[Ffmpeg-cvslog] CVS: ffmpeg/libavcodec Makefile, 1.243, 1.244 dsputil.h, 1.128, 1.129 fft.c, 1.10, 1.11

[Corey Hickey CVS]

Update of /cvsroot/ffmpeg/ffmpeg/libavcodec
In directory mail:/var2/tmp/cvs-serv22424/libavcodec

Modified Files:
	Makefile dsputil.h fft.c 
Log Message:
3DNow! & Extended 3DNow! versions of FFT

Patch by Zuxy Meng, zuxy <<dot>> meng >>at<< gmail <<dot>> com
Minor non-functional diff-related fixes by me.


Index: Makefile
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/Makefile,v
retrieving revision 1.243
retrieving revision 1.244
diff -u -d -r1.243 -r1.244
--- Makefile	17 Feb 2006 03:17:42 -0000	1.243
+++ Makefile	8 Mar 2006 04:13:55 -0000	1.244
@@ -330,7 +330,7 @@
 	i386/dsputil_mmx.o i386/mpegvideo_mmx.o \
 	i386/idct_mmx.o i386/motion_est_mmx.o \
 	i386/simple_idct_mmx.o i386/fft_sse.o i386/vp3dsp_mmx.o \
-	i386/vp3dsp_sse2.o
+	i386/vp3dsp_sse2.o i386/fft_3dn.o i386/fft_3dn2.o
 ifeq ($(CONFIG_GPL),yes)
 OBJS += i386/idct_mmx_xvid.o
 endif
@@ -338,6 +338,10 @@
 i386/fft_sse.o: CFLAGS+= -msse
 depend: CFLAGS+= -msse
 endif
+ifdef TARGET_BUILTIN_3DNOW
+i386/fft_3dn.o: CFLAGS+= -m3dnow
+i386/fft_3dn2.o: CFLAGS+= -march=athlon
+endif
 endif
 
 # armv4l specific stuff

Index: dsputil.h
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/dsputil.h,v
retrieving revision 1.128
retrieving revision 1.129
diff -u -d -r1.128 -r1.129
--- dsputil.h	10 Feb 2006 06:55:24 -0000	1.128
+++ dsputil.h	8 Mar 2006 04:13:55 -0000	1.129
@@ -564,6 +564,8 @@
 void ff_fft_permute(FFTContext *s, FFTComplex *z);
 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
+void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
+void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 
 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)

Index: fft.c
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/fft.c,v
retrieving revision 1.10
retrieving revision 1.11
diff -u -d -r1.10 -r1.11
--- fft.c	12 Jan 2006 22:43:15 -0000	1.10
+++ fft.c	8 Mar 2006 04:13:55 -0000	1.11
@@ -57,12 +57,12 @@
     s->exptab1 = NULL;
 
     /* compute constant table for HAVE_SSE version */
-#if (defined(HAVE_MMX) && defined(HAVE_BUILTIN_VECTOR)) || defined(HAVE_ALTIVEC)
+#if (defined(HAVE_MMX) && (defined(HAVE_BUILTIN_VECTOR) || defined(HAVE_MM3DNOW))) || defined(HAVE_ALTIVEC)
     {
         int has_vectors = 0;
 
 #if defined(HAVE_MMX)
-        has_vectors = mm_support() & MM_SSE;
+        has_vectors = mm_support() & (MM_3DNOW | MM_3DNOWEXT | MM_SSE | MM_SSE2);
 #endif
 #if defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE)
         has_vectors = mm_support() & MM_ALTIVEC;
@@ -94,8 +94,24 @@
             } while (nblocks != 0);
             av_freep(&s->exptab);
 #if defined(HAVE_MMX)
-            s->fft_calc = ff_fft_calc_sse;
-#else
+#ifdef HAVE_MM3DNOW
+            if (has_vectors & MM_3DNOWEXT)
+                /* 3DNowEx for Athlon(XP) */
+                s->fft_calc = ff_fft_calc_3dn2;
+            else if (has_vectors & MM_3DNOW)
+                /* 3DNow! for K6-2/3 */
+                s->fft_calc = ff_fft_calc_3dn;
+#endif
+#ifdef HAVE_BUILTIN_VECTOR
+            if (has_vectors & MM_SSE2)
+                /* SSE for P4/K8 */
+                s->fft_calc = ff_fft_calc_sse;
+            else if ((has_vectors & MM_SSE) &&
+                     s->fft_calc == ff_fft_calc_c)
+                /* SSE for P3 */
+                s->fft_calc = ff_fft_calc_sse;
+#endif
+#else /* HAVE_MMX */
             s->fft_calc = ff_fft_calc_altivec;
 #endif
         }