[FFmpeg-devel,v1,1/2] avcodec/v410dec: add the slice threading support

Submitted by lance.lmwang@gmail.com on Aug. 30, 2019, 3:53 p.m.

Details

Message ID 20190830155357.27747-1-lance.lmwang@gmail.com
State New
Headers show

Commit Message

lance.lmwang@gmail.com Aug. 30, 2019, 3:53 p.m.
From: Limin Wang <lance.lmwang@gmail.com>

Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
---
 libavcodec/v410dec.c | 77 ++++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 57 insertions(+), 20 deletions(-)

Comments

Paul B Mahol Aug. 30, 2019, 4:32 p.m.
All this patches are missing number showing speedup gains.

On 8/30/19, lance.lmwang@gmail.com <lance.lmwang@gmail.com> wrote:
> From: Limin Wang <lance.lmwang@gmail.com>
>
> Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
> ---
>  libavcodec/v410dec.c | 77
> ++++++++++++++++++++++++++++++++++++++--------------
>  1 file changed, 57 insertions(+), 20 deletions(-)
>
> diff --git a/libavcodec/v410dec.c b/libavcodec/v410dec.c
> index 48fab68..a9b17a0 100644
> --- a/libavcodec/v410dec.c
> +++ b/libavcodec/v410dec.c
> @@ -24,6 +24,14 @@
>  #include "libavutil/intreadwrite.h"
>  #include "avcodec.h"
>  #include "internal.h"
> +#include "thread.h"
> +
> +#define MAX_SLICES 32
> +typedef struct ThreadData {
> +    AVFrame *frame;
> +    uint8_t *buf;
> +    int stride;
> +} ThreadData;
>
>  static av_cold int v410_decode_init(AVCodecContext *avctx)
>  {
> @@ -42,31 +50,30 @@ 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 nb_jobs)
>  {
> -    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, MAX_SLICES);
> +    int slice_h = avctx->height / thread_count;
> +    int slice_m = avctx->height % thread_count;
> +    int slice_start = jobnr * slice_h;
> +    int slice_end = slice_start + slice_h;
> +    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);
> -    }
> +    int i, j;
>
> -    if ((ret = ff_get_buffer(avctx, pic, 0)) < 0)
> -        return ret;
> +     /* add the remaining slice for the last job */
> +    if (jobnr == thread_count - 1)
> +        slice_end += slice_m;
>
> -    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 < avctx->height; i++) {
>          for (j = 0; j < avctx->width; j++) {
>              val = AV_RL32(src);
>
> @@ -82,6 +89,34 @@ 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;
> +    AVFrame *pic = data;
> +    uint8_t *src = avpkt->data;
> +    int ret;
> +    int thread_count = av_clip(avctx->thread_count, 1, MAX_SLICES);
> +
> +    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_get_buffer(avctx, pic, 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 +129,7 @@ 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,
> +    .caps_internal= FF_CODEC_CAP_INIT_THREADSAFE |
> +                      FF_CODEC_CAP_INIT_CLEANUP,
>  };
> --
> 2.6.4
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Patch hide | download patch | download mbox

diff --git a/libavcodec/v410dec.c b/libavcodec/v410dec.c
index 48fab68..a9b17a0 100644
--- a/libavcodec/v410dec.c
+++ b/libavcodec/v410dec.c
@@ -24,6 +24,14 @@ 
 #include "libavutil/intreadwrite.h"
 #include "avcodec.h"
 #include "internal.h"
+#include "thread.h"
+
+#define MAX_SLICES 32
+typedef struct ThreadData {
+    AVFrame *frame;
+    uint8_t *buf;
+    int stride;
+} ThreadData;
 
 static av_cold int v410_decode_init(AVCodecContext *avctx)
 {
@@ -42,31 +50,30 @@  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 nb_jobs)
 {
-    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, MAX_SLICES);
+    int slice_h = avctx->height / thread_count;
+    int slice_m = avctx->height % thread_count;
+    int slice_start = jobnr * slice_h;
+    int slice_end = slice_start + slice_h;
+    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);
-    }
+    int i, j;
 
-    if ((ret = ff_get_buffer(avctx, pic, 0)) < 0)
-        return ret;
+     /* add the remaining slice for the last job */
+    if (jobnr == thread_count - 1)
+        slice_end += slice_m;
 
-    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 < avctx->height; i++) {
         for (j = 0; j < avctx->width; j++) {
             val = AV_RL32(src);
 
@@ -82,6 +89,34 @@  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;
+    AVFrame *pic = data;
+    uint8_t *src = avpkt->data;
+    int ret;
+    int thread_count = av_clip(avctx->thread_count, 1, MAX_SLICES);
+
+    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_get_buffer(avctx, pic, 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 +129,7 @@  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,
+    .caps_internal= FF_CODEC_CAP_INIT_THREADSAFE |
+                      FF_CODEC_CAP_INIT_CLEANUP,
 };