Felix Jentzsch

RTL ConvolutionInputGenerator

1

## Overview Adds an RTL Convolution Input Generator / Sliding Window Generator (SWG) that aims to replace all 16 "ConvolutionInputGenerator_*" [function variants in finn-hlslib](https://github.com/Xilinx/finn-hlslib/blob/vitis-hls/slidingwindow.h), extend functionality, and improve resource &...

**Depends on:** https://github.com/Xilinx/finn-hlslib/pull/65

Related Gitter discussion: https://gitter.im/xilinx-finn/community?at=60dad875effd491d8c8ec7bc For some (?) users, the finn pip install in "finn_entrypoint.sh" (`pip install --user -e $FINN_ROOT`) re-installs finn-base from the official main branch even though a different...

### Problem: In test_end2end_bnn_pynq.py, the "ipstitch_rtlsim" test and the rtlsim-based part of the "validate_top1" test fail. The simulation runs through but yields incorrect results (e.g. a top1-accuracy of ~6% for...

### Summary Exposes control registers to enable re-programming of the RTL SWG during runtime. The RTL component is generated once for the largest required configuration in terms of buffer depth...

### Overview Introduces support for additional parallelism along the spatial feature map dimension, i.e., processing multiple 'pixels' or 'samples' simultaneously. For some layers (e.g., MVAU), this functionality already existed in...

Restores MMV support for the VVAU, which was non-functional after the recent addition of SIMD parallelism. This is part of a larger FINN PR and might be extended by MMV...

In some cases clang memory usage during the initial HLS phase explodes, resulting in the clang process being killed. The user receives `ERROR: [HLS 200-1715] Encountered problem during source synthesis`...

**Work-in-progress** Extends the characterization-based FIFO sizing algorithm to work better with non-linear models (e.g., ResNets). To minimize rtlsim, the goal is to rely mostly on the new [analytical approach](https://github.com/Xilinx/finn/pull/1185) to...