| Version 2 (modified by , 8 hours ago) ( diff ) |
|---|
Kinara AI
Create an account at https://support.kinara.ai/
Hardware
Download the ara2-sdk onto your host and extract it. It should create a folder named 'art'
Instructions are in the README_linux_aarch64.txt file inside the extracted folder.
Copy the 'art' folder to the Gateworks SBC with something like scp.
Verify the Kinara card is detected on the PCIe bus:
root@noble-venice:~/art/linux/aarch64/hw_utils# lspci 81:00.0 Processing accelerators: Device 1e58:0002 (rev 02)
Insert the uiodma module:
insmod art/linux/drivers/uiodma/uiodma.ko sudo echo "1e58 0002" > /sys/bus/pci/drivers/uiodma/new_id
Get Chip info:
root@noble-venice:~/art/linux# ./aarch64/hw_utils/bins/chip_info_aarch64 -e 0 Kinara Inc. Silicon utilities version 0.5.9 Run directory: /root/art/linux ./aarch64/hw_utils/bins/chip_info_aarch64 -e 0 [I:20260123:00:02:36:178031] [] [kinara_main_635][IODeviceManager] enumerate ara2 pci devices [I:20260123:00:02:36:405024] [] [kinara_main_635][IODeviceManager] path= /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/0000:02:03.0/0000:81:00.0/ [I:20260123:00:02:36:405187] [] [kinara_main_635][IODeviceManager] instance name: IODeviceManager, found an ara2 pci device at: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/0000:02:03.0/0000:81:00.0/ [I:20260123:00:02:36:405372] [] [kinara_main_635][pci_io_00:81.0] cache management 0 [I:20260123:00:02:36:405398] [] [kinara_main_635][pci_io_00:81.0] instance name: Ara2PcieIO_0, ara2 pci device folder path: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/0000:02:03.0/0000:81:00.0/ [I:20260123:00:02:36:405521] [] [kinara_main_635][pci_io_00:81.0] instance name: Ara2PcieIO_0, phy0=0x18700000 phy1=0x18500000 phy2=0x18600000 [I:20260123:00:02:36:411041] [] [kinara_main_635][pci_io_00:81.0] enabling pcie_ste_dma for device [W:20260123:00:02:36:411150] [] [kinara_main_635][pci_io_00:81.0] instance name: Ara2PcieIO_0, uiodma addr is : 0x7b000000 [I:20260123:00:02:36:411293] [] [kinara_main_635][pci_io_00:81.0] successfully enabled pcie_ste_dma for pcie device [I:20260123:00:02:36:421550] [] [kinara_main_635][pci_io_00:81.0] instance name: Ara2PcieIO_0, vendor=0x1e58 product=0x2 [I:20260123:00:02:36:422136] [] [kinara_main_635][pci_io_00:81.0] Calling pcie tl clock programming, removing pcie clock from bypass [W:20260123:00:02:36:425541] [ChipInfo] [kinara_main_635][ChipInfo] ara_physical_id: 0 [I:20260123:00:02:36:426262] [] [task_request_queue_processing_task_runner0_640][ChipInfoRequest] BOOT SUCCESS [I:20260123:00:02:36:426304] [] [task_request_queue_processing_task_runner0_640][ChipInfoRequest] HIF IS ENABLED [I:20260123:00:02:36:527551] [ChipInfo] [kinara_main_635][ChipInfo] Request Done [I:20260123:00:02:36:532629] [ChipInfo] [kinara_main_635][ChipInfo] chip info for the devices connected +-----------+---------------+----------------------------------------------+ | Tool | Device Number | Info | +-----------+---------------+----------------------------------------------+ | Chip Info | 0, strap:0 | +-----------------------------+------------+ | | | | | Parameter | Value | | | | | +-----------------------------+------------+ | | | | | chip_id | Ara2 A01 | | | | | | bus_id | 0:129:0 | | | | | | if_type | PCIE | | | | | | chip_physical_id | 0 | | | | | | pcie_present | 1 | | | | | | usb_present | 0 | | | | | | ddr_present | 1 | | | | | | jtag_present | | | | | | | sys_frequency | 25 MHz | | | | | | neural_processor_frequency | 25 MHz | | | | | | control_processor_frequency | 25 MHz | | | | | | ddr_frequency | 25 MHz | | | | | | ddr_mfr | | | | | | | ddr_size | NA | | | | | | flash_mfr | NA | | | | | | flash_size | NA | | | | | | num_flash_images | NA | | | | | | dev_temp1 | 21.07 C | | | | | | voltage1 | 0.788796 V | | | | | | last power state | NA | | | | | | current power state | L1 | | | | | | boot_status | 1 | | | | | | internal_mem_size | 8 MB | | | | | | life_cycle | TESTED | | | | | | firmware_version(raw) | 32779 | | | | | | firmware_version | 0.128.11 | | | | | +-----------------------------+------------+ | | | | | +-----------+---------------+----------------------------------------------+ root@noble-venice:~/art/linux#
Program the Flash (one time only):
./bins/program_flash_aarch64 -e 0 -f ./boot_img/willow_therm.hex <reboot the host/ reset ara device to get new firmware loaded on ara device>
Activate the chip:
root@noble-venice:~/art/linux/aarch64/hw_utils# ./bins/active_enable_aarch64 -e 0 -m 2
Kinara Inc. Silicon utilities version 0.5.9
Run directory: /root/art/linux/aarch64/hw_utils
./bins/active_enable_aarch64 -e 0 -m 2
[I:20260129:17:16:09:595108] [] [kinara_main_678][Active_Enable] In run phase of Flash image
[I:20260129:17:16:09:595458] [] [kinara_main_678][Active_Enable] Device Found Done ara-physical-id= 0
[I:20260129:17:16:09:595487] [] [kinara_main_678][IODeviceManager] enumerate ara2 pci devices
[I:20260129:17:16:09:794621] [] [kinara_main_678][IODeviceManager] path= /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/
[I:20260129:17:16:09:794732] [] [kinara_main_678][IODeviceManager] instance name: IODeviceManager, found an ara2 pci device at: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/
[I:20260129:17:16:09:794880] [] [kinara_main_678][pci_io_00:01.0] cache management 0
[I:20260129:17:16:09:794901] [] [kinara_main_678][pci_io_00:01.0] instance name: Ara2PcieIO_0, ara2 pci device folder path: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/
[I:20260129:17:16:09:795008] [] [kinara_main_678][pci_io_00:01.0] instance name: Ara2PcieIO_0, phy0=0x18300000 phy1=0x18100000 phy2=0x18200000
[I:20260129:17:16:09:798594] [] [kinara_main_678][pci_io_00:01.0] enabling pcie_ste_dma for device
[W:20260129:17:16:09:798675] [] [kinara_main_678][pci_io_00:01.0] instance name: Ara2PcieIO_0, uiodma addr is : 0xf0a00000
[I:20260129:17:16:09:798777] [] [kinara_main_678][pci_io_00:01.0] successfully enabled pcie_ste_dma for pcie device
[I:20260129:17:16:09:805173] [] [kinara_main_678][pci_io_00:01.0] instance name: Ara2PcieIO_0, vendor=0x1e58 product=0x2
[I:20260129:17:16:09:807783] [] [kinara_main_678][pci_io_00:01.0] Calling pcie tl clock programming, removing pcie clock from bypass
[I:20260129:17:16:09:808154] [] [kinara_main_678][Active_Enable] [CTEST]: SBP Ready for Runtime Command
[I:20260129:17:16:09:808182] [] [kinara_main_678][Active_Enable] [CTEST]: HIF is SETUP
[I:20260129:17:16:09:808197] [] [kinara_main_678][Active_Enable] Entering State Full PRD 2
[I:20260129:17:16:09:808214] [] [kinara_main_678][Active_Enable] {} 4
[I:20260129:17:16:09:808251] [] [kinara_main_678][Active_Enable] [CTEST]: Pushed HIF req Type : 211 1 1
[I:20260129:17:16:09:808265] [] [kinara_main_678][Active_Enable] [CTEST]: Pushed HIF req occ : 32
[I:20260129:17:16:09:808629] [] [kinara_main_678][Active_Enable] {} 4
[I:20260129:17:16:09:808673] [] [kinara_main_678][Active_Enable] [CTEST]: HIF response 0
[I:20260129:17:16:09:808693] [] [kinara_main_678][Active_Enable] [CTEST]: CMD_TYPE 211
[I:20260129:17:16:09:808711] [] [kinara_main_678][Active_Enable] [CTEST]: CMD_ID 1
[I:20260129:17:16:09:808728] [] [kinara_main_678][Active_Enable] [CTEST]: QID_ID 1
[I:20260129:17:16:09:808746] [] [kinara_main_678][Active_Enable] [CTEST]: HIF PS Read Response
[I:20260129:17:16:09:808765] [] [kinara_main_678][Active_Enable] [CTEST]: Status : 0
[I:20260129:17:16:09:808782] [] [kinara_main_678][Active_Enable] [CTEST]: PS Response Data Status : 0
[I:20260129:17:16:09:808799] [] [kinara_main_678][Active_Enable] [CTEST]: PS Response Data Current Power State : 2
[I:20260129:17:16:09:808816] [] [kinara_main_678][Active_Enable] [CTEST]: PS Response Data PVT_LOCK : 0
[I:20260129:17:16:09:808836] [] [kinara_main_678][Active_Enable] [CTEST]: TEST PASSED
Configure the DRAM:
root@noble-venice:~/art/linux/aarch64/hw_utils# ./bins/ddr_unified_config_aarch64 -e 0 -s 1 -g 1 -l 3 -w 2 -f 0 -n 0 -u 0 -m 0 -r 0 -b willow -o ./ddr_config/ddr_cfg_35.bin Kinara Inc. Silicon utilities version 0.5.9 Run directory: /root/art/linux/aarch64/hw_utils ./bins/ddr_unified_config_aarch64 -e 0 -s 1 -g 1 -l 3 -w 2 -f 0 -n 0 -u 0 -m 0 -r 0 -b willow -o ./ddr_config/ddr_cfg_35.bin [I:20260129:17:17:05:155740] [] [kinara_main_706][IODeviceManager] enumerate ara2 pci devices [I:20260129:17:17:05:343281] [] [kinara_main_706][IODeviceManager] path= /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/ [I:20260129:17:17:05:343392] [] [kinara_main_706][IODeviceManager] instance name: IODeviceManager, found an ara2 pci device at: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/ [I:20260129:17:17:05:343538] [] [kinara_main_706][pci_io_00:01.0] cache management 0 [I:20260129:17:17:05:343558] [] [kinara_main_706][pci_io_00:01.0] instance name: Ara2PcieIO_0, ara2 pci device folder path: /sys/devices/platform/soc@0/33800000.pcie/pci0000:00/0000:00:00.0/0000:01:00.0/ [I:20260129:17:17:05:343663] [] [kinara_main_706][pci_io_00:01.0] instance name: Ara2PcieIO_0, phy0=0x18300000 phy1=0x18100000 phy2=0x18200000 [I:20260129:17:17:05:346269] [] [kinara_main_706][pci_io_00:01.0] enabling pcie_ste_dma for device [W:20260129:17:17:05:346327] [] [kinara_main_706][pci_io_00:01.0] instance name: Ara2PcieIO_0, uiodma addr is : 0xf0a00000 [I:20260129:17:17:05:346428] [] [kinara_main_706][pci_io_00:01.0] successfully enabled pcie_ste_dma for pcie device [I:20260129:17:17:05:351159] [] [kinara_main_706][pci_io_00:01.0] instance name: Ara2PcieIO_0, vendor=0x1e58 product=0x2 [I:20260129:17:17:05:353537] [] [kinara_main_706][pci_io_00:01.0] Calling pcie tl clock programming, removing pcie clock from bypass [I:20260129:17:17:05:353868] [] [kinara_main_706][ddr_unified_config] Configuring DDR now. Approxmiate time to complete: 60 seconds. [I:20260129:17:17:05:353894] [] [kinara_main_706][ddr_unified_config] UNIFIED CFG BIN LOADING [I:20260129:17:17:36:009795] [] [kinara_main_706][ddr_unified_config] DDR configuation successful. DDR BIST: PASSED
Activate Proxy: (change config/proxy_config.yaml if you want to run in background)
root@noble-venice:~/art/linux/aarch64/proxy# sudo ./proxy_aarch64 -c ../../config/proxy_config.yaml Kinara Inc. Silicon utilities version 0.5.9 Run directory: /root/art/linux/aarch64/proxy ./proxy_aarch64 -c ../../config/proxy_config.yaml [I:20260129:17:23:04:425472] [] [kinara_main_764][GlobalLogger] root@noble-venice:~/art/linux/aarch64/proxy# ps -aef | grep prox root 766 1 9 17:23 ? 00:00:00 ./proxy_aarch64 -c ../../config/proxy_config.yaml
Download a sample yolov8n model (.dvm file) and untar:
- Download URL (may require access to support.kinara.ai portal)
- Transfer to Gateworks SBC and untar:
root@noble-venice:~/art/linux/aarch64/nnapp# wget 172.24.20.26/kinara/sdk_model_compile.tar.gz --2026-01-29 17:26:52-- http://192.168.1.1/kinara/sdk_model_compile.tar.gz Connecting to 172.24.20.26:80... connected. HTTP request sent, awaiting response... 200 OK Length: 23754978 (23M) [application/x-gtar-compressed] Saving to: ‘sdk_model_compile.tar.gz’ sdk_model_compile.tar.gz 100%[=====================================================================================================>] 22.65M 112MB/s in 0.2s 2026-01-29 17:26:52 (112 MB/s) - ‘sdk_model_compile.tar.gz’ saved [23754978/23754978] root@noble-venice:~/art/linux/aarch64/nnapp# tar -xvf sdk_model_compile.tar.gz
Either change the input path to the model, or move the model to /tmp
Modify the ../../config/nnapp_config.yaml and change the input_path in the async section:
async:
# if this flag is TRUE, then the input directory of each model is iterated only once.
# else, the input directory is iterated for model_param.batch_size * model_param.iterations times
run_once_and_for_all: False
utilization_stats: False
# dump outputs with timestamp
dump_with_timestamp: False
# this section contains all the model related params for infer async mode
models:
0:
name: "model1"
path: "/tmp/model.dvm"
input_path: ""
output_path: ""
Execute the sample model run:
root@noble-venice:~/art/linux/aarch64/nnapp# ./nnapp_aarch64 --config ../../config/nnapp_config.yaml [W:20260129:17:29:41:832406] [] [n/a_-1][config] default YAML file not found ./config.yaml , specify yaml path Kinara, Inc. *---------------------------------------------------------------* * Neural Network Inference Application * * Copyright(c) 2018-23, Kinara, Inc. All rights reserved. * *---------------------------------------------------------------* Run: working directory: /root/art/linux/aarch64/nnapp ./nnapp_aarch64 --config ../../config/nnapp_config.yaml the app expects the inputs as raw tensors. The inputs are assumed to be preprocessed and written to a file the inputs have to be in the below format: <input_image_name>/quantized/layer<layer_id>-<blob_name>-<blob_id>_input_dv.dat and the golden inputs in the below format: <input_image_name>/quantized/<layer_name>-<layer_id>-<blob_name>-<blob_id> [I:20260129:17:29:41:898866] [] [main_807][model_parameters] empty input_path for model= model1 , setting output_path and golden_output_path to none [I:20260129:17:29:41:903483] [] [main_807][BaseApp] connected to proxy socket: IP_Address= 127.0.0.1 port= 5000 [I:20260129:17:29:41:904161] [] [main_807][BaseApp] +------------+-----------------+ | Product | Current Version | +------------+-----------------+ | client_lib | 1.3.0.0 | | firmware | 0.128.11.0 | | proxy | 1.3.0.0 | | sysapi | 1.1.61.0 | +------------+-----------------+ [I:20260129:17:29:41:904950] [] [main_807][BaseApp] +------------+--------------------+ | Product | Supported Versions | +------------+--------------------+ | client_lib | 1.0.0.0 | | client_lib | 1.1.1.0 | | client_lib | 1.1.2.0 | | client_lib | 1.3.0.0 | | cnn_model | 2.0.0.0 | | cnn_model | 2.1.0.0 | | firmware | 0.5.2.0 | | firmware | 1.1.2.0 | | llm_model | 3.0.0.0 | | llm_model | 3.1.0.0 | | pci_driver | 1.0.4.0 | | pci_driver | 1.0.6.6 | | proxy | 0.8.0.0 | | proxy | 0.9.0.0 | | proxy | 1.1.1.0 | | proxy | 1.3.0.0 | +------------+--------------------+ [I:20260129:17:29:41:905463] [] [main_807][BaseApp] found endpoints: count= 1 [I:20260129:17:29:41:905908] [] [main_807][BaseApp] creation of endpoint group successful: model= model1 [I:20260129:17:29:41:984637] [] [main_807][BaseApp] model load successful: model= model1 [I:20260129:17:29:42:011215] [] [task_print_device_stats_812][Stats] temperature= 36.2557 ep status: DV_ENDPOINT_STATE_ACTIVE 0x1000608c: 15d0200 [I:20260129:17:29:42:072253] [] [main_807][AsyncApp] completed iteration= 1 [I:20260129:17:29:42:145357] [] [main_807][AsyncApp] completed iteration= 2 [I:20260129:17:29:42:218729] [] [main_807][AsyncApp] completed iteration= 3 [I:20260129:17:29:42:291966] [] [main_807][AsyncApp] completed iteration= 4 [I:20260129:17:29:42:365203] [] [main_807][AsyncApp] completed iteration= 5 [I:20260129:17:29:42:437872] [] [main_807][AsyncApp] completed iteration= 6 [I:20260129:17:29:42:510537] [] [main_807][AsyncApp] completed iteration= 7 [I:20260129:17:29:42:584005] [] [main_807][AsyncApp] completed iteration= 8 [I:20260129:17:29:42:656563] [] [main_807][AsyncApp] completed iteration= 9 [I:20260129:17:29:42:729354] [] [main_807][AsyncApp] completed iteration= 10 [I:20260129:17:29:42:802121] [] [main_807][AsyncApp] completed iteration= 11 [I:20260129:17:29:42:874931] [] [main_807][AsyncApp] completed iteration= 12 [I:20260129:17:29:42:947354] [] [main_807][AsyncApp] completed iteration= 13 [I:20260129:17:29:43:012893] [] [task_print_device_stats_812][Stats] temperature= 36.2557 ep status: DV_ENDPOINT_STATE_ACTIVE 0x1000608c: 15d0200 [I:20260129:17:29:43:020108] [] [main_807][AsyncApp] completed iteration= 14 [I:20260129:17:29:43:094043] [] [main_807][AsyncApp] completed iteration= 15 [I:20260129:17:29:43:166269] [] [main_807][AsyncApp] completed iteration= 16 [I:20260129:17:29:43:238760] [] [main_807][AsyncApp] completed iteration= 17 [I:20260129:17:29:43:311065] [] [main_807][AsyncApp] completed iteration= 18 [I:20260129:17:29:43:383388] [] [main_807][AsyncApp] completed iteration= 19 [I:20260129:17:29:43:455880] [] [main_807][AsyncApp] completed iteration= 20 [I:20260129:17:29:43:456773] [] [main_807][Stats] summary stats ╔════════╤════════════════════════════╤═════════════════════════╤═══════════════════════════╤══════════════════╗ ║ Model │ hw_time (ms) │ ip_time (ms) │ op_time (ms) │ app_time (ms) ║ ╠════════╪════════╤═════════╤═════════╪═══════╤═══════╤═════════╪════════╤════════╤═════════╪═════╤═════╤══════╣ ║ │ max │ min │ mean │ max │ min │ mean │ max │ min │ mean │ max │ min │ mean ║ ╟────────┼────────┼─────────┼─────────┼───────┼───────┼─────────┼────────┼────────┼─────────┼─────┼─────┼──────╢ ║ model1 │ 3.1052 │ 3.08395 │ 3.09515 │ 4.365 │ 2.346 │ 3.17945 │ 12.187 │ 10.369 │ 11.3023 │ 76 │ 71 │ 71.9 ║ ╚════════╧════════╧═════════╧═════════╧═══════╧═══════╧═════════╧════════╧════════╧═════════╧═════╧═════╧══════╝ model_name=model1 iteration=20 batch_size=5 num_ip=1 num_op=1 total_ip_size=122880000 total_op_size=120960000 num_inferences=100 [I:20260129:17:29:43:457176] [] [main_807][Stats] endpoint load distribution: [0,100] [I:20260129:17:29:43:457195] [] [main_807][Stats] total number of inferences: 100 [I:20260129:17:29:43:457221] [] [main_807][Stats] Performance: HW IPS= 323.086 App IPS= 69.541 root@noble-venice:~/art/linux/aarch64/nnapp#
Note the final line that shows Performance: HW IPS= 323.086.
Note:
See TracWiki
for help on using the wiki.
