rockbox/apps/codecs/Tremor/asm_mcf5249.h
Pedro Vasconcelos 1b3622d589 Use asm optimised memset() in vorbis_apply_window.
Slight speed improvement. "Toughest" Q10 encodings play without skipping now.


git-svn-id: svn://svn.rockbox.org/rockbox/trunk@6792 a1c6a512-1295-4272-9138-f99709370657
2005-06-21 20:14:18 +00:00

354 lines
12 KiB
C

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
*
* Copyright (C) 2005 by Pedro Vasconcelos
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
/* asm routines for wide math on the MCF5249 */
#include "os_types.h"
#if CONFIG_CPU == MCF5249 && !defined(SIMULATOR)
/* attribute for 16-byte alignment */
#define LINE_ATTR __attribute__ ((aligned (16)))
#ifndef _V_WIDE_MATH
#define _V_WIDE_MATH
#define MB()
static inline void mcf5249_init_mac(void) {
asm volatile ("move.l #0x20, %macsr;"); /* frac, truncate, no saturation */
}
static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
asm volatile ("mac.l %[x], %[y], %%acc0;" /* multiply & shift */
"movclr.l %%acc0, %[x];" /* move & clear acc */
"asr.l #1, %[x];" /* no overflow test */
: [x] "+&d" (x)
: [y] "r" (y)
: "cc");
return x;
}
static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
asm volatile ("mac.l %[x], %[y], %%acc0;" /* multiply */
"movclr.l %%acc0, %[x];" /* move and clear */
: [x] "+&r" (x)
: [y] "r" (y)
: "cc");
return x;
}
static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
ogg_int32_t r;
asm volatile ("mac.l %[x], %[y], %%acc0;" /* multiply */
"movclr.l %%acc0, %[r];" /* get higher half */
"mulu.l %[y], %[x];" /* get lower half */
"asl.l #8, %[r];" /* hi<<16, plus one free */
"asl.l #8, %[r];"
"lsr.l #8, %[x];" /* (unsigned)lo >> 15 */
"lsr.l #7, %[x];"
"or.l %[x], %[r];" /* logical-or results */
: [r] "=&d" (r), [x] "+d" (x)
: [y] "d" (y)
: "cc");
return r;
}
static inline
void XPROD31(ogg_int32_t a, ogg_int32_t b,
ogg_int32_t t, ogg_int32_t v,
ogg_int32_t *x, ogg_int32_t *y)
{
asm volatile ("mac.l %[a], %[t], %%acc0;"
"mac.l %[b], %[v], %%acc0;"
"mac.l %[b], %[t], %%acc1;"
"msac.l %[a], %[v], %%acc1;"
"movclr.l %%acc0, %[a];"
"move.l %[a], (%[x]);"
"movclr.l %%acc1, %[a];"
"move.l %[a], (%[y]);"
: [a] "+&r" (a)
: [x] "a" (x), [y] "a" (y),
[b] "r" (b), [t] "r" (t), [v] "r" (v)
: "cc", "memory");
}
static inline
void XNPROD31(ogg_int32_t a, ogg_int32_t b,
ogg_int32_t t, ogg_int32_t v,
ogg_int32_t *x, ogg_int32_t *y)
{
asm volatile ("mac.l %[a], %[t], %%acc0;"
"msac.l %[b], %[v], %%acc0;"
"mac.l %[b], %[t], %%acc1;"
"mac.l %[a], %[v], %%acc1;"
"movclr.l %%acc0, %[a];"
"move.l %[a], (%[x]);"
"movclr.l %%acc1, %[a];"
"move.l %[a], (%[y]);"
: [a] "+&r" (a)
: [x] "a" (x), [y] "a" (y),
[b] "r" (b), [t] "r" (t), [v] "r" (v)
: "cc", "memory");
}
#if 0 /* canonical Tremor definition */
#define XPROD32(_a, _b, _t, _v, _x, _y) \
{ (_x)=MULT32(_a,_t)+MULT32(_b,_v); \
(_y)=MULT32(_b,_t)-MULT32(_a,_v); }
#endif
/* this could lose the LSB by overflow, but i don't think it'll ever happen.
if anyone think they can hear a bug caused by this, please try the above
version. */
#define XPROD32(_a, _b, _t, _v, _x, _y) \
asm volatile ("mac.l %[a], %[t], %%acc0;" \
"mac.l %[b], %[v], %%acc0;" \
"mac.l %[b], %[t], %%acc1;" \
"msac.l %[a], %[v], %%acc1;" \
"movclr.l %%acc0, %[x];" \
"asr.l #1, %[x];" \
"movclr.l %%acc1, %[y];" \
"asr.l #1, %[y];" \
: [x] "=&d" (_x), [y] "=&d" (_y) \
: [a] "r" (_a), [b] "r" (_b), \
[t] "r" (_t), [v] "r" (_v) \
: "cc");
/* asm versions of vector operations for block.c, window.c */
/* assumes MAC is initialized & accumulators cleared */
static inline
void mcf5249_vect_add(ogg_int32_t *x, ogg_int32_t *y, int n)
{
/* align to 16 bytes */
while(n>0 && (int)x&16) {
*x++ += *y++;
n--;
}
asm volatile ("bra 1f;"
"0:" /* loop start */
"movem.l (%[x]), %%d0-%%d3;" /* fetch values */
"movem.l (%[y]), %%a0-%%a3;"
/* add */
"add.l %%a0, %%d0;"
"add.l %%a1, %%d1;"
"add.l %%a2, %%d2;"
"add.l %%a3, %%d3;"
/* store and advance */
"movem.l %%d0-%%d3, (%[x]);"
"lea.l (4*4, %[x]), %[x];"
"lea.l (4*4, %[y]), %[y];"
"subq.l #4, %[n];" /* done 4 elements */
"1: cmpi.l #4, %[n];"
"bge 0b;"
: [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
: : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
"cc", "memory");
/* add final elements */
while (n>0) {
*x++ += *y++;
n--;
}
}
static inline
void mcf5249_vect_copy(ogg_int32_t *x, ogg_int32_t *y, int n)
{
/* align to 16 bytes */
while(n>0 && (int)x&16) {
*x++ = *y++;
n--;
}
asm volatile ("bra 1f;"
"0:" /* loop start */
"movem.l (%[y]), %%d0-%%d3;" /* fetch values */
"movem.l %%d0-%%d3, (%[x]);" /* store */
"lea.l (4*4, %[x]), %[x];" /* advance */
"lea.l (4*4, %[y]), %[y];"
"subq.l #4, %[n];" /* done 4 elements */
"1: cmpi.l #4, %[n];"
"bge 0b;"
: [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
: : "%d0", "%d1", "%d2", "%d3", "cc", "memory");
/* copy final elements */
while (n>0) {
*x++ = *y++;
n--;
}
}
static inline
void mcf5249_vect_mult_fw(ogg_int32_t *data, LOOKUP_T *window, int n)
{
/* ensure data is aligned to 16-bytes */
while(n>0 && (int)data%16) {
*data = MULT31(*data, *window);
data++;
window++;
n--;
}
asm volatile ("movem.l (%[d]), %%d0-%%d3;" /* loop start */
"movem.l (%[w]), %%a0-%%a3;" /* pre-fetch registers */
"lea.l (4*4, %[w]), %[w];"
"bra 1f;" /* jump to loop condition */
"0:" /* loop body */
/* multiply and load next window values */
"mac.l %%d0, %%a0, (%[w])+, %%a0, %%acc0;"
"mac.l %%d1, %%a1, (%[w])+, %%a1, %%acc1;"
"mac.l %%d2, %%a2, (%[w])+, %%a2, %%acc2;"
"mac.l %%d3, %%a3, (%[w])+, %%a3, %%acc3;"
"movclr.l %%acc0, %%d0;" /* get the products */
"movclr.l %%acc1, %%d1;"
"movclr.l %%acc2, %%d2;"
"movclr.l %%acc3, %%d3;"
/* store and advance */
"movem.l %%d0-%%d3, (%[d]);"
"lea.l (4*4, %[d]), %[d];"
"movem.l (%[d]), %%d0-%%d3;"
"subq.l #4, %[n];" /* done 4 elements */
"1: cmpi.l #4, %[n];"
"bge 0b;"
/* multiply final elements */
"tst.l %[n];"
"beq 1f;" /* n=0 */
"mac.l %%d0, %%a0, %%acc0;"
"movclr.l %%acc0, %%d0;"
"move.l %%d0, (%[d])+;"
"subq.l #1, %[n];"
"beq 1f;" /* n=1 */
"mac.l %%d1, %%a1, %%acc0;"
"movclr.l %%acc0, %%d1;"
"move.l %%d1, (%[d])+;"
"subq.l #1, %[n];"
"beq 1f;" /* n=2 */
/* otherwise n = 3 */
"mac.l %%d2, %%a2, %%acc0;"
"movclr.l %%acc0, %%d2;"
"move.l %%d2, (%[d])+;"
"1:"
: [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
: : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
"cc", "memory");
}
static inline
void mcf5249_vect_mult_bw(ogg_int32_t *data, LOOKUP_T *window, int n)
{
/* ensure at least data is aligned to 16-bytes */
while(n>0 && (int)data%16) {
*data = MULT31(*data, *window);
data++;
window--;
n--;
}
asm volatile ("lea.l (-3*4, %[w]), %[w];" /* loop start */
"movem.l (%[d]), %%d0-%%d3;" /* pre-fetch registers */
"movem.l (%[w]), %%a0-%%a3;"
"bra 1f;" /* jump to loop condition */
"0:" /* loop body */
/* multiply and load next window value */
"mac.l %%d0, %%a3, -(%[w]), %%a3, %%acc0;"
"mac.l %%d1, %%a2, -(%[w]), %%a2, %%acc1;"
"mac.l %%d2, %%a1, -(%[w]), %%a1, %%acc2;"
"mac.l %%d3, %%a0, -(%[w]), %%a0, %%acc3;"
"movclr.l %%acc0, %%d0;" /* get the products */
"movclr.l %%acc1, %%d1;"
"movclr.l %%acc2, %%d2;"
"movclr.l %%acc3, %%d3;"
/* store and advance */
"movem.l %%d0-%%d3, (%[d]);"
"lea.l (4*4, %[d]), %[d];"
"movem.l (%[d]), %%d0-%%d3;"
"subq.l #4, %[n];" /* done 4 elements */
"1: cmpi.l #4, %[n];"
"bge 0b;"
/* multiply final elements */
"tst.l %[n];"
"beq 1f;" /* n=0 */
"mac.l %%d0, %%a3, %%acc0;"
"movclr.l %%acc0, %%d0;"
"move.l %%d0, (%[d])+;"
"subq.l #1, %[n];"
"beq 1f;" /* n=1 */
"mac.l %%d1, %%a2, %%acc0;"
"movclr.l %%acc0, %%d1;"
"move.l %%d1, (%[d])+;"
"subq.l #1, %[n];"
"beq 1f;" /* n=2 */
/* otherwise n = 3 */
"mac.l %%d2, %%a1, %%acc0;"
"movclr.l %%acc0, %%d2;"
"move.l %%d2, (%[d])+;"
"1:"
: [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
: : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
"cc", "memory");
}
#if 0
/* this routine is subsumed by Jens' asm optimised memset which appears
to be more efficient anyway; will eventually be deleted */
static inline
void mcf5249_vect_zero(ogg_int32_t *ptr, int n)
{
/* ensure ptr is aligned to 16-bytes */
while(n>0 && (int)ptr%16) {
*ptr++ = 0;
n--;
}
asm volatile ("clr.l %%d0;"
"clr.l %%d1;"
"clr.l %%d2;"
"clr.l %%d3;"
/* loop start */
"tst.l %[n];"
"bra 1f;"
"0: movem.l %%d0-%%d3, (%[ptr]);"
"lea (4*4, %[ptr]), %[ptr];"
"subq.l #4, %[n];" /* done 4 elements */
"1: bgt 0b;"
: [n] "+d" (n), [ptr] "+a" (ptr)
:
: "%d0","%d1","%d2","%d3","cc","memory");
/* clear remaining elements */
while(n>0) {
*ptr++ = 0;
n--;
}
}
#endif
#endif
#ifndef _V_CLIP_MATH
#define _V_CLIP_MATH
/* this is portable C and simple; why not use this as default? */
static inline ogg_int32_t CLIP_TO_15(register ogg_int32_t x) {
register ogg_int32_t hi=32767, lo=-32768;
return (x>=hi ? hi : (x<=lo ? lo : x));
}
#endif
#else
#define LINE_ATTR
#endif