[RFC PATCH v3 08/11] [media] vimc: Optimize frame generation through the pipe
From: Helen Koike
Date: Fri Jun 02 2017 - 23:01:30 EST
Add a parameter called vsen_tpg, if true then vimc will work as before:
frames will be generated in the sensor nodes then propagated through the
pipe and processed by each node until a capture node is reached.
If vsen_tpg is false, then the frame is not generated by the sensor, but
directly from the capture node, thus saving intermediate memory buffers
and process time, allowing a higher frame rate.
Signed-off-by: Helen Koike <helen.koike@xxxxxxxxxxxxx>
---
Changes in v3:
[media] vimc: Optimize frame generation through the pipe
- This is a new patch in the series
Changes in v2: None
---
drivers/media/platform/vimc/vimc-capture.c | 178 +++++++++++++++++++++--------
drivers/media/platform/vimc/vimc-common.h | 2 +
drivers/media/platform/vimc/vimc-sensor.c | 30 ++++-
3 files changed, 156 insertions(+), 54 deletions(-)
diff --git a/drivers/media/platform/vimc/vimc-capture.c b/drivers/media/platform/vimc/vimc-capture.c
index e943267..b5da0ea 100644
--- a/drivers/media/platform/vimc/vimc-capture.c
+++ b/drivers/media/platform/vimc/vimc-capture.c
@@ -15,7 +15,10 @@
*
*/
+#include <linux/freezer.h>
+#include <linux/kthread.h>
#include <media/v4l2-ioctl.h>
+#include <media/v4l2-tpg.h>
#include <media/videobuf2-core.h>
#include <media/videobuf2-vmalloc.h>
@@ -38,6 +41,8 @@ struct vimc_cap_device {
struct mutex lock;
u32 sequence;
struct media_pipeline pipe;
+ struct tpg_data tpg;
+ struct task_struct *kthread_cap;
};
static const struct v4l2_pix_format fmt_default = {
@@ -246,6 +251,91 @@ static void vimc_cap_return_all_buffers(struct vimc_cap_device *vcap,
spin_unlock(&vcap->qlock);
}
+static void vimc_cap_process_frame(struct vimc_ent_device *ved,
+ struct media_pad *sink, const void *frame)
+{
+ struct vimc_cap_device *vcap = container_of(ved, struct vimc_cap_device,
+ ved);
+ struct vimc_cap_buffer *vimc_buf;
+ void *vbuf;
+
+ spin_lock(&vcap->qlock);
+
+ /* Get the first entry of the list */
+ vimc_buf = list_first_entry_or_null(&vcap->buf_list,
+ typeof(*vimc_buf), list);
+ if (!vimc_buf) {
+ spin_unlock(&vcap->qlock);
+ return;
+ }
+
+ /* Remove this entry from the list */
+ list_del(&vimc_buf->list);
+
+ spin_unlock(&vcap->qlock);
+
+ /* Fill the buffer */
+ vimc_buf->vb2.vb2_buf.timestamp = ktime_get_ns();
+ vimc_buf->vb2.sequence = vcap->sequence++;
+ vimc_buf->vb2.field = vcap->format.field;
+
+ vbuf = vb2_plane_vaddr(&vimc_buf->vb2.vb2_buf, 0);
+
+ if (sink)
+ memcpy(vbuf, frame, vcap->format.sizeimage);
+ else
+ tpg_fill_plane_buffer(&vcap->tpg, V4L2_STD_PAL, 0, vbuf);
+
+ /* Set it as ready */
+ vb2_set_plane_payload(&vimc_buf->vb2.vb2_buf, 0,
+ vcap->format.sizeimage);
+ vb2_buffer_done(&vimc_buf->vb2.vb2_buf, VB2_BUF_STATE_DONE);
+}
+
+
+static int vimc_cap_tpg_thread(void *data)
+{
+ struct vimc_cap_device *vcap = data;
+
+ set_freezable();
+ set_current_state(TASK_UNINTERRUPTIBLE);
+
+ for (;;) {
+ try_to_freeze();
+ if (kthread_should_stop())
+ break;
+
+ vimc_cap_process_frame(&vcap->ved, NULL, NULL);
+
+ /* 60 frames per second */
+ schedule_timeout(HZ/60);
+ }
+
+ return 0;
+}
+
+static void vimc_cap_tpg_s_format(struct vimc_cap_device *vcap)
+{
+ const struct vimc_pix_map *vpix =
+ vimc_pix_map_by_pixelformat(vcap->format.pixelformat);
+
+ tpg_reset_source(&vcap->tpg, vcap->format.width, vcap->format.height,
+ vcap->format.field);
+ tpg_s_bytesperline(&vcap->tpg, 0, vcap->format.width * vpix->bpp);
+ tpg_s_buf_height(&vcap->tpg, vcap->format.height);
+ tpg_s_fourcc(&vcap->tpg, vpix->pixelformat);
+ /*
+ * TODO: check why the tpg_s_field need this third argument if
+ * it is already receiving the field
+ */
+ tpg_s_field(&vcap->tpg, vcap->format.field,
+ vcap->format.field == V4L2_FIELD_ALTERNATE);
+ tpg_s_colorspace(&vcap->tpg, vcap->format.colorspace);
+ tpg_s_ycbcr_enc(&vcap->tpg, vcap->format.ycbcr_enc);
+ tpg_s_quantization(&vcap->tpg, vcap->format.quantization);
+ tpg_s_xfer_func(&vcap->tpg, vcap->format.xfer_func);
+}
+
static int vimc_cap_start_streaming(struct vb2_queue *vq, unsigned int count)
{
struct vimc_cap_device *vcap = vb2_get_drv_priv(vq);
@@ -256,20 +346,43 @@ static int vimc_cap_start_streaming(struct vb2_queue *vq, unsigned int count)
/* Start the media pipeline */
ret = media_pipeline_start(entity, &vcap->pipe);
- if (ret) {
- vimc_cap_return_all_buffers(vcap, VB2_BUF_STATE_QUEUED);
- return ret;
- }
+ if (ret)
+ goto err_ret_all_buffs;
/* Enable streaming from the pipe */
ret = vimc_pipeline_s_stream(&vcap->vdev.entity, 1);
- if (ret) {
- media_pipeline_stop(entity);
- vimc_cap_return_all_buffers(vcap, VB2_BUF_STATE_QUEUED);
- return ret;
+ if (ret < 0)
+ goto err_mpipe_stop;
+
+ if (ret == VIMC_PIPE_OPT) {
+ tpg_init(&vcap->tpg, vcap->format.width, vcap->format.height);
+ ret = tpg_alloc(&vcap->tpg, VIMC_FRAME_MAX_WIDTH);
+ if (ret)
+ /* We don't need to call vimc_pipeline_s_stream(e, 0) */
+ goto err_mpipe_stop;
+
+ vimc_cap_tpg_s_format(vcap);
+ vcap->kthread_cap = kthread_run(vimc_cap_tpg_thread, vcap,
+ "%s-cap", vcap->vdev.v4l2_dev->name);
+ if (IS_ERR(vcap->kthread_cap)) {
+ dev_err(vcap->vdev.v4l2_dev->dev,
+ "%s: kernel_thread() failed\n",
+ vcap->vdev.name);
+ ret = PTR_ERR(vcap->kthread_cap);
+ goto err_tpg_free;
+ }
}
return 0;
+
+err_tpg_free:
+ tpg_free(&vcap->tpg);
+err_mpipe_stop:
+ media_pipeline_stop(entity);
+err_ret_all_buffs:
+ vimc_cap_return_all_buffers(vcap, VB2_BUF_STATE_QUEUED);
+
+ return ret;
}
/*
@@ -280,8 +393,15 @@ static void vimc_cap_stop_streaming(struct vb2_queue *vq)
{
struct vimc_cap_device *vcap = vb2_get_drv_priv(vq);
- /* Disable streaming from the pipe */
- vimc_pipeline_s_stream(&vcap->vdev.entity, 0);
+ if (vcap->kthread_cap) {
+ /* Stop image generator */
+ kthread_stop(vcap->kthread_cap);
+ vcap->kthread_cap = NULL;
+ tpg_free(&vcap->tpg);
+ } else {
+ /* Disable streaming from the pipe */
+ vimc_pipeline_s_stream(&vcap->vdev.entity, 0);
+ }
/* Stop the media pipeline */
media_pipeline_stop(&vcap->vdev.entity);
@@ -361,44 +481,6 @@ static void vimc_cap_destroy(struct vimc_ent_device *ved)
kfree(vcap);
}
-static void vimc_cap_process_frame(struct vimc_ent_device *ved,
- struct media_pad *sink, const void *frame)
-{
- struct vimc_cap_device *vcap = container_of(ved, struct vimc_cap_device,
- ved);
- struct vimc_cap_buffer *vimc_buf;
- void *vbuf;
-
- spin_lock(&vcap->qlock);
-
- /* Get the first entry of the list */
- vimc_buf = list_first_entry_or_null(&vcap->buf_list,
- typeof(*vimc_buf), list);
- if (!vimc_buf) {
- spin_unlock(&vcap->qlock);
- return;
- }
-
- /* Remove this entry from the list */
- list_del(&vimc_buf->list);
-
- spin_unlock(&vcap->qlock);
-
- /* Fill the buffer */
- vimc_buf->vb2.vb2_buf.timestamp = ktime_get_ns();
- vimc_buf->vb2.sequence = vcap->sequence++;
- vimc_buf->vb2.field = vcap->format.field;
-
- vbuf = vb2_plane_vaddr(&vimc_buf->vb2.vb2_buf, 0);
-
- memcpy(vbuf, frame, vcap->format.sizeimage);
-
- /* Set it as ready */
- vb2_set_plane_payload(&vimc_buf->vb2.vb2_buf, 0,
- vcap->format.sizeimage);
- vb2_buffer_done(&vimc_buf->vb2.vb2_buf, VB2_BUF_STATE_DONE);
-}
-
struct vimc_ent_device *vimc_cap_create(struct v4l2_device *v4l2_dev,
const char *const name,
u16 num_pads,
diff --git a/drivers/media/platform/vimc/vimc-common.h b/drivers/media/platform/vimc/vimc-common.h
index 2189fd6..5b2691e 100644
--- a/drivers/media/platform/vimc/vimc-common.h
+++ b/drivers/media/platform/vimc/vimc-common.h
@@ -27,6 +27,8 @@
#define VIMC_FRAME_MIN_WIDTH 16
#define VIMC_FRAME_MIN_HEIGHT 16
+#define VIMC_PIPE_OPT 1
+
/**
* struct vimc_pix_map - maps media bus code with v4l2 pixel format
*
diff --git a/drivers/media/platform/vimc/vimc-sensor.c b/drivers/media/platform/vimc/vimc-sensor.c
index 90c41c6..df89ee7 100644
--- a/drivers/media/platform/vimc/vimc-sensor.c
+++ b/drivers/media/platform/vimc/vimc-sensor.c
@@ -15,6 +15,7 @@
*
*/
+#include <linux/module.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include <linux/v4l2-mediabus.h>
@@ -24,6 +25,13 @@
#include "vimc-sensor.h"
+static bool vsen_tpg;
+module_param(vsen_tpg, bool, 0000);
+MODULE_PARM_DESC(vsen_tpg, " generate image from sensor node\n"
+ "If set to false, then the pipe will be optimized and image will be "
+ "generated directly in the capture node instead of going through "
+ "the whole pipe");
+
struct vimc_sen_device {
struct vimc_ent_device ved;
struct v4l2_subdev sd;
@@ -239,6 +247,13 @@ static int vimc_sen_s_stream(struct v4l2_subdev *sd, int enable)
container_of(sd, struct vimc_sen_device, sd);
int ret;
+ if (!vsen_tpg)
+ /*
+ * If we are not generating the frames, then inform the caller
+ * to generate the frame in shallow level of the pipe
+ */
+ return VIMC_PIPE_OPT;
+
if (enable) {
const struct vimc_pix_map *vpix;
unsigned int frame_size;
@@ -306,7 +321,8 @@ static void vimc_sen_destroy(struct vimc_ent_device *ved)
container_of(ved, struct vimc_sen_device, ved);
vimc_ent_sd_unregister(ved, &vsen->sd);
- tpg_free(&vsen->tpg);
+ if (vsen_tpg)
+ tpg_free(&vsen->tpg);
kfree(vsen);
}
@@ -344,11 +360,13 @@ struct vimc_ent_device *vimc_sen_create(struct v4l2_device *v4l2_dev,
vsen->mbus_format = fmt_default;
/* Initialize the test pattern generator */
- tpg_init(&vsen->tpg, vsen->mbus_format.width,
- vsen->mbus_format.height);
- ret = tpg_alloc(&vsen->tpg, VIMC_FRAME_MAX_WIDTH);
- if (ret)
- goto err_unregister_ent_sd;
+ if (vsen_tpg) {
+ tpg_init(&vsen->tpg, vsen->mbus_format.width,
+ vsen->mbus_format.height);
+ ret = tpg_alloc(&vsen->tpg, VIMC_FRAME_MAX_WIDTH);
+ if (ret)
+ goto err_unregister_ent_sd;
+ }
return &vsen->ved;
--
2.7.4