MMX opt

git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3570 b3059339-0415-0410-9bf9-f77b7e298cf2

3 files changed, 285 insertions, 0 deletions

diff --git a/liba52/resample.c b/liba52/resample.c
index 7d774bc26e..99645fce73 100644
--- a/liba52/resample.c
+++ b/liba52/resample.c
@@ -7,6 +7,14 @@
 
 // a52_resample() should do the conversion.
 
+// MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
+
+/* optimization TODO / NOTES 
+    movntq is slightly faster (0.5% with the current test.c benchmark) 
+	(but thats just test.c so that needs to be testd in reallity)
+	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
+*/
+
 #include <inttypes.h>
 #include "a52.h"
 #include "../config.h"
@@ -108,12 +116,61 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_2F2R:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"1:				\n\t"
+		"movq (%1, %%esi), %%mm0	\n\t"
+		"movq 8(%1, %%esi), %%mm1	\n\t"
+		"movq 1024(%1, %%esi), %%mm2	\n\t"
+		"movq 1032(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq 2048(%1, %%esi), %%mm3	\n\t"
+		"movq 2056(%1, %%esi), %%mm4	\n\t"
+		"movq 3072(%1, %%esi), %%mm5	\n\t"
+		"movq 3080(%1, %%esi), %%mm6	\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"psubd %%mm7, %%mm4		\n\t"
+		"psubd %%mm7, %%mm5		\n\t"
+		"psubd %%mm7, %%mm6		\n\t"
+		"packssdw %%mm4, %%mm3		\n\t"
+		"packssdw %%mm6, %%mm5		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"movq %%mm3, %%mm4		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"punpcklwd %%mm5, %%mm3		\n\t"
+		"punpckhwd %%mm5, %%mm4		\n\t"
+		"movq %%mm0, %%mm2		\n\t"
+		"movq %%mm1, %%mm5		\n\t"
+		"punpckldq %%mm3, %%mm0		\n\t"
+		"punpckhdq %%mm3, %%mm2		\n\t"
+		"punpckldq %%mm4, %%mm1		\n\t"
+		"punpckhdq %%mm4, %%mm5		\n\t"
+		"movq %%mm0, (%0, %%esi,2)	\n\t"
+		"movq %%mm2, 8(%0, %%esi,2)	\n\t"
+		"movq %%mm1, 16(%0, %%esi,2)	\n\t"
+		"movq %%mm5, 24(%0, %%esi,2)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1024), "r" (f+256)
+		:"%esi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[4*i] = convert (f[i]);
 	    s16[4*i+1] = convert (f[i+256]);
 	    s16[4*i+2] = convert (f[i+512]);
 	    s16[4*i+3] = convert (f[i+768]);
 	}
+#endif	
 	break;
     case A52_3F2R:
 	for (i = 0; i < 256; i++) {
@@ -125,11 +182,49 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_MONO | A52_LFE:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"pxor %%mm6, %%mm6		\n\t"
+		"1:				\n\t"
+		"movq 1024(%1, %%esi), %%mm0	\n\t"
+		"movq 1032(%1, %%esi), %%mm1	\n\t"
+		"movq (%1, %%esi), %%mm2	\n\t"
+		"movq 8(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"leal (%%esi, %%esi, 2), %%edi	\n\t"
+		"movq %%mm6, (%0, %%edi)	\n\t"
+		"movd %%mm0, 8(%0, %%edi)	\n\t"
+		"punpckhdq %%mm0, %%mm0		\n\t"
+		"movq %%mm6, 12(%0, %%edi)	\n\t"
+		"movd %%mm0, 20(%0, %%edi)	\n\t"
+		"movq %%mm6, 24(%0, %%edi)	\n\t"
+		"movd %%mm1, 32(%0, %%edi)	\n\t"
+		"punpckhdq %%mm1, %%mm1		\n\t"
+		"movq %%mm6, 36(%0, %%edi)	\n\t"
+		"movd %%mm1, 44(%0, %%edi)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1536), "r" (f+256)
+		:"%esi", "%edi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
 	    s16[6*i+4] = convert (f[i+256]);
 	    s16[6*i+5] = convert (f[i]);
 	}
+#endif
 	break;
     case A52_CHANNEL | A52_LFE:
     case A52_STEREO | A52_LFE:
diff --git a/liba52/resample_c.c b/liba52/resample_c.c
index 7d774bc26e..99645fce73 100644
--- a/liba52/resample_c.c
+++ b/liba52/resample_c.c
@@ -7,6 +7,14 @@
 
 // a52_resample() should do the conversion.
 
+// MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
+
+/* optimization TODO / NOTES 
+    movntq is slightly faster (0.5% with the current test.c benchmark) 
+	(but thats just test.c so that needs to be testd in reallity)
+	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
+*/
+
 #include <inttypes.h>
 #include "a52.h"
 #include "../config.h"
@@ -108,12 +116,61 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_2F2R:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"1:				\n\t"
+		"movq (%1, %%esi), %%mm0	\n\t"
+		"movq 8(%1, %%esi), %%mm1	\n\t"
+		"movq 1024(%1, %%esi), %%mm2	\n\t"
+		"movq 1032(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq 2048(%1, %%esi), %%mm3	\n\t"
+		"movq 2056(%1, %%esi), %%mm4	\n\t"
+		"movq 3072(%1, %%esi), %%mm5	\n\t"
+		"movq 3080(%1, %%esi), %%mm6	\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"psubd %%mm7, %%mm4		\n\t"
+		"psubd %%mm7, %%mm5		\n\t"
+		"psubd %%mm7, %%mm6		\n\t"
+		"packssdw %%mm4, %%mm3		\n\t"
+		"packssdw %%mm6, %%mm5		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"movq %%mm3, %%mm4		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"punpcklwd %%mm5, %%mm3		\n\t"
+		"punpckhwd %%mm5, %%mm4		\n\t"
+		"movq %%mm0, %%mm2		\n\t"
+		"movq %%mm1, %%mm5		\n\t"
+		"punpckldq %%mm3, %%mm0		\n\t"
+		"punpckhdq %%mm3, %%mm2		\n\t"
+		"punpckldq %%mm4, %%mm1		\n\t"
+		"punpckhdq %%mm4, %%mm5		\n\t"
+		"movq %%mm0, (%0, %%esi,2)	\n\t"
+		"movq %%mm2, 8(%0, %%esi,2)	\n\t"
+		"movq %%mm1, 16(%0, %%esi,2)	\n\t"
+		"movq %%mm5, 24(%0, %%esi,2)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1024), "r" (f+256)
+		:"%esi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[4*i] = convert (f[i]);
 	    s16[4*i+1] = convert (f[i+256]);
 	    s16[4*i+2] = convert (f[i+512]);
 	    s16[4*i+3] = convert (f[i+768]);
 	}
+#endif	
 	break;
     case A52_3F2R:
 	for (i = 0; i < 256; i++) {
@@ -125,11 +182,49 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_MONO | A52_LFE:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"pxor %%mm6, %%mm6		\n\t"
+		"1:				\n\t"
+		"movq 1024(%1, %%esi), %%mm0	\n\t"
+		"movq 1032(%1, %%esi), %%mm1	\n\t"
+		"movq (%1, %%esi), %%mm2	\n\t"
+		"movq 8(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"leal (%%esi, %%esi, 2), %%edi	\n\t"
+		"movq %%mm6, (%0, %%edi)	\n\t"
+		"movd %%mm0, 8(%0, %%edi)	\n\t"
+		"punpckhdq %%mm0, %%mm0		\n\t"
+		"movq %%mm6, 12(%0, %%edi)	\n\t"
+		"movd %%mm0, 20(%0, %%edi)	\n\t"
+		"movq %%mm6, 24(%0, %%edi)	\n\t"
+		"movd %%mm1, 32(%0, %%edi)	\n\t"
+		"punpckhdq %%mm1, %%mm1		\n\t"
+		"movq %%mm6, 36(%0, %%edi)	\n\t"
+		"movd %%mm1, 44(%0, %%edi)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1536), "r" (f+256)
+		:"%esi", "%edi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
 	    s16[6*i+4] = convert (f[i+256]);
 	    s16[6*i+5] = convert (f[i]);
 	}
+#endif
 	break;
     case A52_CHANNEL | A52_LFE:
     case A52_STEREO | A52_LFE:
diff --git a/liba52/resample_mmx.c b/liba52/resample_mmx.c
index 7d774bc26e..99645fce73 100644
--- a/liba52/resample_mmx.c
+++ b/liba52/resample_mmx.c
@@ -7,6 +7,14 @@
 
 // a52_resample() should do the conversion.
 
+// MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
+
+/* optimization TODO / NOTES 
+    movntq is slightly faster (0.5% with the current test.c benchmark) 
+	(but thats just test.c so that needs to be testd in reallity)
+	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
+*/
+
 #include <inttypes.h>
 #include "a52.h"
 #include "../config.h"
@@ -108,12 +116,61 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_2F2R:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"1:				\n\t"
+		"movq (%1, %%esi), %%mm0	\n\t"
+		"movq 8(%1, %%esi), %%mm1	\n\t"
+		"movq 1024(%1, %%esi), %%mm2	\n\t"
+		"movq 1032(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq 2048(%1, %%esi), %%mm3	\n\t"
+		"movq 2056(%1, %%esi), %%mm4	\n\t"
+		"movq 3072(%1, %%esi), %%mm5	\n\t"
+		"movq 3080(%1, %%esi), %%mm6	\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"psubd %%mm7, %%mm4		\n\t"
+		"psubd %%mm7, %%mm5		\n\t"
+		"psubd %%mm7, %%mm6		\n\t"
+		"packssdw %%mm4, %%mm3		\n\t"
+		"packssdw %%mm6, %%mm5		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"movq %%mm3, %%mm4		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"punpcklwd %%mm5, %%mm3		\n\t"
+		"punpckhwd %%mm5, %%mm4		\n\t"
+		"movq %%mm0, %%mm2		\n\t"
+		"movq %%mm1, %%mm5		\n\t"
+		"punpckldq %%mm3, %%mm0		\n\t"
+		"punpckhdq %%mm3, %%mm2		\n\t"
+		"punpckldq %%mm4, %%mm1		\n\t"
+		"punpckhdq %%mm4, %%mm5		\n\t"
+		"movq %%mm0, (%0, %%esi,2)	\n\t"
+		"movq %%mm2, 8(%0, %%esi,2)	\n\t"
+		"movq %%mm1, 16(%0, %%esi,2)	\n\t"
+		"movq %%mm5, 24(%0, %%esi,2)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1024), "r" (f+256)
+		:"%esi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[4*i] = convert (f[i]);
 	    s16[4*i+1] = convert (f[i+256]);
 	    s16[4*i+2] = convert (f[i+512]);
 	    s16[4*i+3] = convert (f[i+768]);
 	}
+#endif	
 	break;
     case A52_3F2R:
 	for (i = 0; i < 256; i++) {
@@ -125,11 +182,49 @@ int a52_resample(float * _f, int16_t * s16)
 	}
 	break;
     case A52_MONO | A52_LFE:
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl $-1024, %%esi		\n\t"
+		"movq magicF2W, %%mm7		\n\t"
+		"pxor %%mm6, %%mm6		\n\t"
+		"1:				\n\t"
+		"movq 1024(%1, %%esi), %%mm0	\n\t"
+		"movq 1032(%1, %%esi), %%mm1	\n\t"
+		"movq (%1, %%esi), %%mm2	\n\t"
+		"movq 8(%1, %%esi), %%mm3	\n\t"
+		"psubd %%mm7, %%mm0		\n\t"
+		"psubd %%mm7, %%mm1		\n\t"
+		"psubd %%mm7, %%mm2		\n\t"
+		"psubd %%mm7, %%mm3		\n\t"
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"movq %%mm0, %%mm1		\n\t"
+		"punpcklwd %%mm2, %%mm0		\n\t"
+		"punpckhwd %%mm2, %%mm1		\n\t"
+		"leal (%%esi, %%esi, 2), %%edi	\n\t"
+		"movq %%mm6, (%0, %%edi)	\n\t"
+		"movd %%mm0, 8(%0, %%edi)	\n\t"
+		"punpckhdq %%mm0, %%mm0		\n\t"
+		"movq %%mm6, 12(%0, %%edi)	\n\t"
+		"movd %%mm0, 20(%0, %%edi)	\n\t"
+		"movq %%mm6, 24(%0, %%edi)	\n\t"
+		"movd %%mm1, 32(%0, %%edi)	\n\t"
+		"punpckhdq %%mm1, %%mm1		\n\t"
+		"movq %%mm6, 36(%0, %%edi)	\n\t"
+		"movd %%mm1, 44(%0, %%edi)	\n\t"
+		"addl $16, %%esi		\n\t"
+		" jnz 1b			\n\t"
+		"emms				\n\t"
+		:: "r" (s16+1536), "r" (f+256)
+		:"%esi", "%edi", "memory"
+	);
+#else
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
 	    s16[6*i+4] = convert (f[i+256]);
 	    s16[6*i+5] = convert (f[i]);
 	}
+#endif
 	break;
     case A52_CHANNEL | A52_LFE:
     case A52_STEREO | A52_LFE: