akaashrp opened a new pull request, #20166: URL: https://github.com/apache/tvm/pull/20166
Building on #20156's borrowed tensor-cache record views, upload aligned records that require no decoding directly into their final WebGPU tensors. This avoids allocating a temporary CPU tensor, passing the record through the FFI decoder, and copying the CPU tensor to WebGPU. BF16, unaligned, and CPU records retain the existing decoder path. Also: - Validate exact tensor storage sizes, including packed sub-byte dtypes - Honor tensor byte offsets during WebGPU writes - Read `DLTensor::byte_offset` as an aligned unsigned 64-bit value and reject values outside JavaScript’s safe integer range - Dispose partially constructed tensors when loading fails -- 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]
