#2345 Regression test gcc-torture-execute-divcmp-1 fails on NetBSD 6.1 Sparc64

[Erik Petrich]

gcc-torture-execute-divcmp-1 fails to compile and aborts with an illegal instruction exception for at least the stm8 and s08 backends on NetBSD 6.1 on Sparc64.

Debugging the dumped core shows:

Reading symbols from /home/epetrich/sdccbuild/bin/sdcc...done.
[New process 1]
Core was generated by `sdcc'.
Program terminated with signal 4, Illegal instruction.
#0  tree_dec_ralloc_nodes<boost::adjacency_list<boost::vecS, boost::vecS, boost::bidirectionalS, tree_dec_node>, boost::adjacency_list<boost::vecS, boost::vecS, boost::bidirectionalS, cfg_node>, boost::adjacency_matrix<boost::undirectedS, con_node> > (T=..., t=<optimized out>, G=..., I=..., ac=..., 
    assignment_optimal=0x100000000)
    at ../../../sdcc/src/stm8/../SDCCralloc.hpp:979
979     static void tree_dec_ralloc_nodes(T_t &T, typename boost::graph_traits<T_t>::vertex_descriptor t, const G_t &G, const I_t &I, const assignment& ac, bool *const assignment_optimal)
(gdb) bt
#0  tree_dec_ralloc_nodes<boost::adjacency_list<boost::vecS, boost::vecS, boost::bidirectionalS, tree_dec_node>, boost::adjacency_list<boost::vecS, boost::vecS, boost::bidirectionalS, cfg_node>, boost::adjacency_matrix<boost::undirectedS, con_node> > (T=..., t=<optimized out>, G=..., I=..., ac=..., 
    assignment_optimal=0x100000000)
    at ../../../sdcc/src/stm8/../SDCCralloc.hpp:979
#1  0x0000000100000008 in ?? ()
#2  0x0000000100000008 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
(gdb)

[Ben Shi]

I also reproduce this crash of sdcc(.exe) on cygwin64.

$ make test-ucz80
Running ucz80 regression tests
sdcc: Calling preprocessor...
sdcc: sdcpp -nostdinc -Wall -DNO_VARARGS -I../../.. -DREENTRANT= -I../../../support/regression/fwk/include -I../../../support/regression/tests -I../../../device/include -obj-ext=.rel -D__SDCC_STACK_AUTO -D__SDCC_INT_LONG_REENT -D__SDCC_FLOAT_REENT -D__SDCC=3_4_3 -D__SDCC_REVISION=9224 -D__SDCC_z80 -D__STDC_NO_COMPLEX__ -D__STDC_NO_THREADS__ -D__STDC_NO_ATOMICS__ -D__STDC_NO_VLA__  gen/ucz80/gcc-torture-execute-divcmp-1/gcc-torture-execute-divcmp-1.c
sdcc: Generating code...
Caught signal 11: SIGSEGV
../../../support/regression/ports/ucz80/spec.mk:53: recipe for target 'gen/ucz80/gcc-torture-execute-divcmp-1/gcc-torture-execute-divcmp-1.rel' failed
make[3]: *** [gen/ucz80/gcc-torture-execute-divcmp-1/gcc-torture-execute-divcmp-1.rel] Error 1
Makefile:277: recipe for target 'results/ucz80/gcc-torture-execute-divcmp-1.out' failed
make[2]: *** [results/ucz80/gcc-torture-execute-divcmp-1.out] Error 2

[Ben Shi]

on cygwin64, it crashed at the recursive call of tree_dec_ralloc_nodes, and the nested level is too deep. but actually cygwin32 was ok with such depth.

[Ben Shi]

This test case on cygwin64 was fixed in reversion #9300.

[Ben Shi]

The stack size is enlarged to 4MB in runtime, but I am not sure if it is OK for other platforms. Otherwise I will roll back it if there are failures in tomorrow's regression tests.

[Philipp Klaus Krause]

Does anyone have a NetBSD sparc64 machine to check if this bug still exists in current SDCC?

Philipp

[Erik Petrich]

It will take me a little bit to set it back up, but I do have a sparc64 with NetBSD. I will check if this bug is still reproducable.

[Philipp Klaus Krause]

Does this problem still exist in current SDCC?

[Philipp Klaus Krause]

It has been 8 years since the bug report. Apparently no one has a NetBSD on sparc64 machine around. SDCC has changed much since, so I'll close this as out of date. Is someone can reproduce it using current SDCC, please reopen the ticket.