undefined local_port fix by (Giancarlo Formicuccia <ilsensine at inwind dot it>)
[libav.git] / libavcodec / ppc / gmc_altivec.c
CommitLineData
e629ab68 1/*
744ac4be 2 * GMC (Global Motion Compensation)
e629ab68
RD
3 * AltiVec-enabled
4 * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
20
21#include "../dsputil.h"
22
a9a07762
MN
23#include "gcc_fixes.h"
24
e629ab68
RD
25#include "dsputil_altivec.h"
26
27/*
28 altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8,
29 to preserve proper dst alignement.
30*/
0c1a9eda 31void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
e629ab68 32{
35e5fb06 33POWERPC_TBL_DECLARE(altivec_gmc1_num, h == 8);
db40a39a 34#ifdef ALTIVEC_USE_REFERENCE_C_CODE
e629ab68
RD
35 const int A=(16-x16)*(16-y16);
36 const int B=( x16)*(16-y16);
37 const int C=(16-x16)*( y16);
38 const int D=( x16)*( y16);
e629ab68 39 int i;
db40a39a 40
35e5fb06 41POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
e629ab68
RD
42
43 for(i=0; i<h; i++)
44 {
45 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
46 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
47 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;
48 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8;
49 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8;
50 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8;
51 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8;
52 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8;
53 dst+= stride;
54 src+= stride;
55 }
db40a39a 56
35e5fb06 57POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
db40a39a
MN
58
59#else /* ALTIVEC_USE_REFERENCE_C_CODE */
e629ab68
RD
60 const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
61 {rounder, rounder, rounder, rounder,
62 rounder, rounder, rounder, rounder};
63 const unsigned short __attribute__ ((aligned(16))) ABCD[8] =
64 {
65 (16-x16)*(16-y16), /* A */
66 ( x16)*(16-y16), /* B */
67 (16-x16)*( y16), /* C */
68 ( x16)*( y16), /* D */
69 0, 0, 0, 0 /* padding */
70 };
3b991c54
RD
71 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
72 register const vector unsigned short vcsr8 = (const vector unsigned short)vec_splat_u16(8);
e629ab68
RD
73 register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
74 register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
75 int i;
76 unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
77 unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
78
db40a39a 79
35e5fb06 80POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
db40a39a 81
e629ab68
RD
82 tempA = vec_ld(0, (unsigned short*)ABCD);
83 Av = vec_splat(tempA, 0);
84 Bv = vec_splat(tempA, 1);
85 Cv = vec_splat(tempA, 2);
86 Dv = vec_splat(tempA, 3);
87
88 rounderV = vec_ld(0, (unsigned short*)rounder_a);
89
90 // we'll be able to pick-up our 9 char elements
91 // at src from those 32 bytes
92 // we load the first batch here, as inside the loop
93 // we can re-use 'src+stride' from one iteration
94 // as the 'src' of the next.
95 src_0 = vec_ld(0, src);
96 src_1 = vec_ld(16, src);
97 srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
98
99 if (src_really_odd != 0x0000000F)
100 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
101 srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
102 }
103 else
104 {
105 srcvB = src_1;
106 }
107 srcvA = vec_mergeh(vczero, srcvA);
108 srcvB = vec_mergeh(vczero, srcvB);
109
110 for(i=0; i<h; i++)
111 {
112 dst_odd = (unsigned long)dst & 0x0000000F;
113 src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
114
115 dstv = vec_ld(0, dst);
116
117 // we we'll be able to pick-up our 9 char elements
118 // at src + stride from those 32 bytes
119 // then reuse the resulting 2 vectors srvcC and srcvD
120 // as the next srcvA and srcvB
121 src_0 = vec_ld(stride + 0, src);
122 src_1 = vec_ld(stride + 16, src);
123 srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
124
125 if (src_really_odd != 0x0000000F)
126 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
127 srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
128 }
129 else
130 {
131 srcvD = src_1;
132 }
133
134 srcvC = vec_mergeh(vczero, srcvC);
135 srcvD = vec_mergeh(vczero, srcvD);
136
137
138 // OK, now we (finally) do the math :-)
139 // those four instructions replaces 32 int muls & 32 int adds.
140 // isn't AltiVec nice ?
141 tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
142 tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
143 tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
144 tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
145
146 srcvA = srcvC;
147 srcvB = srcvD;
148
149 tempD = vec_sr(tempD, vcsr8);
150
151 dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
152
153 if (dst_odd)
154 {
155 dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
156 }
157 else
158 {
159 dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
160 }
161
162 vec_st(dstv2, 0, dst);
163
164 dst += stride;
165 src += stride;
166 }
db40a39a 167
35e5fb06 168POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
db40a39a
MN
169
170#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
e629ab68 171}