comparison liba52/liba52_changes.diff @ 18721:722ac20fac5f

sync with liba52 0.7.4, patch by Emanuele Giaquinta >emanuele.giaquinta ! gmail * com< part 2: cosmetic changes

18720:4bad7f00556e

 void a52_bit_allocate (a52_state_t * state, ba_t * ba, int bndstart,
 		       int start, int end, int fastleak, int slowleak,
 		       expbap_t * expbap);
 
 int a52_downmix_init (int input, int flags, sample_t * level,
 		  sample_t clev, sample_t slev);
+void downmix_accel_init(uint32_t mm_accel);
 int a52_downmix_coeff (sample_t * coeff, int acmod, int output, sample_t level,
 		   sample_t clev, sample_t slev);
-void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
+extern void (*a52_downmix) (sample_t * samples, int acmod, int output, sample_t bias,
 	      sample_t clev, sample_t slev);
-void a52_upmix (sample_t * samples, int acmod, int output);
+extern void (*a52_upmix) (sample_t * samples, int acmod, int output);
 
 void a52_imdct_init (uint32_t mm_accel);
 void a52_imdct_256 (sample_t * data, sample_t * delay, sample_t bias);

+}
+#    endif
+#endif //!ARCH_X86
+
+#endif
+ 
 /* (stolen from the kernel) */
 #ifdef WORDS_BIGENDIAN
 
@@ -28,7 +74,7 @@
 

+    if(mm_accel & MM_ACCEL_X86_MMX) a52_upmix= upmix_MMX;
+    if(mm_accel & MM_ACCEL_X86_SSE) a52_downmix= downmix_SSE;
+    if(mm_accel & MM_ACCEL_X86_3DNOW) a52_downmix= downmix_3dnow;
+#endif
+}
+   
 int a52_downmix_init (int input, int flags, sample_t * level,
 		  sample_t clev, sample_t slev)
 {
@@ -451,7 +480,7 @@
 	samples[i] = 0;
 }
 
-void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
+void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
 	      sample_t clev, sample_t slev)
 {
     switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
@@ -563,7 +592,7 @@
 	break;
 

+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];
+#endif    
+    
 /* Root values for IFFT */
 static sample_t roots16[3];
 static sample_t roots32[7];
@@ -245,7 +322,7 @@
     ifft_pass (buf, roots128 - 32, 32);
     }
 
-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;

+		"movlps %%xmm2, (%1, %%"REG_d", 8)	\n\t"
+		"movhps %%xmm2, (%1, %%"REG_BP", 8)	\n\t"
+		"add $16, %%"REG_S"			\n\t"
+		"add $2, %%"REG_a"			\n\t" // avoid complex addressing for P4 crap
+		"sub $16, %%"REG_D"			\n\t"
+		" jnc 1b				\n\t"
+		"pop %%"REG_BP"				\n\t"//no we didnt touch ebp *g*
+		:: "b" (data), "c" (buf)
+		: "%"REG_S, "%"REG_D, "%"REG_a, "%"REG_d
+	);
+

+		"subps %%xmm2, %%xmm0			\n\t"
+		"movaps %%xmm1, (%%"REG_S", %%"REG_D")	\n\t"
+		"movaps %%xmm0, (%%"REG_d", %%"REG_D")	\n\t"
+		"add $16, %%"REG_D"			\n\t"
+		"cmp %3, %%"REG_D"			\n\t" //FIXME (opt) count against 0 
+		" jb 2b					\n\t"
+		"add %2, %%"REG_S"			\n\t"
+		"cmp %1, %%"REG_S"			\n\t"
+		" jb 1b					\n\t"
+		:: "g" (buf), "m" (buf_offset), "m" (two_m_plus_one_shl3), "r" (two_m<<3),
+		   "r" (sseW[m])

 {
     int i, k;
@@ -368,7 +1153,7 @@
 
 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 */
@@ -420,6 +1205,99 @@
 	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));
+	xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
+    }
+    for (i = 0; i < 7; i++) {

+#if defined(ARCH_X86) || defined(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;

+        if (mm_accel & MM_ACCEL_PPC_ALTIVEC)
+	{
+	  fprintf(stderr, "Using AltiVec optimized IMDCT transform\n");
+          a52_imdct_512 = imdct_do_512_altivec;
+	}
+        else
+#endif
 
 #ifdef LIBA52_DJBFFT
     if (mm_accel & MM_ACCEL_DJBFFT) {
@@ -430,7 +1308,5 @@

+#endif
+    if(((int)state->samples%16) && (mm_accel&MM_ACCEL_X86_SSE)){
+      mm_accel &=~MM_ACCEL_X86_SSE;
+      fprintf(stderr, "liba52: unable to get 16 byte aligned memory disabling usage of SSE instructions\n");
+    }   
+    
     if (state->samples == NULL) {
 	free (state);
 	return NULL;
@@ -74,6 +92,7 @@
     state->lfsr_state = 1;

18721:722ac20fac5f

 void a52_bit_allocate (a52_state_t * state, ba_t * ba, int bndstart,
 		       int start, int end, int fastleak, int slowleak,
 		       expbap_t * expbap);
 
 int a52_downmix_init (int input, int flags, sample_t * level,
 		      sample_t clev, sample_t slev);
+void downmix_accel_init(uint32_t mm_accel);
 int a52_downmix_coeff (sample_t * coeff, int acmod, int output, sample_t level,
 		       sample_t clev, sample_t slev);
-void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
+extern void (*a52_downmix) (sample_t * samples, int acmod, int output, sample_t bias,
 		  sample_t clev, sample_t slev);
-void a52_upmix (sample_t * samples, int acmod, int output);
+extern void (*a52_upmix) (sample_t * samples, int acmod, int output);
 
 void a52_imdct_init (uint32_t mm_accel);
 void a52_imdct_256 (sample_t * data, sample_t * delay, sample_t bias);

+}
+#    endif
+#endif //!ARCH_X86
+
+#endif
+
 /* (stolen from the kernel) */
 #ifdef WORDS_BIGENDIAN
 
@@ -28,7 +74,7 @@
 

+    if(mm_accel & MM_ACCEL_X86_MMX) a52_upmix= upmix_MMX;
+    if(mm_accel & MM_ACCEL_X86_SSE) a52_downmix= downmix_SSE;
+    if(mm_accel & MM_ACCEL_X86_3DNOW) a52_downmix= downmix_3dnow;
+#endif
+}
+ 
 int a52_downmix_init (int input, int flags, sample_t * level,
 		      sample_t clev, sample_t slev)
 {
@@ -451,7 +480,7 @@
 	samples[i] = 0;
 }
 
-void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
+void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
 		  sample_t clev, sample_t slev)
 {
     switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
@@ -563,7 +592,7 @@
 	break;
 

+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];
+#endif
+
 /* Root values for IFFT */
 static sample_t roots16[3];
 static sample_t roots32[7];
@@ -245,7 +322,7 @@
     ifft_pass (buf, roots128 - 32, 32);
 }
 
-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;

+		"movlps %%xmm2, (%1, %%"REG_d", 8)	\n\t"
+		"movhps %%xmm2, (%1, %%"REG_BP", 8)	\n\t"
+		"add $16, %%"REG_S"			\n\t"
+		"add $2, %%"REG_a"			\n\t" // avoid complex addressing for P4 crap
+		"sub $16, %%"REG_D"			\n\t"
+		"jnc 1b				 	\n\t"
+		"pop %%"REG_BP"				\n\t"//no we didnt touch ebp *g*
+		:: "b" (data), "c" (buf)
+		: "%"REG_S, "%"REG_D, "%"REG_a, "%"REG_d
+	);
+

+		"subps %%xmm2, %%xmm0			\n\t"
+		"movaps %%xmm1, (%%"REG_S", %%"REG_D")	\n\t"
+		"movaps %%xmm0, (%%"REG_d", %%"REG_D")	\n\t"
+		"add $16, %%"REG_D"			\n\t"
+		"cmp %3, %%"REG_D"			\n\t" //FIXME (opt) count against 0 
+		"jb 2b					\n\t"
+		"add %2, %%"REG_S"			\n\t"
+		"cmp %1, %%"REG_S"			\n\t"
+		" jb 1b					\n\t"
+		:: "g" (buf), "m" (buf_offset), "m" (two_m_plus_one_shl3), "r" (two_m<<3),
+		   "r" (sseW[m])

 {
     int i, k;
@@ -368,7 +1153,7 @@
 
 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 */
@@ -420,6 +1205,99 @@
 	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));
+	xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
+    }
+    for (i = 0; i < 7; i++) {

+#if defined(ARCH_X86) || defined(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;

+        if (mm_accel & MM_ACCEL_PPC_ALTIVEC)
+	{
+	  fprintf(stderr, "Using AltiVec optimized IMDCT transform\n");
+          a52_imdct_512 = imdct_do_512_altivec;
+	}
+	else
+#endif
 
 #ifdef LIBA52_DJBFFT
     if (mm_accel & MM_ACCEL_DJBFFT) {
@@ -430,7 +1308,5 @@

+#endif
+    if(((int)state->samples%16) && (mm_accel&MM_ACCEL_X86_SSE)){
+      mm_accel &=~MM_ACCEL_X86_SSE;
+      fprintf(stderr, "liba52: unable to get 16 byte aligned memory disabling usage of SSE instructions\n");
+    }   
+ 
     if (state->samples == NULL) {
 	free (state);
 	return NULL;
@@ -74,6 +92,7 @@
     state->lfsr_state = 1;