[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#762640: gfortran-4.9: generates buggy assembler at -O2 on armhf



Package: gfortran-4.9
Version: 4.9.1-15
Severity: important

nwchem FTBFS on armhf, see e.g.
https://buildd.debian.org/status/fetch.php?pkg=nwchem&arch=armhf&ver=6.3%2Br2-1&stamp=1410744209

|gfortran -c -g -O2 -ffast-math -fno-aggressive-loop-optimizations  -O2  
|       -ffast-math -Wuninitialized -I. -I../include -I./include
|       -I/«BUILDDIR»/nwchem-6.3+r2/src/include
|       -I/«BUILDDIR»/nwchem-6.3+r2/src/tools/install/include -DLINUX
|	-DGCC4 -DGCC46 -DGFORTRAN -DPARALLEL_DIAG -DMRCC_METHODS
|	tce_mrcc_energy.F
|/tmp/ccmB6kme.s: Assembler messages:
|/tmp/ccmB6kme.s:6881: Error: offset out of range
|/tmp/ccmB6kme.s:6885: Error: offset out of range
|make[5]: ***
|[/«BUILDDIR»/nwchem-6.3+r2/lib/LINUX/libtce.a(tce_mrcc_energy.o)]
|Error 1

This is a regression from earlier versions of gfortran-4.9, other arches
seem fine.

If -O2 is replaced with -O1, it builds fine.

The preprocessed source can be found here:

https://people.debian.org/~mbanck/tce_mrcc_energy.f

It should (not) build like this: "gfortran -O2 -c tce_mrcc_energy.f"

The assembler output for -O1 and -O2, respectively, can be found here:

https://people.debian.org/~mbanck/tce_mrcc_energy.s.ok
https://people.debian.org/~mbanck/tce_mrcc_energy.s.wrong


Michael


Reply to: