Unify pragmas mm kernel

[L-roro]

This file was apparently forgotten in #2248.

[jgmelber]

A couple questions but otherwise LGTM

[L-roro]

Clang's pragmas are supposed to be before the loop, the same way it is done in the outer loop [see here]. The chess compiler user manual that explains chess' pragmas is not public and I do not have access to it (link), so I can only imitate what is being done in other examples (such as the previously mentioned pr #2248) or in the public AIE API user guide (here). In those examples, chess' pragmas are between the loop header and body. So I assumed that was the expected behavior.

In any case, do note that:
-When using peano, the AIE_LOOP_FLATTEN and AIE_PREPARE_FOR_PIPELINING macros are simply being ignored in our repo (look here).
-When using chess, this is just identical to the AIE API example. Putting the pragma before in the first loop actually differs from that, but I cannot check for correctness of this in the guide without having access.

Additionally it looks like adding the clang prepare for pipelining pragma (when using Peano then) breaks other flags, in particular the emulate_bfloat16_mmul_with_bfp16 one and outputs garbage. This might be an error from peano since chess does work with the equivalent flags and pragmas.

TLDR:
-I thought this would be trivial, but I need to ask someone about it, changing the pr back to draft.
-Might need to open a peano issue
-No public information about chess' pragmas makes it impossible for me to answer your question confidently

[jgmelber]

@jackl-xilinx can you help?