Failed to Compile a fp32 Gemmini about gemmini HOT 3 CLOSED

By the way, could you please clarify the Floating point datatype a bit more? Is Float(8, 24) a typical Float32 or is Float(8, 32) a typical Float32?

from gemmini.

I think a 32-bit fp should be Float(8,24). You can find its definition in arithmatic.scala.
I tried to use floating point before but with no success. The function overload for fp operation is not completely defined in arithmatic.scala. Even after I finished defining those functions for fp, I could not synthesize it. It seemed that all those fp units need to be synthesized into logic instead of using some floating point dsp, which was taking a hell lot of time and I just gave up. I used fixed-point instead which is better supported by chisel.
If you have any success please let me know because I am also interested in a floating point implementation.

from gemmini.

Sorry for the late reply. Floating-point types are working now in the dev branch. You can look for commonly-used floating point configurations in ConfigsFP.scala, where FP32, BF16 and other configurations are defined.

Also, if you use the dev branch, then be sure to check the CHIPYARD.hash and SPIKE.hash files, which will let you know which Chipyard and Spike commits you need to be on to match your Gemmini version.

from gemmini.