tqchen commented on code in PR #671:
URL: https://github.com/apache/tvm-ffi/pull/671#discussion_r3591822406
##########
python/tvm_ffi/cython/device.pxi:
##########
@@ -115,13 +115,8 @@ cdef class Device:
}
_DEVICE_NAME_TO_TYPE = {
- "llvm": DLDeviceType.kDLCPU,
"cpu": DLDeviceType.kDLCPU,
- "c": DLDeviceType.kDLCPU,
- "test": DLDeviceType.kDLCPU,
"cuda": DLDeviceType.kDLCUDA,
- "nvptx": DLDeviceType.kDLCUDA,
- "cl": DLDeviceType.kDLOpenCL,
"opencl": DLDeviceType.kDLOpenCL,
"vulkan": DLDeviceType.kDLVulkan,
"metal": DLDeviceType.kDLMetal,
Review Comment:
Good catch on the mismatch. This change intentionally makes device parsing
canonical-only, so adding aliases to Python would reintroduce the compatibility
surface being removed. I updated the C++ parser in 68cc76b to accept only
`metal` and `webgpu`, and moved the existing `mps`/`wgpu` expectations to the
rejection cases. Direct Python construction and typed FFI conversion now follow
the same canonical-name policy.
--
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]
