SPEC2000 applu.f generates internal compiler error

[frank smith]

Using Linaro's 2014.04 release of the arm-linux-gnueabihf compiler I get the following failure when compiling applu.f which is part of the SPEC2000 floating-point suite.

/opt/gcc-linaro-arm-linux-gnueabihf-4.8-2014.04_linux/bin/arm-linux-gnueabihf-gfortran -c -o applu.o -static -O3 -march=armv7-a -mfpu=vfpv3-d16 -fno-use-linker-plugin -flto -marm -mfloat-abi=hard -ffast-math -ftree-vectorize -floop-interchange -g -static applu.f

GNU MP: Cannot allocate memory (size=4202504)
applu.f: In function 'setbv_':
applu.f:331:0: internal compiler error: Aborted
       subroutine setbv
 ^
no stack trace because unwind library not available
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://bugs.launchpad.net/gcc-linaro> for instructions.
specmake: *** [applu.o] Error 1

[frank smith (smithfra)]

I was able to build all of SPEC200 CINT2000 and all of SPEC2000 CFP2000 with the exception of applu.f in the floating point suite.

[frank smith (smithfra)]

I re-built SPEC cpu2000 with the same flags as above using the 2014.05 release of gcc-linaro-arm-linux-gnueabihf (GCC 4.9) and Applu compiled with no problems.

[Kugan Vivekanandarajah (kugan-vivekanandarajah)]

I can not reproduce the ICE, but it goes into infinite loop in
isl_vec_mat_product,
....
 while ((off = row_first_non_zero(left->row+first,
     left->n_row-first, row)) != -1) {
....

I can also reproduce this in trunk with cloog-0.18.1.tar.gz, isl-0.12.2.tar.bz2, gmp-4.3.2.tar.bz2, mpc-0.8.1.tar.gz and mpfr-2.4.2.tar.bz2.

seems to be a graphite issue and the mininmal argument thar reproduces this is -O3 -floop-interchange. -floop-interchange part of the graphite is what is casuing the trouble.

[Kugan Vivekanandarajah (kugan-vivekanandarajah)]

minimal Fortran code to reproduce should contain subroutine setbv and exact.

[Kuganv (kuganv)]

arm-linux-gnueabihf-gfortran -c -floop-interchange applu.f goes into infinite loop.

It goes into infinite loop in
isl_vec_mat_product,
....
 while ((off = row_first_non_zero(left->row+first,
     left->n_row-first, row)) != -1) {
....

Happens in trunk with cloog-0.18.1.tar.gz, isl-0.12.2.tar.bz2, gmp-4.3.2.tar.bz2, mpc-0.8.1.tar.gz and mpfr-2.4.2.tar.bz2.

minimal Fortran code to reproduce this should contain subroutine setbv and exact.

gcc -v output:
arm-none-linux-gnueabi-gfortran -v
Using built-in specs.
COLLECT_GCC=/home/kugan/work/builds/gcc-fsf-trunk/tools/bin/arm-none-linux-gnueabi-gfortran
COLLECT_LTO_WRAPPER=/home/kugan/work/builds/gcc-fsf-trunk/tools/libexec/gcc/arm-none-linux-gnueabi/4.10.0/lto-wrapper
Target: arm-none-linux-gnueabi
Configured with: /home/kugan/work/sources/gcc-fsf/trunk/configure --target=arm-none-linux-gnueabi --prefix=/home/kugan/work/builds/gcc-fsf-trunk/tools --with-sysroot=/home/kugan/work/builds/gcc-fsf-trunk/sysroot-arm-none-linux-gnueabi --disable-libssp --disable-libgomp --disable-libmudflap --enable-languages=c,c++,fortran --with-arch=armv7-a --with-fpu=vfpv3-d16 --with-float=softfp --with-thumb
Thread model: posix
gcc version 4.10.0 20140710 (experimental) (GCC)

[Dominiq (dominiq)]

> minimal Fortran code to reproduce this should contain subroutine setbv and exact.

Such code, or better, a reduced reproducer, should be provided. From comment 0, it seems that this PR a graphite problem on arm processor, thus not a gfortran bug.

[Dominiq (dominiq)]

No reproducer, no feedback for more than four months. Closing as INVALID.