Mercurial > SDL_sound_CoreAudio

diff decoders/libmpg123/dct64_sse.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/dct64_sse.S	Fri Jan 30 02:44:47 2009 -0500
@@ -0,0 +1,567 @@
+/*
+	dct64_sse: MMX/SSE optimized dct64
+
+	copyright 2006-2007 by Zuxy Meng <zuxy.meng@gmail.com> / 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 the mysterious higway for MMX (apparently)
+	then developed into SSE opt by Zuxy Meng, also building on Romain Dolbeau's AltiVec
+	Both have agreed to distribution under LGPL 2.1 .
+
+	Transformed back into standalone asm, with help of
+	gcc -S -DHAVE_CONFIG_H -I.  -march=pentium3 -O3 -Wall -pedantic -fno-strict-aliasing -DREAL_IS_FLOAT -c -o dct64_sse.{S,c}
+
+	Original comment from MPlayer source follows:
+*/
+
+/*
+ * Discrete Cosine Tansform (DCT) for SSE
+ * based upon code from mp3lib/dct64.c, mp3lib/dct64_altivec.c
+ * and mp3lib/dct64_MMX.c
+ */
+
+#include "mangle.h"
+
+#ifndef __APPLE__
+	.section	.rodata
+#else
+	.data
+#endif
+	ALIGN16
+	/* .type	nnnn, @object
+	   .size	nnnn, 16 */
+nnnn:
+	.long	-2147483648
+	.long	-2147483648
+	.long	-2147483648
+	.long	-2147483648
+	ALIGN16
+	/* .type	ppnn, @object
+	   .size	ppnn, 16 */
+ppnn:
+	.long	0
+	.long	0
+	.long	-2147483648
+	.long	-2147483648
+	ALIGN16
+	/* .type	pnpn, @object
+	   .size	pnpn, 16 */
+pnpn:
+	.long	0
+	.long	-2147483648
+	.long	0
+	.long	-2147483648
+	ALIGN4
+	/* .type	one.4748, @object
+	   .size	one.4748, 4 */
+one.4748:
+	.long	1065353216
+
+	.text
+	ALIGN16,,15
+.globl ASM_NAME(dct64_sse)
+	/* .type	ASM_NAME(dct64_sse), @function */
+ASM_NAME(dct64_sse):
+	pushl	%ebp
+	movl	%esp, %ebp
+	/* stack from ebp: 0=ebp 4=back 8=arg0 12=arg1 16=arg2 */
+#define ARG(n) (8+n*4)(%ebp)
+	andl	$-16, %esp /* align the stack at 16 bytes */
+	subl	$256, %esp /* reserve space for local b1 and b2 */
+	pushl	%ebx
+/* stack from esp: 0=ebx 4...131=b2 132...259=b1 */
+#define B1OFF 132
+#define B2OFF 4
+#define B1(n) (B1OFF+n)(%esp)
+#define B2(n) (B2OFF+n)(%esp)
+
+	movl	ARG(2), %eax
+	movl	ARG(0), %ecx
+/* APP */
+/* for (i = 0; i < 0x20 / 2; i += 4) cycle 1 */
+	movaps    ASM_NAME(costab_mmxsse), %xmm3
+	shufps    $27, %xmm3, %xmm3
+	MOVUAPS    (%eax), %xmm1
+	movaps    %xmm1, %xmm4
+	MOVUAPS    112(%eax), %xmm2
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm2, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B1(0)
+	subps     %xmm2, %xmm4
+	mulps     %xmm3, %xmm4
+	movaps    %xmm4, B1(112)
+	
+/* NO_APP */
+	movl	ARG(1), %ebx
+/* APP */
+/* for (i = 0; i < 0x20 / 2; i += 4) cycle 2 */
+	movaps    ASM_NAME(costab_mmxsse)+16, %xmm3
+	shufps    $27, %xmm3, %xmm3
+	MOVUAPS    16(%eax), %xmm1
+	movaps    %xmm1, %xmm4
+	MOVUAPS    96(%eax), %xmm2
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm2, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B1(16)
+	subps     %xmm2, %xmm4
+	mulps     %xmm3, %xmm4
+	movaps    %xmm4, B1(96)
+	
+/* for (i = 0; i < 0x20 / 2; i += 4) cycle 3 */
+	movaps    ASM_NAME(costab_mmxsse)+32, %xmm3
+	shufps    $27, %xmm3, %xmm3
+	MOVUAPS    32(%eax), %xmm1
+	movaps    %xmm1, %xmm4
+	MOVUAPS    80(%eax), %xmm2
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm2, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B1(32)
+	subps     %xmm2, %xmm4
+	mulps     %xmm3, %xmm4
+	movaps    %xmm4, B1(80)
+	
+/* for (i = 0; i < 0x20 / 2; i += 4) cycle 4 */
+	movaps    ASM_NAME(costab_mmxsse)+48, %xmm3
+	shufps    $27, %xmm3, %xmm3
+	MOVUAPS    48(%eax), %xmm1
+	movaps    %xmm1, %xmm4
+	MOVUAPS    64(%eax), %xmm2
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm2, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B1(48)
+	subps     %xmm2, %xmm4
+	mulps     %xmm3, %xmm4
+	movaps    %xmm4, B1(64)
+	
+	movaps    B1(0), %xmm1
+	movaps    B1(16), %xmm3
+	movaps    B1(32), %xmm4
+	movaps    B1(48), %xmm6
+	movaps    %xmm1, %xmm7
+	shufps    $27, %xmm7, %xmm7
+	movaps    %xmm3, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	movaps    %xmm4, %xmm2
+	shufps    $27, %xmm2, %xmm2
+	movaps    %xmm6, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B2(0)
+	addps     %xmm2, %xmm3
+	movaps    %xmm3, B2(16)
+	subps     %xmm4, %xmm5
+	movaps    %xmm5, B2(32)
+	subps     %xmm6, %xmm7
+	movaps    %xmm7, B2(48)
+	
+	movaps    B1(64), %xmm1
+	movaps    B1(80), %xmm3
+	movaps    B1(96), %xmm4
+	movaps    B1(112), %xmm6
+	movaps    %xmm1, %xmm7
+	shufps    $27, %xmm7, %xmm7
+	movaps    %xmm3, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	movaps    %xmm4, %xmm2
+	shufps    $27, %xmm2, %xmm2
+	movaps    %xmm6, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	addps     %xmm0, %xmm1
+	movaps    %xmm1, B2(64)
+	addps     %xmm2, %xmm3
+	movaps    %xmm3, B2(80)
+	subps     %xmm4, %xmm5
+	movaps    %xmm5, B2(96)
+	subps     %xmm6, %xmm7
+	movaps    %xmm7, B2(112)
+	
+	movaps    B2(32), %xmm0
+	movaps    B2(48), %xmm1
+	movaps    ASM_NAME(costab_mmxsse)+64, %xmm4
+	xorps     %xmm6, %xmm6
+	shufps    $27, %xmm4, %xmm4
+	mulps     %xmm4, %xmm1
+	movaps    ASM_NAME(costab_mmxsse)+80, %xmm2
+	xorps     %xmm7, %xmm7
+	shufps    $27, %xmm2, %xmm2
+	mulps     %xmm2, %xmm0
+	movaps    %xmm0, B2(32)
+	movaps    %xmm1, B2(48)
+	movaps    B2(96), %xmm3
+	mulps     %xmm2, %xmm3
+	subps     %xmm3, %xmm6
+	movaps    %xmm6, B2(96)
+	movaps    B2(112), %xmm5
+	mulps     %xmm4, %xmm5
+	subps     %xmm5, %xmm7
+	movaps    %xmm7, B2(112)
+	
+	movaps    ASM_NAME(costab_mmxsse)+96, %xmm0
+	shufps    $27, %xmm0, %xmm0
+	movaps    nnnn, %xmm5
+	movaps    %xmm5, %xmm6
+	
+	movaps    B2(0), %xmm2
+	movaps    B2(16), %xmm3
+	movaps    %xmm2, %xmm4
+	xorps     %xmm5, %xmm6
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm3, %xmm1
+	shufps    $27, %xmm1, %xmm1
+	addps     %xmm1, %xmm2
+	movaps    %xmm2, B1(0)
+	subps     %xmm3, %xmm4
+	xorps     %xmm6, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B1(16)
+	
+	movaps    B2(32), %xmm2
+	movaps    B2(48), %xmm3
+	movaps    %xmm2, %xmm4
+	xorps     %xmm5, %xmm6
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm3, %xmm1
+	shufps    $27, %xmm1, %xmm1
+	addps     %xmm1, %xmm2
+	movaps    %xmm2, B1(32)
+	subps     %xmm3, %xmm4
+	xorps     %xmm6, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B1(48)
+	
+	movaps    B2(64), %xmm2
+	movaps    B2(80), %xmm3
+	movaps    %xmm2, %xmm4
+	xorps     %xmm5, %xmm6
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm3, %xmm1
+	shufps    $27, %xmm1, %xmm1
+	addps     %xmm1, %xmm2
+	movaps    %xmm2, B1(64)
+	subps     %xmm3, %xmm4
+	xorps     %xmm6, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B1(80)
+	
+	movaps    B2(96), %xmm2
+	movaps    B2(112), %xmm3
+	movaps    %xmm2, %xmm4
+	xorps     %xmm5, %xmm6
+	shufps    $27, %xmm4, %xmm4
+	movaps    %xmm3, %xmm1
+	shufps    $27, %xmm1, %xmm1
+	addps     %xmm1, %xmm2
+	movaps    %xmm2, B1(96)
+	subps     %xmm3, %xmm4
+	xorps     %xmm6, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B1(112)
+	
+	movss     one.4748, %xmm1
+	movss     ASM_NAME(costab_mmxsse)+112, %xmm0
+	movaps    %xmm1, %xmm3
+	unpcklps  %xmm0, %xmm3
+	movss     ASM_NAME(costab_mmxsse)+116, %xmm2
+	movaps    %xmm1, %xmm0
+	unpcklps  %xmm2, %xmm0
+	unpcklps  %xmm3, %xmm0
+	movaps    ppnn, %xmm2
+	
+	movaps    B1(0), %xmm3
+	movaps    %xmm3, %xmm4
+	shufps    $20, %xmm4, %xmm4
+	shufps    $235, %xmm3, %xmm3
+	xorps     %xmm2, %xmm3
+	addps     %xmm3, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B2(0)
+	movaps    B1(16), %xmm6
+	movaps    %xmm6, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	xorps     %xmm2, %xmm5
+	addps     %xmm5, %xmm6
+	mulps     %xmm0, %xmm6
+	movaps    %xmm6, B2(16)
+	
+	movaps    B1(32), %xmm3
+	movaps    %xmm3, %xmm4
+	shufps    $20, %xmm4, %xmm4
+	shufps    $235, %xmm3, %xmm3
+	xorps     %xmm2, %xmm3
+	addps     %xmm3, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B2(32)
+	movaps    B1(48), %xmm6
+	movaps    %xmm6, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	xorps     %xmm2, %xmm5
+	addps     %xmm5, %xmm6
+	mulps     %xmm0, %xmm6
+	movaps    %xmm6, B2(48)
+	
+	movaps    B1(64), %xmm3
+	movaps    %xmm3, %xmm4
+	shufps    $20, %xmm4, %xmm4
+	shufps    $235, %xmm3, %xmm3
+	xorps     %xmm2, %xmm3
+	addps     %xmm3, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B2(64)
+	movaps    B1(80), %xmm6
+	movaps    %xmm6, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	xorps     %xmm2, %xmm5
+	addps     %xmm5, %xmm6
+	mulps     %xmm0, %xmm6
+	movaps    %xmm6, B2(80)
+	
+	movaps    B1(96), %xmm3
+	movaps    %xmm3, %xmm4
+	shufps    $20, %xmm4, %xmm4
+	shufps    $235, %xmm3, %xmm3
+	xorps     %xmm2, %xmm3
+	addps     %xmm3, %xmm4
+	mulps     %xmm0, %xmm4
+	movaps    %xmm4, B2(96)
+	movaps    B1(112), %xmm6
+	movaps    %xmm6, %xmm5
+	shufps    $27, %xmm5, %xmm5
+	xorps     %xmm2, %xmm5
+	addps     %xmm5, %xmm6
+	mulps     %xmm0, %xmm6
+	movaps    %xmm6, B2(112)
+	
+	movss     ASM_NAME(costab_mmxsse)+120, %xmm0
+	movaps    %xmm1, %xmm2
+	movaps    %xmm0, %xmm7
+	unpcklps  %xmm1, %xmm2
+	unpcklps  %xmm0, %xmm7
+	movaps    pnpn, %xmm0
+	unpcklps  %xmm7, %xmm2
+	
+	movaps    B2(32), %xmm1
+	movaps    %xmm1, %xmm3
+	shufps    $224, %xmm3, %xmm3
+	shufps    $181, %xmm1, %xmm1
+	xorps     %xmm0, %xmm1
+	addps     %xmm1, %xmm3
+	mulps     %xmm2, %xmm3
+	movaps    %xmm3, B1(32)
+	movaps    B2(48), %xmm4
+	movaps    %xmm4, %xmm5
+	shufps    $224, %xmm5, %xmm5
+	shufps    $181, %xmm4, %xmm4
+	xorps     %xmm0, %xmm4
+	addps     %xmm4, %xmm5
+	mulps     %xmm2, %xmm5
+	movaps    %xmm5, B1(48)
+	
+	movaps    B2(64), %xmm1
+	movaps    %xmm1, %xmm3
+	shufps    $224, %xmm3, %xmm3
+	shufps    $181, %xmm1, %xmm1
+	xorps     %xmm0, %xmm1
+	addps     %xmm1, %xmm3
+	mulps     %xmm2, %xmm3
+	movaps    %xmm3, B1(64)
+	movaps    B2(80), %xmm4
+	movaps    %xmm4, %xmm5
+	shufps    $224, %xmm5, %xmm5
+	shufps    $181, %xmm4, %xmm4
+	xorps     %xmm0, %xmm4
+	addps     %xmm4, %xmm5
+	mulps     %xmm2, %xmm5
+	movaps    %xmm5, B1(80)
+	
+	movaps    B2(96), %xmm1
+	movaps    %xmm1, %xmm3
+	shufps    $224, %xmm3, %xmm3
+	shufps    $181, %xmm1, %xmm1
+	xorps     %xmm0, %xmm1
+	addps     %xmm1, %xmm3
+	mulps     %xmm2, %xmm3
+	movaps    %xmm3, B1(96)
+	movaps    B2(112), %xmm4
+	movaps    %xmm4, %xmm5
+	shufps    $224, %xmm5, %xmm5
+	shufps    $181, %xmm4, %xmm4
+	xorps     %xmm0, %xmm4
+	addps     %xmm4, %xmm5
+	mulps     %xmm2, %xmm5
+	movaps    %xmm5, B1(112)
+	
+/* NO_APP */
+	flds	B1(40)
+	movl	%esp, %edx
+	addl	$B1OFF, %edx
+	movl	%esp, %eax
+	addl	$B2OFF, %eax
+	fadds	B1(44)
+	fstps	B1(40)
+	flds	B1(56)
+	fadds	B1(60)
+	flds	B1(48)
+	fadd	%st(1), %st
+	fstps	B1(48)
+	fadds	B1(52)
+	fstps	B1(56)
+	flds	B1(52)
+	fadds	B1(60)
+	fstps	B1(52)
+	flds	B1(72)
+	fadds	B1(76)
+	fstps	B1(72)
+	flds	B1(88)
+	fadds	B1(92)
+	flds	B1(80)
+	fadd	%st(1), %st
+	fstps	B1(80)
+	fadds	B1(84)
+	fstps	B1(88)
+	flds	B1(84)
+	fadds	B1(92)
+	fstps	B1(84)
+	flds	B1(104)
+	fadds	B1(108)
+	fstps	B1(104)
+	flds	B1(120)
+	fadds	B1(124)
+	flds	B1(112)
+	fadd	%st(1), %st
+	fstps	B1(112)
+	fadds	B1(116)
+	fstps	B1(120)
+	flds	B1(116)
+	fadds	B1(124)
+	fstps	B1(116)
+/* APP */
+	flds       ASM_NAME(costab_mmxsse)+120
+	flds     (%eax)
+	fadds   4(%eax)
+	fistp 512(%ecx)
+	flds     (%eax)
+	fsubs   4(%eax)
+	fmul  %st(1)
+	fistp    (%ecx)
+	flds   12(%eax)
+	fsubs   8(%eax)
+	fmul  %st(1)
+	fist  256(%ebx)
+	fadds  12(%eax)
+	fadds   8(%eax)
+	fistp 256(%ecx)
+	flds   16(%eax)
+	fsubs  20(%eax)
+	fmul  %st(1)
+	flds   28(%eax)
+	fsubs  24(%eax)
+	fmul  %st(2)
+	fist  384(%ebx)
+	fld   %st(0)
+	fadds  24(%eax)
+	fadds  28(%eax)
+	fld   %st(0)
+	fadds  16(%eax)
+	fadds  20(%eax)
+	fistp 384(%ecx)
+	fadd  %st(2)
+	fistp 128(%ecx)
+	faddp %st(1)
+	fistp 128(%ebx)
+	flds   32(%edx)
+	fadds  48(%edx)
+	fistp 448(%ecx)
+	flds   48(%edx)
+	fadds  40(%edx)
+	fistp 320(%ecx)
+	flds   40(%edx)
+	fadds  56(%edx)
+	fistp 192(%ecx)
+	flds   56(%edx)
+	fadds  36(%edx)
+	fistp  64(%ecx)
+	flds   36(%edx)
+	fadds  52(%edx)
+	fistp  64(%ebx)
+	flds   52(%edx)
+	fadds  44(%edx)
+	fistp 192(%ebx)
+	flds   60(%edx)
+	fist  448(%ebx)
+	fadds  44(%edx)
+	fistp 320(%ebx)
+	flds   96(%edx)
+	fadds 112(%edx)
+	fld   %st(0)
+	fadds  64(%edx)
+	fistp 480(%ecx)
+	fadds  80(%edx)
+	fistp 416(%ecx)
+	flds  112(%edx)
+	fadds 104(%edx)
+	fld   %st(0)
+	fadds  80(%edx)
+	fistp 352(%ecx)
+	fadds  72(%edx)
+	fistp 288(%ecx)
+	flds  104(%edx)
+	fadds 120(%edx)
+	fld   %st(0)
+	fadds  72(%edx)
+	fistp 224(%ecx)
+	fadds  88(%edx)
+	fistp 160(%ecx)
+	flds  120(%edx)
+	fadds 100(%edx)
+	fld   %st(0)
+	fadds  88(%edx)
+	fistp  96(%ecx)
+	fadds  68(%edx)
+	fistp  32(%ecx)
+	flds  100(%edx)
+	fadds 116(%edx)
+	fld   %st(0)
+	fadds  68(%edx)
+	fistp  32(%ebx)
+	fadds  84(%edx)
+	fistp  96(%ebx)
+	flds  116(%edx)
+	fadds 108(%edx)
+	fld   %st(0)
+	fadds  84(%edx)
+	fistp 160(%ebx)
+	fadds  76(%edx)
+	fistp 224(%ebx)
+	flds  108(%edx)
+	fadds 124(%edx)
+	fld   %st(0)
+	fadds  76(%edx)
+	fistp 288(%ebx)
+	fadds  92(%edx)
+	fistp 352(%ebx)
+	flds  124(%edx)
+	fist  480(%ebx)
+	fadds  92(%edx)
+	fistp 416(%ebx)
+	ffreep %st(0)
+	
+/* NO_APP */
+	movzwl	(%ecx), %eax
+	movw	%ax, (%ebx)
+	popl	%ebx
+	movl	%ebp, %esp
+	popl	%ebp
+	ret
+	/* .size	ASM_NAME(dct64_sse), .-ASM_NAME(dct64_sse) */
+
+/* Mark non-executable stack. */
+#if defined(__linux__) && defined(__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif