/*

	decode_3dnow.s - 3DNow! optimized synth_1to1()

	copyright ?-2007 by the mpg123 project - free software under the terms of the LGPL 2.1

	see COPYING and AUTHORS files in distribution or http://mpg123.org

	initially written by Syuuhei Kashiyama

	This code based 'decode_3dnow.s' by Syuuhei Kashiyama

	<squash@mb.kcom.ne.jp>,only two types of changes have been made:

	- remove PREFETCH instruction for speedup

	- change function name for support 3DNow! automatic detect

	- femms moved to before 'call dct64_3dnow'

	You can find Kashiyama's original 3dnow! support patch

	(for mpg123-0.59o) at

	http://user.ecc.u-tokyo.ac.jp/~g810370/linux-simd/ (Japanese).

	by KIMURA Takuhiro <kim@hannah.ipc.miyakyo-u.ac.jp> - until 31.Mar.1999

                  	<kim@comtec.co.jp>               - after  1.Apr.1999

	Replacement of synth_1to1() with AMD's 3DNow! SIMD operations support

	Syuuhei Kashiyama <squash@mb.kcom.ne.jp>

	The author of this program disclaim whole expressed or implied

	warranties with regard to this program, and in no event shall the

	author of this program liable to whatever resulted from the use of

	this program. Use it at your own risk.

*/

#include "mangle.h"

#ifdef ACCURATE_ROUNDING

#ifndef __APPLE__

	.section	.rodata

#else

	.data

#endif

	ALIGN8

max_s16:

	.long   1191181824 /* 32767.0 */

	.long   1191181824

min_s16:

	.long   -956301312 /* -32768.0 */

	.long   -956301312

ftoi_magic:

	.long	1262485504 /* 2^23 + 2^22 */

	.long	1262485504

#endif

	.text

	ALIGN16

.globl ASM_NAME(synth_1to1_3dnow_asm)

/* int synth_1to1_3dnow_asm(real *bandPtr, int channel, unsigned char *out, unsigned char *buffs, int *bo, real *decwin); */

ASM_NAME(synth_1to1_3dnow_asm):

	subl $24,%esp

	pushl %ebp

	pushl %edi

	xorl %ebp,%ebp

	pushl %esi

	pushl %ebx

/* stack old: 0=ebx 4=esi 8=edi 12=ebp 16,20,24,28,32,36=local 40=back 44=bandptr 48=channel 52=out 56=pnt */

/* stack new: 0=ebx 4=esi 8=edi 12=ebp 16,20,24,28,32,36=local 40=back 44=bandptr 48=channel 52=out 56=buffs 60=bo 64=decwin */

#define OUT     52(%esp)

#define CHANNEL 48(%esp)

#define BANDPTR 44(%esp)

#define BUFFS   56(%esp)

#define BO      60(%esp)

#define DECWIN  64(%esp)

#define LOCAL0  16(%esp)

#define LOCAL1  20(%esp)

#define EBXSAVE 24(%esp)

#define LOCAL5  36(%esp)

#ifdef ACCURATE_ROUNDING

	#undef _EBX_

	#define _EBX_ %eax

	GET_GOT

	/* FIXME */

#if PIC

	movl _EBX_, EBXSAVE

#endif

#endif

	movl OUT,%esi

	movl %esi,LOCAL0 /* save buffer start (samples pointer) to another local var */

	movl CHANNEL,%ebx

	movl BO,%esi     /* bo address */

	movl (%esi),%edx /* bo value */

	femms

	testl %ebx,%ebx

	jne .L26

/* if(!channel) */

	decl %edx   /* --bo */

	andl $15,%edx

	movl %edx,(%esi) /* save bo */

	movl BUFFS,%ecx

	jmp .L27

.L26: /* if(channel) */

	addl $2,LOCAL0   /* samples++ */

	movl BUFFS,%ecx

	addl $2176,%ecx

.L27:

/* edx (and it's lower end) still holds bo value */

	testb $1,%dl  /* bo & 0x1 */

	je .L28

	movl %edx,LOCAL5

	movl %ecx,%ebx

	movl BANDPTR,%esi

	movl %edx,%edi

	pushl %esi

	sall $2,%edi

	movl %ebx,%eax

	movl %edi,24(%esp) /* LOCAL1, actually */

	addl %edi,%eax

	pushl %eax

	movl %edx,%eax

	incl %eax

	andl $15,%eax

	leal 1088(,%eax,4),%eax

	addl %ebx,%eax

	pushl %eax

	call FUNC(dct64_3dnow)

	addl $12,%esp

	jmp .L29

.L28:

	leal 1(%edx),%esi

	movl BANDPTR,%edi

	movl %esi,LOCAL5

	leal 1092(%ecx,%edx,4),%eax

	pushl %edi

	leal 1088(%ecx),%ebx

	pushl %eax

	sall $2,%esi

	leal (%ecx,%edx,4),%eax

	pushl %eax

	call FUNC(dct64_3dnow)

	addl $12,%esp

	movl %esi,LOCAL1

.L29:

	movl DECWIN,%edx

	addl $64,%edx

	movl $16,%ecx

	subl LOCAL1,%edx

	movl LOCAL0,%edi

	pcmpeqb %mm7,%mm7

	pslld $31,%mm7

	movq (%edx),%mm0

	movq (%ebx),%mm1

	ALIGN32

.L33:

#if defined(ACCURATE_ROUNDING) && defined(PIC)

	movl EBXSAVE, _EBX_

#endif

	movq 8(%edx),%mm3

	pfmul %mm1,%mm0

	movq 8(%ebx),%mm4

	movq 16(%edx),%mm5

	pfmul %mm4,%mm3

	movq 16(%ebx),%mm6

	pfadd %mm3,%mm0

	movq 24(%edx),%mm1

	pfmul %mm6,%mm5

	movq 24(%ebx),%mm2

	pfadd %mm5,%mm0

	movq 32(%edx),%mm3

	pfmul %mm2,%mm1

	movq 32(%ebx),%mm4

	pfadd %mm1,%mm0

	movq 40(%edx),%mm5

	pfmul %mm4,%mm3

	movq 40(%ebx),%mm6

	pfadd %mm3,%mm0

	movq 48(%edx),%mm1

	pfmul %mm6,%mm5

	movq 48(%ebx),%mm2

	pfadd %mm0,%mm5

	movq 56(%edx),%mm3

	pfmul %mm1,%mm2

	movq 56(%ebx),%mm4

	pfadd %mm5,%mm2

	addl $64,%ebx

	subl $-128,%edx

	movq (%edx),%mm0

	pfmul %mm4,%mm3

	movq (%ebx),%mm1

	pfadd %mm3,%mm2

	movq %mm2,%mm3

	psrlq $32,%mm3

	pfsub %mm3,%mm2

	incl %ebp

#ifdef ACCURATE_ROUNDING

	pfmin LOCAL_VAR(max_s16),%mm2

	pfmax LOCAL_VAR(min_s16),%mm2

	pfadd LOCAL_VAR(ftoi_magic),%mm2

#else

	pf2id %mm2,%mm2

	packssdw %mm2,%mm2

#endif

	movd %mm2,%eax

	movw %ax,0(%edi)

	addl $4,%edi

	decl %ecx

	jnz .L33

#if defined(ACCURATE_ROUNDING) && defined(PIC)

	movl EBXSAVE, _EBX_

#endif

	movd (%ebx),%mm0

	movd (%edx),%mm1

	punpckldq 8(%ebx),%mm0

	punpckldq 8(%edx),%mm1

	movd 16(%ebx),%mm3

	movd 16(%edx),%mm4

	pfmul %mm1,%mm0

	punpckldq 24(%ebx),%mm3

	punpckldq 24(%edx),%mm4

	movd 32(%ebx),%mm5

	movd 32(%edx),%mm6

	pfmul %mm4,%mm3

	punpckldq 40(%ebx),%mm5

	punpckldq 40(%edx),%mm6

	pfadd %mm3,%mm0

	movd 48(%ebx),%mm1

	movd 48(%edx),%mm2

	pfmul %mm6,%mm5

	punpckldq 56(%ebx),%mm1

	punpckldq 56(%edx),%mm2

	pfadd %mm5,%mm0

	pfmul %mm2,%mm1

	pfadd %mm1,%mm0

	pfacc %mm1,%mm0

#ifdef ACCURATE_ROUNDING

	pfmin LOCAL_VAR(max_s16),%mm0

	pfmax LOCAL_VAR(min_s16),%mm0

	pfadd LOCAL_VAR(ftoi_magic),%mm0

#else

	pf2id %mm0,%mm0

	packssdw %mm0,%mm0

#endif

	movd %mm0,%eax

	movw %ax,0(%edi)

	incl %ebp

	movl LOCAL5,%esi

	addl $-64,%ebx

	movl $15,%ebp

	addl $4,%edi

	leal -128(%edx,%esi,8),%edx

	movl $15,%ecx

	movd (%ebx),%mm0

	movd -4(%edx),%mm1

	punpckldq 4(%ebx),%mm0

	punpckldq -8(%edx),%mm1

	ALIGN32

.L46:

#if defined(ACCURATE_ROUNDING) && defined(PIC)

	movl EBXSAVE, _EBX_

#endif

	movd 8(%ebx),%mm3

	movd -12(%edx),%mm4

	pfmul %mm1,%mm0

	punpckldq 12(%ebx),%mm3

	punpckldq -16(%edx),%mm4

	movd 16(%ebx),%mm5

	movd -20(%edx),%mm6

	pfmul %mm4,%mm3

	punpckldq 20(%ebx),%mm5

	punpckldq -24(%edx),%mm6

	pfadd %mm3,%mm0

	movd 24(%ebx),%mm1

	movd -28(%edx),%mm2

	pfmul %mm6,%mm5

	punpckldq 28(%ebx),%mm1

	punpckldq -32(%edx),%mm2

	pfadd %mm5,%mm0

	movd 32(%ebx),%mm3

	movd -36(%edx),%mm4

	pfmul %mm2,%mm1

	punpckldq 36(%ebx),%mm3

	punpckldq -40(%edx),%mm4

	pfadd %mm1,%mm0

	movd 40(%ebx),%mm5

	movd -44(%edx),%mm6

	pfmul %mm4,%mm3

	punpckldq 44(%ebx),%mm5

	punpckldq -48(%edx),%mm6

	pfadd %mm3,%mm0

	movd 48(%ebx),%mm1

	movd -52(%edx),%mm2

	pfmul %mm6,%mm5

	punpckldq 52(%ebx),%mm1

	punpckldq -56(%edx),%mm2

	pfadd %mm0,%mm5

	movd 56(%ebx),%mm3

	movd -60(%edx),%mm4

	pfmul %mm2,%mm1

	punpckldq 60(%ebx),%mm3

	punpckldq (%edx),%mm4

	pfadd %mm1,%mm5

	addl $-128,%edx

	addl $-64,%ebx

	movd (%ebx),%mm0

	movd -4(%edx),%mm1

	pfmul %mm4,%mm3

	punpckldq 4(%ebx),%mm0

	punpckldq -8(%edx),%mm1

	pfadd %mm5,%mm3

	pfacc %mm3,%mm3

	incl %ebp

	pxor %mm7,%mm3

#ifdef ACCURATE_ROUNDING

	pfmin LOCAL_VAR(max_s16),%mm3

	pfmax LOCAL_VAR(min_s16),%mm3

	pfadd LOCAL_VAR(ftoi_magic),%mm3

#else

	pf2id %mm3,%mm3

	packssdw %mm3,%mm3

#endif

	movd %mm3,%eax

	movw %ax,(%edi)

	addl $4,%edi

	decl %ecx

	jnz .L46

	femms

	movl %ebp,%eax

	popl %ebx

	popl %esi

	popl %edi

	popl %ebp

	addl $24,%esp

	ret

#if defined(PIC) && defined(__APPLE__)

	.section __IMPORT,__jump_table,symbol_stubs,self_modifying_code+pure_instructions,5

L_dct64_3dnow:

	.indirect_symbol ASM_NAME(dct64_3dnow)

	hlt ; hlt ; hlt ; hlt ; hlt

#endif

NONEXEC_STACK