floating point AAN IDCT

Originally committed as revision 12233 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/Makefile

@@ -13,6 +13,7 @@ OBJS = allcodecs.o \
        dsputil.o \
        error_resilience.o \
        eval.o \
+       faanidct.o \
        imgconvert.o \
        jrevdct.o \
        mpeg12.o \

libavcodec/avcodec.h

@@ -1326,6 +1326,7 @@ typedef struct AVCodecContext {
 #define FF_IDCT_SIMPLEARMV6   17
 #define FF_IDCT_SIMPLEVIS     18
 #define FF_IDCT_WMV2          19
+#define FF_IDCT_FAAN          20
 
     /**
      * slice count

libavcodec/dct-test.c

@@ -36,6 +36,7 @@
 
 #include "simple_idct.h"
 #include "faandct.h"
+#include "faanidct.h"
 
 #ifndef MAX
 #define MAX(a, b)  (((a) > (b)) ? (a) : (b))
@@ -87,6 +88,7 @@ struct algo {
 struct algo algos[] = {
   DCT_ERROR( "REF-DBL",        0, fdct,               fdct, NO_PERM),
   DCT_ERROR("FAAN",            0, ff_faandct,         fdct, FAAN_SCALE),
+  DCT_ERROR("FAANI",           1, ff_faanidct,        idct, NO_PERM),
   DCT_ERROR("IJG-AAN-INT",     0, fdct_ifast,         fdct, SCALE_PERM),
   DCT_ERROR("IJG-LLM-INT",     0, ff_jpeg_fdct_islow, fdct, NO_PERM),
   DCT_ERROR("REF-DBL",         1, idct,               idct, NO_PERM),

libavcodec/dsputil.c

@@ -32,6 +32,7 @@
 #include "mpegvideo.h"
 #include "simple_idct.h"
 #include "faandct.h"
+#include "faanidct.h"
 #include "h263.h"
 #include "snow.h"
 
@@ -4046,6 +4047,11 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
             c->idct_add= ff_wmv2_idct_add_c;
             c->idct    = ff_wmv2_idct_c;
             c->idct_permutation_type= FF_NO_IDCT_PERM;
+        }else if(avctx->idct_algo==FF_IDCT_FAAN){
+            c->idct_put= ff_faanidct_put;
+            c->idct_add= ff_faanidct_add;
+            c->idct    = ff_faanidct;
+            c->idct_permutation_type= FF_NO_IDCT_PERM;
         }else{ //accurate/default
             c->idct_put= ff_simple_idct_put;
             c->idct_add= ff_simple_idct_add;

libavcodec/faanidct.c

@@ -0,0 +1,166 @@
+/*
+ * Floating point AAN IDCT
+ * Copyright (c) 2008 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+#include "mpegvideo.h"
+
+#define FLOAT float
+
+#define B0 1.0000000000000000000000
+#define B1 1.3870398453221474618216 // cos(pi*1/16)sqrt(2)
+#define B2 1.3065629648763765278566 // cos(pi*2/16)sqrt(2)
+#define B3 1.1758756024193587169745 // cos(pi*3/16)sqrt(2)
+#define B4 1.0000000000000000000000 // cos(pi*4/16)sqrt(2)
+#define B5 0.7856949583871021812779 // cos(pi*5/16)sqrt(2)
+#define B6 0.5411961001461969843997 // cos(pi*6/16)sqrt(2)
+#define B7 0.2758993792829430123360 // cos(pi*7/16)sqrt(2)
+
+#define A4 0.70710678118654752438 // cos(pi*4/16)
+#define A2 0.92387953251128675613 // cos(pi*2/16)
+
+static const FLOAT prescale[64]={
+B0*B0/8, B0*B1/8, B0*B2/8, B0*B3/8, B0*B4/8, B0*B5/8, B0*B6/8, B0*B7/8,
+B1*B0/8, B1*B1/8, B1*B2/8, B1*B3/8, B1*B4/8, B1*B5/8, B1*B6/8, B1*B7/8,
+B2*B0/8, B2*B1/8, B2*B2/8, B2*B3/8, B2*B4/8, B2*B5/8, B2*B6/8, B2*B7/8,
+B3*B0/8, B3*B1/8, B3*B2/8, B3*B3/8, B3*B4/8, B3*B5/8, B3*B6/8, B3*B7/8,
+B4*B0/8, B4*B1/8, B4*B2/8, B4*B3/8, B4*B4/8, B4*B5/8, B4*B6/8, B4*B7/8,
+B5*B0/8, B5*B1/8, B5*B2/8, B5*B3/8, B5*B4/8, B5*B5/8, B5*B6/8, B5*B7/8,
+B6*B0/8, B6*B1/8, B6*B2/8, B6*B3/8, B6*B4/8, B6*B5/8, B6*B6/8, B6*B7/8,
+B7*B0/8, B7*B1/8, B7*B2/8, B7*B3/8, B7*B4/8, B7*B5/8, B7*B6/8, B7*B7/8,
+};
+
+static inline void p8idct(DCTELEM data[64], FLOAT temp[64], uint8_t *dest, int stride, int x, int y, int type){
+    int i;
+    FLOAT tmp0, tmp1;
+    FLOAT s04, d04, s17, d17, s26, d26, s53, d53;
+    FLOAT os07, os16, os25, os34;
+    FLOAT od07, od16, od25, od34;
+
+    for(i=0; i<y*8; i+=y){
+        s17= temp[1*x + i] + temp[7*x + i];
+        d17= temp[1*x + i] - temp[7*x + i];
+        s53= temp[5*x + i] + temp[3*x + i];
+        d53= temp[5*x + i] - temp[3*x + i];
+
+        od07=  s17 + s53;
+        od25= (s17 - s53)*(2*A4);
+
+#if 0 //these 2 are equivalent
+        tmp0= (d17 + d53)*(2*A2);
+        od34=  d17*( 2*B6) - tmp0;
+        od16=  d53*(-2*B2) + tmp0;
+#else
+        od34=  d17*(2*(B6-A2)) - d53*(2*A2);
+        od16=  d53*(2*(A2-B2)) + d17*(2*A2);
+#endif
+
+        od16 -= od07;
+        od25 -= od16;
+        od34 += od25;
+
+        s26 = temp[2*x + i] + temp[6*x + i];
+        d26 = temp[2*x + i] - temp[6*x + i];
+        tmp1= d26*(2*A4) - s26;
+
+        s04= temp[0*x + i] + temp[4*x + i];
+        d04= temp[0*x + i] - temp[4*x + i];
+
+        os07= s04 + s26;
+        os34= s04 - s26;
+        os16= d04 + tmp1;
+        os25= d04 - tmp1;
+
+        if(type==0){
+            temp[0*x + i]= os07 + od07;
+            temp[7*x + i]= os07 - od07;
+            temp[1*x + i]= os16 + od16;
+            temp[6*x + i]= os16 - od16;
+            temp[2*x + i]= os25 + od25;
+            temp[5*x + i]= os25 - od25;
+            temp[3*x + i]= os34 - od34;
+            temp[4*x + i]= os34 + od34;
+        }else if(type==1){
+            data[0*x + i]= lrintf(os07 + od07);
+            data[7*x + i]= lrintf(os07 - od07);
+            data[1*x + i]= lrintf(os16 + od16);
+            data[6*x + i]= lrintf(os16 - od16);
+            data[2*x + i]= lrintf(os25 + od25);
+            data[5*x + i]= lrintf(os25 - od25);
+            data[3*x + i]= lrintf(os34 - od34);
+            data[4*x + i]= lrintf(os34 + od34);
+        }else if(type==2){
+            dest[0*stride + i]= av_clip_uint8(((int)dest[0*stride + i]) + lrintf(os07 + od07));
+            dest[7*stride + i]= av_clip_uint8(((int)dest[7*stride + i]) + lrintf(os07 - od07));
+            dest[1*stride + i]= av_clip_uint8(((int)dest[1*stride + i]) + lrintf(os16 + od16));
+            dest[6*stride + i]= av_clip_uint8(((int)dest[6*stride + i]) + lrintf(os16 - od16));
+            dest[2*stride + i]= av_clip_uint8(((int)dest[2*stride + i]) + lrintf(os25 + od25));
+            dest[5*stride + i]= av_clip_uint8(((int)dest[5*stride + i]) + lrintf(os25 - od25));
+            dest[3*stride + i]= av_clip_uint8(((int)dest[3*stride + i]) + lrintf(os34 - od34));
+            dest[4*stride + i]= av_clip_uint8(((int)dest[4*stride + i]) + lrintf(os34 + od34));
+        }else{
+            dest[0*stride + i]= av_clip_uint8(lrintf(os07 + od07));
+            dest[7*stride + i]= av_clip_uint8(lrintf(os07 - od07));
+            dest[1*stride + i]= av_clip_uint8(lrintf(os16 + od16));
+            dest[6*stride + i]= av_clip_uint8(lrintf(os16 - od16));
+            dest[2*stride + i]= av_clip_uint8(lrintf(os25 + od25));
+            dest[5*stride + i]= av_clip_uint8(lrintf(os25 - od25));
+            dest[3*stride + i]= av_clip_uint8(lrintf(os34 - od34));
+            dest[4*stride + i]= av_clip_uint8(lrintf(os34 + od34));
+        }
+    }
+}
+
+void ff_faanidct(DCTELEM block[64]){
+    FLOAT temp[64];
+    int i;
+
+    emms_c();
+
+    for(i=0; i<64; i++)
+        temp[i] = block[i] * prescale[i];
+
+    p8idct(block, temp, NULL, 0, 1, 8, 0);
+    p8idct(block, temp, NULL, 0, 8, 1, 1);
+}
+
+void ff_faanidct_add(uint8_t *dest, int line_size, DCTELEM block[64]){
+    FLOAT temp[64];
+    int i;
+
+    emms_c();
+
+    for(i=0; i<64; i++)
+        temp[i] = block[i] * prescale[i];
+
+    p8idct(block, temp, NULL,         0, 1, 8, 0);
+    p8idct(NULL , temp, dest, line_size, 8, 1, 2);
+}
+
+void ff_faanidct_put(uint8_t *dest, int line_size, DCTELEM block[64]){
+    FLOAT temp[64];
+    int i;
+
+    emms_c();
+
+    for(i=0; i<64; i++)
+        temp[i] = block[i] * prescale[i];
+
+    p8idct(block, temp, NULL,         0, 1, 8, 0);
+    p8idct(NULL , temp, dest, line_size, 8, 1, 3);
+}

libavcodec/faanidct.h

@@ -0,0 +1,3 @@
+void ff_faanidct(DCTELEM block[64]);
+void ff_faanidct_add(uint8_t *dest, int line_size, DCTELEM block[64]);
+void ff_faanidct_put(uint8_t *dest, int line_size, DCTELEM block[64]);

libavcodec/utils.c

@@ -595,6 +595,7 @@ static const AVOption options[]={
 {"vp3", NULL, 0, FF_OPT_TYPE_CONST, FF_IDCT_VP3, INT_MIN, INT_MAX, V|E|D, "idct"},
 {"ipp", NULL, 0, FF_OPT_TYPE_CONST, FF_IDCT_IPP, INT_MIN, INT_MAX, V|E|D, "idct"},
 {"xvidmmx", NULL, 0, FF_OPT_TYPE_CONST, FF_IDCT_XVIDMMX, INT_MIN, INT_MAX, V|E|D, "idct"},
+{"faani", "floating point AAN IDCT", 0, FF_OPT_TYPE_CONST, FF_IDCT_FAAN, INT_MIN, INT_MAX, V|D|E, "idct"},
 {"slice_count", NULL, OFFSET(slice_count), FF_OPT_TYPE_INT, DEFAULT, INT_MIN, INT_MAX},
 {"ec", "set error concealment strategy", OFFSET(error_concealment), FF_OPT_TYPE_FLAGS, 3, INT_MIN, INT_MAX, V|D, "ec"},
 {"guess_mvs", "iterative motion vector (MV) search (slow)", 0, FF_OPT_TYPE_CONST, FF_EC_GUESS_MVS, INT_MIN, INT_MAX, V|D, "ec"},