diff mbox series

[v3,4/5] media: meson: vdec: add VP9 input support

Message ID 20200116133437.2443-5-narmstrong@baylibre.com (mailing list archive)
State New, archived
Headers show
Series media: meson: vdec: Add VP9 decoding support | expand

Commit Message

Neil Armstrong Jan. 16, 2020, 1:34 p.m. UTC
From: Maxime Jourdan <mjourdan@baylibre.com>

Amlogic VP9 decoder requires an additional 16-byte payload before every
frame header.

The source buffer is updated in-place, then given to the Parser FIFO DMA.

The FIFO DMA copies the blocks into the 16MiB parser ring buffer, then parses
and copies the slice into the decoder "workspace".

Signed-off-by: Maxime Jourdan <mjourdan@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
---
 drivers/staging/media/meson/vdec/esparser.c | 142 +++++++++++++++++++-
 1 file changed, 138 insertions(+), 4 deletions(-)

Comments

Maxime Jourdan Jan. 24, 2020, 11:18 a.m. UTC | #1
On Thu, Jan 16, 2020 at 2:34 PM Neil Armstrong <narmstrong@baylibre.com> wrote:
>
> From: Maxime Jourdan <mjourdan@baylibre.com>
>
> Amlogic VP9 decoder requires an additional 16-byte payload before every
> frame header.
>
> The source buffer is updated in-place, then given to the Parser FIFO DMA.
>
> The FIFO DMA copies the blocks into the 16MiB parser ring buffer, then parses
> and copies the slice into the decoder "workspace".
>
> Signed-off-by: Maxime Jourdan <mjourdan@baylibre.com>
> Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
> ---

[...]

> @@ -186,13 +294,27 @@ esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)
>         int ret;
>         struct vb2_buffer *vb = &vbuf->vb2_buf;
>         struct amvdec_core *core = sess->core;
> +       struct amvdec_codec_ops *codec_ops = sess->fmt_out->codec_ops;
>         u32 payload_size = vb2_get_plane_payload(vb, 0);
>         dma_addr_t phy = vb2_dma_contig_plane_dma_addr(vb, 0);
> +       u32 num_dst_bufs = 0;
>         u32 offset;
>         u32 pad_size;
>
> -       if (esparser_vififo_get_free_space(sess) < payload_size)
> +       if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9) {
> +               if (codec_ops->num_pending_bufs)
> +                       num_dst_bufs = codec_ops->num_pending_bufs(sess);
> +
> +               num_dst_bufs += v4l2_m2m_num_dst_bufs_ready(sess->m2m_ctx);
> +               if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9)
> +                       num_dst_bufs -= 2;

With the changes that happened in the updated series where one more
ref frame is held by VP9, this should be -= 3 to prevent a shortage of
CAPTURE buffers on the decoder side.

For the future, a good enhancement of the way this is handled could be
to notify new capture buffers to the decoding modules (codec_*.c), so
that they could pause when there is no capture buffer available and
resume on this notification.

> +
> +               if (esparser_vififo_get_free_space(sess) < payload_size ||
> +                   atomic_read(&sess->esparser_queued_bufs) >= num_dst_bufs)
> +                       return -EAGAIN;
> +       } else if (esparser_vififo_get_free_space(sess) < payload_size) {
>                 return -EAGAIN;
> +       }
>
>         v4l2_m2m_src_buf_remove_by_buf(sess->m2m_ctx, vbuf);
>
> @@ -206,7 +328,19 @@ esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)
>         vbuf->field = V4L2_FIELD_NONE;
>         vbuf->sequence = sess->sequence_out++;
>
> -       pad_size = esparser_pad_start_code(core, vb);
> +       if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9) {
> +               payload_size = vp9_update_header(core, vb);
> +
> +               /* If unable to alter buffer to add headers */
> +               if (payload_size == 0) {
> +                       amvdec_remove_ts(sess, vb->timestamp);
> +                       v4l2_m2m_buf_done(vbuf, VB2_BUF_STATE_ERROR);
> +
> +                       return 0;
> +               }
> +       }
> +
> +       pad_size = esparser_pad_start_code(core, vb, payload_size);
>         ret = esparser_write_data(core, phy, payload_size + pad_size);
>
>         if (ret <= 0) {
> --
> 2.22.0
>
diff mbox series

Patch

diff --git a/drivers/staging/media/meson/vdec/esparser.c b/drivers/staging/media/meson/vdec/esparser.c
index adc5c1e81a4c..aeb68f6c732a 100644
--- a/drivers/staging/media/meson/vdec/esparser.c
+++ b/drivers/staging/media/meson/vdec/esparser.c
@@ -52,6 +52,7 @@ 
 #define PARSER_VIDEO_HOLE	0x90
 
 #define SEARCH_PATTERN_LEN	512
+#define VP9_HEADER_SIZE		16
 
 static DECLARE_WAIT_QUEUE_HEAD(wq);
 static int search_done;
@@ -74,14 +75,121 @@  static irqreturn_t esparser_isr(int irq, void *dev)
 	return IRQ_HANDLED;
 }
 
+/**
+ * VP9 frame headers need to be appended by a 16-byte long
+ * Amlogic custom header
+ */
+static int vp9_update_header(struct amvdec_core *core, struct vb2_buffer *buf)
+{
+	u8 *dp;
+	u8 marker;
+	int dsize;
+	int num_frames, cur_frame;
+	int cur_mag, mag, mag_ptr;
+	int frame_size[8], tot_frame_size[8];
+	int total_datasize = 0;
+	int new_frame_size;
+	unsigned char *old_header = NULL;
+
+	dp = (uint8_t *)vb2_plane_vaddr(buf, 0);
+	dsize = vb2_get_plane_payload(buf, 0);
+
+	if (dsize == vb2_plane_size(buf, 0)) {
+		dev_warn(core->dev, "%s: unable to update header\n", __func__);
+		return 0;
+	}
+
+	marker = dp[dsize - 1];
+	if ((marker & 0xe0) == 0xc0) {
+		num_frames = (marker & 0x7) + 1;
+		mag = ((marker >> 3) & 0x3) + 1;
+		mag_ptr = dsize - mag * num_frames - 2;
+		if (dp[mag_ptr] != marker)
+			return 0;
+
+		mag_ptr++;
+		for (cur_frame = 0; cur_frame < num_frames; cur_frame++) {
+			frame_size[cur_frame] = 0;
+			for (cur_mag = 0; cur_mag < mag; cur_mag++) {
+				frame_size[cur_frame] |=
+					(dp[mag_ptr] << (cur_mag * 8));
+				mag_ptr++;
+			}
+			if (cur_frame == 0)
+				tot_frame_size[cur_frame] =
+					frame_size[cur_frame];
+			else
+				tot_frame_size[cur_frame] =
+					tot_frame_size[cur_frame - 1] +
+					frame_size[cur_frame];
+			total_datasize += frame_size[cur_frame];
+		}
+	} else {
+		num_frames = 1;
+		frame_size[0] = dsize;
+		tot_frame_size[0] = dsize;
+		total_datasize = dsize;
+	}
+
+	new_frame_size = total_datasize + num_frames * VP9_HEADER_SIZE;
+
+	if (new_frame_size >= vb2_plane_size(buf, 0)) {
+		dev_warn(core->dev, "%s: unable to update header\n", __func__);
+		return 0;
+	}
+
+	for (cur_frame = num_frames - 1; cur_frame >= 0; cur_frame--) {
+		int framesize = frame_size[cur_frame];
+		int framesize_header = framesize + 4;
+		int oldframeoff = tot_frame_size[cur_frame] - framesize;
+		int outheaderoff =  oldframeoff + cur_frame * VP9_HEADER_SIZE;
+		u8 *fdata = dp + outheaderoff;
+		u8 *old_framedata = dp + oldframeoff;
+
+		memmove(fdata + VP9_HEADER_SIZE, old_framedata, framesize);
+
+		fdata[0] = (framesize_header >> 24) & 0xff;
+		fdata[1] = (framesize_header >> 16) & 0xff;
+		fdata[2] = (framesize_header >> 8) & 0xff;
+		fdata[3] = (framesize_header >> 0) & 0xff;
+		fdata[4] = ((framesize_header >> 24) & 0xff) ^ 0xff;
+		fdata[5] = ((framesize_header >> 16) & 0xff) ^ 0xff;
+		fdata[6] = ((framesize_header >> 8) & 0xff) ^ 0xff;
+		fdata[7] = ((framesize_header >> 0) & 0xff) ^ 0xff;
+		fdata[8] = 0;
+		fdata[9] = 0;
+		fdata[10] = 0;
+		fdata[11] = 1;
+		fdata[12] = 'A';
+		fdata[13] = 'M';
+		fdata[14] = 'L';
+		fdata[15] = 'V';
+
+		if (!old_header) {
+			/* nothing */
+		} else if (old_header > fdata + 16 + framesize) {
+			dev_dbg(core->dev, "%s: data has gaps, setting to 0\n",
+				__func__);
+			memset(fdata + 16 + framesize, 0,
+			       (old_header - fdata + 16 + framesize));
+		} else if (old_header < fdata + 16 + framesize) {
+			dev_err(core->dev, "%s: data overwritten\n", __func__);
+		}
+		old_header = fdata;
+	}
+
+	return new_frame_size;
+}
+
 /* Pad the packet to at least 4KiB bytes otherwise the VDEC unit won't trigger
  * ISRs.
  * Also append a start code 000001ff at the end to trigger
  * the ESPARSER interrupt.
  */
-static u32 esparser_pad_start_code(struct amvdec_core *core, struct vb2_buffer *vb)
+static u32 esparser_pad_start_code(struct amvdec_core *core,
+				   struct vb2_buffer *vb,
+				   u32 payload_size)
 {
-	u32 payload_size = vb2_get_plane_payload(vb, 0);
 	u32 pad_size = 0;
 	u8 *vaddr = vb2_plane_vaddr(vb, 0);
 
@@ -186,13 +294,27 @@  esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)
 	int ret;
 	struct vb2_buffer *vb = &vbuf->vb2_buf;
 	struct amvdec_core *core = sess->core;
+	struct amvdec_codec_ops *codec_ops = sess->fmt_out->codec_ops;
 	u32 payload_size = vb2_get_plane_payload(vb, 0);
 	dma_addr_t phy = vb2_dma_contig_plane_dma_addr(vb, 0);
+	u32 num_dst_bufs = 0;
 	u32 offset;
 	u32 pad_size;
 
-	if (esparser_vififo_get_free_space(sess) < payload_size)
+	if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9) {
+		if (codec_ops->num_pending_bufs)
+			num_dst_bufs = codec_ops->num_pending_bufs(sess);
+
+		num_dst_bufs += v4l2_m2m_num_dst_bufs_ready(sess->m2m_ctx);
+		if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9)
+			num_dst_bufs -= 2;
+
+		if (esparser_vififo_get_free_space(sess) < payload_size ||
+		    atomic_read(&sess->esparser_queued_bufs) >= num_dst_bufs)
+			return -EAGAIN;
+	} else if (esparser_vififo_get_free_space(sess) < payload_size) {
 		return -EAGAIN;
+	}
 
 	v4l2_m2m_src_buf_remove_by_buf(sess->m2m_ctx, vbuf);
 
@@ -206,7 +328,19 @@  esparser_queue(struct amvdec_session *sess, struct vb2_v4l2_buffer *vbuf)
 	vbuf->field = V4L2_FIELD_NONE;
 	vbuf->sequence = sess->sequence_out++;
 
-	pad_size = esparser_pad_start_code(core, vb);
+	if (sess->fmt_out->pixfmt == V4L2_PIX_FMT_VP9) {
+		payload_size = vp9_update_header(core, vb);
+
+		/* If unable to alter buffer to add headers */
+		if (payload_size == 0) {
+			amvdec_remove_ts(sess, vb->timestamp);
+			v4l2_m2m_buf_done(vbuf, VB2_BUF_STATE_ERROR);
+
+			return 0;
+		}
+	}
+
+	pad_size = esparser_pad_start_code(core, vb, payload_size);
 	ret = esparser_write_data(core, phy, payload_size + pad_size);
 
 	if (ret <= 0) {