liba52_changes.diff

自己移植的linux下的流媒体播放器原代码,支持mms协议,支持ftp和http协议.

+    case CONVERT (A52_3F, A52_STEREO):
+    case CONVERT (A52_3F, A52_DOLBY):
+    mix_3to2_3dnow:
+	mix3to2_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_2F1R, A52_STEREO):
+	if (slev == 0)
+	    break;
+	mix21to2_3dnow (samples, samples + 256, bias);
+	break;
+
+    case CONVERT (A52_2F1R, A52_DOLBY):
+	mix21toS_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_3F1R, A52_STEREO):
+	if (slev == 0)
+	    goto mix_3to2_3dnow;
+	mix31to2_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_3F1R, A52_DOLBY):
+	mix31toS_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_2F2R, A52_STEREO):
+	if (slev == 0)
+	    break;
+	mix2to1_3dnow (samples, samples + 512, bias);
+	mix2to1_3dnow (samples + 256, samples + 768, bias);
+	break;
+
+    case CONVERT (A52_2F2R, A52_DOLBY):
+	mix22toS_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_3F2R, A52_STEREO):
+	if (slev == 0)
+	    goto mix_3to2_3dnow;
+	mix32to2_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_3F2R, A52_DOLBY):
+	mix32toS_3dnow (samples, bias);
+	break;
+
+    case CONVERT (A52_3F1R, A52_3F):
+	if (slev == 0)
+	    break;
+	mix21to2_3dnow (samples, samples + 512, bias);
+	break;
+
+    case CONVERT (A52_3F2R, A52_3F):
+	if (slev == 0)
+	    break;
+	mix2to1_3dnow (samples, samples + 768, bias);
+	mix2to1_3dnow (samples + 512, samples + 1024, bias);
+	break;
+
+    case CONVERT (A52_3F1R, A52_2F1R):
+	mix3to2_3dnow (samples, bias);
+	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
+	break;
+
+    case CONVERT (A52_2F2R, A52_2F1R):
+	mix2to1_3dnow (samples + 512, samples + 768, bias);
+	break;
+
+    case CONVERT (A52_3F2R, A52_2F1R):
+	mix3to2_3dnow (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
+	move2to1_3dnow (samples + 768, samples + 512, bias);
+	break;
+
+    case CONVERT (A52_3F2R, A52_3F1R):
+	mix2to1_3dnow (samples + 768, samples + 1024, bias);
+	break;
+
+    case CONVERT (A52_2F1R, A52_2F2R):
+	memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
+	break;
+
+    case CONVERT (A52_3F1R, A52_2F2R):
+	mix3to2_3dnow (samples, bias);
+	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
+	break;
+
+    case CONVERT (A52_3F2R, A52_2F2R):
+	mix3to2_3dnow (samples, bias);
+	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
+	memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
+	break;
+
+    case CONVERT (A52_3F1R, A52_3F2R):
+	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
+	break;
+    }
+    __asm __volatile("femms":::"memory");
+}
+
+#endif //ARCH_X86
--- liba52/imdct.c	2005-03-22 19:59:35.000000000 +0100
+++ imdct.c	2004-04-26 22:00:57.000000000 +0200
@@ -17,17 +23,32 @@
  * You should have received a copy of the GNU General Public License
  * along with this program; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ * SSE optimizations from Michael Niedermayer ([e-mail])
+ * 3DNOW optimizations from Nick Kurshev <[e-mail]>
+ *   michael did port them from libac3 (untested, perhaps totally broken)
+ * AltiVec optimizations from Romain Dolbeau ([e-mail])
  */
 
 #include "config.h"
 
-#include <inttypes.h>
 #include <math.h>
 #include <stdio.h>
+#ifndef M_PI
+#define M_PI 3.1415926535897932384626433832795029
+#endif
+#include <inttypes.h>
 
 #include "a52.h"
 #include "a52_internal.h"
 #include "mm_accel.h"
+#include "mangle.h"
+
+#ifdef RUNTIME_CPUDETECT
+#undef HAVE_3DNOWEX
+#endif
+
+#define USE_AC3_C
 
 void (* imdct_256) (sample_t data[], sample_t delay[], sample_t bias);
 void (* imdct_512) (sample_t data[], sample_t delay[], sample_t bias);
@@ -37,9 +58,22 @@
     sample_t imag;
 } complex_t;
 
+static void fft_128p(complex_t *a);
+
+static const int pm128[128] attribute_used __attribute__((aligned(16))) =
+{
+	0, 16, 32, 48, 64, 80,  96, 112,  8, 40, 72, 104, 24, 56,  88, 120,
+	4, 20, 36, 52, 68, 84, 100, 116, 12, 28, 44,  60, 76, 92, 108, 124,
+	2, 18, 34, 50, 66, 82,  98, 114, 10, 42, 74, 106, 26, 58,  90, 122,
+	6, 22, 38, 54, 70, 86, 102, 118, 14, 46, 78, 110, 30, 62,  94, 126,
+	1, 17, 33, 49, 65, 81,  97, 113,  9, 41, 73, 105, 25, 57,  89, 121,
+	5, 21, 37, 53, 69, 85, 101, 117, 13, 29, 45,  61, 77, 93, 109, 125,
+	3, 19, 35, 51, 67, 83,  99, 115, 11, 43, 75, 107, 27, 59,  91, 123,
+	7, 23, 39, 55, 71, 87, 103, 119, 15, 31, 47,  63, 79, 95, 111, 127
+}; 
 
 /* 128 point bit-reverse LUT */
-static uint8_t bit_reverse_512[] = {
+static uint8_t attribute_used bit_reverse_512[] = {
 	0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70, 
 	0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78, 
 	0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74, 
@@ -67,23 +101,42 @@
 	0x03, 0x23, 0x13, 0x33, 0x0b, 0x2b, 0x1b, 0x3b, 
 	0x07, 0x27, 0x17, 0x37, 0x0f, 0x2f, 0x1f, 0x3f};
 
-static complex_t buf[128];
+#ifdef ARCH_X86
+// NOTE: SSE needs 16byte alignment or it will segfault 
+// 
+static complex_t __attribute__((aligned(16))) buf[128];
+static float __attribute__((aligned(16))) sseSinCos1c[256];
+static float __attribute__((aligned(16))) sseSinCos1d[256];
+static float attribute_used __attribute__((aligned(16))) ps111_1[4]={1,1,1,-1};
+//static float __attribute__((aligned(16))) sseW0[4];
+static float __attribute__((aligned(16))) sseW1[8];
+static float __attribute__((aligned(16))) sseW2[16];
+static float __attribute__((aligned(16))) sseW3[32];
+static float __attribute__((aligned(16))) sseW4[64];
+static float __attribute__((aligned(16))) sseW5[128];
+static float __attribute__((aligned(16))) sseW6[256];
+static float __attribute__((aligned(16))) *sseW[7]=
+	{NULL /*sseW0*/,sseW1,sseW2,sseW3,sseW4,sseW5,sseW6};
+static float __attribute__((aligned(16))) sseWindow[512];
+#else
+static complex_t  __attribute__((aligned(16))) buf[128];
+#endif
 
 /* Twiddle factor LUT */
-static complex_t w_1[1];
-static complex_t w_2[2];
-static complex_t w_4[4];
-static complex_t w_8[8];
-static complex_t w_16[16];
-static complex_t w_32[32];
-static complex_t w_64[64];
-static complex_t * w[7] = {w_1, w_2, w_4, w_8, w_16, w_32, w_64};
+static complex_t __attribute__((aligned(16))) w_1[1];
+static complex_t __attribute__((aligned(16))) w_2[2];
+static complex_t __attribute__((aligned(16))) w_4[4];
+static complex_t __attribute__((aligned(16))) w_8[8];
+static complex_t __attribute__((aligned(16))) w_16[16];
+static complex_t __attribute__((aligned(16))) w_32[32];
+static complex_t __attribute__((aligned(16))) w_64[64];
+static complex_t __attribute__((aligned(16))) * w[7] = {w_1, w_2, w_4, w_8, w_16, w_32, w_64};
 
 /* Twiddle factors for IMDCT */
-static sample_t xcos1[128];
-static sample_t xsin1[128];
-static sample_t xcos2[64];
-static sample_t xsin2[64];
+static sample_t __attribute__((aligned(16))) xcos1[128];
+static sample_t __attribute__((aligned(16))) xsin1[128];
+static sample_t __attribute__((aligned(16))) xcos2[64];
+static sample_t __attribute__((aligned(16))) xsin2[64];
 
 /* Windowing function for Modified DCT - Thank you acroread */
 sample_t imdct_window[] = {
@@ -145,16 +198,19 @@
 void
 imdct_do_512(sample_t data[],sample_t delay[], sample_t bias)
 {
-    int i,k;
+    int i;
+#ifndef USE_AC3_C
+	int k;
     int p,q;
     int m;
     int two_m;
     int two_m_plus_one;
 
-    sample_t tmp_a_i;
-    sample_t tmp_a_r;
     sample_t tmp_b_i;
     sample_t tmp_b_r;
+#endif
+    sample_t tmp_a_i;
+    sample_t tmp_a_r;
 
     sample_t *data_ptr;
     sample_t *delay_ptr;
@@ -162,22 +218,21 @@
 	
     /* 512 IMDCT with source and dest data in 'data' */
 	
-    /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
+    /* Pre IFFT complex multiply plus IFFT cmplx conjugate & reordering*/
     for( i=0; i < 128; i++) {
 	/* z[i] = (X[256-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */ 
-	buf[i].real =         (data[256-2*i-1] * xcos1[i])  -  (data[2*i]       * xsin1[i]);
-	buf[i].imag = -1.0 * ((data[2*i]       * xcos1[i])  +  (data[256-2*i-1] * xsin1[i]));
-    }
-
-    /* Bit reversed shuffling */
-    for(i=0; i<128; i++) {
-	k = bit_reverse_512[i];
-	if (k < i)
-	    swap_cmplx(&buf[i],&buf[k]);
+#ifdef USE_AC3_C
+	int j= pm128[i];
+#else
+	int j= bit_reverse_512[i];
+#endif	
+	buf[i].real =         (data[256-2*j-1] * xcos1[j])  -  (data[2*j]       * xsin1[j]);
+	buf[i].imag = -1.0 * ((data[2*j]       * xcos1[j])  +  (data[256-2*j-1] * xsin1[j]));
     }
 
     /* FFT Merge */
-    for (m=0; m < 7; m++) {
+/* unoptimized variant
+    for (m=1; m < 7; m++) {
 	if(m)
 	    two_m = (1 << m);
 	else
@@ -185,8 +240,8 @@
 
 	two_m_plus_one = (1 << (m+1));
 
-	for(k = 0; k < two_m; k++) {
-	    for(i = 0; i < 128; i += two_m_plus_one) {
+	for(i = 0; i < 128; i += two_m_plus_one) {
+	    for(k = 0; k < two_m; k++) {
 		p = k + i;
 		q = p + two_m;
 		tmp_a_r = buf[p].real;
@@ -200,7 +255,102 @@
 	    }
 	}
     }
+*/
+#ifdef USE_AC3_C
+	fft_128p (&buf[0]);
+#else
+
+    /* 1. iteration */
+    for(i = 0; i < 128; i += 2) {
+	tmp_a_r = buf[i].real;
+	tmp_a_i = buf[i].imag;
+	tmp_b_r = buf[i+1].real;
+	tmp_b_i = buf[i+1].imag;
+	buf[i].real = tmp_a_r + tmp_b_r;
+	buf[i].imag =  tmp_a_i + tmp_b_i;
+	buf[i+1].real = tmp_a_r - tmp_b_r;
+	buf[i+1].imag =  tmp_a_i - tmp_b_i;
+    }
+        
+    /* 2. iteration */
+	// Note w[1]={{1,0}, {0,-1}}
+    for(i = 0; i < 128; i += 4) {
+	tmp_a_r = buf[i].real;
+	tmp_a_i = buf[i].imag;
+	tmp_b_r = buf[i+2].real;
+	tmp_b_i = buf[i+2].imag;
+	buf[i].real = tmp_a_r + tmp_b_r;
+	buf[i].imag =  tmp_a_i + tmp_b_i;
+	buf[i+2].real = tmp_a_r - tmp_b_r;
+	buf[i+2].imag =  tmp_a_i - tmp_b_i;
+	tmp_a_r = buf[i+1].real;
+	tmp_a_i = buf[i+1].imag;
+	tmp_b_r = buf[i+3].imag;
+	tmp_b_i = buf[i+3].real;
+	buf[i+1].real = tmp_a_r + tmp_b_r;
+	buf[i+1].imag =  tmp_a_i - tmp_b_i;
+	buf[i+3].real = tmp_a_r - tmp_b_r;
+	buf[i+3].imag =  tmp_a_i + tmp_b_i;
+    }
 
+    /* 3. iteration */
+    for(i = 0; i < 128; i += 8) {
+		tmp_a_r = buf[i].real;
+		tmp_a_i = buf[i].imag;
+		tmp_b_r = buf[i+4].real;
+		tmp_b_i = buf[i+4].imag;
+		buf[i].real = tmp_a_r + tmp_b_r;
+		buf[i].imag =  tmp_a_i + tmp_b_i;
+		buf[i+4].real = tmp_a_r - tmp_b_r;
+		buf[i+4].imag =  tmp_a_i - tmp_b_i;
+		tmp_a_r = buf[1+i].real;
+		tmp_a_i = buf[1+i].imag;
+		tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
+		tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
+		buf[1+i].real = tmp_a_r + tmp_b_r;
+		buf[1+i].imag =  tmp_a_i + tmp_b_i;
+		buf[i+5].real = tmp_a_r - tmp_b_r;
+		buf[i+5].imag =  tmp_a_i - tmp_b_i;
+		tmp_a_r = buf[i+2].real;
+		tmp_a_i = buf[i+2].imag;
+		tmp_b_r = buf[i+6].imag;
+		tmp_b_i = - buf[i+6].real;
+		buf[i+2].real = tmp_a_r + tmp_b_r;
+		buf[i+2].imag =  tmp_a_i + tmp_b_i;
+		buf[i+6].real = tmp_a_r - tmp_b_r;
+		buf[i+6].imag =  tmp_a_i - tmp_b_i;
+		tmp_a_r = buf[i+3].real;
+		tmp_a_i = buf[i+3].imag;
+		tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
+		tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
+		buf[i+3].real = tmp_a_r + tmp_b_r;
+		buf[i+3].imag =  tmp_a_i + tmp_b_i;
+		buf[i+7].real = tmp_a_r - tmp_b_r;
+		buf[i+7].imag =  tmp_a_i - tmp_b_i;
+     }
+    
+    /* 4-7. iterations */
+    for (m=3; m < 7; m++) {
+        two_m = (1 << m);
+
+	two_m_plus_one = two_m<<1;
+
+	for(i = 0; i < 128; i += two_m_plus_one) {
+	    for(k = 0; k < two_m; k++) {
+		int p = k + i;
+		int q = p + two_m;
+		tmp_a_r = buf[p].real;
+		tmp_a_i = buf[p].imag;
+		tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
+		tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
+		buf[p].real = tmp_a_r + tmp_b_r;
+		buf[p].imag =  tmp_a_i + tmp_b_i;
+		buf[q].real = tmp_a_r - tmp_b_r;
+		buf[q].imag =  tmp_a_i - tmp_b_i;
+	    }
+	}
+    }
+#endif    
     /* Post IFFT complex multiply  plus IFFT complex conjugate*/
     for( i=0; i < 128; i++) {
 	/* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
@@ -219,12 +369,12 @@
 	*data_ptr++   = -buf[64+i].imag   * *window_ptr++ + *delay_ptr++ + bias; 
 	*data_ptr++   =  buf[64-i-1].real * *window_ptr++ + *delay_ptr++ + bias; 
     }
-
+    
     for(i=0; i< 64; i++) { 
 	*data_ptr++  = -buf[i].real       * *window_ptr++ + *delay_ptr++ + bias; 
 	*data_ptr++  =  buf[128-i-1].imag * *window_ptr++ + *delay_ptr++ + bias; 
     }
-
+    
     /* The trailing edge of the window goes into the delay line */
     delay_ptr = delay;
 
@@ -232,13 +382,717 @@
 	*delay_ptr++  = -buf[64+i].real   * *--window_ptr; 
 	*delay_ptr++  =  buf[64-i-1].imag * *--window_ptr; 
     }
-
+    
     for(i=0; i<64; i++) {
 	*delay_ptr++  =  buf[i].imag       * *--window_ptr; 
 	*delay_ptr++  = -buf[128-i-1].real * *--window_ptr; 
     }
 }
 
+#ifdef HAVE_ALTIVEC
+
+#ifndef SYS_DARWIN
+#include <altivec.h>
+#endif
+
+// used to build registers permutation vectors (vcprm)
+// the 's' are for words in the _s_econd vector
+#define WORD_0 0x00,0x01,0x02,0x03
+#define WORD_1 0x04,0x05,0x06,0x07
+#define WORD_2 0x08,0x09,0x0a,0x0b
+#define WORD_3 0x0c,0x0d,0x0e,0x0f
+#define WORD_s0 0x10,0x11,0x12,0x13
+#define WORD_s1 0x14,0x15,0x16,0x17
+#define WORD_s2 0x18,0x19,0x1a,0x1b
+#define WORD_s3 0x1c,0x1d,0x1e,0x1f
+
+#ifdef SYS_DARWIN
+#define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
+#else
+#define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
+#endif
+
+// vcprmle is used to keep the same index as in the SSE version.
+// it's the same as vcprm, with the index inversed
+// ('le' is Little Endian)
+#define vcprmle(a,b,c,d) vcprm(d,c,b,a)
+
+// used to build inverse/identity vectors (vcii)
+// n is _n_egative, p is _p_ositive
+#define FLOAT_n -1.
+#define FLOAT_p 1.
+
+#ifdef SYS_DARWIN
+#define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
+#else
+#define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}