[Bug jit/106312] New: jit tests fail on powerpc64le

[Bug jit/106312] New: jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

            Bug ID: 106312
           Summary: jit tests fail on powerpc64le
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: testsuite-fail
          Severity: normal
          Priority: P3
         Component: jit
          Assignee: dmalcolm at gcc dot gnu.org
          Reporter: redi at gcc dot gnu.org
  Target Milestone: ---
            Target: powerpc64le-unknown-linux-gnu

When built with --disable-multilib --enable-languages=c,c++,jit
--enable-host-shared

make check-gcc gives more than 100 failures like this:

FAIL: tut01-hello-world.c.exe killed: 128208 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: tut01-hello-world.cc.exe killed: 130474 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: tut02-square.c.exe killed: 130611 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: tut02-square.cc.exe killed: 131024 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: tut03-sum-of-squares.c.exe killed: 131625 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: tut03-sum-of-squares.cc.exe killed: 132893 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: toyvm.c.exe killed: 133014 exp8 0 0 CHILDKILLED SIGSEGV {segmentation
violation}
FAIL: toyvm.cc.exe killed: 133290 exp8 0 0 CHILDKILLED SIGSEGV {segmentation
violation}
FAIL: tut05-bf.c.exe killed: 133398 exp8 0 0 CHILDKILLED SIGSEGV {segmentation
violation}
FAIL: emit-alphabet.bf.exe does not have executable bit set
ERROR: ./emit-alphabet.bf.exe does not exist in unix_load.
UNRESOLVED: emit-alphabet.bf.exe execution test
FAIL: test-accessing-bitfield.c.exe killed: 133501 exp8 0 0 CHILDKILLED SIGSEGV
{segmentation violation}
FAIL: test-error-pr63969-missing-driver.c.exe killed: 134470 exp8 0 0
CHILDKILLED SIGSEGV {segmentation violation}
FAIL: test-error-register-variable-bad-name.c.exe killed: 134698 exp8 0 0
CHILDKILLED SIGSEGV {segmentation violation}
FAIL: test-error-register-variable-size-mismatch.c.exe killed: 134973 exp8 0 0
CHILDKILLED SIGSEGV {segmentation violation}



These failures don't occur for the same configuration on x86_64-pc-linux-gnu.

[Bug jit/106312] jit tests fail on powerpc64le

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #1 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Seems to work for me (on gcc112); is there a particular machine you tried this
on?

[Bug jit/106312] jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Created attachment 53303
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53303&action=edit
top-level config.log (gzip compressed)

Odd, I'm using gcc112 as well.

Full configure command is:

/home/jwakely/src/gcc/configure --prefix=/home/jwakely/gcc/13 --with-gnu-as
--with-gnu-ld --disable-multilib --disable-nls --enable-languages=c,c++,jit
--enable-host-shared

PATH=/opt/cfarm/git-latest/bin:/opt/cfarm/dejagnu-latest/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/home/jwakely/.local/bin:/home/jwakely/bin

gmp, mpfr and mpc are built in-tree after running
./contrib/download_prerequisites --no-isl

[Bug jit/106312] jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Created attachment 53304
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53304&action=edit
gcc/config.log (gzip compressed)

[Bug jit/106312] jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Valgrind shows:

==29655== Process terminating with default action of signal 11 (SIGSEGV):
dumping core
==29655==  Bad permissions for mapped region at address 0x64322A8
==29655==    at 0x5D061A4: mpfr_set_emin (in
/home/jwakely/build/gcc/libgccjit.so.0.0.1)
==29655==    by 0x42A3D27: toplev::main(int, char**) (in
/home/jwakely/build/gcc/libgccjit.so.0.0.1)
==29655==    by 0x42DEC7B: gcc::jit::recording::context::compile() (in
/home/jwakely/build/gcc/libgccjit.so.0.0.1)
==29655==    by 0x42C4DDB: gcc_jit_context_compile (in
/home/jwakely/build/gcc/libgccjit.so.0.0.1)
==29655==    by 0x1000117B: main (in
/home/jwakely/build/gcc/testsuite/jit/tut01-hello-world.c.exe)

[Bug jit/106312] jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Ah, I have quite old versions of the prerequisite libs:

lrwxrwxrwx. 1 jwakely jwakely 12 Feb 17  2017 gmp -> ./gmp-6.1.0/
lrwxrwxrwx. 1 jwakely jwakely 12 Feb 17  2017 mpc -> ./mpc-1.0.3/
lrwxrwxrwx. 1 jwakely jwakely 13 Feb 17  2017 mpfr -> ./mpfr-3.1.4/

I've removed those symlinks and run contrib/download_prerequisites again and am
rebuilding now.

If this fixes it, we might need to document that mpfr 3.1.0 is no longer
sufficient for GCC: https://gcc.gnu.org/install/prerequisites.html

[Bug jit/106312] jit tests fail on powerpc64le

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106312

--- Comment #6 from Jonathan Wakely <redi at gcc dot gnu.org> ---
I still get the same failures.