Re: [apache/tvm-rfcs] Additional Target Hooks RFC (#10)

@mbs-octoml / @areusch apologies for the delay, there was a fair amount to tweak in here to reflect all the discussions around this RFC. Please could you take another look as I think this is lined up now :smile_cat: -- You are receiving this because you are subscribed to this thread. Reply to

[apache/tvm-rfcs] Fix formatting in CMSIS NN RFC and remove plan (#35)

This fixes the formatting of the Markdown so that it properly renders code snippets, links and some of the lists. It also removes the upstreaming plan in favour of the GitHub issue. Co-authored-by: Ashutosh Parkhi You can view, comment on, or merge this pull request onli

[apache/tvm] BUG #8013: Remove register_alter_op_layout example from dev/use_pass_infra.py (#9076)

This tutorial registers a global layout transformation for conv2d for all targets which is not well-formed. Later uses of conv2d in the tutorials pick that layout up then assert fail in the conv2d type-relation. Better would be to register a transform for an entirely fake target, but that is beyon

Re: [apache/tvm-rfcs] [RFC][TIR] TIR Non-scalar Constants (#22)

@junrushao1994 so the concern is embedding e.g. the output of `NDArray::repr` into TIR text format, correct? That makes sense to me. It seems like small tir.constant (e.g. perhaps we could constraint this as: tir.constant with shape that can be squeezed to rank 2 and, when doing so, can reasonab

Re: [apache/tvm-rfcs] [RFC][TIR] TIR Non-scalar Constants (#22)

@areusch Thanks for the discussion! Yep, I agree with you that we need to address these points, and it's certainly doable under the current design. @manupa-arm would you like to respond to Andrew's question? Thanks a lot! -- You are receiving this because you are subscribed to this thread. Repl

Re: [apache/tvm-rfcs] [RFC][TIR] TIR Non-scalar Constants (#22)

Since we have relay.Constants having the same need to parse in constants, it would be appreciated not to block progress on deciding on the mechanics of parsing in NDArrays in to IRModule. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view

Re: [apache/tvm-rfcs] [RFC][TIR] Adding annotation field to tir.allocate (#23)

There is generally a tradeoff between the: - A0: The desire to preserve information during transformation. - A1: A growing set of attributes that is impossible to keep track of(thus preserve) during transformations. >From the A1's pov, allowing set of attributes that changes the semantics of >th