Support for RDNA 2/3/3.5 and future UDNA?

[Looong01]

Hello and thanks for amd engineers' devote for Taichi lang!

I am a ROCM developer and a RDNA GPU user. I want to know whether Taichi support RDNA 2/3/3.5 and future UDNA? If not yet, whether it adds these support in the future?

Thanks!

[yaoliu13]

Our team currently focus on MI300X. I would like to collect some information from you, and I will reach out to other teams.

1. What specific RDNA GPU are you using?
2. What company or university are you working for?

[Looong01]

Our team currently focus on MI300X. I would like to collect some information from you, and I will reach out to other teams.

1. What specific RDNA GPU are you using?
2. What company or university are you working for?

1. 7900 XTX & Pro W7900
2. University of Amsterdam

[Looong01]

Up to now, I meet this error:

Describe the bug
import taichi as ti. Then, free(): invalid pointer. Stopped (core dumped) without any other logs.
I try to debug Taichi itself and I found that it crush at from taichi._lib.core import taichi_python as core of taichi/_lib/utils.py with trying to import something from taichi/_lib/core/taichi_python.cpython-310-x86_64-linux-gnu.so

To Reproduce
import taichi as ti
Then, free(): invalid pointer. Stopped (core dumped) without any other logs.

Log/Screenshots

Additional comments
Ubuntu 24.04 LTS
AMD Radeon 7900XTX
ROCm 6.4.3
GLIBC 2.39
Python 3.10&3.11

[yaoliu13]

https://rocm.docs.amd.com/en/latest/compatibility/ml-compatibility/taichi-compatibility.html
Taichi is supported on ROCm 6.3.2. There is support through the ROCm software stack for all Taichi GPU features on AMD Instinct MI250X and MI210X series GPUs with the exception of Taichi’s GPU rendering system, CGUI.

Could you try ROCm 6.3.2? If it still doesn't work, I will reach out to other teams for you.