comparison liba52/liba52_changes.diff @ 29601:cd3ae709054f

Disable liba52 SSE imdct for x86_64+PIC (e.g. OSX 64 bit) since it will not compile for that combination.

29600:577c86f00dd1

-void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias)
+void imdct_do_512 (sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
     sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
@@ -285,6 +366,702 @@
     }
 }

+#if HAVE_ALTIVEC
+

+#include "imdct_3dnow.h"
+#undef HAVE_AMD3DNOWEXT
+#define HAVE_AMD3DNOWEXT 1
+#include "imdct_3dnow.h"
+
+void
+imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
+{
+/*	int i,k;
+    int p,q;*/

+		" jb 1b					\n\t"
+		:: "r" (buf), "r" (delay_ptr)
+		: "%"REG_S, "%"REG_D
+	);
+}
+#endif // ARCH_X86 || ARCH_X86_64
+
 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
@@ -364,7 +1141,7 @@

 void a52_imdct_init (uint32_t mm_accel)
 {
-    int i, k;
+    int i, j, k;
     double sum;

     /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
@@ -416,6 +1193,99 @@
 	post2[i].real = cos ((M_PI / 128) * (i + 0.5));
 	post2[i].imag = sin ((M_PI / 128) * (i + 0.5));
     }
+    for (i = 0; i < 128; i++) {
+	xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));

+	a52_imdct_512 = imdct_do_512;
+	ifft128 = ifft128_c;
+	ifft64 = ifft64_c;
+
+#if ARCH_X86 || ARCH_X86_64
+	if(mm_accel & MM_ACCEL_X86_SSE)
+	{
+	  fprintf (stderr, "Using SSE optimized IMDCT transform\n");
+	  a52_imdct_512 = imdct_do_512_sse;
+	}
+	else
+	if(mm_accel & MM_ACCEL_X86_3DNOWEXT)
+	{
+	  fprintf (stderr, "Using 3DNowEx optimized IMDCT transform\n");
+	  a52_imdct_512 = imdct_do_512_3dnowex;
+	}

+	else
+#endif

 #ifdef LIBA52_DJBFFT
     if (mm_accel & MM_ACCEL_DJBFFT) {
@@ -426,7 +1296,5 @@
 #endif
     {
 	fprintf (stderr, "No accelerated IMDCT transform found\n");
-	ifft128 = ifft128_c;
-	ifft64 = ifft64_c;

29601:cd3ae709054f

-void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias)
+void imdct_do_512 (sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
     sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
@@ -285,6 +366,704 @@
     }
 }

+#if HAVE_ALTIVEC
+

+#include "imdct_3dnow.h"
+#undef HAVE_AMD3DNOWEXT
+#define HAVE_AMD3DNOWEXT 1
+#include "imdct_3dnow.h"
+
+#if !ARCH_X86_64 || !defined(PIC)
+void
+imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
+{
+/*	int i,k;
+    int p,q;*/

+		" jb 1b					\n\t"
+		:: "r" (buf), "r" (delay_ptr)
+		: "%"REG_S, "%"REG_D
+	);
+}
+#endif
+#endif // ARCH_X86 || ARCH_X86_64
+
 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
@@ -364,7 +1143,7 @@

 void a52_imdct_init (uint32_t mm_accel)
 {
-    int i, k;
+    int i, j, k;
     double sum;

     /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
@@ -416,6 +1195,101 @@
 	post2[i].real = cos ((M_PI / 128) * (i + 0.5));
 	post2[i].imag = sin ((M_PI / 128) * (i + 0.5));
     }
+    for (i = 0; i < 128; i++) {
+	xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));

+	a52_imdct_512 = imdct_do_512;
+	ifft128 = ifft128_c;
+	ifft64 = ifft64_c;
+
+#if ARCH_X86 || ARCH_X86_64
+#if !ARCH_X86_64 || !defined(PIC)
+	if(mm_accel & MM_ACCEL_X86_SSE)
+	{
+	  fprintf (stderr, "Using SSE optimized IMDCT transform\n");
+	  a52_imdct_512 = imdct_do_512_sse;
+	}
+	else
+#endif
+	if(mm_accel & MM_ACCEL_X86_3DNOWEXT)
+	{
+	  fprintf (stderr, "Using 3DNowEx optimized IMDCT transform\n");
+	  a52_imdct_512 = imdct_do_512_3dnowex;
+	}

+	else
+#endif

 #ifdef LIBA52_DJBFFT
     if (mm_accel & MM_ACCEL_DJBFFT) {
@@ -426,7 +1300,5 @@
 #endif
     {
 	fprintf (stderr, "No accelerated IMDCT transform found\n");
-	ifft128 = ifft128_c;
-	ifft64 = ifft64_c;