Re: Code crashes icaros [Bug Reports]

Forum Index

Board index » Development » Bug Reports » Code crashes icaros

Browsing this Thread: 1 Anonymous Users

Bottom

Previous Topic

Next Topic Register To Post

Kalamatee

Re: Code crashes icaros

Home away from home

are you able to test a small change?

in compiler/mlib/s_rint.c

line 41


double w,t;

change to ..


volatile double w;

    double t;

and let me know if that fixes it? it seems that "w" gets optimised away.

Posted on: 2013/7/26 22:06

ncafferkey

Re: Code crashes icaros

Home away from home

Quote:

Kalamatee wrote:

Excuse my ignorance - but what is mlib optimisation?

If you mean gcc optimising the mlib code - then probably yes (unless you can see why it breaks?)

Yes, I just meant adding -O0 to CFLAGS.

I think it crashes somewhere in this function (compiler/mlib/s_rint.c):


double

rint(double x)

{

        int32_t i0,j0,sx;

        u_int32_t i,i1;

        double w,t;

        EXTRACT_WORDS(i0,i1,x);

        sx = (i0>>31)&1;

        j0 = ((i0>>20)&0x7ff)-0x3ff;

        if(j0<20) {

            if(j0<0) {

                if(((i0&0x7fffffff)|i1)==0) return x;

                i1 |= (i0&0x0fffff);

                i0 &= 0xfffe0000;

                i0 |= ((i1|-i1)>>12)&0x80000;

                SET_HIGH_WORD(x,i0);

                w = TWO52[sx]+x;

                t =  w-TWO52[sx];

                GET_HIGH_WORD(i0,t);

                SET_HIGH_WORD(t,(i0&0x7fffffff)|(sx<<31));

                return t;      

            } else {

                i = (0x000fffff)>>j0;

                if(((i0&i)|i1)==0) return x; /* x is integral */

                i>>=1;

                if(((i0&i)|i1)!=0) {

                    /*

                     * Some bit is set after the 0.5 bit.  To avoid the

                     * possibility of errors from double rounding in

                     * w = TWO52[sx]+x, adjust the 0.25 bit to a lower

                     * guard bit.  We do this for all j0<=51.  The

                     * adjustment is trickiest for j0==18 and j0==19

                     * since then it spans the word boundary.

                     */

                    if(j0==19) i1 = 0x40000000; else

                    if(j0==18) i1 = 0x80000000; else

                    i0 = (i0&(~i))|((0x20000)>>j0);

                }

            }

        } else if (j0>51) {

            if(j0==0x400) return x+x;   /* inf or NaN */

            else return x;              /* x is integral */

        } else {

            i = ((u_int32_t)(0xffffffff))>>(j0-20);

            if((i1&i)==0) return x;     /* x is integral */

            i>>=1;

            if((i1&i)!=0) i1 = (i1&(~i))|((0x40000000)>>(j0-20));

        }

        INSERT_WORDS(x,i0,i1);

        *(volatile double *)&w = TWO52[sx]+x;   /* clip any extra precision */

        return w-TWO52[sx];

}

Posted on: 2013/7/26 20:15

x303

Re: Code crashes icaros

Just can't stay away

With the included lrint() it doesn't crash anymore, even with -O3 set. I only commented out the math.h coz else it wouldn't compile correctly (round() issue).

Attach file:

Link only for registered users

Posted on: 2013/7/26 18:53

Kalamatee

Re: Code crashes icaros

Home away from home

Quote:

ncafferkey wrote:
It crashes in lrint(). It can be worked around by turning off optimisation for mlib.

@Kalamatee and other developers:
Should I turn off mlib optimisation for ABIv0?

Excuse my ignorance - but what is mlib optimisation?

If you mean gcc optimising the mlib code - then probably yes (unless you can see why it breaks?)

If its gcc's ability to replace math with more efficient routines, then it might be worth finding out what routines are failing.

It would be nice to know exactly why it fails though so the same problem doesn't crop up again.

Posted on: 2013/7/26 18:43

ncafferkey

Re: Code crashes icaros

Home away from home

It crashes in lrint(). It can be worked around by turning off optimisation for mlib.

@Kalamatee and other developers:
Should I turn off mlib optimisation for ABIv0?

Posted on: 2013/7/26 17:28

x303

Re: Code crashes icaros

Just can't stay away

Abi V1 uses a newer version of gcc, which could be the solution. I couldn't compile it myself because of an error with gcc, libgcc.a (file format) doesn't get recognised. The error is also there under nightly abi 0. Also no problem with vbcc aros.

Posted on: 2013/7/26 10:54

Kalamatee

Re: Code crashes icaros

Home away from home

I tried under ABI v1 and it works correctly there so it must be specific to ABI v0.

You could always try in AROS hosted and see if you can use GDB to identify where/why it crashes.

Posted on: 2013/7/25 23:39

x303

Code crashes icaros

Just can't stay away

When you compile and run the supplied code, then vmware and vbox (with icaros 1.5.2) are reset completely (starting from the beginning with the grub window). I found the code crashing when running the configure file from libsndfile. Hope someone can check it out and fix this one.

Attach file:

Link only for registered users

Posted on: 2013/7/25 21:01