[GDB]: gdb can't load up simple test case with usage of altivec vector bringing internal error.

[kas1e]

There very simple test case:

#include <stdio.h>
#include <altivec.h>

void add_vectors(vector signed int* a, vector signed int* b, vector signed int* result) {
   // *result = vec_add(*a, *b);
}

int main() {
    return 0;
}

Compile it like:

ppc-amigaos-gcc -maltivec -gstabs test_vector.c -o test_vector

Once we tried to run it in GDB (that one Pegasos2 of course, with real Altivec in) , GDB bring this warning:

../../gdb/gdbtypes.c:939: internal-error: create_range_type: Assertion 'index_type->length () > 0` failed.
A problem internal to GDB has been detected.
futher debugging may prove unrelable.
Unable to dump core

[migthymax]

I tried to track it down. The reason why it fails is that gdb has trouble reading the gstab debug information for the array )A vector is I think internally handled as an array by the compiler).

Dumping the gstab .section gives the following information

> objdump -G  test_vector
 
test_vector:     file format elf32-powerpc
 
Contents of .stab section:
 
Symnum n_type n_othr n_desc n_value  n_strx String
 
-1     HdrSym 0      15     000000b2 1     
 0      SO     0      2      010004c0 1      test_vector.c
 1      OPT    0      0      00000000 15     gcc2_compiled.
 2      FUN    0      0      010004c0 30     add_vectors:F(0,1)=(0,1)
 3      LSYM   0      0      00000000 55     void:t(0,1)
 4      PSYM   0      0      00000008 67     a:p(0,2)=*(0,3)=ar(0,0);0;3;(0,4)=r(0,4);-2147483648;2147483647;
 5      PSYM   0      0      0000000c 132    b:p(0,2)
 6      PSYM   0      0      00000010 141    result:p(0,2)
 7      LSYM   0      0      00000000 155    int:t(0,4)
 8      SLINE  0      4      00000000 0      
 9      SLINE  0      6      00000018 0      
 10     FUN    0      0      010004ec 166    main:F(0,4)
 11     SLINE  0      8      00000000 0      
 12     SLINE  0      9      0000000c 0      
 13     SLINE  0      10     00000010 0      
 14     SO     0      0      01000510 0      

This line makes the trouble:

...
 4      PSYM   0      0      00000008 67     a:p(0,2)=*(0,3)=ar(0,0);0;3;(0,4)=r(0,4);-2147483648;2147483647;
...

The question which arise is, if this is a general gdb bug, or specific because of my port. The later I don't think so, because I don't know how to influence the parsing of gstab. But it even might be that the gcc is producing buggy gstab information.

So if you have or know someone with a linux installation on a PowerPC with an altivec engines, which could verify if it works under linux or fails, would be helpful.