Add patches for block-level quant in theora, -O2 on build, and statically link pthread on Windows

r18986_theora_block_quant.patch

+commit bab197002b561fb53abfa6e24def881f6ec86c9b
+Author: conrad <conrad@9553f0bf-9b14-0410-a0b8-cfaf0461ba5b>
+Date:   Fri May 29 21:43:22 2009 +0000
+
+    Support block-level quantization in Theora
+    
+    git-svn-id: file:///var/local/repositories/ffmpeg/trunk@18986 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b
+
+diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c
+index 9f7dfb6..1646dd9 100644
+--- a/libavcodec/vp3.c
++++ b/libavcodec/vp3.c
+@@ -60,6 +60,7 @@ typedef struct Vp3Fragment {
+     uint8_t coding_method;
+     int8_t motion_x;
+     int8_t motion_y;
++    uint8_t qpi;
+ } Vp3Fragment;
+ 
+ #define SB_NOT_CODED        0
+@@ -134,10 +135,9 @@ typedef struct Vp3DecodeContext {
+     DSPContext dsp;
+     int flipped_image;
+ 
+-    int qis[3];
+-    int nqis;
+-    int quality_index;
+-    int last_quality_index;
++    int qps[3];
++    int nqps;
++    int last_qps[3];
+ 
+     int superblock_count;
+     int y_superblock_width;
+@@ -191,7 +191,7 @@ typedef struct Vp3DecodeContext {
+ 
+     /* these arrays need to be on 16-byte boundaries since SSE2 operations
+      * index into them */
+-    DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]);        //<qmat[is_inter][plane]
++    DECLARE_ALIGNED_16(int16_t, qmat[3][2][3][64]);     //<qmat[qpi][is_inter][plane]
+ 
+     /* This table contains superblock_count * 16 entries. Each set of 16
+      * numbers corresponds to the fragment indexes 0..15 of the superblock.
+@@ -467,6 +467,7 @@ static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
+         s->all_fragments[i].motion_x = 127;
+         s->all_fragments[i].motion_y = 127;
+         s->all_fragments[i].next_coeff= NULL;
++        s->all_fragments[i].qpi = 0;
+         s->coeffs[i].index=
+         s->coeffs[i].coeff=0;
+         s->coeffs[i].next= NULL;
+@@ -477,10 +478,10 @@ static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
+  * This function sets up the dequantization tables used for a particular
+  * frame.
+  */
+-static void init_dequantizer(Vp3DecodeContext *s)
++static void init_dequantizer(Vp3DecodeContext *s, int qpi)
+ {
+-    int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
+-    int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
++    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
++    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
+     int i, plane, inter, qri, bmi, bmj, qistart;
+ 
+     for(inter=0; inter<2; inter++){
+@@ -488,27 +489,29 @@ static void init_dequantizer(Vp3DecodeContext *s)
+             int sum=0;
+             for(qri=0; qri<s->qr_count[inter][plane]; qri++){
+                 sum+= s->qr_size[inter][plane][qri];
+-                if(s->quality_index <= sum)
++                if(s->qps[qpi] <= sum)
+                     break;
+             }
+             qistart= sum - s->qr_size[inter][plane][qri];
+             bmi= s->qr_base[inter][plane][qri  ];
+             bmj= s->qr_base[inter][plane][qri+1];
+             for(i=0; i<64; i++){
+-                int coeff= (  2*(sum    -s->quality_index)*s->base_matrix[bmi][i]
+-                            - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
++                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
++                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
+                             + s->qr_size[inter][plane][qri])
+                            / (2*s->qr_size[inter][plane][qri]);
+ 
+                 int qmin= 8<<(inter + !i);
+                 int qscale= i ? ac_scale_factor : dc_scale_factor;
+ 
+-                s->qmat[inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
++                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
+             }
++            // all DC coefficients use the same quant so as not to interfere with DC prediction
++            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
+         }
+     }
+ 
+-    memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
++    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
+ }
+ 
+ /*
+@@ -521,7 +524,7 @@ static void init_loop_filter(Vp3DecodeContext *s)
+     int filter_limit;
+     int x;
+ 
+-    filter_limit = s->filter_limit_values[s->quality_index];
++    filter_limit = s->filter_limit_values[s->qps[0]];
+ 
+     /* set up the bounding values */
+     memset(s->bounding_values_array, 0, 256 * sizeof(int));
+@@ -963,6 +966,47 @@ static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
+     return 0;
+ }
+ 
++static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
++{
++    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
++    int num_blocks = s->coded_fragment_list_index;
++
++    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
++        i = blocks_decoded = num_blocks_at_qpi = 0;
++
++        bit = get_bits1(gb);
++
++        do {
++            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
++            if (run_length == 34)
++                run_length += get_bits(gb, 12);
++            blocks_decoded += run_length;
++
++            if (!bit)
++                num_blocks_at_qpi += run_length;
++
++            for (j = 0; j < run_length; i++) {
++                if (i > s->coded_fragment_list_index)
++                    return -1;
++
++                if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
++                    s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
++                    j++;
++                }
++            }
++
++            if (run_length == 4129)
++                bit = get_bits1(gb);
++            else
++                bit ^= 1;
++        } while (blocks_decoded < num_blocks);
++
++        num_blocks -= num_blocks_at_qpi;
++    }
++
++    return 0;
++}
++
+ /*
+  * This function is called by unpack_dct_coeffs() to extract the VLCs from
+  * the bitstream. The VLCs encode tokens which are used to unpack DCT
+@@ -1394,9 +1438,9 @@ static void render_slice(Vp3DecodeContext *s, int slice)
+                                 motion_source + stride + 1 + d,
+                                 stride, 8);
+                         }
+-                        dequantizer = s->qmat[1][plane];
++                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
+                     }else{
+-                        dequantizer = s->qmat[0][plane];
++                        dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
+                     }
+ 
+                     /* dequantize the DCT coefficients */
+@@ -1648,7 +1692,8 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
+ 
+     /* initialize to an impossible value which will force a recalculation
+      * in the first frame decode */
+-    s->quality_index = -1;
++    for (i = 0; i < 3; i++)
++        s->qps[i] = -1;
+ 
+     s->y_superblock_width = (s->width + 31) / 32;
+     s->y_superblock_height = (s->height + 31) / 32;
+@@ -1819,24 +1864,29 @@ static int vp3_decode_frame(AVCodecContext *avctx,
+     s->keyframe = !get_bits1(&gb);
+     if (!s->theora)
+         skip_bits(&gb, 1);
+-    s->last_quality_index = s->quality_index;
++    for (i = 0; i < 3; i++)
++        s->last_qps[i] = s->qps[i];
+ 
+-    s->nqis=0;
++    s->nqps=0;
+     do{
+-        s->qis[s->nqis++]= get_bits(&gb, 6);
+-    } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
+-
+-    s->quality_index= s->qis[0];
++        s->qps[s->nqps++]= get_bits(&gb, 6);
++    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
++    for (i = s->nqps; i < 3; i++)
++        s->qps[i] = -1;
+ 
+     if (s->avctx->debug & FF_DEBUG_PICT_INFO)
+         av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
+-            s->keyframe?"key":"", counter, s->quality_index);
++            s->keyframe?"key":"", counter, s->qps[0]);
+     counter++;
+ 
+-    if (s->quality_index != s->last_quality_index) {
+-        init_dequantizer(s);
++    if (s->qps[0] != s->last_qps[0])
+         init_loop_filter(s);
+-    }
++
++    for (i = 0; i < s->nqps; i++)
++        // reinit all dequantizers if the first one changed, because
++        // the DC of the first quantizer must be used for all matrices
++        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
++            init_dequantizer(s, i);
+ 
+     if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
+         return buf_size;
+@@ -1916,6 +1966,10 @@ static int vp3_decode_frame(AVCodecContext *avctx,
+         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
+         return -1;
+     }
++    if (unpack_block_qpis(s, &gb)){
++        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
++        return -1;
++    }
+     if (unpack_dct_coeffs(s, &gb)){
+         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
+         return -1;