Fix ALU fmin/fmax to use float, not int comparison (!23) · Merge requests · OpenHW Group / backup-20241120 / cv32e40p

Eclipse Webmaster requested to merge github/fork/TobiasKaiser/pulpinov1 into pulpinov1 Oct 27, 2017

Created by: TobiasKaiser

The fmin.s / fmax.s instructions gave wrong results, because they used the integer comparison flag, not the float comparison flag.

A minimal example to demonstrate the unintended behavior of of fmax.s is attached. In this example fmax.s(-200, -190) gave -200 as a result. fmax_min_example.c.txt

I found this problem when a C program that uses floats returned wrong results when using the RI5CY core with FPU. The ri5cy toolchain gcc makes frequent use of the fmax.s when optimizing with -O3.

Fix ALU fmin/fmax to use float, not int comparison

Merge request reports