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2, |
38 int size) { | |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
39 int i, size16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
40 vector signed char vpix1; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
41 vector signed short vpix2, vdiff, vpix1l,vpix1h; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
42 union { vector signed int vscore; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
43 int32_t score[4]; |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
7204
diff
changeset
|
44 } u; |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
45 u.vscore = vec_splat_s32(0); |
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 //XXX lazy way, fix it later |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
48 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
49 #define vec_unaligned_load(b) \ |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
50 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
|
51 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
52 size16 = size >> 4; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
53 while(size16) { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
54 // score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
55 //load pix1 and the first batch of pix2 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
56 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
57 vpix1 = vec_unaligned_load(pix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
58 vpix2 = vec_unaligned_load(pix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
59 pix2 += 8; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
60 //unpack |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
61 vpix1h = vec_unpackh(vpix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
62 vdiff = vec_sub(vpix1h, vpix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
63 vpix1l = vec_unpackl(vpix1); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
64 // load another batch from pix2 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
65 vpix2 = vec_unaligned_load(pix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
66 u.vscore = vec_msum(vdiff, vdiff, u.vscore); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
67 vdiff = vec_sub(vpix1l, vpix2); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
68 u.vscore = vec_msum(vdiff, vdiff, u.vscore); |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
69 pix1 += 16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
70 pix2 += 8; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
71 size16--; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
72 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
73 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
|
74 |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
75 size %= 16; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
76 for (i = 0; i < size; i++) { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
77 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
|
78 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
79 return u.score[3]; |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
80 } |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
81 |
7204 | 82 static void add_int16_altivec(int16_t * v1, int16_t * v2, int order) |
83 { | |
84 int i; | |
8494 | 85 register vec_s16 vec, *pv; |
7204 | 86 |
87 for(i = 0; i < order; i += 8){ | |
8494 | 88 pv = (vec_s16*)v2; |
7204 | 89 vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2)); |
90 vec_st(vec_add(vec_ld(0, v1), vec), 0, v1); | |
91 v1 += 8; | |
92 v2 += 8; | |
93 } | |
94 } | |
95 | |
96 static void sub_int16_altivec(int16_t * v1, int16_t * v2, int order) | |
97 { | |
98 int i; | |
8494 | 99 register vec_s16 vec, *pv; |
7204 | 100 |
101 for(i = 0; i < order; i += 8){ | |
8494 | 102 pv = (vec_s16*)v2; |
7204 | 103 vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2)); |
104 vec_st(vec_sub(vec_ld(0, v1), vec), 0, v1); | |
105 v1 += 8; | |
106 v2 += 8; | |
107 } | |
108 } | |
109 | |
110 static int32_t scalarproduct_int16_altivec(int16_t * v1, int16_t * v2, int order, const int shift) | |
111 { | |
112 int i; | |
113 LOAD_ZERO; | |
8494 | 114 register vec_s16 vec1, *pv; |
115 register vec_s32 res = vec_splat_s32(0), t; | |
116 register vec_u32 shifts; | |
10082 | 117 int32_t ires; |
7204 | 118 |
119 shifts = zero_u32v; | |
120 if(shift & 0x10) shifts = vec_add(shifts, vec_sl(vec_splat_u32(0x08), vec_splat_u32(0x1))); | |
121 if(shift & 0x08) shifts = vec_add(shifts, vec_splat_u32(0x08)); | |
122 if(shift & 0x04) shifts = vec_add(shifts, vec_splat_u32(0x04)); | |
123 if(shift & 0x02) shifts = vec_add(shifts, vec_splat_u32(0x02)); | |
124 if(shift & 0x01) shifts = vec_add(shifts, vec_splat_u32(0x01)); | |
125 | |
126 for(i = 0; i < order; i += 8){ | |
8494 | 127 pv = (vec_s16*)v1; |
7204 | 128 vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1)); |
129 t = vec_msum(vec1, vec_ld(0, v2), zero_s32v); | |
130 t = vec_sr(t, shifts); | |
131 res = vec_sums(t, res); | |
132 v1 += 8; | |
133 v2 += 8; | |
134 } | |
135 res = vec_splat(res, 3); | |
136 vec_ste(res, 0, &ires); | |
137 return ires; | |
138 } | |
139 | |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
140 void int_init_altivec(DSPContext* c, AVCodecContext *avctx) |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
141 { |
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
142 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec; |
7204 | 143 c->add_int16 = add_int16_altivec; |
144 c->sub_int16 = sub_int16_altivec; | |
145 c->scalarproduct_int16 = scalarproduct_int16_altivec; | |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
diff
changeset
|
146 } |