Enable timestamp reordering by using the
calculated timestamps maintained in a
sorted order and associate with output
buffers. This would resolve the issue
if the parser sends incorect timestamps.
Change-Id: I870fd8ed25e05b8e86c677e79182e9c7785656a8
Signed-off-by: Darshana Patil <darshana@codeaurora.org>
Check if the non-pix and pix prefetched memory is sufficient
Change-Id: I71561959a3ed1e9acdb654650144f35656230345
Signed-off-by: Akshay Chandrashekhar Kalghatgi <akalghat@codeaurora.org>
Secure memory allocations take a long time. Use memory prefetch and
drain operations to reduce allocation times.
Change-Id: Iaa86a97d7a8b1d6eb1683d273ddb27ffe8d891ae
Signed-off-by: Akshay Chandrashekhar Kalghatgi <akalghat@codeaurora.org>
Calculate decoder framerate based on input buffer
timestamps and maintain a sorted order list for
timestamps and framerate in a sliding window of
size 32. Send calculated framerate to client in
every dqbuf output buffer. Use the maximum framerate
available in the list for bandwidth and clockrate
calculations.
Change-Id: I5fa357645cb806bf97ccd30ed68bd784d526e233
Signed-off-by: Darshana Patil <darshana@codeaurora.org>
In qbuf (input) store input_tag to etb_data list and send it to
firmware in etb. Firmware will return input_tag in ftb_done,
store it in fbd_data list and return the same when client does
dqbuf (output).
Change-Id: I3589e5a3017cac49000fc64191f9f756bc4011f4
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
1. is_batching_allowed api is not considering platform
check(core->resources.decode_batching).
2. increase both input & output buffers count if
current session supports batching.
Change-Id: Iabe7479b4664a78b637ee981d67c604f7d0d5d56
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
For sub-frame encode sessions, input buffer done is received only after
all sub-frames have been received. In these sub-frame encode cases,
input tags are not fetched for sub-frame outputs for first input and
buffers are dropped in HAL in absence of apt input tags. Limit buffer
counter check from input tag fetch condition for decode sessions only.
Change-Id: I8c584834369be921d8699da49cfa1fec65fb993d
CRs-Fixed: 2547055
Signed-off-by: Amit Shekhar <ashekhar@codeaurora.org>
Add support for vidioc_querymenu for userspace to
query menu type controls.
Change-Id: Idffb7e978e7f6eba9bf2744bcf87bafcd919decf
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
1. Set perf mode for image session buffers so that
they will be processed quickly.
2. Set max frequency for any turbo sessions.
Change-Id: Icf9b0af5cffcd395356783a9ff7a9421a309f52c
Signed-off-by: Shi Zhongbo <zhongbos@codeaurora.org>
Add new ioctl to query the supported ROI map type.
Based on vpu version, report whether support 2-bit type
ROI map, or 2-byte type ROI map.
Change-Id: I6db970f5f36e4cc61365867c2a07d63aaf9365ea
Signed-off-by: Qiwei Liu <qiweil@codeaurora.org>
Signed-off-by: Mihir Ganu <mganu@codeaurora.org>
Currently setting signal_info, only if colorspace info
is set properly. But sometimes client can set only range.
So need to send signal_info always to firmware.
Change-Id: I5282ec01b473485ec39f39abbf7160b9c5cc0256
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
- Made the changes in video module to compile successfully.
- Removed cvp and v4l2 private files.
Change-Id: Iaf5f29d6042a87d5b96b1fa4abb985ee3e592f0e
Signed-off-by: Karthikeyan Periasamy <kperiasa@codeaurora.org>
Introduced CVP skip ratio property to indicate the number of
buffers skipped for CVP processing. This serves as a hint to Video
firmware to know the number of frames for which CVP metadata has to be
reused from earlier frame to which CVP metadata was sent
Change-Id: I58ad5daf9a9c6ae7a0aa48e3bd1d60e7dc9229e6
Signed-off-by: Akshata Sahukar <asahukar@codeaurora.org>
[1] Maintain log_ctxt cookie to track debug information.
[2] Print codec type and session type by default in logs.
Change-Id: I6c01f3238ba868fdc525f5c60f58545a07bdef3a
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
Print session_id details in all possible call flow paths.
Change-Id: I0c583dc99d4ebee7fbcb4e36d917ba9120fa670e
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
Remove freq list, which is unused in current dcvs design.
Remove unused dcvs load, load_low, load_norm, load_high,
as current dcvs design only use min_freq and dcvs_flags.
Change-Id: Ifbf251a9506286f2f34eca6eeb1af79bd07c0f43
Signed-off-by: Qiwei Liu <qiweil@codeaurora.org>
For encode usecase, video driver operates cache always on
the encoded data length rather than the buffer length.
This is an optimization and saves cache operation on unused
memory region.
VB index is used to identify a buffer given to client and fed
again to encoder. Based on that, the filled length of that
buffer is then operated for cache operations.
For certain usecases, it is difficult to identify the buffer
based on VB index. Hence keep a counter to track the max filled
length during the usecase and operate cache on that length
Change-Id: I62cfe876077786ec9d14f541a0080cd2fc124b01
Signed-off-by: Vikash Garodia <vgarodia@codeaurora.org>
V4L controls are made to be part of a cluster when they
are interdependent. If any one control in that cluster
is updated, all the remaining controls get updated as well
with the current control value.
Existing video usecase does not require any v4l controls
to be clustered. Hence removing the cluster controls.
Change-Id: I98dfc4af5291837054a8a28b00942fea6866f0a9
Signed-off-by: Vikash Garodia <vgarodia@codeaurora.org>
- Possiblilty of performance hit due to memory
alloc/free at frame level.
- Reuse same buffer_tag instead of reallocating.
Change-Id: Ifb8efc43907feea4d6da5e8a0bf0aeaa47c8e1d7
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
1. Client needs input etb to output fbd association. Introducing
input_tag and input_tag2 fields for this. Client will send
input_tag via etb reserved fields and client will get back
those as part of fbd reserved fields.
2. Convey sub-frame information to client from firmware.
CRs-Fixed: 2494945
Change-Id: Ic64131c7f8f3e3035cd6ff3997fe483171d71fc4
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
1. Don't allow qbuf (etb & ftb), if flush is issued to
(encoder & decoder) input port.
2. Don't allow qbuf (etb & ftb), if flush is issued to encoder
output port
3. Don't qbuf (ftb), if flush is issued to decoder output port
4. Allow qbuf (etb), if flush is issued to decoder output port
to handle decoder reconfig.
Change-Id: I08a6a10612cc1a14ad164c55c5c8e54550c84845
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
Set VBR as default rate control instead of RC_OFF.
Change-Id: Ie89f0109a9fe38a6c18b4713e8e8e905beb40b70
Signed-off-by: Mihir Ganu <mganu@codeaurora.org>
If kernel to kernel CVP is enabled, initially driver disables
CVP extradata to firmware and then later, re-enables it.
Introduced a new flag to usage of CVP kernel to kernel metadata
to save the usage. As part of start streaming, driver will send
out CVP extradata enabled once based on overall usage.
Also, disabled kernel to kernel CVP usage for superframe.
Change-Id: I5590bd2274e27243da99f207d61fb00fc6fc82e0
Signed-off-by: Akshata Sahukar <asahukar@codeaurora.org>
Increase input buffers count for first four elgible
sessions to avoid high memory usage. Also made changes
to extra buffer calculations.
Change-Id: Id50791f6800390ad27abc6f361e54dcc7b61a8c8
Signed-off-by: Darshana Patil <darshana@codeaurora.org>
Parameter that affect BW change during ETB and EBD.
During FTB BW calculation and voting is redundant.
Change-Id: I418d422f614670c2543a19709b36b3f39fd0d0e1
Signed-off-by: Chinmay Sawarkar <chinmays@codeaurora.org>
Input and output buffer size must be 4K aligned for both
encoder and decoder session. So reject the session during
msm_vidc_qbuf, if alloc_len is not 4K aligned.
Change-Id: If555f26f1658ce50e12a8533cb9d59134da0aa81
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
For heic/hevc image session, check respective capabilities to
allow or reject.
Change-Id: I4e565e6614a7863f75ecd4719480b850e1af43ae
Signed-off-by: Amit Shekhar <ashekhar@codeaurora.org>
1. Add all intra encoding capability;
2. Disable LTR, HP, Slice mode and IR settings;
3. Use WorkMode2, HQ and Pipeline 4 by force;
4. Ignore dynamic switch between IPPP and all intra;
5. Reject sessions if static/dynamic frame rate
larger than 240;
6. Reject any non HEVC or AVC sessions;
7. Reject any HEVC Main10 profile sessions;
8. Reset all intra flag to false when stream off;
Change-Id: I9f405a7712c84bc6f8319025a7fb2a0a7f9064df
Signed-off-by: Shi Zhongbo <zhongbos@codeaurora.org>
Client expects all the buffers to be returned in certain scenarios
but driver does not return due to decode batching feature. Add
delayed_work functionality to queue batched buffers to video hardware
upon timeout, so that all the buffers will be returned to client.
Change-Id: I926de1da93d941ff2a64bfec1bbd351de2f3a601
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
v4l2_ctrl_get_name may return NULL that leads to NULL
ptr dereference. So added necessay check to avoid ptr
dereferencing if it is NULL.
Change-Id: If1a4ee8fe22e240b278f31888637d22b81b334e1
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
Client is not supposed to queue buffer while driver
is in flush. Return error if client queue buffer
in flush to avoid possible deadlock issues in driver
due to release buffer reference event processing.
Change-Id: Ica2ff40e428f82b9f6d3aa1314f7bf7d322375bd
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Possibility of dereferencing userspace ptr in kernel for
invalid cmd. So added check to return error if unsupported
cmd is given as input to ioctl.
Change-Id: I3466fbd06e5b600f748824b9e16bcfdb4438bdef
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
Remove msm_vidc_calculate_output_buffer_count function
calls while g_ctrl event to avoid overwriting of
buffer count to old buffer count value that is set by
client.
Change-Id: I76135a88948ad7e2845c054a48e9f148bf367abb
Signed-off-by: Akshata Sahukar <asahukar@codeaurora.org>
FW provides UBWC_CR_STATS_INFO with every EBD which can be used
to accurately calculate BW requirements. Also renamed reconbufs
to refbufs, which is common to encode and decode.
CRs-Fixed: 2467542
Change-Id: I29afdcd466bd61f455507d30ab7ca951d0e9911c
Signed-off-by: Chinmay Sawarkar <chinmays@codeaurora.org>
driver returns min buffer count to client which includes
extra buffers count which is required for batching even
though batching may be disabled in start_streaming() later.
So decide batching in set format (configuration) to avoid
adding extra buffers count to reduce memory usage in
non-batching cases.
Change-Id: I147f7aaf928074a23c9dcda2dbf69a744cec382d
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Clean up debug log levels as:
1. Revise VIDC_WARN to VIDC_ERR;
2. Revise VIDC_PROF as VIDC_PERF;
3. Mark some one-time logs, e.g most of logs
in initialization and deinitialization as VIDC_HIGH;
4. Keep VIDC_PKT;
5. All other logs change to VIDC_LOW;
Change-Id: I8fc30f97dc3424da8418aab00e8af074ec8d4ef9
Signed-off-by: Shi Zhongbo <zhongbos@codeaurora.org>
