syncing ffmpeg (with some compilation fixes)

- fixes wma bugs
- mace, huffyuv and mp3 decoders imported (but not enabled)
tested: wma (v1 and v2), mpeg4, msmpeg4 v1, v2 and v3, divx3, divx4, divx5, xvid and
dv decoders. everything looks fine.

CVS patchset: 3828
CVS date: 2003/01/08 13:18:42

6 files changed, 168 insertions, 46 deletions

diff --git a/src/libffmpeg/libavcodec/ppc/Makefile.am b/src/libffmpeg/libavcodec/ppc/Makefile.am
index d791fe4a8..a623a96f3 100644
--- a/src/libffmpeg/libavcodec/ppc/Makefile.am
+++ b/src/libffmpeg/libavcodec/ppc/Makefile.am
@@ -14,6 +14,7 @@ noinst_LTLIBRARIES = libavcodec_ppc.la
 
 libavcodec_ppc_src =  dsputil_altivec.c \
 		      dsputil_ppc.c \
+		      fft_altivec.c \
 		      idct_altivec.c \
 		      mpegvideo_altivec.c \
 		      mpegvideo_ppc.c
diff --git a/src/libffmpeg/libavcodec/ppc/dsputil_altivec.c b/src/libffmpeg/libavcodec/ppc/dsputil_altivec.c
index ed34a2d92..5f14ed0eb 100644
--- a/src/libffmpeg/libavcodec/ppc/dsputil_altivec.c
+++ b/src/libffmpeg/libavcodec/ppc/dsputil_altivec.c
@@ -343,48 +343,6 @@ int pix_norm1_altivec(uint8_t *pix, int line_size)
     return s;
 }
 
-
-int pix_norm_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
-{
-    int s, i;
-    vector unsigned char *tv, zero;
-    vector unsigned char pix1v, pix2v, t5;
-    vector unsigned int sv;
-    vector signed int sum;
-
-    zero = vec_splat_u8(0);
-    sv = vec_splat_u32(0);
-    s = 0;
-    for (i = 0; i < 16; i++) {
-        /* Read in the potentially unaligned pixels */
-        tv = (vector unsigned char *) pix1;
-        pix1v = vec_perm(tv[0], tv[1], vec_lvsl(0, pix1));
-
-        tv = (vector unsigned char *) pix2;
-        pix2v = vec_perm(tv[0], tv[1], vec_lvsl(0, pix2));
-
-        /*
-           Since we want to use unsigned chars, we can take advantage
-           of the fact that abs(a-b)^2 = (a-b)^2.
-        */
-        
-        /* Calculate a sum of abs differences vector */
-        t5 = vec_sub(vec_max(pix1v, pix2v), vec_min(pix1v, pix2v));
-
-        /* Square the values and add them to our sum */
-        sv = vec_msum(t5, t5, sv);
-        
-        pix1 += line_size;
-        pix2 += line_size;
-    }
-    /* Sum up the four partial sums, and put the result into s */
-    sum = vec_sums((vector signed int) sv, (vector signed int) zero);
-    sum = vec_splat(sum, 3);
-    vec_ste(sum, 0, &s);
-    return s;
-}
-
-
 int pix_sum_altivec(UINT8 * pix, int line_size)
 {
 
diff --git a/src/libffmpeg/libavcodec/ppc/dsputil_altivec.h b/src/libffmpeg/libavcodec/ppc/dsputil_altivec.h
index 94fe3a023..d4d259d9e 100644
--- a/src/libffmpeg/libavcodec/ppc/dsputil_altivec.h
+++ b/src/libffmpeg/libavcodec/ppc/dsputil_altivec.h
@@ -23,7 +23,6 @@ extern int pix_abs16x16_xy2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
 extern int pix_abs16x16_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
 extern int pix_abs8x8_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
 extern int pix_norm1_altivec(uint8_t *pix, int line_size);
-extern int pix_norm_altivec(uint8_t *pix1, uint8_t *pix2, int line_size);
 extern int pix_sum_altivec(UINT8 * pix, int line_size);
 extern void diff_pixels_altivec(DCTELEM* block, const UINT8* s1, const UINT8* s2, int stride);
 extern void get_pixels_altivec(DCTELEM* block, const UINT8 * pixels, int line_size);
diff --git a/src/libffmpeg/libavcodec/ppc/dsputil_ppc.c b/src/libffmpeg/libavcodec/ppc/dsputil_ppc.c
index ffe3ce063..733d0c156 100644
--- a/src/libffmpeg/libavcodec/ppc/dsputil_ppc.c
+++ b/src/libffmpeg/libavcodec/ppc/dsputil_ppc.c
@@ -42,7 +42,6 @@ void dsputil_init_ppc(DSPContext* c, unsigned mask)
         c->pix_abs16x16 = pix_abs16x16_altivec;
         c->pix_abs8x8 = pix_abs8x8_altivec;
         c->pix_norm1 = pix_norm1_altivec;
-        c->pix_norm = pix_norm_altivec;
         c->pix_sum = pix_sum_altivec;
         c->diff_pixels = diff_pixels_altivec;
         c->get_pixels = get_pixels_altivec;
diff --git a/src/libffmpeg/libavcodec/ppc/fft_altivec.c b/src/libffmpeg/libavcodec/ppc/fft_altivec.c
new file mode 100644
index 000000000..1a926b77c
--- /dev/null
+++ b/src/libffmpeg/libavcodec/ppc/fft_altivec.c
@@ -0,0 +1,166 @@
+/*
+ * FFT/IFFT transforms
+ * AltiVec-enabled
+ * Copyright (c) 2002 Romain Dolbeau <romain@dolbeau.org>
+ * Based on code Copyright (c) 2002 Fabrice Bellard.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+#include "../dsputil.h"
+
+#include "dsputil_altivec.h"
+
+// used to build registers permutation vectors (vcprm)
+// the 's' are for words in the _s_econd vector
+#define WORD_0 0x00,0x01,0x02,0x03
+#define WORD_1 0x04,0x05,0x06,0x07
+#define WORD_2 0x08,0x09,0x0a,0x0b
+#define WORD_3 0x0c,0x0d,0x0e,0x0f
+#define WORD_s0 0x10,0x11,0x12,0x13
+#define WORD_s1 0x14,0x15,0x16,0x17
+#define WORD_s2 0x18,0x19,0x1a,0x1b
+#define WORD_s3 0x1c,0x1d,0x1e,0x1f
+
+#define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
+
+// 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.
+
+#define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
+
+/**
+ * Do a complex FFT with the parameters defined in fft_init(). The
+ * input data must be permuted before with s->revtab table. No
+ * 1.0/sqrt(n) normalization is done.
+ * AltiVec-enabled
+ * This code assumes that the 'z' pointer is 16 bytes-aligned
+ * It also assumes all FFTComplex are 8 bytes-aligned pair of float
+ * The code is exactly the same as the SSE version, except
+ * that successive MUL + ADD/SUB have been fusionned into
+ * fused multiply-add ('vec_madd' in altivec)
+ *
+ * To test this code you can use fft-test in libavcodec ; use
+ * the following line in libavcodec to compile (MacOS X):
+ * #####
+ * gcc -I. -Ippc -no-cpp-precomp -pipe -O3 -fomit-frame-pointer -mdynamic-no-pic -Wall
+ *     -faltivec -DARCH_POWERPC -DHAVE_ALTIVEC -DCONFIG_DARWIN fft-test.c fft.c
+ *     ppc/fft_altivec.c ppc/dsputil_altivec.c mdct.c -DHAVE_LRINTF -o fft-test
+ * #####
+ */
+void fft_calc_altivec(FFTContext *s, FFTComplex *z)
+{
+    register const vector float vczero = (vector float)( 0., 0., 0., 0.);
+    
+    int ln = s->nbits;
+    int	j, np, np2;
+    int	nblocks, nloops;
+    register FFTComplex *p, *q;
+    FFTComplex *cptr, *cptr1;
+    int k;
+
+    np = 1 << ln;
+
+    {
+        vector float *r, a, b, a1, c1, c2;
+
+        r = (vector float *)&z[0];
+
+        c1 = vcii(p,p,n,n);
+        
+        if (s->inverse)
+            {
+                c2 = vcii(p,p,n,p);
+            }
+        else
+            {
+                c2 = vcii(p,p,p,n);
+            }
+        
+        j = (np >> 2);
+        do {
+            a = vec_ld(0, r);
+            a1 = vec_ld(sizeof(vector float), r);
+            
+            b = vec_perm(a,a,vcprmle(1,0,3,2));
+            a = vec_madd(a,c1,b);
+            /* do the pass 0 butterfly */
+            
+            b = vec_perm(a1,a1,vcprmle(1,0,3,2));
+            b = vec_madd(a1,c1,b);
+            /* do the pass 0 butterfly */
+            
+            /* multiply third by -i */
+            b = vec_perm(b,b,vcprmle(2,3,1,0));
+            
+            /* do the pass 1 butterfly */
+            vec_st(vec_madd(b,c2,a), 0, r);
+            vec_st(vec_nmsub(b,c2,a), sizeof(vector float), r);
+            
+            r += 2;
+        } while (--j != 0);
+    }
+    /* pass 2 .. ln-1 */
+
+    nblocks = np >> 3;
+    nloops = 1 << 2;
+    np2 = np >> 1;
+
+    cptr1 = s->exptab1;
+    do {
+        p = z;
+        q = z + nloops;
+        j = nblocks;
+        do {
+            cptr = cptr1;
+            k = nloops >> 1;
+            do {
+                vector float a,b,c,t1;
+
+                a = vec_ld(0, (float*)p);
+                b = vec_ld(0, (float*)q);
+                
+                /* complex mul */
+                c = vec_ld(0, (float*)cptr);
+                /*  cre*re cim*re */
+                t1 = vec_madd(c, vec_perm(b,b,vcprmle(2,2,0,0)),vczero);
+                c = vec_ld(sizeof(vector float), (float*)cptr);
+                /*  -cim*im cre*im */
+                b = vec_madd(c, vec_perm(b,b,vcprmle(3,3,1,1)),t1);
+                
+                /* butterfly */
+                vec_st(vec_add(a,b), 0, (float*)p);
+                vec_st(vec_sub(a,b), 0, (float*)q);
+                
+                p += 2;
+                q += 2;
+                cptr += 4;
+            } while (--k);
+            
+            p += nloops;
+            q += nloops;
+        } while (--j);
+        cptr1 += nloops * 2;
+        nblocks = nblocks >> 1;
+        nloops = nloops << 1;
+    } while (nblocks != 0);
+}
+
diff --git a/src/libffmpeg/libavcodec/ppc/mpegvideo_ppc.c b/src/libffmpeg/libavcodec/ppc/mpegvideo_ppc.c
index 18888c8f4..94d608b63 100644
--- a/src/libffmpeg/libavcodec/ppc/mpegvideo_ppc.c
+++ b/src/libffmpeg/libavcodec/ppc/mpegvideo_ppc.c
@@ -16,10 +16,9 @@
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  */

  

-#include <time.h>

-#include "../../config.h"

 #include "../dsputil.h"

 #include "../mpegvideo.h"

+#include <time.h>

 

 #ifdef HAVE_ALTIVEC

 #include "dsputil_altivec.h"