Common Commands for DDR and NPU on AllWinner MR536

In the AllWinner MR536 Tina5 Linux system environment, the commonly used debugging commands on the board are as follows:1. Mount debugfs (the /sys/kernel/debug directory will depend on this)

mount -t debugfs none /sys/kernel/debug

2. Get DDR frequency

cat /sys/kernel/debug/clk/clk_summary | grep ddr

After execution, it may return the following content:

   pll-ddr                           5        5        0  2784000000          0     0  50000         Y

This indicates that the DDR clock frequency is 2784000000/2=1392MHz, meaning the clock frequency is half of the transfer rate.3. Get and set NPU frequency

cat /sys/kernel/debug/clk/clk_summary | grep npu

After execution, it will return the following content:

npu                            0        0        0   696000000          0     0  50000         N

This indicates that the NPU frequency is 696000000=696MHz. When we load the NPU driver vipcore.ko, the corresponding /sys/kernel/debug/viplite directory will appear, and at this point, the above command will return the following content:

npu                            0        1        0   864000000          0     0  50000         N

When the NPU driver version is v2.0.3 or a newer version, the NPU frequency can be read and set using the following commands:Read:

cat /sys/kernel/debug/viplite/clk_freq

Set:

echo 328000000 > /sys/kernel/debug/viplite/clk_freq

4. Read NPU Loading data

cat /sys/kernel/debug/viplite/core_loading

5. Additional Information1) The relevant descriptions for AllWinner NPU are as follows:

The model for the V85x and R853s platforms is VIP9000PICO_PID0XEE

The model for the MR527, AI985, and T527 platforms is VIP9000NANOSI_PLUS_PID0X10000016

The model for the MR536 and T536 platforms is VIP9000NANODI_PLUS_PID0X1000003B

2) The printed information when loading vipcore.ko is as follows:

/lib/modules/5.15.147 # insmod vipcore.ko[ 5038.716931] npu[6243][6243] vipcore, platform driver init[ 5038.717602] npu[6243][6243] vipcore, insmod param cpu_physical=0x0[ 5038.718346] npu[6243][6243] vipcore, insmod param cpu_physical=0x0[ 5038.719089] npu[6243][6243] vipcore, insmod param cpu_physical=0x0[ 5038.719832] npu[6243][6243] vipcore, insmod param cpu_physical=0x0[ 5038.720591] npu[6243][6243] vipcore, platform device compatible=allwinner,npu[ 5038.721693] npu[6243][6243] vipcore, platform driver device=0xffffff800153bc10[ 5038.722582] npu[6243][6243] vipcore irq number is 102.[ 5038.723202] vipdrv_drv_adjust_param 825 SUCCESS[ 5038.723759] NPU AXI CLK NULL[ 5038.724108] NPU AHB CLK NULL[ 5038.725127] npu[6243][6243] EXTRA: 0xf05a, SID: 0x0, max_freq: 1000000000, default_freq: 1000000000[ 5038.726440] npu[6243][6243] current vf index: 0[ 5038.728033] npu[6243][6243] Supported vf list:[ 5038.729152] npu[6243][6243]     freq: 480000000, vol: 920000[ 5038.729862] npu[6243][6243]     freq: 492000000, vol: 1000000[ 5038.732073] npu[6243][6243]     freq: 648000000, vol: 1000000[ 5038.732760] npu[6243][6243]     freq: 864000000, vol: 1150000[ 5038.733451] npu[6243][6243] NPU Use VF0000, use freq 864 MHz[ 5038.734227] npu[6243][6243] Want set pclk rate(864000000) support(864000000) real(864000000)[ 5038.735222] npu[6243][6243] Want set mclk rate(864000000) support(864000000) real(864000000)[ 5038.742916] npu[6243][6243] Want set npu vol(1150000) now vol(1150000)[ 5038.742935] npu[6243][6243] vipcore, device_cnt=1, core_cnt=1[ 5038.744728] npu[6243][6243] core_0, request irqline=102, name=vipcore_0[ 5038.745580] npu[6243][6243] vipcore, allocate page for video memory, size: 0x400000 bytes[ 5038.746589] npu[6243][6243] vipcore, heap_cnt=0, reserved video memory cpu_phy=0x44000000, vip_phy=0x44000000, size=0x400000[ 5038.748006] npu[6243][6243] VIPLite driver version 2.0.3.1-AW-2024-08-08

Related posts

  1. The Three Giants of Computing: A Hard-Core Showdown Between CPU, GPU, and NPU
  2. Qualcomm’s On-Device Generative AI: Innovations in Hexagon NPU Architecture Achieving 98% Performance Improvement, 0.6 Seconds Image Generation, and 20 Tokens/s LLM Inference
  3. TinyML: An Overview of MCU Manufacturers’ Support for Edge AI
  4. Shenzhen’s ‘Post-00s’ Homemade Drone Sets World Record! How Hardcore is the ‘Aerial High-Speed Rail’ at 340 km/h?
  5. Nuvoton Technology Launches Efficient AI MCU Deployment Tool NuML Toolkit
  6. The Rise of Domestic Analog Chip Manufacturers Amidst Market Changes
  7. 6 PCB Design Guidelines Every PCB Engineer Should Know
  8. Why Terminal Resistors are Necessary in CAN Bus Networks
  9. Pitfalls and Pathways in the Embedded Industry
  10. Exploring Smart Factories! The PCB Production Line is Surprisingly Sci-Fi!
  11. How to Use a Multimeter: A Comprehensive Guide
  12. TF ADC Competition Accelerates: Zai Ding Files for Cervical Cancer Approval, Xin Nuo Wei Launches Phase III for Pancreatic Cancer
  13. How to Correctly Open Douyin Commands
  14. Cangjie Language Version of SimpleHttpServer Updated to Support macOS!
  15. Douyin Command Tutorial: Learn 3 Simple Ways to Open Douyin Commands and Usage Rules in 1 Minute
  16. Can’t Open Douyin Command? Unlock 3 Simple Techniques in 1 Minute
  17. Douyin Command Opening Tutorial: Learn 3 Simple Ways to Open Douyin Commands and Usage Rules in 1 Minute
  18. Linux System Security and Applications: Is It Just Simple “Firewall Configuration”?
  19. Linux Shell Variables: A Comprehensive Guide from Beginner to Expert

Leave a Comment