avcodec/v410dec: add support for frame and slice threading

1, Test server configure:
[root@localhost ~]# cat /proc/cpuinfo  |grep "model name"
model name	: Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz
model name	: Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz
...

[root@localhost ~]# free -h
              total        used        free      shared  buff/cache   available
              Mem:           102G        1.1G        100G         16M        657M        100G
              Swap:          4.0G          0B        4.0G

2, Test result:
encode the v410 input data for testing:
 ./ffmpeg -y -i 4k_422.ts  -c:v v410 -vframes 10 test.avi

master:
./ffmpeg -y -stream_loop 1000 -i ./test.avi -benchmark -f null -
frame=10010 fps= 37 q=-0.0 Lsize=N/A time=00:38:26.30 bitrate=N/A speed= 8.6x
video:5240kB audio:432432kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
bench: utime=166.016s stime=102.192s rtime=268.120s
bench: maxrss=273400kB

patch applied:
./ffmpeg -y -threads 2 -thread_type slice -stream_loop 1000 -i ./test.avi -benchmark -f null -
frame=10010 fps= 53 q=-0.0 Lsize=N/A time=00:38:26.30 bitrate=N/A speed=12.3x
video:5240kB audio:432432kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
bench: utime=165.135s stime=100.456s rtime=187.994s
bench: maxrss=275476kB

./ffmpeg -y -threads 2 -thread_type frame -stream_loop 1000 -i ./test.avi -benchmark -f null -
frame=10010 fps= 61 q=-0.0 Lsize=N/A time=00:38:26.30 bitrate=N/A speed=14.1x
video:5240kB audio:432432kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
bench: utime=171.386s stime=122.102s rtime=163.637s
bench: maxrss=340308kB
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/v410dec.c

@@ -24,6 +24,13 @@
 #include "libavutil/intreadwrite.h"
 #include "avcodec.h"
 #include "internal.h"
+#include "thread.h"
+
+typedef struct ThreadData {
+    AVFrame *frame;
+    uint8_t *buf;
+    int stride;
+} ThreadData;
 
 static av_cold int v410_decode_init(AVCodecContext *avctx)
 {
@@ -42,31 +49,24 @@ static av_cold int v410_decode_init(AVCodecContext *avctx)
     return 0;
 }
 
-static int v410_decode_frame(AVCodecContext *avctx, void *data,
-                             int *got_frame, AVPacket *avpkt)
+static int v410_decode_slice(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
 {
-    AVFrame *pic = data;
-    uint8_t *src = avpkt->data;
+    ThreadData *td = arg;
+    AVFrame *pic = td->frame;
+    int stride = td->stride;
+    int thread_count = av_clip(avctx->thread_count, 1, avctx->height/4);
+    int slice_start = (avctx->height *  jobnr) / thread_count;
+    int slice_end = (avctx->height * (jobnr+1)) / thread_count;
+    const uint8_t *src = td->buf + stride * slice_start;
     uint16_t *y, *u, *v;
     uint32_t val;
-    int i, j, ret;
-
-    if (avpkt->size < 4 * avctx->height * avctx->width) {
-        av_log(avctx, AV_LOG_ERROR, "Insufficient input data.\n");
-        return AVERROR(EINVAL);
-    }
-
-    if ((ret = ff_get_buffer(avctx, pic, 0)) < 0)
-        return ret;
+    int i, j;
 
-    pic->key_frame = 1;
-    pic->pict_type = AV_PICTURE_TYPE_I;
+    y = (uint16_t*)pic->data[0] + slice_start * (pic->linesize[0] >> 1);
+    u = (uint16_t*)pic->data[1] + slice_start * (pic->linesize[1] >> 1);
+    v = (uint16_t*)pic->data[2] + slice_start * (pic->linesize[2] >> 1);
 
-    y = (uint16_t *)pic->data[0];
-    u = (uint16_t *)pic->data[1];
-    v = (uint16_t *)pic->data[2];
-
-    for (i = 0; i < avctx->height; i++) {
+    for (i = slice_start; i < slice_end; i++) {
         for (j = 0; j < avctx->width; j++) {
             val = AV_RL32(src);
 
@@ -82,6 +82,35 @@ static int v410_decode_frame(AVCodecContext *avctx, void *data,
         v += pic->linesize[2] >> 1;
     }
 
+    return 0;
+}
+
+static int v410_decode_frame(AVCodecContext *avctx, void *data,
+                             int *got_frame, AVPacket *avpkt)
+{
+    ThreadData td;
+    ThreadFrame frame = { .f = data };
+    AVFrame *pic = data;
+    uint8_t *src = avpkt->data;
+    int ret;
+    int thread_count = av_clip(avctx->thread_count, 1, avctx->height/4);
+
+    td.stride = avctx->width * 4;
+    if (avpkt->size < 4 * avctx->height * avctx->width) {
+        av_log(avctx, AV_LOG_ERROR, "Insufficient input data.\n");
+        return AVERROR(EINVAL);
+    }
+
+    if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
+        return ret;
+
+    pic->key_frame = 1;
+    pic->pict_type = AV_PICTURE_TYPE_I;
+
+    td.buf = src;
+    td.frame = pic;
+    avctx->execute2(avctx, v410_decode_slice, &td, NULL, thread_count);
+
     *got_frame = 1;
 
     return avpkt->size;
@@ -94,5 +123,6 @@ AVCodec ff_v410_decoder = {
     .id           = AV_CODEC_ID_V410,
     .init         = v410_decode_init,
     .decode       = v410_decode_frame,
-    .capabilities = AV_CODEC_CAP_DR1,
+    .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS |
+                    AV_CODEC_CAP_FRAME_THREADS
 };