[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20220107093455.73766-1-wenst@chromium.org>
Date: Fri, 7 Jan 2022 17:34:47 +0800
From: Chen-Yu Tsai <wenst@...omium.org>
To: Ezequiel Garcia <ezequiel@...guardiasur.com.ar>,
Philipp Zabel <p.zabel@...gutronix.de>,
Mauro Carvalho Chehab <mchehab@...nel.org>,
Hans Verkuil <hverkuil-cisco@...all.nl>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc: Chen-Yu Tsai <wenst@...omium.org>, linux-media@...r.kernel.org,
linux-rockchip@...ts.infradead.org, linux-staging@...ts.linux.dev,
linux-kernel@...r.kernel.org
Subject: [PATCH RFT v2 0/8] media: hantro: jpeg: Various improvements
Hi everyone,
Here are some improvements to the Hantro JPEG encoder driver. This
finishes two of the TODO items.
Changes since v1:
- New patch 8 cleaning up JPEG quantization table dimension magic
numbers and sanity checking of various tables
- Add sanity checks for JPEG header size and alignment
- Drop linux/dma-mapping.h header now that the DMA bounce buffer is gone
- Make JPEG_ACTIVE_MARKER control read only
Patch 1 cleans up the final register write sequence in the JPEG encoder
driver. This particular bit was a bit confusing and hard to understand
given the lack of context around the original wmb(). Was it used to
force all the register writes to finish or to make sure memory writes
were completed? In the end I stuck with what the other hantro decoders
were doing.
Patch 2 fixes a misleading register name.
Patch 3 implements cropping on the output queue with the selection API
for the JPEG encoder. This allows specifying the visible area slightly
smaller than the macroblock-aligned coded size. This bit can be reused
by other stateless encoders once they are implemented.
Patch 4 adds a JFIF APP0 segment to the JPEG encoder output.
Patch 5 adds a COM segment to the JPEG encoder output. This is used to
align the SOS segment payload area.
Patch 6 implements the V4L2_CID_JPEG_ACTIVE_MARKER control. This is only
used to signal the segments added to userspace. The driver ignores any
changes requested.
Patch 7 lets the encoder output directly into the capture buffer,
getting rid of the DMA bounce buffer.
Patch 8 cleans up some of the magic number 64 instances around the
quantization table code. Sanity checks are also added.
Please have a look and test. I only tested this on the RK3399 with
gstreamer, and with Chromium's jpeg_encode_accelerator_unittest with the
patches backported to v5.10 on a RK3399 Kevin Chromebook. The H1 variant
is untested by me.
To test the selection API bits with gstreamer, the v4l2videoenc plugin
needs to be patched. A gst_v4l2_object_set_crop() call should be
inserted after gst_v4l2_object_set_format() in
gst_v4l2_video_enc_set_format().
Regards
ChenYu
Chen-Yu Tsai (8):
media: hantro: jpeg: Relax register writes before write starting
hardware
media: hantro: Fix overfill bottom register field name
media: hantro: Support cropping visible area for encoders
media: hantro: jpeg: Add JFIF APP0 segment to JPEG encoder output
media: hantro: jpeg: Add COM segment to JPEG header to align image
scan
media: hantro: Implement V4L2_CID_JPEG_ACTIVE_MARKER control
media: hantro: output encoded JPEG content directly to capture buffers
media: hantro: jpeg: Remove open-coded size in quantization table code
drivers/staging/media/hantro/TODO | 7 --
drivers/staging/media/hantro/hantro.h | 1 -
drivers/staging/media/hantro/hantro_drv.c | 41 ++++++---
.../staging/media/hantro/hantro_h1_jpeg_enc.c | 44 ++++++----
drivers/staging/media/hantro/hantro_h1_regs.h | 2 +-
drivers/staging/media/hantro/hantro_hw.h | 11 ---
drivers/staging/media/hantro/hantro_jpeg.c | 86 ++++++++++---------
drivers/staging/media/hantro/hantro_jpeg.h | 2 +-
drivers/staging/media/hantro/hantro_v4l2.c | 77 +++++++++++++++++
.../media/hantro/rockchip_vpu2_hw_jpeg_enc.c | 47 +++++-----
.../staging/media/hantro/rockchip_vpu_hw.c | 6 --
11 files changed, 206 insertions(+), 118 deletions(-)
--
2.34.1.575.g55b058a8bb-goog
Powered by blists - more mailing lists