Alexander Efimov

[luci-micro] Remove C++ runtime from micro interpreter

7

## Goal - Investigate how C++ library affects flash footprint of luci-micro interpreter - Verify that licu-interpreter could run without C++ runtime - Modify luci-micro and/or it's CMakeListst.txt accordingly ##...

[mir] Do we still need mir-related components?

3

## What? I propose to delete https://github.com/Samsung/ONE/tree/master/compiler/mir and related components from source code. ## Why? This is an old part of the project, and I am not sure if someone...

[luci-micro] Speedup GRU/LSTM operations

11

## What Need to implement optimized kernels for GRU and LSTM operations in interpreter for MCU ## Why For now these operations are not optimized, works slow and consumes a...

[luci-interpreter] Benchmark modes from tflite-micro

31

This is continuation of #5080, but with more specific goals. ### Goal To measure luci-interpreter performance and memory consumption for models delivered with tflite-micro: https://github.com/tensorflow/tflite-micro/tree/main/tensorflow/lite/micro/benchmarks Currently it contains two models:...

Support "trans" epilogue for asymmetrical tensors. Example of crash: M==16, N==64, i.e. z_tri of shape [16, 64]

[AMD] [triton-opt] Add AMD passes to triton-opt

5

This PR adds AMD specific passes to triton-opt.

[AMD] Generalize optimize epilogue pattern

7

This PR adds support of following pattern in optimize epilogue pass: ->ConvertLayout -> ElementwiseOp -> StoreOp

[AMD] Fix kwidth printing and parsing

3

This PR fixes: - parsing and printing of kWidth attribute for MFMA and WMMA layouts - comment in 06-fused-attention.p tutorial

[AMD] OptimizeLDSUsage pass

8

This PR inroduces OptimizeLDSUsage pass which generalizes LDS optimization, which was part of DecomposeUnsupportedLayouts pass.

[AMD][WMMA] Support dot3d

7

This PR enables support of 3d dot and fixes tests in test_core.py