comparison liba52/liba52_changes.diff @ 25995:236ab58453f7

Refactor AltiVec macros as done for FFmpeg.

25994:32e21d1beb48

+    }
+    __asm __volatile("femms":::"memory");
+}
+
+#endif // ARCH_X86 || ARCH_X86_64
--- liba52/imdct.c	2006-06-12 15:18:27.000000000 +0200
+++ liba52/imdct.c	2006-06-12 19:18:39.000000000 +0200
@@ -26,6 +26,11 @@
  * 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])

-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 +365,714 @@
     }
 }
 
+#ifdef HAVE_ALTIVEC
+

+#define WORD_s1 0x14,0x15,0x16,0x17
+#define WORD_s2 0x18,0x19,0x1a,0x1b
+#define WORD_s3 0x1c,0x1d,0x1e,0x1f
+
+#ifdef __APPLE_CC__
+#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 __APPLE_CC__
+#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}
+#endif
+
+#ifdef __APPLE_CC__
+#define FOUROF(a) (a)
+#else
+#define FOUROF(a) {a,a,a,a}
+#endif
+
+
+void
+imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
+{

+#endif // ARCH_X86 || ARCH_X86_64
+
 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
@@ -364,7 +1152,7 @@
 
 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 +1204,99 @@
 	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));

+	else
+#endif
 
 #ifdef LIBA52_DJBFFT
     if (mm_accel & MM_ACCEL_DJBFFT) {
@@ -426,7 +1307,5 @@
 #endif
     {
 	fprintf (stderr, "No accelerated IMDCT transform found\n");
-	ifft128 = ifft128_c;
-	ifft64 = ifft64_c;

25995:236ab58453f7

+    }
+    __asm __volatile("femms":::"memory");
+}
+
+#endif // ARCH_X86 || ARCH_X86_64
--- liba52/imdct.c	2008-02-19 00:18:33.000000000 +0100
+++ liba52/imdct.c	2008-02-19 00:16:40.000000000 +0100
@@ -22,6 +26,11 @@
  * 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])

-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 +365,707 @@
     }
 }
 
+#ifdef HAVE_ALTIVEC
+

+#define WORD_s1 0x14,0x15,0x16,0x17
+#define WORD_s2 0x18,0x19,0x1a,0x1b
+#define WORD_s3 0x1c,0x1d,0x1e,0x1f
+
+#ifdef __APPLE_CC__
+#define AVV(x...) (x)
+#else
+#define AVV(x...) {x}
+#endif
+
+#define vcprm(a,b,c,d) (const vector unsigned char)AVV(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
+#define vcii(a,b,c,d) (const vector float)AVV(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
+
+#define FOUROF(a) AVV(a,a,a,a)
+
+// 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.
+
+
+void
+imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
+{

+#endif // ARCH_X86 || ARCH_X86_64
+
 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
 {
     int i, k;
@@ -364,7 +1145,7 @@
 
 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 +1197,99 @@
 	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));

+	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;