porting vp3 idct over to lavc idct api
authorMichael Niedermayer <michaelni@gmx.at>
Tue, 17 May 2005 11:12:04 +0000 (11:12 +0000)
committerMichael Niedermayer <michaelni@gmx.at>
Tue, 17 May 2005 11:12:04 +0000 (11:12 +0000)
Originally committed as revision 4257 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/avcodec.h
libavcodec/dsputil.c
libavcodec/dsputil.h
libavcodec/vp3.c
libavcodec/vp3dsp.c

index b9fccc4..f849f6b 100644 (file)
@@ -1177,6 +1177,7 @@ typedef struct AVCodecContext {
 #define FF_IDCT_SH4          9
 #define FF_IDCT_SIMPLEARM    10
 #define FF_IDCT_H264         11
+#define FF_IDCT_VP3          12
 
     /**
      * slice count.
index a19f79d..5505dff 100644 (file)
@@ -3643,6 +3643,11 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
             c->idct_add= ff_jref_idct_add;
             c->idct    = j_rev_dct;
             c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
+        }else if(avctx->idct_algo==FF_IDCT_VP3){
+            c->idct_put= ff_vp3_idct_put_c;
+            c->idct_add= ff_vp3_idct_add_c;
+            c->idct    = ff_vp3_idct_c;
+            c->idct_permutation_type= FF_NO_IDCT_PERM;
         }else{ //accurate/default
             c->idct_put= simple_idct_put;
             c->idct_add= simple_idct_add;
@@ -3653,10 +3658,6 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
 
     c->h264_idct_add= ff_h264_idct_add_c;
 
-    /* VP3 DSP support */
-    c->vp3_dsp_init = vp3_dsp_init_c;
-    c->vp3_idct = vp3_idct_c;
-
     c->get_pixels = get_pixels_c;
     c->diff_pixels = diff_pixels_c;
     c->put_pixels_clamped = put_pixels_clamped_c;
index 5754743..002e284 100644 (file)
@@ -68,17 +68,9 @@ extern uint32_t squareTbl[512];
 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
 
 /* VP3 DSP functions */
-void vp3_dsp_init_c(void);
-void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, DCTELEM *output_data);
-
-void vp3_dsp_init_mmx(void);
-void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, DCTELEM *output_data);
-
-void vp3_dsp_init_sse2(void);
-void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, DCTELEM *output_data);
+void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
+void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
+void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 
 /* minimum alignment rules ;)
 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
index 7923229..da4f531 100644 (file)
@@ -2053,7 +2053,7 @@ static void render_fragments(Vp3DecodeContext *s,
                              int height,
                              int plane /* 0 = Y, 1 = U, 2 = V */) 
 {
-    int x, y;
+    int x, y, j;
     int m, n;
     int i = first_fragment;
     int16_t *dequantizer;
@@ -2188,21 +2188,32 @@ av_log(s->avctx, AV_LOG_ERROR, " help! got beefy vector! (%X, %X)\n", motion_x,
                     i, s->all_fragments[i].coding_method, 
                     s->all_fragments[i].coeffs[0], dequantizer[0]);
 
+                if(s->avctx->idct_algo==FF_IDCT_VP3){
+                    for (j = 0; j < 64; j++) {
+                        s->all_fragments[i].coeffs[j] *= dequantizer[j];
+                    }
+                }else{
+                    for (j = 0; j < 64; j++) {
+                        s->all_fragments[i].coeffs[j]= (dequantizer[j] * s->all_fragments[i].coeffs[j] + 2) >> 2;
+                    }
+                }
+
                 /* invert DCT and place (or add) in final output */
-                s->dsp.vp3_idct(s->all_fragments[i].coeffs,
-                    dequantizer,
-                    s->all_fragments[i].coeff_count,
-                    output_samples);
-                memset(s->all_fragments[i].coeffs, 0, 64*sizeof(DCTELEM));
+                
                 if (s->all_fragments[i].coding_method == MODE_INTRA) {
-                    s->dsp.put_signed_pixels_clamped(output_samples,
+                    if(s->avctx->idct_algo!=FF_IDCT_VP3)
+                        s->all_fragments[i].coeffs[0] += 128<<3;
+                    s->dsp.idct_put(
                         output_plane + s->all_fragments[i].first_pixel,
-                        stride);
+                        stride,
+                        s->all_fragments[i].coeffs);
                 } else {
-                    s->dsp.add_pixels_clamped(output_samples,
+                    s->dsp.idct_add(
                         output_plane + s->all_fragments[i].first_pixel,
-                        stride);
+                        stride,
+                        s->all_fragments[i].coeffs);
                 }
+                memset(s->all_fragments[i].coeffs, 0, 64*sizeof(DCTELEM));
 
                 debug_idct("block after idct_%s():\n",
                     (s->all_fragments[i].coding_method == MODE_INTRA)?
@@ -2496,8 +2507,9 @@ static int vp3_decode_init(AVCodecContext *avctx)
 #endif
     avctx->pix_fmt = PIX_FMT_YUV420P;
     avctx->has_b_frames = 0;
+    if(avctx->idct_algo==FF_IDCT_AUTO)
+        avctx->idct_algo=FF_IDCT_VP3;
     dsputil_init(&s->dsp, avctx);
-    s->dsp.vp3_dsp_init();
 
     /* initialize to an impossible value which will force a recalculation
      * in the first frame decode */
index 8cf77a2..015f57b 100644 (file)
 #define xC6S2 25080
 #define xC7S1 12785
 
-void vp3_dsp_init_c(void)
+static always_inline void idct(uint8_t *dst, int stride, int16_t *input, int type)
 {
-    /* nop */
-}
-
-void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, int16_t *output_data)
-{
-    int32_t dequantized_data[64];
-    int32_t *ip = dequantized_data;
-    int16_t *op = output_data;
+    int16_t *ip = input;
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;
 
-    int32_t A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_;
-    int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
-    int32_t t1, t2;
+    int A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_;
+    int _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
+    int t1, t2;
 
     int i, j;
     
-    /* de-zigzag and dequantize */
-    for (i = 0; i < coeff_count; i++) {
-        dequantized_data[i] = dequant_matrix[i] * input_data[i];
-    }
-
     /* Inverse DCT on the rows now */
     for (i = 0; i < 8; i++) {
         /* Check for non-zero values */
@@ -131,28 +119,28 @@ void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
             _Hd = _Bd + H_;
 
             /*  Final sequence of operations over-write original inputs. */
-            ip[0] = (int16_t)((_Gd + _Cd )   >> 0);
-            ip[7] = (int16_t)((_Gd - _Cd )   >> 0);
+            ip[0] = _Gd + _Cd ;
+            ip[7] = _Gd - _Cd ;
 
-            ip[1] = (int16_t)((_Add + _Hd )  >> 0);
-            ip[2] = (int16_t)((_Add - _Hd )  >> 0);
+            ip[1] = _Add + _Hd;
+            ip[2] = _Add - _Hd;
 
-            ip[3] = (int16_t)((_Ed + _Dd )   >> 0);
-            ip[4] = (int16_t)((_Ed - _Dd )   >> 0);
+            ip[3] = _Ed + _Dd ;
+            ip[4] = _Ed - _Dd ;
 
-            ip[5] = (int16_t)((_Fd + _Bdd )  >> 0);
-            ip[6] = (int16_t)((_Fd - _Bdd )  >> 0);
+            ip[5] = _Fd + _Bdd;
+            ip[6] = _Fd - _Bdd;
 
         }
 
         ip += 8;            /* next row */
     }
-
-    ip = dequantized_data;
+    
+    ip = input;
 
     for ( i = 0; i < 8; i++) {
         /* Check for non-zero values (bitwise or faster than ||) */
-        if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
+        if ( ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
              ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
 
             t1 = (int32_t)(xC1S7 * ip[1*8]);
@@ -222,37 +210,103 @@ void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
             _Fd = F_ - _Ad;
             _Hd = _Bd + H_;
 
+            if(type==1){  //HACK
+                _Gd += 16*128;
+                _Add+= 16*128;
+                _Ed += 16*128;
+                _Fd += 16*128;
+            }
             _Gd += IdctAdjustBeforeShift;
             _Add += IdctAdjustBeforeShift;
             _Ed += IdctAdjustBeforeShift;
             _Fd += IdctAdjustBeforeShift;
 
             /* Final sequence of operations over-write original inputs. */
-            op[0*8] = (int16_t)((_Gd + _Cd )   >> 4);
-            op[7*8] = (int16_t)((_Gd - _Cd )   >> 4);
-
-            op[1*8] = (int16_t)((_Add + _Hd )  >> 4);
-            op[2*8] = (int16_t)((_Add - _Hd )  >> 4);
-
-            op[3*8] = (int16_t)((_Ed + _Dd )   >> 4);
-            op[4*8] = (int16_t)((_Ed - _Dd )   >> 4);
-
-            op[5*8] = (int16_t)((_Fd + _Bdd )  >> 4);
-            op[6*8] = (int16_t)((_Fd - _Bdd )  >> 4);
+            if(type==0){
+                ip[0*8] = (_Gd + _Cd )  >> 4;
+                ip[7*8] = (_Gd - _Cd )  >> 4;
+    
+                ip[1*8] = (_Add + _Hd ) >> 4;
+                ip[2*8] = (_Add - _Hd ) >> 4;
+    
+                ip[3*8] = (_Ed + _Dd )  >> 4;
+                ip[4*8] = (_Ed - _Dd )  >> 4;
+    
+                ip[5*8] = (_Fd + _Bdd ) >> 4;
+                ip[6*8] = (_Fd - _Bdd ) >> 4;
+            }else if(type==1){
+                dst[0*stride] = cm[(_Gd + _Cd )  >> 4];
+                dst[7*stride] = cm[(_Gd - _Cd )  >> 4];
+    
+                dst[1*stride] = cm[(_Add + _Hd ) >> 4];
+                dst[2*stride] = cm[(_Add - _Hd ) >> 4];
+    
+                dst[3*stride] = cm[(_Ed + _Dd )  >> 4];
+                dst[4*stride] = cm[(_Ed - _Dd )  >> 4];
+    
+                dst[5*stride] = cm[(_Fd + _Bdd ) >> 4];
+                dst[6*stride] = cm[(_Fd - _Bdd ) >> 4];
+            }else{
+                dst[0*stride] = cm[dst[0*stride] + ((_Gd + _Cd )  >> 4)];
+                dst[7*stride] = cm[dst[7*stride] + ((_Gd - _Cd )  >> 4)];
+    
+                dst[1*stride] = cm[dst[1*stride] + ((_Add + _Hd ) >> 4)];
+                dst[2*stride] = cm[dst[2*stride] + ((_Add - _Hd ) >> 4)];
+    
+                dst[3*stride] = cm[dst[3*stride] + ((_Ed + _Dd )  >> 4)];
+                dst[4*stride] = cm[dst[4*stride] + ((_Ed - _Dd )  >> 4)];
+    
+                dst[5*stride] = cm[dst[5*stride] + ((_Fd + _Bdd ) >> 4)];
+                dst[6*stride] = cm[dst[6*stride] + ((_Fd - _Bdd ) >> 4)];
+            }
 
         } else {
-
-            op[0*8] = 0;
-            op[7*8] = 0;
-            op[1*8] = 0;
-            op[2*8] = 0;
-            op[3*8] = 0;
-            op[4*8] = 0;
-            op[5*8] = 0;
-            op[6*8] = 0;
+            if(type==0){
+                ip[0*8] = 
+                ip[1*8] = 
+                ip[2*8] = 
+                ip[3*8] = 
+                ip[4*8] = 
+                ip[5*8] = 
+                ip[6*8] =
+                ip[7*8] = ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20);
+            }else if(type==1){
+                dst[0*stride]=
+                dst[1*stride]=
+                dst[2*stride]=
+                dst[3*stride]=
+                dst[4*stride]=
+                dst[5*stride]=
+                dst[6*stride]=
+                dst[7*stride]= 128 + ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20);
+            }else{
+                if(ip[0*8]){
+                    int v= ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20);
+                    dst[0*stride] = cm[dst[0*stride] + v];
+                    dst[1*stride] = cm[dst[1*stride] + v];
+                    dst[2*stride] = cm[dst[2*stride] + v];
+                    dst[3*stride] = cm[dst[3*stride] + v];
+                    dst[4*stride] = cm[dst[4*stride] + v];
+                    dst[5*stride] = cm[dst[5*stride] + v];
+                    dst[6*stride] = cm[dst[6*stride] + v];
+                    dst[7*stride] = cm[dst[7*stride] + v];
+                }
+            }
         }
 
         ip++;            /* next column */
-        op++;
+        dst++;
     }
 }
+
+void ff_vp3_idct_c(DCTELEM *block/* align 16*/){
+    idct(NULL, 0, block, 0);
+}
+    
+void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){
+    idct(dest, line_size, block, 1);
+}
+
+void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){
+    idct(dest, line_size, block, 2);
+}