mirror of
git://nv-tegra.nvidia.com/tegra/nv-sci-src/nvsci_samples.git
synced 2025-12-22 09:21:21 +03:00
a3f1b7ea33
2eba699906039d6615aae4967f6ea79bfe44a40a - event_sample_app/block_pool.c f3abb0a884f0647204ad32ff51255c4712e52120 - event_sample_app/Makefile 9ee49033e077ac5c8bf458a04c91dd3dbed9633d - event_sample_app/event_loop.h b33adce6eb1bbc7af23f6c37b6a635479e18a66a - event_sample_app/block_returnsync.c a56041c06b6bc1d3812b72b399d7d78dd7895485 - event_sample_app/block_limiter.c ca34c957759f7a010f0cbbbf9bedc03a2c98092b - event_sample_app/block_c2c.c 8d6d0ec3aa8e374a1d2a5fedc9dd24ff7bbdb731 - event_sample_app/block_multicast.c a76149a2531899e35843d939f60ad8979d8cf65f - event_sample_app/block_consumer_uc1.c 9da8763e4af4b4b7278507a3ebfe2c68a7a24585 - event_sample_app/util.h 2bf7e1383d6e8913c9b0a6a8bdd48fe63d8098d0 - event_sample_app/block_producer_uc1.c a54abf82eaa2d888e379ab4596ba68ce264e80b5 - event_sample_app/block_info.h 080a6efe263be076c7046e70e31098c2bbed0f6d - event_sample_app/block_presentsync.c 7dd10e5ea71f0c4a09bbe1f9f148f67a13ee098c - event_sample_app/util.c bc1a6f9017b28e5707c166a658a35e6b3986fdf4 - event_sample_app/usecase1.h 317f43efc59638bf1eae8303f0c79eafb059241a - event_sample_app/block_ipc.c 40361c8f0b68f7d5207db2466ce08c19c0bf1c90 - event_sample_app/event_loop_service.c efad113d0107e5d8f90146f3102a7c0ed22f1a35 - event_sample_app/event_loop_threads.c 2908615cebcf36330b9850c57e8745bf324867b2 - event_sample_app/block_queue.c 36ed68eca1a7800cf0d94e763c9fc352ee8cda1e - event_sample_app/block_common.c 675f75d61bd0226625a8eaaf0e503c9e976c8d61 - event_sample_app/main.c c3b26619dd07d221e953fc5dc29a50dcb95a8b97 - rawstream/Makefile 1fbb82e2281bb2e168c87fd20903bbed898ca160 - rawstream/rawstream_cuda.c 1d96498fe3c922f143f7e50e0a32b099714060ad - rawstream/rawstream_consumer.c d077dafc9176686f6d081026225325c2a303a60e - rawstream/rawstream_producer.c 54ae655edddda7dcabe22fbf0b27c3f617978851 - rawstream/rawstream.h d5ffeef3c7ad2af6f6f31385db7917b5ef9a7438 - rawstream/rawstream_ipc_linux.c 81e3d6f8ff5252797a7e9e170b74df6255f54f1b - rawstream/rawstream_main.c Change-Id: I0f4e671693eb0addfe8d0e6532cc8f240cb6c778
259 lines
13 KiB
Plaintext
259 lines
13 KiB
Plaintext
SPDX-FileCopyrightText: Copyright (c) 2021-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
SPDX-License-Identifier: LicenseRef-NvidiaProprietary
|
|
|
|
NVIDIA CORPORATION, its affiliates and licensors retain all intellectual
|
|
property and proprietary rights in and to this material, related
|
|
documentation and any modifications thereto. Any use, reproduction,
|
|
disclosure or distribution of this material and related documentation
|
|
without an express license agreement from NVIDIA CORPORATION or
|
|
its affiliates is strictly prohibited.
|
|
|
|
NvSciStream Event Loop Driven Sample App - README
|
|
|
|
---
|
|
# nvscistream_event_sample - NvSciStream Sample App
|
|
|
|
## Description
|
|
|
|
This directory contains an NvSciStream sample application that
|
|
supports a variety of use cases, using an event-loop driven model.
|
|
Once the stream is fully connected, all further setup and streaming
|
|
operations are triggered by events, processed either by a single
|
|
NvSciEvent-driven thread or separate threads which wait for events
|
|
on each block. The former is the preferred approach for implementing
|
|
NvSciStream applications. In addition to those events which NvSci
|
|
itself generates, any other event which can be bound to an NvSciEvent
|
|
can be added to the event loop. This allows for robust applications
|
|
which can handle events regardless of the order in which they occur.
|
|
|
|
To use this sample for writing your own applications:
|
|
|
|
* See main.c for examples of how to do top level application setup and
|
|
how to select the blocks needed for your use case and connect them
|
|
all together.
|
|
* See the descriptions in the usecase*.h files to determine which use cases
|
|
involve the producer and consumer engines that you are interested in.
|
|
* See the appropriate block_*.c files for examples of creating the
|
|
necessary blocks and handling the events that they encounter.
|
|
See the block_producer_*.c and block_consumer_*.c files for examples of how
|
|
to map the relevant engines to and from NvSci.
|
|
* See the appropriate event_loop_*.c file for your chosen event handling
|
|
method.
|
|
|
|
## Build the application
|
|
|
|
The NvSciStream sample includes source code and a Makefile.
|
|
Navigate to the sample application directory to build the application:
|
|
|
|
make clean
|
|
make
|
|
|
|
## Examples of how to run the sample application:
|
|
|
|
* NOTE:
|
|
* Inter-process and inter-chip test cases must be run with sudo.
|
|
* NvMedia/CUDA stream (use case 2) of the sample application is not supported
|
|
on x86 and Jetson Linux devices.
|
|
* Inter-chip use cases are not supported on Jetson Linux devices.
|
|
* Update the NvIpc/PCIe endpoint accordingly.
|
|
|
|
Single-process, single-consumer CUDA/CUDA stream that uses the default event
|
|
service:
|
|
|
|
./nvscistream_event_sample
|
|
|
|
Single-process, single-consumer stream that uses the threaded event handling:
|
|
|
|
./nvscistream_event_sample -e t
|
|
|
|
Single-process NvMedia/CUDA stream with yuv format:
|
|
./nvscistream_event_sample -u 2 -s y
|
|
|
|
Single-process NvMedia/CUDA stream with three consumers, and the second uses
|
|
the mailbox mode:
|
|
|
|
./nvscistream_event_sample -u 2 -m 3 -q 1 m
|
|
|
|
Multi-process CUDA/CUDA stream with three consumers, one in the same
|
|
process as the producer, and the other two in separate processes. The
|
|
first and the third consumers use the mailbox mode:
|
|
|
|
./nvscistream_event_sample -m 3 -p -c 0 -q 0 m &
|
|
./nvscistream_event_sample -c 1 -c 2 -q 2 m
|
|
|
|
Multi-process CUDA/CUDA stream with three consumers, one in the same
|
|
process as the producer, and the other two in separate processes.
|
|
To simulate the case with a less trusted consumer, one of the consumer
|
|
processes is set with lower priority. A limiter block is used to restrict
|
|
this consumer to hold at most one packet. The total number of packets is
|
|
increased to five.
|
|
|
|
Linux example:
|
|
|
|
./nvscistream_event_sample -m 3 -f 5 -p -c 0 -l 2 1 &
|
|
./nvscistream_event_sample -c 1 &
|
|
nice -n 19 ./nvscistream_event_sample -c 2 &
|
|
# Makes the third process as nice as possible.
|
|
|
|
QNX example:
|
|
|
|
./nvscistream_event_sample -m 3 -f 5 -p -c 0 -l 2 1 &
|
|
./nvscistream_event_sample -c 1 &
|
|
nice -n 1 ./nvscistream_event_sample -c 2 &
|
|
# Reduces the priority level of the third process by 1.
|
|
|
|
Multi-process CUDA/CUDA stream with two consumers, one in the same
|
|
process as the producer, and the other in a separate processe. Both
|
|
processes enable the endpoint information option:
|
|
|
|
./nvscistream_event_sample -m 2 -p -c 0 -i &
|
|
./nvscistream_event_sample -c 1 -i
|
|
|
|
Multi-process CUDA/CUDA stream with extra validation steps for ASIL-D process
|
|
(Not support on x86 or Jetson Linux devices):
|
|
./nvscistream_event_sample -u 3 -p &
|
|
./nvscistream_event_sample -u 3 -c 0
|
|
|
|
Multi-process CUDA/CUDA stream using external event service to handle internal
|
|
I/O messages acroess process boundary:
|
|
./nvscistream_event_sample -p -E &
|
|
./nvscistream_event_sample -c 0 -E
|
|
|
|
Multi-process CUDA/CUDA stream with one consumer on another SoC.
|
|
The consumer has the FIFO queue attached to the C2C IpcSrc block, and
|
|
a three-packet pool attached to the C2C IpcDst block. It uses IPC channel
|
|
nvscic2c_pcie_s0_c5_1 <-> nvscic2c_pcie_s0_c6_1 for C2C communication.
|
|
|
|
./nvscistream_event_sample -P 0 nvscic2c_pcie_s0_c5_1 -Q 0 f
|
|
# Run below command on another OS running on peer SOC.
|
|
./nvscistream_event_sample -C 0 nvscic2c_pcie_s0_c6_1 -F 0 3
|
|
|
|
Multi-process CUDA/CUDA stream with four consumers, one in the same
|
|
process as the producer, one in another process but in the same OS as the
|
|
producer, and two in another process on another OS running in a peer SoC.
|
|
The third and fourth consumers have a mailbox queue attached to the C2C
|
|
IpcSrc block, and a five-packet pool attached to the C2C IpcDst block.
|
|
The third consumer uses nvscic2c_pcie_s0_c5_1 <-> nvscic2c_pcie_s0_c6_1 for
|
|
C2C communication. The 4th consumer uses nvscic2c_pcie_s0_c5_2 <->
|
|
nvscic2c_pcie_s0_c6_2 for C2C communication.
|
|
|
|
./nvscistream_event_sample -m 4 -c 0 -q 0 m -Q 2 m -Q 3 m -P 2 nvscic2c_pcie_s0_c5_1 -P 3 nvscic2c_pcie_s0_c5_2 &
|
|
./nvscistream_event_sample -c 1 -q 1 m
|
|
# Run below command on another OS running on peer SOC.
|
|
./nvscistream_event_sample -C 2 nvscic2c_pcie_s0_c6_1 -q 2 f -F 2 5 -C 3 nvscic2c_pcie_s0_c6_2 -q 3 m -F 3 5
|
|
|
|
#Example commands for inter-process late attach usecase
|
|
Multi-process CUDA/CUDA stream with one early consumer and one late-attached consumer
|
|
Producer and early consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached consumer process is configured to receive 10000 frames.
|
|
# Run the below commands to launch producer and early consumer processes.
|
|
./nvscistream_event_sample -m 2 -r 1 -p &
|
|
./nvscistream_event_sample -c 0 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -c 1 -k 1 10000 &
|
|
|
|
Multi-process CUDA/CUDA stream with one early consumer and two late-attached consumers
|
|
Producer and early consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached consumer process one is configured to receive 10000 frames and
|
|
the late-attached consumer process two is configured to receive 50000 frames
|
|
# Run the below commands to launch producer and early consumer processes.
|
|
./nvscistream_event_sample -m 3 -r 2 -p &
|
|
./nvscistream_event_sample -c 0 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process one.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -c 1 -k 1 10000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process two.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -c 2 -k 2 50000 &
|
|
|
|
#Example commands for inter-process re-attach usecase
|
|
Multi-process CUDA/CUDA stream with one early consumer and two late-attached consumers
|
|
Producer and early consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached consumer process one is configured to receive 10000 frames and
|
|
the late-attached consumer process two is configured to receive 50000 frames.
|
|
Once late-attached consumer process one completes streaming, re-attach it for receiving
|
|
5000 frames.
|
|
# Run the below commands to launch producer and early consumer processes.
|
|
./nvscistream_event_sample -m 3 -r 2 -p &
|
|
./nvscistream_event_sample -c 0 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process one.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -c 1 -k 1 10000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process two.
|
|
sleep 1;
|
|
./nvscistream_event_sample -L -c 2 -k 2 50000 &
|
|
# After late-attached consumer process one completes, re-attach it.
|
|
./nvscistream_event_sample -L -c 1 -k 1 5000 &
|
|
|
|
Limitations with C2C late/re-attach:
|
|
This sample app has the following limitations.
|
|
1. For C2C late/re-attach, this sample app does not support IPC consumer being the only early
|
|
consumer and all the remaining consumers as C2C late-attached. This is due to setting static
|
|
attribute logic for late-attach is not added.
|
|
2. A C2C consumer can acts as an IPC consumer during late-/re-attach but an IPC consumer
|
|
cannot be made as C2C consumer during Late/re-attach.
|
|
|
|
#Example commands for inter-chip late attach usecase
|
|
Multi-process CUDA/CUDA stream with one early C2C consumer and one C2C late-attached consumer
|
|
Producer and early C2C consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached C2C consumer process is configured to receive 10000 frames.
|
|
The early consumer uses nvscic2c_pcie_s0_c5_1 <-> nvscic2c_pcie_s0_c6_1 for
|
|
C2C communication. The late-attached consumer uses nvscic2c_pcie_s0_c5_2 <->
|
|
nvscic2c_pcie_s0_c6_2 for C2C communication.
|
|
|
|
# Run the below commands to launch producer on SOC1
|
|
./nvscistream_event_sample -m 2 -r 1 -P 0 nvscic2c_pcie_s0_c5_1 -P 1 nvscic2c_pcie_s0_c5_2 &
|
|
# Run the below commands to launch early consumer process on SOC2
|
|
./nvscistream_event_sample -C 0 nvscic2c_pcie_s0_c6_1 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process on SOC2
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -C 1 nvscic2c_pcie_s0_c6_2 -k 1 10000 &
|
|
|
|
Multi-process CUDA/CUDA stream with one early C2C consumer and two C2C late-attached consumer
|
|
Producer and early C2C consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached C2C consumer process is one configured to receive 10000 frames and
|
|
the late-attached C2C consumer process is two configured to receive 10000 frames.
|
|
The early consumer uses nvscic2c_pcie_s0_c5_1 <-> nvscic2c_pcie_s0_c6_1 for
|
|
C2C communication. The late-attached consumer one uses nvscic2c_pcie_s0_c5_2 <->
|
|
nvscic2c_pcie_s0_c6_2 for C2C communication and the late-attached consumer two
|
|
uses nvscic2c_pcie_s0_c5_3 <->nvscic2c_pcie_s0_c6_3 for C2C communication.
|
|
|
|
# Run the below commands to launch producer on SOC1
|
|
./nvscistream_event_sample -m 3 -r 2 -P 0 nvscic2c_pcie_s0_c5_1 -P 1 nvscic2c_pcie_s0_c5_2 -P 2 nvscic2c_pcie_s0_c5_3 &
|
|
# Run the below commands to launch early consumer process on SOC2
|
|
./nvscistream_event_sample -C 0 nvscic2c_pcie_s0_c6_1 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -C 1 nvscic2c_pcie_s0_c6_2 -k 1 10000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process.
|
|
sleep 1;
|
|
./nvscistream_event_sample -L -C 2 nvscic2c_pcie_s0_c6_3 -k 2 10000 &
|
|
|
|
#Example commands for inter-chip/process re-attach usecase
|
|
Multi-process CUDA/CUDA stream with one early consumer and two late-attached consumers
|
|
Producer and early consumer processes are configured to stream 100000 frames, where as
|
|
the late-attached consumer process one is configured to receive 10000 frames and
|
|
the late-attached consumer process two is configured to receive 50000 frames.
|
|
Once late-attached consumer process one completes streaming, re-attach it for receiving
|
|
5000 frames.
|
|
Once late-attached consumer process two completes streaming, re-attach it as IPC consumer for receiving
|
|
5000 frames.
|
|
|
|
# Run the below commands to launch producer on SOC1
|
|
./nvscistream_event_sample -m 3 -r 2 -P 0 nvscic2c_pcie_s0_c5_1 -P 1 nvscic2c_pcie_s0_c5_2 -P 2 nvscic2c_pcie_s0_c5_3 &
|
|
# Run the below commands to launch early consumer process on SOC2
|
|
./nvscistream_event_sample -C 0 nvscic2c_pcie_s0_c6_1 -k 0 100000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process.
|
|
sleep 1; # This 1s delay will let producer and consumer to enter into streaming phase.
|
|
./nvscistream_event_sample -L -C 1 nvscic2c_pcie_s0_c6_2 -k 1 10000 &
|
|
# Run the below command after some delay to launch the late-attached consumer process.
|
|
sleep 1;
|
|
./nvscistream_event_sample -L -C 2 nvscic2c_pcie_s0_c6_3 -k 2 50000 &
|
|
# Once late-attached consumer process one completes streaming,
|
|
# re-attach it for receiving 5000 frames.
|
|
./nvscistream_event_sample -L -C 1 nvscic2c_pcie_s0_c6_2 -k 1 5000 &
|
|
# Once late-attached consumer process two completes streaming,
|
|
# re-attach it as IPC consumer on SOC1 for receiving 5000 frames.
|
|
./nvscistream_event_sample -L -c 2 -k 2 5000 &
|