Merge remote branch 'qatar/master'

[ffmpeg] / libavcodec / dnxhdenc.c

diff --git a/libavcodec/dnxhdenc.c b/libavcodec/dnxhdenc.c
index 72378f816de261469cfa866f238c3c21c0090c41..6d1a17fd0adfbc9a189f93d03da9395e4ec692c3 100644 (file)
--- a/libavcodec/dnxhdenc.c
+++ b/libavcodec/dnxhdenc.c
@@ -24,11 +24,20 @@
 //#define DEBUG
 #define RC_VARIANCE 1 // use variance or ssd for fast rc
 
+#include "libavutil/opt.h"
 #include "avcodec.h"
 #include "dsputil.h"
 #include "mpegvideo.h"
 #include "dnxhdenc.h"
 
+#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
+
+static const AVOption options[]={
+    {"nitris_compat", "encode with Avid Nitris compatibility", offsetof(DNXHDEncContext, nitris_compat), FF_OPT_TYPE_INT, 0, 0, 1, VE},
+{NULL}
+};
+static const AVClass class = { "dnxhd", av_default_item_name, options, LIBAVUTIL_VERSION_INT };
+
 int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
 
 #define LAMBDA_FRAC_BITS 10
@@ -146,7 +155,7 @@ static int dnxhd_init_rc(DNXHDEncContext *ctx)
     if (ctx->m.avctx->mb_decision != FF_MB_DECISION_RD)
         FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_cmp, ctx->m.mb_num*sizeof(RCCMPEntry), fail);
 
-    ctx->frame_bits = (ctx->cid_table->coding_unit_size - 640 - 4) * 8;
+    ctx->frame_bits = (ctx->cid_table->coding_unit_size - 640 - 4 - ctx->min_padding) * 8;
     ctx->qscale = 1;
     ctx->lambda = 2<<LAMBDA_FRAC_BITS; // qscale 2
     return 0;
@@ -198,6 +207,10 @@ static int dnxhd_encode_init(AVCodecContext *avctx)
     if (dnxhd_init_qmat(ctx, ctx->m.intra_quant_bias, 0) < 0) // XXX tune lbias/cbias
         return -1;
 
+    // Avid Nitris hardware decoder requires a minimum amount of padding in the coding unit payload
+    if (ctx->nitris_compat)
+        ctx->min_padding = 1600;
+
     if (dnxhd_init_vlc(ctx) < 0)
         return -1;
     if (dnxhd_init_rc(ctx) < 0)
@@ -239,12 +252,12 @@ static int dnxhd_write_header(AVCodecContext *avctx, uint8_t *buf)
     buf[5] = ctx->interlaced ? ctx->cur_field+2 : 0x01;
     buf[6] = 0x80; // crc flag off
     buf[7] = 0xa0; // reserved
-    AV_WB16(buf + 0x18, avctx->height); // ALPF
+    AV_WB16(buf + 0x18, avctx->height>>ctx->interlaced); // ALPF
     AV_WB16(buf + 0x1a, avctx->width);  // SPL
-    AV_WB16(buf + 0x1d, avctx->height); // NAL
+    AV_WB16(buf + 0x1d, avctx->height>>ctx->interlaced); // NAL
 
     buf[0x21] = 0x38; // FIXME 8 bit per comp
-    buf[0x22] = 0x88 + (ctx->frame.interlaced_frame<<2);
+    buf[0x22] = 0x88 + (ctx->interlaced<<2);
     AV_WB32(buf + 0x28, ctx->cid); // CID
     buf[0x2c] = ctx->interlaced ? 0 : 0x80;
 
@@ -398,6 +411,7 @@ static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, i
     DNXHDEncContext *ctx = avctx->priv_data;
     int mb_y = jobnr, mb_x;
     int qscale = ctx->qscale;
+    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
     ctx = ctx->thread[threadnr];
 
     ctx->m.last_dc[0] =
@@ -414,13 +428,12 @@ static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, i
         dnxhd_get_blocks(ctx, mb_x, mb_y);
 
         for (i = 0; i < 8; i++) {
-            DECLARE_ALIGNED_16(DCTELEM, block[64]);
             DCTELEM *src_block = ctx->blocks[i];
             int overflow, nbits, diff, last_index;
             int n = dnxhd_switch_matrix(ctx, i);
 
-            memcpy(block, src_block, sizeof(block));
-            last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
+            memcpy(block, src_block, 64*sizeof(*block));
+            last_index = ctx->m.dct_quantize(&ctx->m, block, i, qscale, &overflow);
             ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
 
             diff = block[0] - ctx->m.last_dc[n];
@@ -465,7 +478,7 @@ static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int
             DCTELEM *block = ctx->blocks[i];
             int last_index, overflow;
             int n = dnxhd_switch_matrix(ctx, i);
-            last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
+            last_index = ctx->m.dct_quantize(&ctx->m, block, i, qscale, &overflow);
             //START_TIMER;
             dnxhd_encode_block(ctx, block, last_index, n);
             //STOP_TIMER("encode_block");
@@ -552,7 +565,7 @@ static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
             if (bits > ctx->frame_bits)
                 break;
         }
-        //dprintf(ctx->m.avctx, "lambda %d, up %u, down %u, bits %d, frame %d\n",
+        //av_dlog(ctx->m.avctx, "lambda %d, up %u, down %u, bits %d, frame %d\n",
         //        lambda, last_higher, last_lower, bits, ctx->frame_bits);
         if (end) {
             if (bits > ctx->frame_bits)
@@ -574,13 +587,15 @@ static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
             last_higher = FFMAX(lambda, last_higher);
             if (last_lower != INT_MAX)
                 lambda = (lambda+last_lower)>>1;
+            else if ((int64_t)lambda + up_step > INT_MAX)
+                return -1;
             else
                 lambda += up_step;
-            up_step *= 5;
+            up_step = FFMIN((int64_t)up_step*5, INT_MAX);
             down_step = 1<<LAMBDA_FRAC_BITS;
         }
     }
-    //dprintf(ctx->m.avctx, "out lambda %d\n", lambda);
+    //av_dlog(ctx->m.avctx, "out lambda %d\n", lambda);
     ctx->lambda = lambda;
     return 0;
 }
@@ -608,7 +623,7 @@ static int dnxhd_find_qscale(DNXHDEncContext *ctx)
             if (bits > ctx->frame_bits)
                 break;
         }
-        //dprintf(ctx->m.avctx, "%d, qscale %d, bits %d, frame %d, higher %d, lower %d\n",
+        //av_dlog(ctx->m.avctx, "%d, qscale %d, bits %d, frame %d, higher %d, lower %d\n",
         //        ctx->m.avctx->frame_number, qscale, bits, ctx->frame_bits, last_higher, last_lower);
         if (bits < ctx->frame_bits) {
             if (qscale == 1)
@@ -638,7 +653,7 @@ static int dnxhd_find_qscale(DNXHDEncContext *ctx)
                 return -1;
         }
     }
-    //dprintf(ctx->m.avctx, "out qscale %d\n", qscale);
+    //av_dlog(ctx->m.avctx, "out qscale %d\n", qscale);
     ctx->qscale = qscale;
     return 0;
 }
@@ -783,7 +798,8 @@ static int dnxhd_encode_picture(AVCodecContext *avctx, unsigned char *buf, int b
     else
         ret = dnxhd_encode_fast(avctx, ctx);
     if (ret < 0) {
-        av_log(avctx, AV_LOG_ERROR, "picture could not fit ratecontrol constraints\n");
+        av_log(avctx, AV_LOG_ERROR,
+               "picture could not fit ratecontrol constraints, increase qmax\n");
         return -1;
     }
 
@@ -845,14 +861,16 @@ static int dnxhd_encode_end(AVCodecContext *avctx)
     return 0;
 }
 
-AVCodec dnxhd_encoder = {
+AVCodec ff_dnxhd_encoder = {
     "dnxhd",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_DNXHD,
     sizeof(DNXHDEncContext),
     dnxhd_encode_init,
     dnxhd_encode_picture,
     dnxhd_encode_end,
+    .capabilities = CODEC_CAP_SLICE_THREADS,
     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUV422P, PIX_FMT_NONE},
     .long_name = NULL_IF_CONFIG_SMALL("VC3/DNxHD"),
+    .priv_class = &class,
 };