]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/mpegvideo_xvmc.c
xsubdec: Convert to the new bitstream reader
[ffmpeg] / libavcodec / mpegvideo_xvmc.c
index ca01b9adeab8d0aeaaf60455915e47467c45c78a..b7de79c5832475b1328b50c336d986688d07d768 100644 (file)
  * XVideo Motion Compensation
  * Copyright (c) 2003 Ivan Kalvachev
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav 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,
+ * Libav 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
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 #include <limits.h>
+#include <X11/extensions/XvMC.h>
 
-//avcodec include
 #include "avcodec.h"
-#include "dsputil.h"
+#include "mpegutils.h"
 #include "mpegvideo.h"
 
 #undef NDEBUG
 #include <assert.h>
 
-//X11 includes are in libavcodec/xvmc.h
-//by replacing it with non-X one
-//XvMC emulation could be performed
-
 #include "xvmc.h"
+#include "xvmc_internal.h"
+#include "version.h"
 
-//#include "xvmc_debug.h"
+#if FF_API_XVMC
 
-//set s->block
-void XVMC_init_block(MpegEncContext *s){
-    struct xvmc_render_state * render;
-    render = (struct xvmc_render_state*)s->current_picture.data[2];
-    assert(render != NULL);
-    if( (render == NULL) || (render->magic != MP_XVMC_RENDER_MAGIC) ){
-        assert(0);
-        return;//make sure that this is a render packet
-    }
-    s->block =(DCTELEM *)(render->data_blocks+(render->next_free_data_block_num)*64);
+/**
+ * Initialize the block field of the MpegEncContext pointer passed as
+ * parameter after making sure that the data is not corrupted.
+ * In order to implement something like direct rendering instead of decoding
+ * coefficients in s->blocks and then copying them, copy them directly
+ * into the data_blocks array provided by xvmc.
+ */
+void ff_xvmc_init_block(MpegEncContext *s)
+{
+    struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
+    assert(render && render->xvmc_id == AV_XVMC_ID);
+
+    s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64);
 }
 
-void XVMC_pack_pblocks(MpegEncContext *s, int cbp){
-int i,j;
-const int mb_block_count = 4+(1<<s->chroma_format);
-
-    j=0;
-    cbp<<= 12-mb_block_count;
-    for(i=0; i<mb_block_count; i++){
-        if(cbp & (1<<11)) {
-           s->pblocks[i] = (short *)(&s->block[(j++)]);
-        }else{
-           s->pblocks[i] = NULL;
-        }
-        cbp+=cbp;
-//        printf("s->pblocks[%d]=%p ,s->block=%p cbp=%d\n",i,s->pblocks[i],s->block,cbp);
+/**
+ * Fill individual block pointers, so there are no gaps in the data_block array
+ * in case not all blocks in the macroblock are coded.
+ */
+void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
+{
+    int i, j = 0;
+    const int mb_block_count = 4 + (1 << s->chroma_format);
+
+    cbp <<= 12-mb_block_count;
+    for (i = 0; i < mb_block_count; i++) {
+        if (cbp & (1 << 11))
+            s->pblocks[i] = &s->block[j++];
+        else
+            s->pblocks[i] = NULL;
+        cbp += cbp;
     }
 }
 
-//These functions should be called on every new field and/or frame.
-//They should be safe if they are called a few times for the same field!
-int XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx){
-    struct xvmc_render_state * render, * last, * next;
-
-    assert(avctx != NULL);
+/**
+ * Find and store the surfaces that are used as reference frames.
+ * This function should be called for every new field and/or frame.
+ * It should be safe to call the function a few times for the same field.
+ */
+int ff_xvmc_field_start(MpegEncContext *s, AVCodecContext *avctx)
+{
+    struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
+    const int mb_block_count = 4 + (1 << s->chroma_format);
+
+    assert(avctx);
+    if (!render || render->xvmc_id != AV_XVMC_ID ||
+        !render->data_blocks || !render->mv_blocks ||
+        (unsigned int)render->allocated_mv_blocks   > INT_MAX/(64*6) ||
+        (unsigned int)render->allocated_data_blocks > INT_MAX/64     ||
+        !render->p_surface) {
+        av_log(avctx, AV_LOG_ERROR,
+               "Render token doesn't look as expected.\n");
+        return -1; // make sure that this is a render packet
+    }
 
-    render = (struct xvmc_render_state*)s->current_picture.data[2];
-    assert(render != NULL);
-    if( (render == NULL) || (render->magic != MP_XVMC_RENDER_MAGIC) )
-        return -1;//make sure that this is render packet
+    if (render->filled_mv_blocks_num) {
+        av_log(avctx, AV_LOG_ERROR,
+               "Rendering surface contains %i unprocessed blocks.\n",
+               render->filled_mv_blocks_num);
+        return -1;
+    }
+    if (render->allocated_mv_blocks   < 1 ||
+        render->allocated_data_blocks <  render->allocated_mv_blocks*mb_block_count ||
+        render->start_mv_blocks_num   >= render->allocated_mv_blocks                ||
+        render->next_free_data_block_num >
+                        render->allocated_data_blocks -
+                        mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) {
+        av_log(avctx, AV_LOG_ERROR,
+               "Rendering surface doesn't provide enough block structures to work with.\n");
+        return -1;
+    }
 
     render->picture_structure = s->picture_structure;
-    render->flags = (s->first_field)? 0: XVMC_SECOND_FIELD;
-
-//make sure that all data is drawn by XVMC_end_frame
-    assert(render->filled_mv_blocks_num==0);
-
-    render->p_future_surface = NULL;
-    render->p_past_surface = NULL;
-
-    switch(s->pict_type){
-        case  FF_I_TYPE:
-            return 0;// no prediction from other frames
-        case  FF_B_TYPE:
-            next = (struct xvmc_render_state*)s->next_picture.data[2];
-            assert(next!=NULL);
-            assert(next->state & MP_XVMC_STATE_PREDICTION);
-            if(next == NULL) return -1;
-            if(next->magic != MP_XVMC_RENDER_MAGIC) return -1;
+    render->flags             = s->first_field ? 0 : XVMC_SECOND_FIELD;
+    render->p_future_surface  = NULL;
+    render->p_past_surface    = NULL;
+
+    switch(s->pict_type) {
+        case  AV_PICTURE_TYPE_I:
+            return 0; // no prediction from other frames
+        case  AV_PICTURE_TYPE_B:
+            next = (struct xvmc_pix_fmt*)s->next_picture.f->data[2];
+            if (!next)
+                return -1;
+            if (next->xvmc_id != AV_XVMC_ID)
+                return -1;
             render->p_future_surface = next->p_surface;
-            //no return here, going to set forward prediction
-        case  FF_P_TYPE:
-            last = (struct xvmc_render_state*)s->last_picture.data[2];
-            if(last == NULL)// && !s->first_field)
-                last = render;//predict second field from the first
-            if(last->magic != MP_XVMC_RENDER_MAGIC) return -1;
-            assert(last->state & MP_XVMC_STATE_PREDICTION);
+            // no return here, going to set forward prediction
+        case  AV_PICTURE_TYPE_P:
+            last = (struct xvmc_pix_fmt*)s->last_picture.f->data[2];
+            if (!last)
+                last = render; // predict second field from the first
+            if (last->xvmc_id != AV_XVMC_ID)
+                return -1;
             render->p_past_surface = last->p_surface;
             return 0;
     }
@@ -111,84 +136,89 @@ int XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx){
 return -1;
 }
 
-void XVMC_field_end(MpegEncContext *s){
-    struct xvmc_render_state * render;
-    render = (struct xvmc_render_state*)s->current_picture.data[2];
-    assert(render != NULL);
+/**
+ * Complete frame/field rendering by passing any remaining blocks.
+ * Normally ff_draw_horiz_band() is called for each slice, however,
+ * some leftover blocks, for example from error_resilience(), may remain.
+ * It should be safe to call the function a few times for the same field.
+ */
+void ff_xvmc_field_end(MpegEncContext *s)
+{
+    struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
+    assert(render);
 
-    if(render->filled_mv_blocks_num > 0){
-//        printf("xvmc: rendering %d left blocks after last slice!!!\n",render->filled_mv_blocks_num );
-        ff_draw_horiz_band(s,0,0);
-    }
+    if (render->filled_mv_blocks_num > 0)
+        ff_mpeg_draw_horiz_band(s, 0, 0);
 }
 
-void XVMC_decode_mb(MpegEncContext *s){
-XvMCMacroBlock * mv_block;
-struct xvmc_render_state * render;
-int i,cbp,blocks_per_mb;
+/**
+ * Synthesize the data needed by XvMC to render one macroblock of data.
+ * Fill all relevant fields, if necessary do IDCT.
+ */
+void ff_xvmc_decode_mb(MpegEncContext *s)
+{
+    XvMCMacroBlock *mv_block;
+    struct xvmc_pix_fmt *render;
+    int i, cbp, blocks_per_mb;
 
-const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
+    const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
 
 
-    if(s->encoding){
+    if (s->encoding) {
         av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
         return;
     }
 
-   //from MPV_decode_mb(),
-    /* update DC predictors for P macroblocks */
+    // from ff_mpv_decode_mb(), update DC predictors for P macroblocks
     if (!s->mb_intra) {
         s->last_dc[0] =
         s->last_dc[1] =
         s->last_dc[2] =  128 << s->intra_dc_precision;
     }
 
-   //MC doesn't skip blocks
+    // MC doesn't skip blocks
     s->mb_skipped = 0;
 
 
-   // Do I need to export quant when I could not perform postprocessing?
-   // Anyway, it doesn't hurt.
+    // Do I need to export quant when I could not perform postprocessing?
+    // Anyway, it doesn't hurt.
     s->current_picture.qscale_table[mb_xy] = s->qscale;
 
-//START OF XVMC specific code
-    render = (struct xvmc_render_state*)s->current_picture.data[2];
-    assert(render!=NULL);
-    assert(render->magic==MP_XVMC_RENDER_MAGIC);
+    // start of XVMC-specific code
+    render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
+    assert(render);
+    assert(render->xvmc_id == AV_XVMC_ID);
     assert(render->mv_blocks);
 
-    //take the next free macroblock
+    // take the next free macroblock
     mv_block = &render->mv_blocks[render->start_mv_blocks_num +
-                                   render->filled_mv_blocks_num ];
-
-// memset(mv_block,0,sizeof(XvMCMacroBlock));
-
-    mv_block->x = s->mb_x;
-    mv_block->y = s->mb_y;
-    mv_block->dct_type = s->interlaced_dct;//XVMC_DCT_TYPE_FRAME/FIELD;
-//    mv_block->motion_type = 0;  //zero to silence warnings
-    if(s->mb_intra){
-        mv_block->macroblock_type = XVMC_MB_TYPE_INTRA;//no MC, all done
-    }else{
+                                  render->filled_mv_blocks_num];
+
+    mv_block->x        = s->mb_x;
+    mv_block->y        = s->mb_y;
+    mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
+    if (s->mb_intra) {
+        mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
+    } else {
         mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
 
-        if(s->mv_dir & MV_DIR_FORWARD){
-            mv_block->macroblock_type|= XVMC_MB_TYPE_MOTION_FORWARD;
-            //pmv[n][dir][xy]=mv[dir][n][xy]
+        if (s->mv_dir & MV_DIR_FORWARD) {
+            mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
+            // PMV[n][dir][xy] = mv[dir][n][xy]
             mv_block->PMV[0][0][0] = s->mv[0][0][0];
             mv_block->PMV[0][0][1] = s->mv[0][0][1];
             mv_block->PMV[1][0][0] = s->mv[0][1][0];
             mv_block->PMV[1][0][1] = s->mv[0][1][1];
         }
-        if(s->mv_dir & MV_DIR_BACKWARD){
-            mv_block->macroblock_type|=XVMC_MB_TYPE_MOTION_BACKWARD;
+        if (s->mv_dir & MV_DIR_BACKWARD) {
+            mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
             mv_block->PMV[0][1][0] = s->mv[1][0][0];
             mv_block->PMV[0][1][1] = s->mv[1][0][1];
             mv_block->PMV[1][1][0] = s->mv[1][1][0];
             mv_block->PMV[1][1][1] = s->mv[1][1][1];
         }
 
-        switch(s->mv_type){
+        switch(s->mv_type) {
             case  MV_TYPE_16X16:
                 mv_block->motion_type = XVMC_PREDICTION_FRAME;
                 break;
@@ -197,32 +227,32 @@ const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
                 break;
             case  MV_TYPE_FIELD:
                 mv_block->motion_type = XVMC_PREDICTION_FIELD;
-                if(s->picture_structure == PICT_FRAME){
-                    mv_block->PMV[0][0][1]<<=1;
-                    mv_block->PMV[1][0][1]<<=1;
-                    mv_block->PMV[0][1][1]<<=1;
-                    mv_block->PMV[1][1][1]<<=1;
+                if (s->picture_structure == PICT_FRAME) {
+                    mv_block->PMV[0][0][1] <<= 1;
+                    mv_block->PMV[1][0][1] <<= 1;
+                    mv_block->PMV[0][1][1] <<= 1;
+                    mv_block->PMV[1][1][1] <<= 1;
                 }
                 break;
             case  MV_TYPE_DMV:
                 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
-                if(s->picture_structure == PICT_FRAME){
+                if (s->picture_structure == PICT_FRAME) {
 
-                    mv_block->PMV[0][0][0] = s->mv[0][0][0];//top from top
-                    mv_block->PMV[0][0][1] = s->mv[0][0][1]<<1;
+                    mv_block->PMV[0][0][0] = s->mv[0][0][0];      // top from top
+                    mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
 
-                    mv_block->PMV[0][1][0] = s->mv[0][0][0];//bottom from bottom
-                    mv_block->PMV[0][1][1] = s->mv[0][0][1]<<1;
+                    mv_block->PMV[0][1][0] = s->mv[0][0][0];      // bottom from bottom
+                    mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
 
-                    mv_block->PMV[1][0][0] = s->mv[0][2][0];//dmv00, top from bottom
-                    mv_block->PMV[1][0][1] = s->mv[0][2][1]<<1;//dmv01
+                    mv_block->PMV[1][0][0] = s->mv[0][2][0];      // dmv00, top from bottom
+                    mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
 
-                    mv_block->PMV[1][1][0] = s->mv[0][3][0];//dmv10, bottom from top
-                    mv_block->PMV[1][1][1] = s->mv[0][3][1]<<1;//dmv11
+                    mv_block->PMV[1][1][0] = s->mv[0][3][0];      // dmv10, bottom from top
+                    mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
 
-                }else{
-                    mv_block->PMV[0][1][0] = s->mv[0][2][0];//dmv00
-                    mv_block->PMV[0][1][1] = s->mv[0][2][1];//dmv01
+                } else {
+                    mv_block->PMV[0][1][0] = s->mv[0][2][0];      // dmv00
+                    mv_block->PMV[0][1][1] = s->mv[0][2][1];      // dmv01
                 }
                 break;
             default:
@@ -231,80 +261,77 @@ const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
 
         mv_block->motion_vertical_field_select = 0;
 
-//set correct field references
-        if(s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8){
-            if( s->field_select[0][0] ) mv_block->motion_vertical_field_select|=1;
-            if( s->field_select[1][0] ) mv_block->motion_vertical_field_select|=2;
-            if( s->field_select[0][1] ) mv_block->motion_vertical_field_select|=4;
-            if( s->field_select[1][1] ) mv_block->motion_vertical_field_select|=8;
+        // set correct field references
+        if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
+            mv_block->motion_vertical_field_select |= s->field_select[0][0];
+            mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
+            mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
+            mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
         }
-    }//!intra
-//time to handle data blocks;
+    } // !intra
+    // time to handle data blocks
     mv_block->index = render->next_free_data_block_num;
 
     blocks_per_mb = 6;
-    if( s->chroma_format >= 2){
-        blocks_per_mb = 4 + (1 << (s->chroma_format));
+    if (s->chroma_format >= 2) {
+        blocks_per_mb = 4 + (1 << s->chroma_format);
     }
 
-//  calculate cbp
+    // calculate cbp
     cbp = 0;
-    for(i=0; i<blocks_per_mb; i++) {
-        cbp+= cbp;
-        if(s->block_last_index[i] >= 0)
+    for (i = 0; i < blocks_per_mb; i++) {
+        cbp += cbp;
+        if (s->block_last_index[i] >= 0)
             cbp++;
     }
 
-    if(s->flags & CODEC_FLAG_GRAY){
-        if(s->mb_intra){//intra frames are always full chroma block
-            for(i=4; i<blocks_per_mb; i++){
-                memset(s->pblocks[i],0,sizeof(short)*8*8);//so we need to clear them
-                if(!render->unsigned_intra)
-                    s->pblocks[i][0] = 1<<10;
+    if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
+        if (s->mb_intra) {                                   // intra frames are always full chroma blocks
+            for (i = 4; i < blocks_per_mb; i++) {
+                memset(s->pblocks[i], 0, sizeof(*s->pblocks[i]));  // so we need to clear them
+                if (!render->unsigned_intra)
+                    *s->pblocks[i][0] = 1 << 10;
             }
-        }else{
-            cbp&= 0xf << (blocks_per_mb - 4);
-            blocks_per_mb = 4;//luminance blocks only
+        } else {
+            cbp &= 0xf << (blocks_per_mb - 4);
+            blocks_per_mb = 4;                               // luminance blocks only
         }
     }
     mv_block->coded_block_pattern = cbp;
-    if(cbp == 0)
+    if (cbp == 0)
         mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
 
-    for(i=0; i<blocks_per_mb; i++){
-        if(s->block_last_index[i] >= 0){
+    for (i = 0; i < blocks_per_mb; i++) {
+        if (s->block_last_index[i] >= 0) {
             // I do not have unsigned_intra MOCO to test, hope it is OK.
-            if( (s->mb_intra) && ( render->idct || (!render->idct && !render->unsigned_intra)) )
-                s->pblocks[i][0]-=1<<10;
-            if(!render->idct){
-                s->dsp.idct(s->pblocks[i]);
-                //!!TODO!clip!!!
+            if (s->mb_intra && (render->idct || !render->unsigned_intra))
+                *s->pblocks[i][0] -= 1 << 10;
+            if (!render->idct) {
+                s->idsp.idct(*s->pblocks[i]);
+                /* It is unclear if MC hardware requires pixel diff values to be
+                 * in the range [-255;255]. TODO: Clipping if such hardware is
+                 * ever found. As of now it would only be an unnecessary
+                 * slowdown. */
             }
-//copy blocks only if the codec doesn't support pblocks reordering
-            if(s->avctx->xvmc_acceleration == 1){
-                memcpy(&render->data_blocks[(render->next_free_data_block_num)*64],
-                        s->pblocks[i],sizeof(short)*8*8);
-            }else{
-/*              if(s->pblocks[i] != &render->data_blocks[
-                        (render->next_free_data_block_num)*64]){
-                   printf("ERROR mb(%d,%d) s->pblocks[i]=%p data_block[]=%p\n",
-                   s->mb_x,s->mb_y, s->pblocks[i],
-                   &render->data_blocks[(render->next_free_data_block_num)*64]);
-                }*/
+            // copy blocks only if the codec doesn't support pblocks reordering
+            if (s->avctx->xvmc_acceleration == 1) {
+                memcpy(&render->data_blocks[render->next_free_data_block_num*64],
+                       s->pblocks[i], sizeof(*s->pblocks[i]));
             }
             render->next_free_data_block_num++;
         }
     }
     render->filled_mv_blocks_num++;
 
-    assert(render->filled_mv_blocks_num <= render->total_number_of_mv_blocks);
-    assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
-
+    assert(render->filled_mv_blocks_num     <= render->allocated_mv_blocks);
+    assert(render->next_free_data_block_num <= render->allocated_data_blocks);
+    /* The above conditions should not be able to fail as long as this function
+     * is used and the following 'if ()' automatically calls a callback to free
+     * blocks. */
 
-    if(render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
-        ff_draw_horiz_band(s,0,0);
-
-// DumpRenderInfo(render);
-// DumpMBlockInfo(mv_block);
 
+    if (render->filled_mv_blocks_num == render->allocated_mv_blocks)
+        ff_mpeg_draw_horiz_band(s, 0, 0);
 }
+
+#endif /* FF_API_XVMC */