This patch adds support for the CODA960 VPU in Freescale i.MX6 SoCs.
It enables h.264 and MPEG4 encoding and decoding support. Besides the usual
register shifting, the CODA960 gains frame memory control and GDI registers
that are set up for linear mapping right now, needs ENC_PIC_SRC_INDEX to be
set beyond the number of internal buffers for some reason, and has subsampling
buffers that need to be set up. Also, the work buffer size is increased to
80 KiB.
The CODA960 firmware spins if there is not enough input data in the bitstream
buffer. To make it continue, buffers need to be copied into the bitstream as
soon as they are queued. As the bitstream fifo is written into from two places,
it must be protected with a mutex. For that, using a threaded interrupt handler
is necessary.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
This patch enables decoding of h.264 and mpeg4 streams on CODA7541.
Queued output buffers are immediately copied into the bitstream
ringbuffer. A device_run can be scheduled whenever there is either
enough compressed bitstream data, or the CODA is in stream end mode.
Each successful device_run, data is read from the bitstream ringbuffer
and a frame is decoded into a free internal framebuffer. Depending on
reordering, a possibly previously decoded frame is marked as display
frame, and at the same time the display frame from the previous run
is copied out into a capture buffer by the rotator hardware.
The dequeued capture buffers are counted to send the EOS signal to
userspace with the last frame. When userspace sends the decoder stop
command or enqueues an empty output buffer, the stream end flag is
set to allow decoding the remaining frames in the bitstream
ringbuffer.
The enum_fmt/try_fmt functions return fixed capture buffer sizes
while the output queue is streaming, to allow better autonegotiation
in userspace.
A per-context buffer mutex is used to lock the picture run against
buffer dequeueing: if a job gets queued, then streamoff dequeues
the last buffer, and then device_run is called, bail out. For that
the interrupt handler has to be threaded.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
Add a bitstream ringbuffer using kfifo. Queued source buffers are to be copied
into the bitstream ringbuffer immediately and marked as done, if possible.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
This patch splits the global workbuf into a global tempbuf and a per-context
workbuf, adds the codec mode aux register, and restores the work buffer
pointer on commands. With the new firmware, there is only a single set of
read/write pointers which need to be restored between context switches.
This allows more than four active contexts at the same time.
All auxiliary buffers are now allocated through a helper function to avoid
code duplication.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
This sets up IRAM areas used as temporary memory for the different
hardware units depending on the frame size.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
This patch adds a list of supported codecs per device type which is used to
determine the allowed pixel formats and maximum frame sizes depending on the
possible codecs. It allows frame sizes larger than 720 x 576 on CODA7 and adds
support for the YVU420 (planar YUV 4:2:0 with switched Cb and Cr) format.
Other uncompressed formats that could be added in the future are the chroma
interleaved NV12 and NV21 formats and the multiplanar variants YUV420M,
YVU420M, NV12M, and NV21M.
Further, rawstreamon is renamed to streamon_out and compstreamon is renamed
to streamon_cap in preparation for decoding support. The maximum encoded
frame buffer size is increased to 1 MiB.
Instead of tracking inst_type and codec across S_FMT calls, set the codec
and inst_type in start_streaming.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
GAMMA_OFFSET is different between CodaDx6 and CODA7.
Also, this is a bitfield.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Kamil Debski <k.debski@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
The hardware can also rotate in 90° steps, but there is no
corresponding V4L2_CID defined yet.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Tested-by: Javier Martin <javier.martin@vista-silicon.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
This uses the ARCH_MXC specific iram_alloc API to allocate a work
buffer in the SoC's on-chip SRAM and sets up the AXI_SRAM_USE
register. In the future, the allocation will be converted to use
the genalloc API.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Tested-by: Javier Martin <javier.martin@vista-silicon.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
The remaining drivers are mostly platform drivers. Name the
dir to reflect it.
It makes sense to latter break it into a few other dirs.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
