Problem with reducers and array notation (__sec_reduce_min)

[leifblaese]

Hi,

I tried to create an example finding the minimum of values in an array using the __sec_reduce_min()-reducer. However, i got various error messages that i believe point out some bugs. I used the latest version of the cilk-branch (downloaded today, 21:00 EST) only with the flags -lcilkrts and -fcilkplus. I am running a Fedora 16 x86-64. 

[cpp]using namespace std;

int main(int argc, char** argv) {
        double* A = new double[5]; 
        for (int i = 0; i < 5; i++) A = i;
        __sec_reduce_min(A[0:5]);
        return 0;
        
}[/cpp]

This code produces the following error:

test.cpp: In function ‘int main(int, char**)’:

test.cpp:18:1: internal compiler error: in gimplify_var_or_parm_decl, at gimplify.c:2055

 }

 ^

test.cpp: In function ‘int main(int, char**)’:test.cpp:18:1: internal compiler error: in gimplify_var_or_parm_decl, at gimplify.c:2055 } ^

If you change the line with the reducer to 

[cpp]double myvar = __sec_reduce_min(A[0:5]);[/cpp]

The error message becomes: 

test.cpp: In function ‘int main(int, char**)’:

test.cpp:15:1: internal compiler error: in get_expr_operands, at tree-ssa-operands.c:1001

 }

 ^

I actually got a third error message in another function. I was trying to build a k-means implementation. Unfortunately, i can't reproduce the error; i used an older version of cilk to compile; now, when i use the latest version, it produces the second error message i wrote. I assume, that the third bug is already fixed. If i find it, i will post it.

[Balaji_I_Intel]

Hello Leif,
    I have reproduced the problem. I will get back to you as soon as I fix it.

Thanks,

Balaji V. Iyer.

[TimP]

1.         double* A = new double[5];   
2.         for (int i = 0; i < 5; i++) A = i;  
3.         __sec_reduce_min(A[0:5]); 

Sorry, I still find the notation difficult to remember, particularly with plain C99 or C++ next to CEAN.  In the notations of other languages, you would exceed the array bounds in this reduction.  Of course, in CEAN this means 5 elements, ending with A[4].

[Balaji_I_Intel]

Hello Tim,
    I am sorry, but I do not understand how we will exceed array bounds..

A[0:5] implies we start at A[0] and access 5 elements starting at A[0], so in essence, we are trying to find the min of A[0], A[1], A[2], A[3] and A[4].

Thanks,

Balaji V. Iyer. 

[Balaji_I_Intel]

Hello Leif,
   I have fixed the problem and checked in a patch to Cilk Plus gcc. Please let me know if you have further issues.

Thanks,

Balaji V. Iyer.