Diffstat (limited to 'core/multimedia/opieplayer/libmad/fixed.h') (more/less context) (ignore whitespace changes)
| -rw-r--r-- | core/multimedia/opieplayer/libmad/fixed.h | 111 |
1 files changed, 67 insertions, 44 deletions
diff --git a/core/multimedia/opieplayer/libmad/fixed.h b/core/multimedia/opieplayer/libmad/fixed.h index c9b98ca..baa7dc5 100644 --- a/core/multimedia/opieplayer/libmad/fixed.h +++ b/core/multimedia/opieplayer/libmad/fixed.h @@ -1,15 +1,15 @@ /* * libmad - MPEG audio decoder library - * Copyright (C) 2000-2001 Robert Leslie + * Copyright (C) 2000-2004 Underbit Technologies, Inc. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program 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 General Public License for more details. * * You should have received a copy of the GNU General Public License @@ -199,54 +199,65 @@ mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y) asm ("addl %4,%2\n\t" \ "adcl %5,%3" \ : "=rm" (__lo_), "=rm" (__hi_) \ : "0" (lo), "1" (hi), \ "ir" (1L << (MAD_F_SCALEBITS - 1)), "ir" (0) \ : "cc"); \ asm ("shrdl %3,%2,%1" \ : "=rm" (__result) \ : "0" (__lo_), "r" (__hi_), "I" (MAD_F_SCALEBITS) \ : "cc"); \ __result; \ }) +# elif defined(OPT_INTEL) +/* + * Alternate Intel scaling that may or may not perform better. + */ +# define mad_f_scale64(hi, lo) \ + ({ mad_fixed_t __result; \ + asm ("shrl %3,%1\n\t" \ + "shll %4,%2\n\t" \ + "orl %2,%1" \ + : "=rm" (__result) \ + : "0" (lo), "r" (hi), \ + "I" (MAD_F_SCALEBITS), "I" (32 - MAD_F_SCALEBITS) \ + : "cc"); \ + __result; \ + }) # else # define mad_f_scale64(hi, lo) \ ({ mad_fixed_t __result; \ asm ("shrdl %3,%2,%1" \ : "=rm" (__result) \ : "0" (lo), "r" (hi), "I" (MAD_F_SCALEBITS) \ : "cc"); \ __result; \ }) # endif /* OPT_ACCURACY */ # define MAD_F_SCALEBITS MAD_F_FRACBITS # endif /* --- ARM ----------------------------------------------------------------- */ # elif defined(FPM_ARM) /* * This ARM V4 version is as accurate as FPM_64BIT but much faster. The * least significant bit is properly rounded at no CPU cycle cost! */ # if 1 /* - * There's a bug somewhere, possibly in the compiler, that sometimes makes - * this necessary instead of the default implementation via MAD_F_MLX and - * mad_f_scale64. It may be related to the use (or lack) of - * -finline-functions and/or -fstrength-reduce. - * - * This is also apparently faster than MAD_F_MLX/mad_f_scale64. + * This is faster than the default implementation via MAD_F_MLX() and + * mad_f_scale64(). */ # define mad_f_mul(x, y) \ ({ mad_fixed64hi_t __hi; \ mad_fixed64lo_t __lo; \ mad_fixed_t __result; \ asm ("smull %0, %1, %3, %4\n\t" \ "movs %0, %0, lsr %5\n\t" \ "adc %2, %0, %1, lsl %6" \ : "=&r" (__lo), "=&r" (__hi), "=r" (__result) \ : "%r" (x), "r" (y), \ "M" (MAD_F_SCALEBITS), "M" (32 - MAD_F_SCALEBITS) \ : "cc"); \ @@ -266,25 +277,25 @@ mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y) # define MAD_F_MLN(hi, lo) \ asm ("rsbs %0, %2, #0\n\t" \ "rsc %1, %3, #0" \ : "=r" (lo), "=r" (hi) \ : "0" (lo), "1" (hi) \ : "cc") # define mad_f_scale64(hi, lo) \ ({ mad_fixed_t __result; \ asm ("movs %0, %1, lsr %3\n\t" \ "adc %0, %0, %2, lsl %4" \ - : "=r" (__result) \ + : "=&r" (__result) \ : "r" (lo), "r" (hi), \ "M" (MAD_F_SCALEBITS), "M" (32 - MAD_F_SCALEBITS) \ : "cc"); \ __result; \ }) # define MAD_F_SCALEBITS MAD_F_FRACBITS /* --- MIPS ---------------------------------------------------------------- */ # elif defined(FPM_MIPS) @@ -334,76 +345,87 @@ mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y) */ # define MAD_F_MLX(hi, lo, x, y) \ asm ("smul %2, %3, %0\n\t" \ "rd %%y, %1" \ : "=r" (lo), "=r" (hi) \ : "%r" (x), "rI" (y)) /* --- PowerPC ------------------------------------------------------------- */ # elif defined(FPM_PPC) /* - * This PowerPC version is tuned for the 4xx embedded processors. It is - * effectively a tuned version of FPM_64BIT. It is a little faster and just - * as accurate. The disposition of the least significant bit depends on - * OPT_ACCURACY via mad_f_scale64(). + * This PowerPC version is fast and accurate; the disposition of the least + * significant bit depends on OPT_ACCURACY via mad_f_scale64(). */ # define MAD_F_MLX(hi, lo, x, y) \ - asm ("mulhw %1, %2, %3\n\t" \ - "mullw %0, %2, %3" \ - : "=&r" (lo), "=&r" (hi) \ - : "%r" (x), "r" (y)) + do { \ + asm ("mullw %0,%1,%2" \ + : "=r" (lo) \ + : "%r" (x), "r" (y)); \ + asm ("mulhw %0,%1,%2" \ + : "=r" (hi) \ + : "%r" (x), "r" (y)); \ + } \ + while (0) -# define MAD_F_MLA(hi, lo, x, y) \ +# if defined(OPT_ACCURACY) +/* + * This gives best accuracy but is not very fast. + */ +# define MAD_F_MLA(hi, lo, x, y) \ ({ mad_fixed64hi_t __hi; \ mad_fixed64lo_t __lo; \ MAD_F_MLX(__hi, __lo, (x), (y)); \ - asm ("addc %0, %2, %3\n\t" \ - "adde %1, %4, %5" \ + asm ("addc %0,%2,%3\n\t" \ + "adde %1,%4,%5" \ : "=r" (lo), "=r" (hi) \ - : "%r" (__lo), "0" (lo), "%r" (__hi), "1" (hi)); \ + : "%r" (lo), "r" (__lo), \ + "%r" (hi), "r" (__hi) \ + : "xer"); \ }) +# endif # if defined(OPT_ACCURACY) /* - * This is accurate and ~2 - 2.5 times slower than the unrounded version. - * - * The __volatile__ improves the generated code by another 5% (fewer spills - * to memory); eventually they should be removed. + * This is slower than the truncating version below it. */ # define mad_f_scale64(hi, lo) \ - ({ mad_fixed_t __result; \ - mad_fixed64hi_t __hi_; \ - mad_fixed64lo_t __lo_; \ - asm __volatile__ ("addc %0, %2, %4\n\t" \ - "addze %1, %3" \ - : "=r" (__lo_), "=r" (__hi_) \ - : "r" (lo), "r" (hi), "r" (1 << (MAD_F_SCALEBITS - 1))); \ - asm __volatile__ ("rlwinm %0, %2,32-%3,0,%3-1\n\t" \ - "rlwimi %0, %1,32-%3,%3,31" \ - : "=&r" (__result) \ - : "r" (__lo_), "r" (__hi_), "I" (MAD_F_SCALEBITS)); \ - __result; \ + ({ mad_fixed_t __result, __round; \ + asm ("rotrwi %0,%1,%2" \ + : "=r" (__result) \ + : "r" (lo), "i" (MAD_F_SCALEBITS)); \ + asm ("extrwi %0,%1,1,0" \ + : "=r" (__round) \ + : "r" (__result)); \ + asm ("insrwi %0,%1,%2,0" \ + : "+r" (__result) \ + : "r" (hi), "i" (MAD_F_SCALEBITS)); \ + asm ("add %0,%1,%2" \ + : "=r" (__result) \ + : "%r" (__result), "r" (__round)); \ + __result; \ }) # else # define mad_f_scale64(hi, lo) \ ({ mad_fixed_t __result; \ - asm ("rlwinm %0, %2,32-%3,0,%3-1\n\t" \ - "rlwimi %0, %1,32-%3,%3,31" \ + asm ("rotrwi %0,%1,%2" \ : "=r" (__result) \ - : "r" (lo), "r" (hi), "I" (MAD_F_SCALEBITS)); \ - __result; \ + : "r" (lo), "i" (MAD_F_SCALEBITS)); \ + asm ("insrwi %0,%1,%2,0" \ + : "+r" (__result) \ + : "r" (hi), "i" (MAD_F_SCALEBITS)); \ + __result; \ }) -# endif /* OPT_ACCURACY */ +# endif # define MAD_F_SCALEBITS MAD_F_FRACBITS /* --- Default ------------------------------------------------------------- */ # elif defined(FPM_DEFAULT) /* * This version is the most portable but it loses significant accuracy. * Furthermore, accuracy is biased against the second argument, so care * should be taken when ordering operands. * @@ -419,26 +441,26 @@ mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y) # endif /* ------------------------------------------------------------------------- */ # else # error "no FPM selected" # endif /* default implementations */ # if !defined(mad_f_mul) # define mad_f_mul(x, y) \ - ({ mad_fixed64hi_t __hi; \ - mad_fixed64lo_t __lo; \ + ({ register mad_fixed64hi_t __hi; \ + register mad_fixed64lo_t __lo; \ MAD_F_MLX(__hi, __lo, (x), (y)); \ mad_f_scale64(__hi, __lo); \ }) # endif # if !defined(MAD_F_MLA) # define MAD_F_ML0(hi, lo, x, y) ((lo) = mad_f_mul((x), (y))) # define MAD_F_MLA(hi, lo, x, y) ((lo) += mad_f_mul((x), (y))) # define MAD_F_MLN(hi, lo) ((lo) = -(lo)) # define MAD_F_MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo)) # endif @@ -460,17 +482,18 @@ mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y) ((((mad_fixed_t) \ (((hi) << (32 - (MAD_F_SCALEBITS - 1))) | \ ((lo) >> (MAD_F_SCALEBITS - 1)))) + 1) >> 1) # else # define mad_f_scale64(hi, lo) \ ((mad_fixed_t) \ (((hi) << (32 - MAD_F_SCALEBITS)) | \ ((lo) >> MAD_F_SCALEBITS))) # endif # define MAD_F_SCALEBITS MAD_F_FRACBITS # endif -/* miscellaneous C routines */ +/* C routines */ mad_fixed_t mad_f_abs(mad_fixed_t); +mad_fixed_t mad_f_div(mad_fixed_t, mad_fixed_t); # endif |