[Bug tree-optimization/86367] New: FRE1 tree pass deletes code in gcc.target/powerpc/nan128-1.c when long double is IEEE 128

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86367

            Bug ID: 86367
           Summary: FRE1 tree pass deletes code in
                    gcc.target/powerpc/nan128-1.c when long double is IEEE
                    128
           Product: gcc
           Version: 9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: meissner at gcc dot gnu.org
  Target Milestone: ---

Created attachment 44339
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44339&action=edit
Test case

I was going through the lists of tests that behave differently when the powerpc
long double type is set to IEEE 128-bit.

The test gcc.target/powerpc/nan128-1.c fails when compiled with
-mabi=ieeelongdouble -Wno-psabi (or configuring the compiler using
--with-long-double-format=ieee).

I tracked this down, and the FRE1 tree pass is deleting all of the code, and
replacing it with a call to abort.