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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include "../dsputil.h"
23 #include "dsputil_altivec.h"
26 altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8,
27 to preserve proper dst alignement.
29 void gmc1_altivec(UINT8 *dst /* align 8 */, UINT8 *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
32 const int A=(16-x16)*(16-y16);
33 const int B=( x16)*(16-y16);
34 const int C=(16-x16)*( y16);
35 const int D=( x16)*( y16);
41 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
42 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
43 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;
44 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8;
45 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8;
46 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8;
47 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8;
48 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8;
53 const unsigned short __attribute__ ((aligned(16))) rounder_a[8] =
54 {rounder, rounder, rounder, rounder,
55 rounder, rounder, rounder, rounder};
56 const unsigned short __attribute__ ((aligned(16))) ABCD[8] =
58 (16-x16)*(16-y16), /* A */
59 ( x16)*(16-y16), /* B */
60 (16-x16)*( y16), /* C */
61 ( x16)*( y16), /* D */
62 0, 0, 0, 0 /* padding */
65 register const vector unsigned char vczero = (const vector unsigned char)(0);
66 register const vector unsigned short vcsr8 = (const vector unsigned short)(8);
67 register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
68 register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
70 unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
71 unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
73 tempA = vec_ld(0, (unsigned short*)ABCD);
74 Av = vec_splat(tempA, 0);
75 Bv = vec_splat(tempA, 1);
76 Cv = vec_splat(tempA, 2);
77 Dv = vec_splat(tempA, 3);
79 rounderV = vec_ld(0, (unsigned short*)rounder_a);
81 // we'll be able to pick-up our 9 char elements
82 // at src from those 32 bytes
83 // we load the first batch here, as inside the loop
84 // we can re-use 'src+stride' from one iteration
85 // as the 'src' of the next.
86 src_0 = vec_ld(0, src);
87 src_1 = vec_ld(16, src);
88 srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
90 if (src_really_odd != 0x0000000F)
91 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
92 srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
98 srcvA = vec_mergeh(vczero, srcvA);
99 srcvB = vec_mergeh(vczero, srcvB);
103 dst_odd = (unsigned long)dst & 0x0000000F;
104 src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
106 dstv = vec_ld(0, dst);
108 // we we'll be able to pick-up our 9 char elements
109 // at src + stride from those 32 bytes
110 // then reuse the resulting 2 vectors srvcC and srcvD
111 // as the next srcvA and srcvB
112 src_0 = vec_ld(stride + 0, src);
113 src_1 = vec_ld(stride + 16, src);
114 srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
116 if (src_really_odd != 0x0000000F)
117 { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
118 srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
125 srcvC = vec_mergeh(vczero, srcvC);
126 srcvD = vec_mergeh(vczero, srcvD);
129 // OK, now we (finally) do the math :-)
130 // those four instructions replaces 32 int muls & 32 int adds.
131 // isn't AltiVec nice ?
132 tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
133 tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
134 tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
135 tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
140 tempD = vec_sr(tempD, vcsr8);
142 dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
146 dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
150 dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
153 vec_st(dstv2, 0, dst);