tqchen commented on code in PR #335: URL: https://github.com/apache/tvm-ffi/pull/335#discussion_r2614218201
########## docs/guides/kernel_library_guide.rst: ########## @@ -94,6 +94,14 @@ However, the tensors allocated by :cpp:func:`tvm::ffi::Tensor::FromNDAlloc` only But in the scenarios of linked runtime libraries and c++ applications, the libraries alive globally throughout the entire lifetime of the process. So :cpp:func:`tvm::ffi::Tensor::FromNDAlloc` works well in these scenarios without the use-after-delete issue above. Otherwise, in general, :cpp:func:`tvm::ffi::Tensor::FromEnvAlloc` is free of this issue, which is more **recommended** in practice. + +FromNDAllocStrided +^^^^^^^^^^^^^^^^^^ + +:cpp:func:`tvm::ffi::Tensor::FromNDAllocStrided` can be used to create a tensor with a custom memory allocator and strided layout (e.g. column major layout). +Note that for tensor memory that will be returned from the kernel library to the caller, we instead recommend using :cpp:func:`tvm::ffi::Tensor::FromEnvAlloc` +followed by :cpp:func:`tvm::ffi::Tensor::as_strided` to create a strided view of the tensor. Review Comment: also added for tensorview -- 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]
