[Bug tree-optimization/122103] New: [15/16 Regression] SVE vectorisation not honoring -fno-trapping-math

[ktkachov at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122103

            Bug ID: 122103
           Summary: [15/16 Regression] SVE vectorisation not honoring
                    -fno-trapping-math
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Keywords: aarch64-sve, wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64

Created attachment 62478
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62478&action=edit
GCC 15 reproducer

Vectorising the testcase in question with SVE leads to a spurious FP trap.

Compiling with e.g. gfortran repro.F90 -O2 -march=armv9-a 
-fdisable-ipa-simdclone and running it I get:
 invalid operation: F
 division by zero: F
 overflow: F
 invalid operation: T // <<<<<<<<<
 division by zero: F
 overflow: F

Whereas with GCC 14 we get:
 invalid operation: F
 division by zero: F
 overflow: F
 invalid operation: F // <<<<<<<<<
 division by zero: F
 overflow: F

I think this is PR96373 reappearing.