Altivec on non darwin systems patch by Romain Dolbeau
[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
23#include "dsputil_altivec.h"
24
25/*
26 altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8,
27 to preserve proper dst alignement.
28*/
29void gmc1_altivec(UINT8 *dst /* align 8 */, UINT8 *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
30{
35e5fb06 31POWERPC_TBL_DECLARE(altivec_gmc1_num, h == 8);
db40a39a 32#ifdef ALTIVEC_USE_REFERENCE_C_CODE
e629ab68
RD
33 const int A=(16-x16)*(16-y16);
34 const int B=( x16)*(16-y16);
35 const int C=(16-x16)*( y16);
36 const int D=( x16)*( y16);
e629ab68 37 int i;
db40a39a 38
35e5fb06 39POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
e629ab68
RD
40
41 for(i=0; i<h; i++)
42 {
43 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
44 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
45 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;
46 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8;
47 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8;
48 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8;
49 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8;
50 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8;
51 dst+= stride;
52 src+= stride;
53 }
db40a39a 54
35e5fb06 55POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
db40a39a
MN
56
57#else /* ALTIVEC_USE_REFERENCE_C_CODE */
e629ab68
RD
58 const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
59 {rounder, rounder, rounder, rounder,
60 rounder, rounder, rounder, rounder};
61 const unsigned short __attribute__ ((aligned(16))) ABCD[8] =
62 {
63 (16-x16)*(16-y16), /* A */
64 ( x16)*(16-y16), /* B */
65 (16-x16)*( y16), /* C */
66 ( x16)*( y16), /* D */
67 0, 0, 0, 0 /* padding */
68 };
3b991c54
RD
69 register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
70 register const vector unsigned short vcsr8 = (const vector unsigned short)vec_splat_u16(8);
e629ab68
RD
71 register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
72 register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
73 int i;
74 unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
75 unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
76
db40a39a 77
35e5fb06 78POWERPC_TBL_START_COUNT(altivec_gmc1_num, h == 8);
db40a39a 79
e629ab68
RD
80 tempA = vec_ld(0, (unsigned short*)ABCD);
81 Av = vec_splat(tempA, 0);
82 Bv = vec_splat(tempA, 1);
83 Cv = vec_splat(tempA, 2);
84 Dv = vec_splat(tempA, 3);
85
86 rounderV = vec_ld(0, (unsigned short*)rounder_a);
87
88 // we'll be able to pick-up our 9 char elements
89 // at src from those 32 bytes
90 // we load the first batch here, as inside the loop
91 // we can re-use 'src+stride' from one iteration
92 // as the 'src' of the next.
93 src_0 = vec_ld(0, src);
94 src_1 = vec_ld(16, src);
95 srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
96
97 if (src_really_odd != 0x0000000F)
98 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
99 srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
100 }
101 else
102 {
103 srcvB = src_1;
104 }
105 srcvA = vec_mergeh(vczero, srcvA);
106 srcvB = vec_mergeh(vczero, srcvB);
107
108 for(i=0; i<h; i++)
109 {
110 dst_odd = (unsigned long)dst & 0x0000000F;
111 src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
112
113 dstv = vec_ld(0, dst);
114
115 // we we'll be able to pick-up our 9 char elements
116 // at src + stride from those 32 bytes
117 // then reuse the resulting 2 vectors srvcC and srcvD
118 // as the next srcvA and srcvB
119 src_0 = vec_ld(stride + 0, src);
120 src_1 = vec_ld(stride + 16, src);
121 srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
122
123 if (src_really_odd != 0x0000000F)
124 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
125 srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
126 }
127 else
128 {
129 srcvD = src_1;
130 }
131
132 srcvC = vec_mergeh(vczero, srcvC);
133 srcvD = vec_mergeh(vczero, srcvD);
134
135
136 // OK, now we (finally) do the math :-)
137 // those four instructions replaces 32 int muls & 32 int adds.
138 // isn't AltiVec nice ?
139 tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
140 tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
141 tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
142 tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
143
144 srcvA = srcvC;
145 srcvB = srcvD;
146
147 tempD = vec_sr(tempD, vcsr8);
148
149 dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
150
151 if (dst_odd)
152 {
153 dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
154 }
155 else
156 {
157 dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
158 }
159
160 vec_st(dstv2, 0, dst);
161
162 dst += stride;
163 src += stride;
164 }
db40a39a 165
35e5fb06 166POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
db40a39a
MN
167
168#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
e629ab68 169}