To use, run "make" in the sw directory. This will create a file
called bldmpy. Running bldmpy with two
numbers as arguments will build a multiply core that multiplies two numbers
together of the bit-widths given. Hence, bldmpy 12 12 will create a multiply
core that multiplies two 12'bit numbers together.
Actually, that's not quite right. It will build two cores--a signed multiply
core and an unsigned multiply core. Both will (by default) be placed into the
rtl directory. Running make in the rtl directory will apply
Verilator to these files. If you
then run make mpy_tb_12x12 in the bench/cpp directory, you'll
build a test bench that (for a 12x12 multiply) will test all combinations of
12x12. (Change the 12x12 for the actual number you built with, if you'd
rather build a test-bench for a different multiply.) Beware, the test
bench is exhaustive: you may not wish to run it on a 32x32 multiply, as it
might take years.
The repository also contains a third core, slowmpy. This one isn't built by the coregen process above. Instead, it contains a multiplication implementation that is designed to be low logic, and hence trades logic for speed. It will take about N+2 clocks to multiply an N-bit number.
This software, and the cores it generates, are licensed under the GPLv3.
Should you find the GPLv3 license insufficient for your needs, other licenses can be purchased from Gisselquist Technology, LLC.