tlopex opened a new pull request, #20259:
URL: https://github.com/apache/tvm/pull/20259
This PR adds a CUDA target tag for NVIDIA Jetson AGX Thor:
```python
tvm.target.Target("nvidia/jetson-agx-thor")
```
The target configuration includes:
- CUDA architecture: `sm_110a`
- Maximum shared memory per block: `232448` bytes
- Registers per block: `65536`
- L2 cache: `32 MiB`
- Host CPU: Arm Neoverse V3AE with 14 cores
The change is limited to the new target tag and its configuration test. It
does not change generic CUDA target detection, CUDA runtime behavior, or
existing targets.
Validation performed on NVIDIA Thor:
- CUDA kernel compilation and execution
- NVCC and NVRTC compilation
- TMA, mbarrier, TMEM, and tcgen05 instructions
- BF16 tcgen05 GEMM correctness
- Dynamic shared-memory launches above 48 KiB
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
