Stelios Tzelepis comments

Results 12 comments of


                                            Stelios Tzelepis

Add Resource, io_parallel, depth_multiplier>1 and Quartus support for (Q)DepthwiseConv

hi @jmitrevs @vloncar I am interested in the `Resource` strategy for DepthwiseConv2D as I want to use it for Separable convolution layers. Anything in mind on how to implement it?

DSP packing Vivado backend

hey @jmitrevs @vloncar if you have any thoughts let me know

DSP packing Vivado backend

had to do some > how are overflows generally handled? either they are detectable on the output and thus one can recover from those with some cost or if one...

Fix initiation interval of pooling and zeropadding layers on Vitis backend

@jmitrevs in the model that I am working with, we only use separable convolutions. if the II=1 for the zeropadding and maxpooling, the depthwise and pointwise convolutions have smaller latecy...

Fix initiation interval of pooling and zeropadding layers on Vitis backend

> For ZeroPadding, the change in fact shouldn't work, but it does. II=1 is not achievable, since the inner loop will write one element to the stream in every iteration....

Depthwise 1D and 2D Resource strategy for io_stream

Some results from vivado synthesis

Depthwise 1D and 2D Resource strategy for io_stream

Pipelining over the kernel size might produce better QoR. Pending to test this hypothesis

Depthwise 1D and 2D Resource strategy for io_stream

thanks @xtreme8000! fixed

FIFO depth optimizer for Vitis backend

hey @vloncar @jmitrevs do you think a unit test is required? To avoid delaying the CI pipeline, I could add a dummy model that verifies synthesis and that the flow...

FIFO depth optimizer for Vitis backend

To the reviewers: when we agree on how we want to test this, I can open an issue to also include the Vivado/VivadoAccelerator FIFO optimization in the tests, as a...