Libav 0.7.1
|
00001 /* 00002 * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org> 00003 * 00004 * This file is part of Libav. 00005 * 00006 * Libav is free software; you can redistribute it and/or 00007 * modify it under the terms of the GNU Lesser General Public 00008 * License as published by the Free Software Foundation; either 00009 * version 2.1 of the License, or (at your option) any later version. 00010 * 00011 * Libav is distributed in the hope that it will be useful, 00012 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00013 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00014 * Lesser General Public License for more details. 00015 * 00016 * You should have received a copy of the GNU Lesser General Public 00017 * License along with Libav; if not, write to the Free Software 00018 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00019 */ 00020 00026 #include "config.h" 00027 #if HAVE_ALTIVEC_H 00028 #include <altivec.h> 00029 #endif 00030 00031 #include "libavcodec/dsputil.h" 00032 00033 #include "dsputil_altivec.h" 00034 00035 #include "types_altivec.h" 00036 00037 static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2, 00038 int size) { 00039 int i, size16; 00040 vector signed char vpix1; 00041 vector signed short vpix2, vdiff, vpix1l,vpix1h; 00042 union { vector signed int vscore; 00043 int32_t score[4]; 00044 } u; 00045 u.vscore = vec_splat_s32(0); 00046 // 00047 //XXX lazy way, fix it later 00048 00049 #define vec_unaligned_load(b) \ 00050 vec_perm(vec_ld(0,b),vec_ld(15,b),vec_lvsl(0, b)); 00051 00052 size16 = size >> 4; 00053 while(size16) { 00054 // score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); 00055 //load pix1 and the first batch of pix2 00056 00057 vpix1 = vec_unaligned_load(pix1); 00058 vpix2 = vec_unaligned_load(pix2); 00059 pix2 += 8; 00060 //unpack 00061 vpix1h = vec_unpackh(vpix1); 00062 vdiff = vec_sub(vpix1h, vpix2); 00063 vpix1l = vec_unpackl(vpix1); 00064 // load another batch from pix2 00065 vpix2 = vec_unaligned_load(pix2); 00066 u.vscore = vec_msum(vdiff, vdiff, u.vscore); 00067 vdiff = vec_sub(vpix1l, vpix2); 00068 u.vscore = vec_msum(vdiff, vdiff, u.vscore); 00069 pix1 += 16; 00070 pix2 += 8; 00071 size16--; 00072 } 00073 u.vscore = vec_sums(u.vscore, vec_splat_s32(0)); 00074 00075 size %= 16; 00076 for (i = 0; i < size; i++) { 00077 u.score[3] += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); 00078 } 00079 return u.score[3]; 00080 } 00081 00082 static int32_t scalarproduct_int16_altivec(const int16_t * v1, const int16_t * v2, int order, const int shift) 00083 { 00084 int i; 00085 LOAD_ZERO; 00086 register vec_s16 vec1, *pv; 00087 register vec_s32 res = vec_splat_s32(0), t; 00088 register vec_u32 shifts; 00089 int32_t ires; 00090 00091 shifts = zero_u32v; 00092 if(shift & 0x10) shifts = vec_add(shifts, vec_sl(vec_splat_u32(0x08), vec_splat_u32(0x1))); 00093 if(shift & 0x08) shifts = vec_add(shifts, vec_splat_u32(0x08)); 00094 if(shift & 0x04) shifts = vec_add(shifts, vec_splat_u32(0x04)); 00095 if(shift & 0x02) shifts = vec_add(shifts, vec_splat_u32(0x02)); 00096 if(shift & 0x01) shifts = vec_add(shifts, vec_splat_u32(0x01)); 00097 00098 for(i = 0; i < order; i += 8){ 00099 pv = (vec_s16*)v1; 00100 vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1)); 00101 t = vec_msum(vec1, vec_ld(0, v2), zero_s32v); 00102 t = vec_sr(t, shifts); 00103 res = vec_sums(t, res); 00104 v1 += 8; 00105 v2 += 8; 00106 } 00107 res = vec_splat(res, 3); 00108 vec_ste(res, 0, &ires); 00109 return ires; 00110 } 00111 00112 static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul) 00113 { 00114 LOAD_ZERO; 00115 vec_s16 *pv1 = (vec_s16*)v1; 00116 vec_s16 *pv2 = (vec_s16*)v2; 00117 vec_s16 *pv3 = (vec_s16*)v3; 00118 register vec_s16 muls = {mul,mul,mul,mul,mul,mul,mul,mul}; 00119 register vec_s16 t0, t1, i0, i1; 00120 register vec_s16 i2 = pv2[0], i3 = pv3[0]; 00121 register vec_s32 res = zero_s32v; 00122 register vec_u8 align = vec_lvsl(0, v2); 00123 int32_t ires; 00124 order >>= 4; 00125 do { 00126 t0 = vec_perm(i2, pv2[1], align); 00127 i2 = pv2[2]; 00128 t1 = vec_perm(pv2[1], i2, align); 00129 i0 = pv1[0]; 00130 i1 = pv1[1]; 00131 res = vec_msum(t0, i0, res); 00132 res = vec_msum(t1, i1, res); 00133 t0 = vec_perm(i3, pv3[1], align); 00134 i3 = pv3[2]; 00135 t1 = vec_perm(pv3[1], i3, align); 00136 pv1[0] = vec_mladd(t0, muls, i0); 00137 pv1[1] = vec_mladd(t1, muls, i1); 00138 pv1 += 2; 00139 pv2 += 2; 00140 pv3 += 2; 00141 } while(--order); 00142 res = vec_splat(vec_sums(res, zero_s32v), 3); 00143 vec_ste(res, 0, &ires); 00144 return ires; 00145 } 00146 00147 void int_init_altivec(DSPContext* c, AVCodecContext *avctx) 00148 { 00149 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec; 00150 c->scalarproduct_int16 = scalarproduct_int16_altivec; 00151 c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_altivec; 00152 }