Seg fault in linearizer after latest patch series

Seg fault in linearizer after latest patch series

[Dibyendu Majumdar]

Hi Luc,

I am getting a segmentation fault in the linearizer - which I
replicated using  your GitHub repository branch llvm-fixes-v2.

Please try -v option when running test-linearizer on this program.
Other programs also cause this. Looks like the offending line is
accessing the def field in a pseudo - and I think one of your recent
changes does not set def anymore?

static double test(double *p)
{
 double x;
 x = *p;
 return x;
}
int main(void)
{

 double u,v;
 v = 42.0;
 u = test(&v);
 return v;
}

Thanks and Regards
Dibyendu

Re: Seg fault in linearizer after latest patch series

[Dibyendu Majumdar]

On 16 March 2017 at 00:24, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> Hi Luc,
>
> I am getting a segmentation fault in the linearizer - which I
> replicated using  your GitHub repository branch llvm-fixes-v2.
>
> Please try -v option when running test-linearizer on this program.
> Other programs also cause this. Looks like the offending line is
> accessing the def field in a pseudo - and I think one of your recent
> changes does not set def anymore?
>

Sorry I mean above that it does set def in argument pseudos. So
perhaps a chek is needed when there is an attempt to access def.

Regards

Re: Seg fault in linearizer after latest patch series

[Luc Van Oostenryck]

On Thu, Mar 16, 2017 at 12:24:03AM +0000, Dibyendu Majumdar wrote:
> Hi Luc,
> 
> I am getting a segmentation fault in the linearizer - which I
> replicated using  your GitHub repository branch llvm-fixes-v2.
> 
> Please try -v option when running test-linearizer on this program.
> Other programs also cause this. Looks like the offending line is
> accessing the def field in a pseudo - and I think one of your recent
> changes does not set def anymore?
> 
> static double test(double *p)
> {
>  double x;
>  x = *p;
>  return x;
> }
> int main(void)
> {
> 
>  double u,v;
>  v = 42.0;
>  u = test(&v);
>  return v;
> }

Hi,

Yes, there is several such crashes that may arise with the -v option,
I mean, even without the llvm serie.

I've a trivial patch for some of the cases since a little while,
I just need a bit more time.

Luc