Libav 0.7.1
|
00001 /* 00002 * Copyright (C) 2009 David Conrad 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 00021 #include "libavcodec/dsputil.h" 00022 #include "util_altivec.h" 00023 #include "types_altivec.h" 00024 #include "dsputil_altivec.h" 00025 00026 static const vec_s16 constants = 00027 {0, 64277, 60547, 54491, 46341, 36410, 25080, 12785}; 00028 static const vec_u8 interleave_high = 00029 {0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29}; 00030 00031 #define IDCT_START \ 00032 vec_s16 A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;\ 00033 vec_s16 Ed, Gd, Add, Bdd, Fd, Hd;\ 00034 vec_s16 eight = vec_splat_s16(8);\ 00035 vec_u16 four = vec_splat_u16(4);\ 00036 \ 00037 vec_s16 C1 = vec_splat(constants, 1);\ 00038 vec_s16 C2 = vec_splat(constants, 2);\ 00039 vec_s16 C3 = vec_splat(constants, 3);\ 00040 vec_s16 C4 = vec_splat(constants, 4);\ 00041 vec_s16 C5 = vec_splat(constants, 5);\ 00042 vec_s16 C6 = vec_splat(constants, 6);\ 00043 vec_s16 C7 = vec_splat(constants, 7);\ 00044 \ 00045 vec_s16 b0 = vec_ld(0x00, block);\ 00046 vec_s16 b1 = vec_ld(0x10, block);\ 00047 vec_s16 b2 = vec_ld(0x20, block);\ 00048 vec_s16 b3 = vec_ld(0x30, block);\ 00049 vec_s16 b4 = vec_ld(0x40, block);\ 00050 vec_s16 b5 = vec_ld(0x50, block);\ 00051 vec_s16 b6 = vec_ld(0x60, block);\ 00052 vec_s16 b7 = vec_ld(0x70, block); 00053 00054 // these functions do (a*C)>>16 00055 // things are tricky because a is signed, but C unsigned. 00056 // M15 is used if C fits in 15 bit unsigned (C6,C7) 00057 // M16 is used if C requires 16 bits unsigned 00058 static inline vec_s16 M15(vec_s16 a, vec_s16 C) 00059 { 00060 return (vec_s16)vec_perm(vec_mule(a,C), vec_mulo(a,C), interleave_high); 00061 } 00062 static inline vec_s16 M16(vec_s16 a, vec_s16 C) 00063 { 00064 return vec_add(a, M15(a, C)); 00065 } 00066 00067 #define IDCT_1D(ADD, SHIFT)\ 00068 A = vec_add(M16(b1, C1), M15(b7, C7));\ 00069 B = vec_sub(M15(b1, C7), M16(b7, C1));\ 00070 C = vec_add(M16(b3, C3), M16(b5, C5));\ 00071 D = vec_sub(M16(b5, C3), M16(b3, C5));\ 00072 \ 00073 Ad = M16(vec_sub(A, C), C4);\ 00074 Bd = M16(vec_sub(B, D), C4);\ 00075 \ 00076 Cd = vec_add(A, C);\ 00077 Dd = vec_add(B, D);\ 00078 \ 00079 E = ADD(M16(vec_add(b0, b4), C4));\ 00080 F = ADD(M16(vec_sub(b0, b4), C4));\ 00081 \ 00082 G = vec_add(M16(b2, C2), M15(b6, C6));\ 00083 H = vec_sub(M15(b2, C6), M16(b6, C2));\ 00084 \ 00085 Ed = vec_sub(E, G);\ 00086 Gd = vec_add(E, G);\ 00087 \ 00088 Add = vec_add(F, Ad);\ 00089 Bdd = vec_sub(Bd, H);\ 00090 \ 00091 Fd = vec_sub(F, Ad);\ 00092 Hd = vec_add(Bd, H);\ 00093 \ 00094 b0 = SHIFT(vec_add(Gd, Cd));\ 00095 b7 = SHIFT(vec_sub(Gd, Cd));\ 00096 \ 00097 b1 = SHIFT(vec_add(Add, Hd));\ 00098 b2 = SHIFT(vec_sub(Add, Hd));\ 00099 \ 00100 b3 = SHIFT(vec_add(Ed, Dd));\ 00101 b4 = SHIFT(vec_sub(Ed, Dd));\ 00102 \ 00103 b5 = SHIFT(vec_add(Fd, Bdd));\ 00104 b6 = SHIFT(vec_sub(Fd, Bdd)); 00105 00106 #define NOP(a) a 00107 #define ADD8(a) vec_add(a, eight) 00108 #define SHIFT4(a) vec_sra(a, four) 00109 00110 void ff_vp3_idct_altivec(DCTELEM block[64]) 00111 { 00112 IDCT_START 00113 00114 IDCT_1D(NOP, NOP) 00115 TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7); 00116 IDCT_1D(ADD8, SHIFT4) 00117 00118 vec_st(b0, 0x00, block); 00119 vec_st(b1, 0x10, block); 00120 vec_st(b2, 0x20, block); 00121 vec_st(b3, 0x30, block); 00122 vec_st(b4, 0x40, block); 00123 vec_st(b5, 0x50, block); 00124 vec_st(b6, 0x60, block); 00125 vec_st(b7, 0x70, block); 00126 } 00127 00128 void ff_vp3_idct_put_altivec(uint8_t *dst, int stride, DCTELEM block[64]) 00129 { 00130 vec_u8 t; 00131 IDCT_START 00132 00133 // pixels are signed; so add 128*16 in addition to the normal 8 00134 vec_s16 v2048 = vec_sl(vec_splat_s16(1), vec_splat_u16(11)); 00135 eight = vec_add(eight, v2048); 00136 00137 IDCT_1D(NOP, NOP) 00138 TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7); 00139 IDCT_1D(ADD8, SHIFT4) 00140 00141 #define PUT(a)\ 00142 t = vec_packsu(a, a);\ 00143 vec_ste((vec_u32)t, 0, (unsigned int *)dst);\ 00144 vec_ste((vec_u32)t, 4, (unsigned int *)dst); 00145 00146 PUT(b0) dst += stride; 00147 PUT(b1) dst += stride; 00148 PUT(b2) dst += stride; 00149 PUT(b3) dst += stride; 00150 PUT(b4) dst += stride; 00151 PUT(b5) dst += stride; 00152 PUT(b6) dst += stride; 00153 PUT(b7) 00154 } 00155 00156 void ff_vp3_idct_add_altivec(uint8_t *dst, int stride, DCTELEM block[64]) 00157 { 00158 LOAD_ZERO; 00159 vec_u8 t, vdst; 00160 vec_s16 vdst_16; 00161 vec_u8 vdst_mask = vec_mergeh(vec_splat_u8(-1), vec_lvsl(0, dst)); 00162 00163 IDCT_START 00164 00165 IDCT_1D(NOP, NOP) 00166 TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7); 00167 IDCT_1D(ADD8, SHIFT4) 00168 00169 #define ADD(a)\ 00170 vdst = vec_ld(0, dst);\ 00171 vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);\ 00172 vdst_16 = vec_adds(a, vdst_16);\ 00173 t = vec_packsu(vdst_16, vdst_16);\ 00174 vec_ste((vec_u32)t, 0, (unsigned int *)dst);\ 00175 vec_ste((vec_u32)t, 4, (unsigned int *)dst); 00176 00177 ADD(b0) dst += stride; 00178 ADD(b1) dst += stride; 00179 ADD(b2) dst += stride; 00180 ADD(b3) dst += stride; 00181 ADD(b4) dst += stride; 00182 ADD(b5) dst += stride; 00183 ADD(b6) dst += stride; 00184 ADD(b7) 00185 }