2
* GMC (Global Motion Compensation)
4
* Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
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.
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.
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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
#include "../dsputil.h"
23
#include "gcc_fixes.h"
25
#include "dsputil_altivec.h"
28
altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8,
29
to preserve proper dst alignement.
31
#define GMC1_PERF_COND (h==8)
32
void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
34
POWERPC_PERF_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
35
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
36
const int A=(16-x16)*(16-y16);
37
const int B=( x16)*(16-y16);
38
const int C=(16-x16)*( y16);
39
const int D=( x16)*( y16);
42
POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
46
dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
47
dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
48
dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;
49
dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8;
50
dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8;
51
dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8;
52
dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8;
53
dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8;
58
POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
60
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
61
const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
62
{rounder, rounder, rounder, rounder,
63
rounder, rounder, rounder, rounder};
64
const unsigned short __attribute__ ((aligned(16))) ABCD[8] =
66
(16-x16)*(16-y16), /* A */
67
( x16)*(16-y16), /* B */
68
(16-x16)*( y16), /* C */
69
( x16)*( y16), /* D */
70
0, 0, 0, 0 /* padding */
72
register const_vector unsigned char vczero = (const_vector unsigned char)vec_splat_u8(0);
73
register const_vector unsigned short vcsr8 = (const_vector unsigned short)vec_splat_u16(8);
74
register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
75
register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
77
unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
78
unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
81
POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
83
tempA = vec_ld(0, (unsigned short*)ABCD);
84
Av = vec_splat(tempA, 0);
85
Bv = vec_splat(tempA, 1);
86
Cv = vec_splat(tempA, 2);
87
Dv = vec_splat(tempA, 3);
89
rounderV = vec_ld(0, (unsigned short*)rounder_a);
91
// we'll be able to pick-up our 9 char elements
92
// at src from those 32 bytes
93
// we load the first batch here, as inside the loop
94
// we can re-use 'src+stride' from one iteration
95
// as the 'src' of the next.
96
src_0 = vec_ld(0, src);
97
src_1 = vec_ld(16, src);
98
srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
100
if (src_really_odd != 0x0000000F)
101
{ // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
102
srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
108
srcvA = vec_mergeh(vczero, srcvA);
109
srcvB = vec_mergeh(vczero, srcvB);
113
dst_odd = (unsigned long)dst & 0x0000000F;
114
src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
116
dstv = vec_ld(0, dst);
118
// we we'll be able to pick-up our 9 char elements
119
// at src + stride from those 32 bytes
120
// then reuse the resulting 2 vectors srvcC and srcvD
121
// as the next srcvA and srcvB
122
src_0 = vec_ld(stride + 0, src);
123
src_1 = vec_ld(stride + 16, src);
124
srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
126
if (src_really_odd != 0x0000000F)
127
{ // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
128
srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
135
srcvC = vec_mergeh(vczero, srcvC);
136
srcvD = vec_mergeh(vczero, srcvD);
139
// OK, now we (finally) do the math :-)
140
// those four instructions replaces 32 int muls & 32 int adds.
141
// isn't AltiVec nice ?
142
tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
143
tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
144
tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
145
tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
150
tempD = vec_sr(tempD, vcsr8);
152
dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
156
dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
160
dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
163
vec_st(dstv2, 0, dst);
169
POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
171
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */