1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libtremor/lib/asm_arm.h Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,245 @@
1.4 +/********************************************************************
1.5 + * *
1.6 + * THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
1.7 + * *
1.8 + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
1.9 + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
1.10 + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
1.11 + * *
1.12 + * THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
1.13 + * BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
1.14 + * *
1.15 + ********************************************************************
1.16 +
1.17 + function: arm7 and later wide math functions
1.18 +
1.19 + ********************************************************************/
1.20 +
1.21 +#ifdef _ARM_ASSEM_
1.22 +
1.23 +#if !defined(_V_WIDE_MATH) && !defined(_LOW_ACCURACY_)
1.24 +#define _V_WIDE_MATH
1.25 +
1.26 +static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
1.27 + int lo,hi;
1.28 + asm volatile("smull\t%0, %1, %2, %3"
1.29 + : "=&r"(lo),"=&r"(hi)
1.30 + : "%r"(x),"r"(y)
1.31 + : "cc");
1.32 + return(hi);
1.33 +}
1.34 +
1.35 +static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
1.36 + return MULT32(x,y)<<1;
1.37 +}
1.38 +
1.39 +static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
1.40 + int lo,hi;
1.41 + asm volatile("smull %0, %1, %2, %3\n\t"
1.42 + "movs %0, %0, lsr #15\n\t"
1.43 + "adc %1, %0, %1, lsl #17\n\t"
1.44 + : "=&r"(lo),"=&r"(hi)
1.45 + : "%r"(x),"r"(y)
1.46 + : "cc");
1.47 + return(hi);
1.48 +}
1.49 +
1.50 +#define MB() asm volatile ("" : : : "memory")
1.51 +
1.52 +static inline void XPROD32(ogg_int32_t a, ogg_int32_t b,
1.53 + ogg_int32_t t, ogg_int32_t v,
1.54 + ogg_int32_t *x, ogg_int32_t *y)
1.55 +{
1.56 + int x1, y1, l;
1.57 + asm( "smull %0, %1, %4, %6\n\t"
1.58 + "smlal %0, %1, %5, %7\n\t"
1.59 + "rsb %3, %4, #0\n\t"
1.60 + "smull %0, %2, %5, %6\n\t"
1.61 + "smlal %0, %2, %3, %7"
1.62 + : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
1.63 + : "3" (a), "r" (b), "r" (t), "r" (v)
1.64 + : "cc" );
1.65 + *x = x1;
1.66 + MB();
1.67 + *y = y1;
1.68 +}
1.69 +
1.70 +static inline void XPROD31(ogg_int32_t a, ogg_int32_t b,
1.71 + ogg_int32_t t, ogg_int32_t v,
1.72 + ogg_int32_t *x, ogg_int32_t *y)
1.73 +{
1.74 + int x1, y1, l;
1.75 + asm( "smull %0, %1, %4, %6\n\t"
1.76 + "smlal %0, %1, %5, %7\n\t"
1.77 + "rsb %3, %4, #0\n\t"
1.78 + "smull %0, %2, %5, %6\n\t"
1.79 + "smlal %0, %2, %3, %7"
1.80 + : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
1.81 + : "3" (a), "r" (b), "r" (t), "r" (v)
1.82 + : "cc" );
1.83 + *x = x1 << 1;
1.84 + MB();
1.85 + *y = y1 << 1;
1.86 +}
1.87 +
1.88 +static inline void XNPROD31(ogg_int32_t a, ogg_int32_t b,
1.89 + ogg_int32_t t, ogg_int32_t v,
1.90 + ogg_int32_t *x, ogg_int32_t *y)
1.91 +{
1.92 + int x1, y1, l;
1.93 + asm( "rsb %2, %4, #0\n\t"
1.94 + "smull %0, %1, %3, %5\n\t"
1.95 + "smlal %0, %1, %2, %6\n\t"
1.96 + "smull %0, %2, %4, %5\n\t"
1.97 + "smlal %0, %2, %3, %6"
1.98 + : "=&r" (l), "=&r" (x1), "=&r" (y1)
1.99 + : "r" (a), "r" (b), "r" (t), "r" (v)
1.100 + : "cc" );
1.101 + *x = x1 << 1;
1.102 + MB();
1.103 + *y = y1 << 1;
1.104 +}
1.105 +
1.106 +#endif
1.107 +
1.108 +#ifndef _V_CLIP_MATH
1.109 +#define _V_CLIP_MATH
1.110 +
1.111 +static inline ogg_int32_t CLIP_TO_15(ogg_int32_t x) {
1.112 + int tmp;
1.113 + asm volatile("subs %1, %0, #32768\n\t"
1.114 + "movpl %0, #0x7f00\n\t"
1.115 + "orrpl %0, %0, #0xff\n"
1.116 + "adds %1, %0, #32768\n\t"
1.117 + "movmi %0, #0x8000"
1.118 + : "+r"(x),"=r"(tmp)
1.119 + :
1.120 + : "cc");
1.121 + return(x);
1.122 +}
1.123 +
1.124 +#endif
1.125 +
1.126 +#ifndef _V_LSP_MATH_ASM
1.127 +#define _V_LSP_MATH_ASM
1.128 +
1.129 +static inline void lsp_loop_asm(ogg_uint32_t *qip,ogg_uint32_t *pip,
1.130 + ogg_int32_t *qexpp,
1.131 + ogg_int32_t *ilsp,ogg_int32_t wi,
1.132 + ogg_int32_t m){
1.133 +
1.134 + ogg_uint32_t qi=*qip,pi=*pip;
1.135 + ogg_int32_t qexp=*qexpp;
1.136 +
1.137 + asm("mov r0,%3;"
1.138 + "movs r1,%5,asr#1;"
1.139 + "add r0,r0,r1,lsl#3;"
1.140 + "beq 2f;\n"
1.141 + "1:"
1.142 +
1.143 + "ldmdb r0!,{r1,r3};"
1.144 + "subs r1,r1,%4;" //ilsp[j]-wi
1.145 + "rsbmi r1,r1,#0;" //labs(ilsp[j]-wi)
1.146 + "umull %0,r2,r1,%0;" //qi*=labs(ilsp[j]-wi)
1.147 +
1.148 + "subs r1,r3,%4;" //ilsp[j+1]-wi
1.149 + "rsbmi r1,r1,#0;" //labs(ilsp[j+1]-wi)
1.150 + "umull %1,r3,r1,%1;" //pi*=labs(ilsp[j+1]-wi)
1.151 +
1.152 + "cmn r2,r3;" // shift down 16?
1.153 + "beq 0f;"
1.154 + "add %2,%2,#16;"
1.155 + "mov %0,%0,lsr #16;"
1.156 + "orr %0,%0,r2,lsl #16;"
1.157 + "mov %1,%1,lsr #16;"
1.158 + "orr %1,%1,r3,lsl #16;"
1.159 + "0:"
1.160 + "cmp r0,%3;\n"
1.161 + "bhi 1b;\n"
1.162 +
1.163 + "2:"
1.164 + // odd filter assymetry
1.165 + "ands r0,%5,#1;\n"
1.166 + "beq 3f;\n"
1.167 + "add r0,%3,%5,lsl#2;\n"
1.168 +
1.169 + "ldr r1,[r0,#-4];\n"
1.170 + "mov r0,#0x4000;\n"
1.171 +
1.172 + "subs r1,r1,%4;\n" //ilsp[j]-wi
1.173 + "rsbmi r1,r1,#0;\n" //labs(ilsp[j]-wi)
1.174 + "umull %0,r2,r1,%0;\n" //qi*=labs(ilsp[j]-wi)
1.175 + "umull %1,r3,r0,%1;\n" //pi*=labs(ilsp[j+1]-wi)
1.176 +
1.177 + "cmn r2,r3;\n" // shift down 16?
1.178 + "beq 3f;\n"
1.179 + "add %2,%2,#16;\n"
1.180 + "mov %0,%0,lsr #16;\n"
1.181 + "orr %0,%0,r2,lsl #16;\n"
1.182 + "mov %1,%1,lsr #16;\n"
1.183 + "orr %1,%1,r3,lsl #16;\n"
1.184 +
1.185 + //qi=(pi>>shift)*labs(ilsp[j]-wi);
1.186 + //pi=(qi>>shift)*labs(ilsp[j+1]-wi);
1.187 + //qexp+=shift;
1.188 +
1.189 + //}
1.190 +
1.191 + /* normalize to max 16 sig figs */
1.192 + "3:"
1.193 + "mov r2,#0;"
1.194 + "orr r1,%0,%1;"
1.195 + "tst r1,#0xff000000;"
1.196 + "addne r2,r2,#8;"
1.197 + "movne r1,r1,lsr #8;"
1.198 + "tst r1,#0x00f00000;"
1.199 + "addne r2,r2,#4;"
1.200 + "movne r1,r1,lsr #4;"
1.201 + "tst r1,#0x000c0000;"
1.202 + "addne r2,r2,#2;"
1.203 + "movne r1,r1,lsr #2;"
1.204 + "tst r1,#0x00020000;"
1.205 + "addne r2,r2,#1;"
1.206 + "movne r1,r1,lsr #1;"
1.207 + "tst r1,#0x00010000;"
1.208 + "addne r2,r2,#1;"
1.209 + "mov %0,%0,lsr r2;"
1.210 + "mov %1,%1,lsr r2;"
1.211 + "add %2,%2,r2;"
1.212 +
1.213 + : "+r"(qi),"+r"(pi),"+r"(qexp)
1.214 + : "r"(ilsp),"r"(wi),"r"(m)
1.215 + : "r0","r1","r2","r3","cc");
1.216 +
1.217 + *qip=qi;
1.218 + *pip=pi;
1.219 + *qexpp=qexp;
1.220 +}
1.221 +
1.222 +static inline void lsp_norm_asm(ogg_uint32_t *qip,ogg_int32_t *qexpp){
1.223 +
1.224 + ogg_uint32_t qi=*qip;
1.225 + ogg_int32_t qexp=*qexpp;
1.226 +
1.227 + asm("tst %0,#0x0000ff00;"
1.228 + "moveq %0,%0,lsl #8;"
1.229 + "subeq %1,%1,#8;"
1.230 + "tst %0,#0x0000f000;"
1.231 + "moveq %0,%0,lsl #4;"
1.232 + "subeq %1,%1,#4;"
1.233 + "tst %0,#0x0000c000;"
1.234 + "moveq %0,%0,lsl #2;"
1.235 + "subeq %1,%1,#2;"
1.236 + "tst %0,#0x00008000;"
1.237 + "moveq %0,%0,lsl #1;"
1.238 + "subeq %1,%1,#1;"
1.239 + : "+r"(qi),"+r"(qexp)
1.240 + :
1.241 + : "cc");
1.242 + *qip=qi;
1.243 + *qexpp=qexp;
1.244 +}
1.245 +
1.246 +#endif
1.247 +#endif
1.248 +
