areusch commented on a change in pull request #8023:
URL: https://github.com/apache/tvm/pull/8023#discussion_r644186315
##########
File path: include/tvm/runtime/module.h
##########
@@ -232,6 +232,8 @@ constexpr const char* tvm_param_prefix = "__tvm_param__";
constexpr const char* tvm_lookup_linked_param = "_lookup_linked_param";
/*! \brief The main AOT executor function */
constexpr const char* tvm_run_func_prefix = "tvm__run_func";
+/*! \brief The entrypoint function to the generated network */
Review comment:
yeah there is some overlap between Module-based Model Runtime and the
firmware-facing interface. the design priority for the firmware-facing
interface is that it makes sense as an embedded-facing thing. however, i'm
trying to keep it to be as similar to Module-based Model Runtime interface
within those constraints.
given this generated function is currently somewhat internal, i think we can
use `tvm_module_main` for now and focus our efforts on the wrapping API code.
should there be a case where multiple models are placed in the same module,
we'll likely have to revisit that later on. but it's not clear we will support
this, nor is it clear we would support it by allowing a single module to mix
models. (there are, however, some compelling use cases, such as running two
unrelated models in parallel).
--
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.
For queries about this service, please contact Infrastructure at:
[email protected]
