michael@0: /******************************************************************** michael@0: * * michael@0: * THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. * michael@0: * * michael@0: * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * michael@0: * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * michael@0: * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * michael@0: * * michael@0: * THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 * michael@0: * BY THE Xiph.Org FOUNDATION http://www.xiph.org/ * michael@0: * * michael@0: ******************************************************************** michael@0: michael@0: function: arm7 and later wide math functions michael@0: michael@0: ********************************************************************/ michael@0: michael@0: #ifdef _ARM_ASSEM_ michael@0: michael@0: #if !defined(_V_WIDE_MATH) && !defined(_LOW_ACCURACY_) michael@0: #define _V_WIDE_MATH michael@0: michael@0: static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) { michael@0: int lo,hi; michael@0: asm volatile("smull\t%0, %1, %2, %3" michael@0: : "=&r"(lo),"=&r"(hi) michael@0: : "%r"(x),"r"(y) michael@0: : "cc"); michael@0: return(hi); michael@0: } michael@0: michael@0: static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) { michael@0: return MULT32(x,y)<<1; michael@0: } michael@0: michael@0: static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) { michael@0: int lo,hi; michael@0: asm volatile("smull %0, %1, %2, %3\n\t" michael@0: "movs %0, %0, lsr #15\n\t" michael@0: "adc %1, %0, %1, lsl #17\n\t" michael@0: : "=&r"(lo),"=&r"(hi) michael@0: : "%r"(x),"r"(y) michael@0: : "cc"); michael@0: return(hi); michael@0: } michael@0: michael@0: #define MB() asm volatile ("" : : : "memory") michael@0: michael@0: static inline void XPROD32(ogg_int32_t a, ogg_int32_t b, michael@0: ogg_int32_t t, ogg_int32_t v, michael@0: ogg_int32_t *x, ogg_int32_t *y) michael@0: { michael@0: int x1, y1, l; michael@0: asm( "smull %0, %1, %4, %6\n\t" michael@0: "smlal %0, %1, %5, %7\n\t" michael@0: "rsb %3, %4, #0\n\t" michael@0: "smull %0, %2, %5, %6\n\t" michael@0: "smlal %0, %2, %3, %7" michael@0: : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a) michael@0: : "3" (a), "r" (b), "r" (t), "r" (v) michael@0: : "cc" ); michael@0: *x = x1; michael@0: MB(); michael@0: *y = y1; michael@0: } michael@0: michael@0: static inline void XPROD31(ogg_int32_t a, ogg_int32_t b, michael@0: ogg_int32_t t, ogg_int32_t v, michael@0: ogg_int32_t *x, ogg_int32_t *y) michael@0: { michael@0: int x1, y1, l; michael@0: asm( "smull %0, %1, %4, %6\n\t" michael@0: "smlal %0, %1, %5, %7\n\t" michael@0: "rsb %3, %4, #0\n\t" michael@0: "smull %0, %2, %5, %6\n\t" michael@0: "smlal %0, %2, %3, %7" michael@0: : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a) michael@0: : "3" (a), "r" (b), "r" (t), "r" (v) michael@0: : "cc" ); michael@0: *x = x1 << 1; michael@0: MB(); michael@0: *y = y1 << 1; michael@0: } michael@0: michael@0: static inline void XNPROD31(ogg_int32_t a, ogg_int32_t b, michael@0: ogg_int32_t t, ogg_int32_t v, michael@0: ogg_int32_t *x, ogg_int32_t *y) michael@0: { michael@0: int x1, y1, l; michael@0: asm( "rsb %2, %4, #0\n\t" michael@0: "smull %0, %1, %3, %5\n\t" michael@0: "smlal %0, %1, %2, %6\n\t" michael@0: "smull %0, %2, %4, %5\n\t" michael@0: "smlal %0, %2, %3, %6" michael@0: : "=&r" (l), "=&r" (x1), "=&r" (y1) michael@0: : "r" (a), "r" (b), "r" (t), "r" (v) michael@0: : "cc" ); michael@0: *x = x1 << 1; michael@0: MB(); michael@0: *y = y1 << 1; michael@0: } michael@0: michael@0: #endif michael@0: michael@0: #ifndef _V_CLIP_MATH michael@0: #define _V_CLIP_MATH michael@0: michael@0: static inline ogg_int32_t CLIP_TO_15(ogg_int32_t x) { michael@0: int tmp; michael@0: asm volatile("subs %1, %0, #32768\n\t" michael@0: "movpl %0, #0x7f00\n\t" michael@0: "orrpl %0, %0, #0xff\n" michael@0: "adds %1, %0, #32768\n\t" michael@0: "movmi %0, #0x8000" michael@0: : "+r"(x),"=r"(tmp) michael@0: : michael@0: : "cc"); michael@0: return(x); michael@0: } michael@0: michael@0: #endif michael@0: michael@0: #ifndef _V_LSP_MATH_ASM michael@0: #define _V_LSP_MATH_ASM michael@0: michael@0: static inline void lsp_loop_asm(ogg_uint32_t *qip,ogg_uint32_t *pip, michael@0: ogg_int32_t *qexpp, michael@0: ogg_int32_t *ilsp,ogg_int32_t wi, michael@0: ogg_int32_t m){ michael@0: michael@0: ogg_uint32_t qi=*qip,pi=*pip; michael@0: ogg_int32_t qexp=*qexpp; michael@0: michael@0: asm("mov r0,%3;" michael@0: "movs r1,%5,asr#1;" michael@0: "add r0,r0,r1,lsl#3;" michael@0: "beq 2f;\n" michael@0: "1:" michael@0: michael@0: "ldmdb r0!,{r1,r3};" michael@0: "subs r1,r1,%4;" //ilsp[j]-wi michael@0: "rsbmi r1,r1,#0;" //labs(ilsp[j]-wi) michael@0: "umull %0,r2,r1,%0;" //qi*=labs(ilsp[j]-wi) michael@0: michael@0: "subs r1,r3,%4;" //ilsp[j+1]-wi michael@0: "rsbmi r1,r1,#0;" //labs(ilsp[j+1]-wi) michael@0: "umull %1,r3,r1,%1;" //pi*=labs(ilsp[j+1]-wi) michael@0: michael@0: "cmn r2,r3;" // shift down 16? michael@0: "beq 0f;" michael@0: "add %2,%2,#16;" michael@0: "mov %0,%0,lsr #16;" michael@0: "orr %0,%0,r2,lsl #16;" michael@0: "mov %1,%1,lsr #16;" michael@0: "orr %1,%1,r3,lsl #16;" michael@0: "0:" michael@0: "cmp r0,%3;\n" michael@0: "bhi 1b;\n" michael@0: michael@0: "2:" michael@0: // odd filter assymetry michael@0: "ands r0,%5,#1;\n" michael@0: "beq 3f;\n" michael@0: "add r0,%3,%5,lsl#2;\n" michael@0: michael@0: "ldr r1,[r0,#-4];\n" michael@0: "mov r0,#0x4000;\n" michael@0: michael@0: "subs r1,r1,%4;\n" //ilsp[j]-wi michael@0: "rsbmi r1,r1,#0;\n" //labs(ilsp[j]-wi) michael@0: "umull %0,r2,r1,%0;\n" //qi*=labs(ilsp[j]-wi) michael@0: "umull %1,r3,r0,%1;\n" //pi*=labs(ilsp[j+1]-wi) michael@0: michael@0: "cmn r2,r3;\n" // shift down 16? michael@0: "beq 3f;\n" michael@0: "add %2,%2,#16;\n" michael@0: "mov %0,%0,lsr #16;\n" michael@0: "orr %0,%0,r2,lsl #16;\n" michael@0: "mov %1,%1,lsr #16;\n" michael@0: "orr %1,%1,r3,lsl #16;\n" michael@0: michael@0: //qi=(pi>>shift)*labs(ilsp[j]-wi); michael@0: //pi=(qi>>shift)*labs(ilsp[j+1]-wi); michael@0: //qexp+=shift; michael@0: michael@0: //} michael@0: michael@0: /* normalize to max 16 sig figs */ michael@0: "3:" michael@0: "mov r2,#0;" michael@0: "orr r1,%0,%1;" michael@0: "tst r1,#0xff000000;" michael@0: "addne r2,r2,#8;" michael@0: "movne r1,r1,lsr #8;" michael@0: "tst r1,#0x00f00000;" michael@0: "addne r2,r2,#4;" michael@0: "movne r1,r1,lsr #4;" michael@0: "tst r1,#0x000c0000;" michael@0: "addne r2,r2,#2;" michael@0: "movne r1,r1,lsr #2;" michael@0: "tst r1,#0x00020000;" michael@0: "addne r2,r2,#1;" michael@0: "movne r1,r1,lsr #1;" michael@0: "tst r1,#0x00010000;" michael@0: "addne r2,r2,#1;" michael@0: "mov %0,%0,lsr r2;" michael@0: "mov %1,%1,lsr r2;" michael@0: "add %2,%2,r2;" michael@0: michael@0: : "+r"(qi),"+r"(pi),"+r"(qexp) michael@0: : "r"(ilsp),"r"(wi),"r"(m) michael@0: : "r0","r1","r2","r3","cc"); michael@0: michael@0: *qip=qi; michael@0: *pip=pi; michael@0: *qexpp=qexp; michael@0: } michael@0: michael@0: static inline void lsp_norm_asm(ogg_uint32_t *qip,ogg_int32_t *qexpp){ michael@0: michael@0: ogg_uint32_t qi=*qip; michael@0: ogg_int32_t qexp=*qexpp; michael@0: michael@0: asm("tst %0,#0x0000ff00;" michael@0: "moveq %0,%0,lsl #8;" michael@0: "subeq %1,%1,#8;" michael@0: "tst %0,#0x0000f000;" michael@0: "moveq %0,%0,lsl #4;" michael@0: "subeq %1,%1,#4;" michael@0: "tst %0,#0x0000c000;" michael@0: "moveq %0,%0,lsl #2;" michael@0: "subeq %1,%1,#2;" michael@0: "tst %0,#0x00008000;" michael@0: "moveq %0,%0,lsl #1;" michael@0: "subeq %1,%1,#1;" michael@0: : "+r"(qi),"+r"(qexp) michael@0: : michael@0: : "cc"); michael@0: *qip=qi; michael@0: *qexpp=qexp; michael@0: } michael@0: michael@0: #endif michael@0: #endif michael@0: