seal 0.1.0

data/mpg123/src/libmpg123/synth_i486.c

@@ -0,0 +1,252 @@
+/*
+	decode_i486.c: i486 decode
+
+	copyright 1998-2006 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 Fabrice Bellard
+
+	One has to see if the modification for non-static memory kills this optimization (cache locality?).
+*/
+
+/* 
+ * Subband Synthesis for MPEG Audio. 
+ *
+ * Version optimized for 80486 by using integer arithmetic,
+ * multiplications by shift and add, and by increasing locality in
+ * order to fit the 8KB L1 cache. This code should be compiled with gcc
+ * 2.7.2 or higher.
+ *
+ * Note: this version does not guaranty a good accuracy. The filter
+ * coefficients are quantified on 14 bits.
+ *
+ * (c) 1998 Fabrice Bellard 
+ */
+
+#include "mpg123lib_intern.h"
+
+#define FIR16_1(pos,c0,c1,c2,c3,c4,c5,c6,c7,c8,c9,c10,c11,c12,c13,c14,c15) \
+{\
+  int sum;\
+  sum=(c0)*b0[0]+(c1)*b0[1]+(c2)*b0[2]+(c3)*b0[3]+\
+  (c4)*b0[4]+(c5)*b0[5]+(c6)*b0[6]+(c7)*b0[7]+\
+  (c8)*b0[8]+(c9)*b0[9]+(c10)*b0[10]+(c11)*b0[11]+\
+  (c12)*b0[12]+(c13)*b0[13]+(c14)*b0[14]+(c15)*b0[15];\
+  sum=(sum+(1 << 13))>>14;\
+  if (sum<-32768) sum=-32768;\
+  else if (sum>32767) sum=32767;\
+  samples[2*(pos)]=sum;\
+  b0+=FIR_BUFFER_SIZE;\
+}
+
+#define FIR16_2(pos1,c0,c1,c2,c3,c4,c5,c6,c7,c8,c9,c10,c11,c12,c13,c14,c15,\
+              pos2,d0,d1,d2,d3,d4,d5,d6,d7,d8,d9,d10,d11,d12,d13,d14,d15) \
+{\
+  int sum1,sum2,v;\
+\
+  v=b0[0];\
+  sum1=(c0)*v;\
+  sum2=(d0)*v;\
+  v=b0[1];\
+  sum1+=(c1)*v;\
+  sum2+=(d1)*v;\
+  v=b0[2];\
+  sum1+=(c2)*v;\
+  sum2+=(d2)*v;\
+  v=b0[3];\
+  sum1+=(c3)*v;\
+  sum2+=(d3)*v;\
+  v=b0[4];\
+  sum1+=(c4)*v;\
+  sum2+=(d4)*v;\
+  v=b0[5];\
+  sum1+=(c5)*v;\
+  sum2+=(d5)*v;\
+  v=b0[6];\
+  sum1+=(c6)*v;\
+  sum2+=(d6)*v;\
+  v=b0[7];\
+  sum1+=(c7)*v;\
+  sum2+=(d7)*v;\
+  v=b0[8];\
+  sum1+=(c8)*v;\
+  sum2+=(d8)*v;\
+  v=b0[9];\
+  sum1+=(c9)*v;\
+  sum2+=(d9)*v;\
+  v=b0[10];\
+  sum1+=(c10)*v;\
+  sum2+=(d10)*v;\
+  v=b0[11];\
+  sum1+=(c11)*v;\
+  sum2+=(d11)*v;\
+  v=b0[12];\
+  sum1+=(c12)*v;\
+  sum2+=(d12)*v;\
+  v=b0[13];\
+  sum1+=(c13)*v;\
+  sum2+=(d13)*v;\
+  v=b0[14];\
+  sum1+=(c14)*v;\
+  sum2+=(d14)*v;\
+  v=b0[15];\
+  sum1+=(c15)*v;\
+  sum2+=(d15)*v;\
+\
+  sum1=(sum1+(1<<13))>>14;\
+  sum2=(sum2+(1<<13))>>14;\
+\
+  if (sum1<-32768) sum1=-32768;\
+  else if (sum1>32767) sum1=32767;\
+  samples[(pos1)*2]=sum1;\
+\
+  if (sum2<-32768) sum2=-32768;\
+  else if (sum2>32767) sum2=32767;\
+  samples[(pos2)*2]=sum2;\
+  b0+=FIR_BUFFER_SIZE;\
+}
+
+int absynth_1to1_i486(real *bandPtr, int channel, mpg123_handle *fr, int nb_blocks)
+{
+  short *samples = (short *) (fr->buffer.data+fr->buffer.fill);
+  int *b0,**buf;
+  int clip = 0; 
+  int block,b,bo_start;
+
+  /* samples address */
+  samples+=channel;
+
+  bo_start=fr->i486bo[channel];
+  buf = fr->int_buffs[channel];
+
+  b=bo_start;
+  for(block=0;block<nb_blocks;block++) {
+
+    /* FIR offset */
+    b++;
+    if (b >= FIR_BUFFER_SIZE) {
+      int *p,*q;
+      int c,i,j;
+      
+      /* we shift the buffers */
+      for(c=0;c<2;c++) {
+        p=&buf[c][0]+1;
+        q=p+(FIR_BUFFER_SIZE-FIR_SIZE);
+        for(i=0;i<17;i++) {
+          for(j=0;j<FIR_SIZE-1;j++) p[j]=q[j];
+          p+=FIR_BUFFER_SIZE;
+          q+=FIR_BUFFER_SIZE;
+        }
+      }
+      /* we update 'bo' accordingly */
+      b=fr->i486bo[channel]=FIR_SIZE;
+    }
+    
+    if(b & 1) {
+      dct64_i486(buf[1]+b,buf[0]+b,bandPtr);
+    } else {
+      dct64_i486(buf[0]+b,buf[1]+b,bandPtr);
+    }
+    bandPtr+=32;
+  }
+  fr->i486bo[channel]=b;
+
+  /* filter bank: part 1 */
+  b=bo_start;
+  for(block=0;block<nb_blocks;block++) {
+    b++;
+    if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
+    if(b & 1) {
+      b0 = buf[0] + b - (FIR_SIZE-1);
+    } else {
+      b0 = buf[1] + b - (FIR_SIZE-1);
+    }
+
+ FIR16_1(0,-7,53,-114,509,-1288,1643,-9372,18759,9372,1643,1288,509,114,53,7,0);
+ FIR16_2(1,-6,52,-100,515,-1197,1783,-8910,18748,9834,1489,1379,500,129,54,7,0,
+ 31,0,-7,54,-129,500,-1379,1489,-9834,18748,8910,1783,1197,515,100,52,6);
+ FIR16_2(2,-6,50,-86,520,-1106,1910,-8447,18714,10294,1322,1469,488,145,55,8,0,
+ 30,0,-8,55,-145,488,-1469,1322,-10294,18714,8447,1910,1106,520,86,50,6);
+ FIR16_2(3,-5,49,-73,521,-1015,2023,-7986,18657,10751,1140,1559,473,161,56,9,0,
+ 29,0,-9,56,-161,473,-1559,1140,-10751,18657,7986,2023,1015,521,73,49,5);              
+    samples+=64;
+  }
+  samples-=64*nb_blocks;
+  
+  /* filter bank: part 2 */
+
+  b=bo_start;
+  for(block=0;block<nb_blocks;block++) {
+    b++;
+    if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
+    if(b & 1) {
+      b0 = buf[0] + b - (FIR_SIZE-1) + 4*FIR_BUFFER_SIZE;
+    } else {
+      b0 = buf[1] + b - (FIR_SIZE-1) + 4*FIR_BUFFER_SIZE;
+    }
+
+ FIR16_2(4,-4,47,-61,521,-926,2123,-7528,18578,11205,944,1647,455,177,56,10,0,
+ 28,0,-10,56,-177,455,-1647,944,-11205,18578,7528,2123,926,521,61,47,4);
+ FIR16_2(5,-4,45,-49,518,-837,2210,-7072,18477,11654,733,1733,434,194,57,11,0,
+ 27,0,-11,57,-194,434,-1733,733,-11654,18477,7072,2210,837,518,49,45,4);
+ FIR16_2(6,-4,44,-38,514,-751,2284,-6620,18353,12097,509,1817,411,212,57,12,0,
+ 26,0,-12,57,-212,411,-1817,509,-12097,18353,6620,2284,751,514,38,44,4);
+ FIR16_2(7,-3,42,-27,508,-665,2347,-6173,18208,12534,270,1899,383,229,56,13,0,
+ 25,0,-13,56,-229,383,-1899,270,-12534,18208,6173,2347,665,508,27,42,3);             
+
+    samples+=64;
+  }
+  samples-=64*nb_blocks;
+
+  /* filter bank: part 3 */
+
+  b=bo_start;
+  for(block=0;block<nb_blocks;block++) {
+    b++;
+    if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
+    if(b & 1) {
+      b0 = buf[0] + b - (FIR_SIZE-1) + 8*FIR_BUFFER_SIZE;
+    } else {
+      b0 = buf[1] + b - (FIR_SIZE-1) + 8*FIR_BUFFER_SIZE;
+    }
+
+ FIR16_2(8,-3,40,-18,500,-582,2398,-5732,18042,12963,17,1977,353,247,56,14,0,
+ 24,0,-14,56,-247,353,-1977,17,-12963,18042,5732,2398,582,500,18,40,3);
+ FIR16_2(9,-2,38,-9,490,-501,2437,-5297,17855,13383,-249,2052,320,266,55,15,0,
+ 23,0,-15,55,-266,320,-2052,-249,-13383,17855,5297,2437,501,490,9,38,2);
+ FIR16_2(10,-2,36,0,479,-423,2465,-4869,17647,13794,-530,2122,282,284,53,17,0,
+ 22,0,-17,53,-284,282,-2122,-530,-13794,17647,4869,2465,423,479,0,36,2);
+ FIR16_2(11,-2,34,7,467,-347,2483,-4449,17419,14194,-825,2188,242,302,52,18,0,
+ 21,0,-18,52,-302,242,-2188,-825,-14194,17419,4449,2483,347,467,-7,34,2);          
+
+    samples+=64;
+  }
+  samples-=64*nb_blocks;
+
+  /* filter bank: part 4 */
+
+  b=bo_start;
+  for(block=0;block<nb_blocks;block++) {
+    b++;
+    if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
+    if(b & 1) {
+      b0 = buf[0] + b - (FIR_SIZE-1) + 12*FIR_BUFFER_SIZE;
+    } else {
+      b0 = buf[1] + b - (FIR_SIZE-1) + 12*FIR_BUFFER_SIZE;
+    }
+
+ FIR16_2(12,-2,33,14,454,-273,2491,-4038,17173,14583,-1133,2249,198,320,50,19,0,
+ 20,0,-19,50,-320,198,-2249,-1133,-14583,17173,4038,2491,273,454,-14,33,2);
+ FIR16_2(13,-1,31,20,439,-203,2489,-3637,16907,14959,-1454,2304,151,339,47,21,-1,
+ 19,-1,-21,47,-339,151,-2304,-1454,-14959,16907,3637,2489,203,439,-20,31,1);
+ FIR16_2(14,-1,29,26,424,-136,2479,-3245,16623,15322,-1788,2354,100,357,44,22,-1,
+ 18,-1,-22,44,-357,100,-2354,-1788,-15322,16623,3245,2479,136,424,-26,29,1);
+ FIR16_2(15,-1,27,31,408,-72,2459,-2863,16322,15671,-2135,2396,46,374,40,24,-1,
+ 17,-1,-24,40,-374,46,-2396,-2135,-15671,16322,2863,2459,72,408,-31,27,1);
+ FIR16_1(16,-1,0,36,0,-11,0,-2493,0,16004,0,2431,0,391,0,26,0);      
+
+    samples+=64;
+  }
+
+  return clip;
+}
+

data/mpg123/src/libmpg123/synth_i586.S

@@ -0,0 +1,336 @@
+/*
+ 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
+
+NONEXEC_STACK