Mercurial > SDL_sound_CoreAudio

diff decoders/libmpg123/decode_i586.S @ 562:7e08477b0fc1

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
MP3 decoder upgrade work. Ripped out SMPEG and mpglib support, replaced it with "mpg123.c" and libmpg123. libmpg123 is a much better version of mpglib, so it should solve all the problems about MP3's not seeking, or most modern MP3's not playing at all, etc. Since you no longer have to make a tradeoff with SMPEG for features, and SMPEG is basically rotting, I removed it from the project. There is still work to be done with libmpg123...there are MMX, 3DNow, SSE, Altivec, etc decoders which we don't have enabled at the moment, and the build system could use some work to make this compile more cleanly, etc. Still: huge win.
author Ryan C. Gordon <icculus@icculus.org>
date Fri, 30 Jan 2009 02:44:47 -0500
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/decoders/libmpg123/decode_i586.S	Fri Jan 30 02:44:47 2009 -0500
@@ -0,0 +1,339 @@
+/*
+ decode_i586: asm synth
+
+ copyright ?-2006 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 Stefan Bieschewski
+ 
+ synth_1to1 works the same way as the c version of this
+ file.  only two types of changes have been made:
+ - reordered floating point instructions to
+   prevent pipline stalls
+ - made WRITE_SAMPLE use integer instead of
+   (slower) floating point
+ all kinds of x86 processors should benefit from these
+ modifications.
+
+ useful sources of information on optimizing x86 code include:
+
+     Intel Architecture Optimization Manual
+     http://www.intel.com/design/pentium/manuals/242816.htm
+
+     Cyrix 6x86 Instruction Set Summary
+     ftp://ftp.cyrix.com/6x86/6x-dbch6.pdf
+
+     AMD-K5 Processor Software Development
+     http://www.amd.com/products/cpg/techdocs/appnotes/20007e.pdf
+
+ Stefan Bieschewski <stb@acm.org>
+
+ $Id: decode_i586.s 1 2004-09-18 13:30:08Z thomas $
+*/
+
+#include "mangle.h"
+
+.data
+#ifndef __APPLE__
+.section .rodata
+#endif
+	ALIGN8
+.LC0:
+	.long 0x0,0x40dfffc0
+	ALIGN8
+.LC1:
+	.long 0x0,0xc0e00000
+	ALIGN8
+.text
+/* int synth_1to1_i586_asm(real *bandPtr, int channel, unsigned char *out, unsigned char *buffs, int *bo, real *decwin); */
+.globl ASM_NAME(synth_1to1_i586_asm)
+ASM_NAME(synth_1to1_i586_asm):
+	subl $12,%esp
+	pushl %ebp
+	pushl %edi
+	pushl %esi
+	pushl %ebx
+/* stack: 0=ebx, 4=esi, 8=edi, 12=ebp, 16,20,24=local, 28=back, 32=bandPtr, 36=channel, 40=out, 44=buffs, 48=bo, 52=decwin */
+	movl 32(%esp),%eax /* *bandPtr */
+	movl 40(%esp),%esi /* *out */
+	movl 48(%esp),%edi /* *bo */
+	movl (%edi),%ebp   /* store bo value in ebp */
+	xorl %edi,%edi
+	cmpl %edi,36(%esp)
+	jne .L48           /* if(!channel) */
+	decl %ebp          /* bo-- */
+	andl $15,%ebp      /* bo &= 0xf */
+	movl 48(%esp),	%edi /* *bo */
+	movl %ebp,(%edi)   /* write back bo */
+	xorl %edi,%edi     /* restore %edi to 0; it's used later */
+	movl 44(%esp),%ecx /* use buffs */
+	jmp .L49
+.L48: /* if(channel) use buffs+2176 */
+	addl $2,%esi
+	movl 44(%esp),%ecx /* *buffs */
+	addl $2176,%ecx
+.L49:
+	testl $1,%ebp
+	je .L50
+	movl %ecx,%ebx
+	movl %ebp,16(%esp)
+	pushl %eax
+	movl 20(%esp),%edx
+	leal (%ebx,%edx,4),%eax
+	pushl %eax
+	movl 24(%esp),%eax
+	incl %eax
+	andl $15,%eax
+	leal 1088(,%eax,4),%eax
+	addl %ebx,%eax
+	jmp .L74
+.L50:
+	leal 1088(%ecx),%ebx
+	leal 1(%ebp),%edx
+	movl %edx,16(%esp)
+	pushl %eax
+	leal 1092(%ecx,%ebp,4),%eax
+	pushl %eax
+	leal (%ecx,%ebp,4),%eax
+.L74:
+	pushl %eax
+	call ASM_NAME(dct64_i386)
+	addl $12,%esp
+/* stack now back on track */
+	movl 16(%esp),%edx
+	leal 0(,%edx,4),%edx
+	movl 52(%esp),%eax /* decwin */
+	addl $64,%eax
+	movl %eax,%ecx
+	subl %edx,%ecx
+	movl $16,%ebp
+.L55:
+	flds (%ecx)
+	fmuls (%ebx)
+	flds 4(%ecx)
+	fmuls 4(%ebx)
+	fxch %st(1)
+	flds 8(%ecx)
+	fmuls 8(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 12(%ecx)
+	fmuls 12(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 16(%ecx)
+	fmuls 16(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 20(%ecx)
+	fmuls 20(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 24(%ecx)
+	fmuls 24(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 28(%ecx)
+	fmuls 28(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 32(%ecx)
+	fmuls 32(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 36(%ecx)
+	fmuls 36(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 40(%ecx)
+	fmuls 40(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 44(%ecx)
+	fmuls 44(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 48(%ecx)
+	fmuls 48(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 52(%ecx)
+	fmuls 52(%ebx)
+	fxch %st(2)         
+	faddp %st,%st(1)
+	flds 56(%ecx)
+	fmuls 56(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds 60(%ecx)
+	fmuls 60(%ebx)
+	fxch %st(2)
+	subl $4,%esp
+	faddp %st,%st(1)
+	fxch %st(1)
+	fsubrp %st,%st(1)
+	fistpl (%esp)
+	popl %eax
+	cmpl $32767,%eax
+	jg 1f
+	cmpl $-32768,%eax
+	jl 2f
+	movw %ax,(%esi)
+	jmp 4f
+1:	movw $32767,(%esi)
+	jmp 3f
+2:	movw $-32768,(%esi)
+3:	incl %edi
+4:
+.L54:
+	addl $64,%ebx
+	subl $-128,%ecx
+	addl $4,%esi
+	decl %ebp
+	jnz .L55
+	flds (%ecx)
+	fmuls (%ebx)
+	flds 8(%ecx)
+	fmuls 8(%ebx)
+	flds 16(%ecx)
+	fmuls 16(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 24(%ecx)
+	fmuls 24(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 32(%ecx)
+	fmuls 32(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 40(%ecx)
+	fmuls 40(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 48(%ecx)
+	fmuls 48(%ebx)
+	fxch %st(2)
+	faddp %st,%st(1)
+	flds 56(%ecx)
+	fmuls 56(%ebx)
+	fxch %st(2)
+	subl $4,%esp
+	faddp %st,%st(1)
+	fxch %st(1)
+	faddp %st,%st(1)
+	fistpl (%esp)
+	popl %eax
+	cmpl $32767,%eax
+	jg 1f
+	cmpl $-32768,%eax
+	jl 2f
+	movw %ax,(%esi)
+	jmp 4f
+1:	movw $32767,(%esi)
+	jmp 3f
+2:	movw $-32768,(%esi)
+3:	incl %edi
+4:
+.L62:
+	addl $-64,%ebx
+	addl $4,%esi
+	movl 16(%esp),%edx
+	leal -128(%ecx,%edx,8),%ecx
+	movl $15,%ebp
+.L68:
+	flds -4(%ecx)
+	fchs
+	fmuls (%ebx)
+	flds -8(%ecx)
+	fmuls 4(%ebx)
+	fxch %st(1)
+	flds -12(%ecx)
+	fmuls 8(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -16(%ecx)
+	fmuls 12(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -20(%ecx)
+	fmuls 16(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -24(%ecx)
+	fmuls 20(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -28(%ecx)
+	fmuls 24(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -32(%ecx)
+	fmuls 28(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -36(%ecx)
+	fmuls 32(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -40(%ecx)
+	fmuls 36(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -44(%ecx)
+	fmuls 40(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -48(%ecx)
+	fmuls 44(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -52(%ecx)
+	fmuls 48(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -56(%ecx)
+	fmuls 52(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds -60(%ecx)
+	fmuls 56(%ebx)
+	fxch %st(2)
+	fsubrp %st,%st(1)
+	flds (%ecx)
+	fmuls 60(%ebx)
+	fxch %st(2)
+	subl $4,%esp
+	fsubrp %st,%st(1)
+	fxch %st(1)
+	fsubrp %st,%st(1)
+	fistpl (%esp)
+	popl %eax
+	cmpl $32767,%eax
+	jg 1f
+	cmpl $-32768,%eax
+	jl 2f
+	movw %ax,(%esi)
+	jmp 4f
+1:	movw $32767,(%esi)
+	jmp 3f
+2:	movw $-32768,(%esi)
+3:	incl %edi
+4:
+.L67:
+	addl $-64,%ebx
+	addl $-128,%ecx
+	addl $4,%esi
+	decl %ebp
+	jnz .L68
+	movl %edi,%eax
+	popl %ebx
+	popl %esi
+	popl %edi
+	popl %ebp
+	addl $12,%esp
+	ret
+
+/* Mark non-executable stack. */
+#if defined(__linux__) && defined(__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif