From e629ab68a7ed341ab7f92b7d18c73509429453f3 Mon Sep 17 00:00:00 2001 From: Romain Dolbeau Date: Sat, 11 Jan 2003 20:51:03 +0000 Subject: altivec jumbo patch by (Romain Dolbeau ) Originally committed as revision 1448 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/ppc/gmc_altivec.c | 159 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 159 insertions(+) create mode 100644 libavcodec/ppc/gmc_altivec.c (limited to 'libavcodec/ppc/gmc_altivec.c') diff --git a/libavcodec/ppc/gmc_altivec.c b/libavcodec/ppc/gmc_altivec.c new file mode 100644 index 0000000000..aef1f54426 --- /dev/null +++ b/libavcodec/ppc/gmc_altivec.c @@ -0,0 +1,159 @@ +/* + * GMC (???) + * AltiVec-enabled + * Copyright (c) 2003 Romain Dolbeau + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#include "../dsputil.h" + +#include "dsputil_altivec.h" + +/* + altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8, + to preserve proper dst alignement. +*/ +void gmc1_altivec(UINT8 *dst /* align 8 */, UINT8 *src /* align1 */, int stride, int h, int x16, int y16, int rounder) +{ +#if 0 + const int A=(16-x16)*(16-y16); + const int B=( x16)*(16-y16); + const int C=(16-x16)*( y16); + const int D=( x16)*( y16); + + int i; + + for(i=0; i>8; + dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; + dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; + dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; + dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; + dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; + dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; + dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; + dst+= stride; + src+= stride; + } +#else + const unsigned short __attribute__ ((aligned(16))) rounder_a[8] = + {rounder, rounder, rounder, rounder, + rounder, rounder, rounder, rounder}; + const unsigned short __attribute__ ((aligned(16))) ABCD[8] = + { + (16-x16)*(16-y16), /* A */ + ( x16)*(16-y16), /* B */ + (16-x16)*( y16), /* C */ + ( x16)*( y16), /* D */ + 0, 0, 0, 0 /* padding */ + }; + + register const vector unsigned char vczero = (const vector unsigned char)(0); + register const vector unsigned short vcsr8 = (const vector unsigned short)(8); + register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD; + register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD; + int i; + unsigned long dst_odd = (unsigned long)dst & 0x0000000F; + unsigned long src_really_odd = (unsigned long)src & 0x0000000F; + + tempA = vec_ld(0, (unsigned short*)ABCD); + Av = vec_splat(tempA, 0); + Bv = vec_splat(tempA, 1); + Cv = vec_splat(tempA, 2); + Dv = vec_splat(tempA, 3); + + rounderV = vec_ld(0, (unsigned short*)rounder_a); + + // we'll be able to pick-up our 9 char elements + // at src from those 32 bytes + // we load the first batch here, as inside the loop + // we can re-use 'src+stride' from one iteration + // as the 'src' of the next. + src_0 = vec_ld(0, src); + src_1 = vec_ld(16, src); + srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src)); + + if (src_really_odd != 0x0000000F) + { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector. + srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src)); + } + else + { + srcvB = src_1; + } + srcvA = vec_mergeh(vczero, srcvA); + srcvB = vec_mergeh(vczero, srcvB); + + for(i=0; i