Reverse engineering the rk3588 npu
License: GNU General Public License v3.0
Reverse engineering the rk3588 npu.
Internals of the NPU, similarity to NVDLA and matrix muplication support covered in blog post. Example integration llama2.c running tinystories covered in blog post
To build :
mkdir build
meson build
cd build
ninja
To run tests (tested against 5.10 kernel) :
ninja -C build test
the description of these registers is vague, and missing bits details, do you have info about these regs
Hi, I'm trying to run your code on rk3568.
but when I run ./matmul_fp16_test 1 32 16 I get "RKNPU_SUBMIT returned -1"
and the output in dmesg:
RKNPU: job: 000000004eb23a7b, iommu domain id: 0, wait_count: 1, continue wait: 0, commit elapse time: 6251358us, wait time: 6251362us, timeout: 6000000us
RKNPU: failed to wait job, task counter: 0, flags: 0x5, ret = 0, elapsed time: 6252728us
RKNPU: job timeout, flags: 0x0:
RKNPU: core 0 irq status: 0x0, raw status: 0x0, require mask: 0x300, task counter: 0x0, elapsed time: 6359758us
I added a register dump to the rknpu.ko driver and run rknn_matmul_api_demo (from rknn toolkit2)
by doing
./rknn_matmul_api_demo 1 1 32 16 0 0 1 1 1,
in the rknpu.ko driver, in rknpu_job.c in the rknpu_job_subcore_commit_pc function, the first_task->regcmd_addr dump looks like this: (full dump in rknn_matmul_api_demo_dump.txt )
[ 302.972445] 0: 0x02010000000e1004 ( op: 0x0201 reg: 0x1004 value: 0x0000000e )
[ 302.973188] 1: 0x04010000000e2004 ( op: 0x0401 reg: 0x2004 value: 0x0000000e )
[ 302.973840] 2: 0x08010000000e3004 ( op: 0x0801 reg: 0x3004 value: 0x0000000e )
[ 302.974668] 3: 0x0201000000711040 ( op: 0x0201 reg: 0x1040 value: 0x00000071 )
[ 302.975319] 4: 0x10010000000e4004 ( op: 0x1001 reg: 0x4004 value: 0x0000000e )
[ 302.975965] 5: 0x020100000220100c ( op: 0x0201 reg: 0x100c value: 0x00000220 )
[ 302.976611] 6: 0x0201000000101010 ( op: 0x0201 reg: 0x1010 value: 0x00000010 )
[ 302.977256] 7: 0x0201000000091014 ( op: 0x0201 reg: 0x1014 value: 0x00000009 )
[ 302.977903] 8: 0x0201000000001018 ( op: 0x0201 reg: 0x1018 value: 0x00000000 )
...
as I understand from your source codes op 0x0201 is OP_REG_CNA,
but what does 0x0401 mean - is there a place where I can read about it?
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.