Element-wise multiplication (hadamard product) support

[rbcarlos]

Is there any way to implement element-wise multiplication (hadamard product) of two tensors?

I have tried to implement this by simply doing:

t3 = t1 * t2

When I try to streamline it however FINN throws an error during the streamlining:

Initializer for mul weights is not set.

[maltanar]

I assume you mean the case where both tensors are dynamic (i.e. not a constant)? That isn't possible right now, but could be enabled by doing the following:

1. Add a new HLS layer to finn-hlslib for multiplying two streams similar to https://github.com/Xilinx/finn-hlslib/blob/master/streamtools.h#L650
2. Add HLSCustomOp wrapper in finn similar to https://github.com/Xilinx/finn/blob/main/src/finn/custom_op/fpgadataflow/addstreams_batch.py
3. Add a convert-to-HLS transformation in finn similar to https://github.com/Xilinx/finn/blob/main/src/finn/transformation/fpgadataflow/convert_to_hls_layers.py#L1088 and a test https://github.com/Xilinx/finn/blob/main/tests/fpgadataflow/test_fpgadataflow_addstreams.py

If you meant the case where one tensor is dynamic and the other is a constant, it should be possible to implement this with existing functionality but might need some massaging (main suspect would be that the constant tensor is the 2nd op and not the 1st)