A fortran testcase for PR 33794

["Uros Bizjak" <ubizjak at gmail dot com>]

Hello!

A bug was exposed by recent optimization that manifested in induct.f90
failure from polyhedron test suite. I was trying to construct a C
testcase from this, but failed miserably.

I would like to ask a fortran speaking person for a bit of help here.
Attached testcase exposes the bug on x87 when compiled with -O2
-ffast-math, but this testcase can only be compiled, it doesn't
produce a runtime test.

I would kindly ask someone if he/she can turn this function into
something appropriate for a fortran testsuite. The input parameters
can be copied from PR and are:

 input:

mu =  1.25663706143591729E-006
turns = 166666.66666666666
l = 3.00000000000000006E-003
r =2.99999999999999989E-002

When compiled correctly, the function should return something like

l_self =  2.13170362012935939E-002

(sorry, I don't know expected tolerances here, but the value shoud be
in the same order of magnitude ;) )

Thanks,
Uros.

Attachment: xx.f90
Description: Binary data