64-bit integer testing is broken

[Clayton Smith]

The icompare function casts all inputs to int before comparing & printing them. This prevents errors in upper bits from being caught, and makes it difficult to diagnose failures.

volk/lib/qa_utils.cc

Lines 463 to 480 in e853e9b

 bool icompare(t* in1, t* in2, unsigned int vlen, unsigned int tol, bool absolute_mode) 
 { 
     bool fail = false; 
     int print_max_errs = 10; 
     for (unsigned int i = 0; i < vlen; i++) { 
         if (((unsigned int)abs(int(((t*)(in1))[i]) - int(((t*)(in2))[i]))) > tol) { 
             fail = true; 
             if (print_max_errs-- > 0) { 
                 std::cout << "offset " << i 
                           << " in1: " << static_cast<int>(t(((t*)(in1))[i])) 
                           << " in2: " << static_cast<int>(t(((t*)(in2))[i])); 
                 std::cout << " tolerance was: " << tol << std::endl; 
             } 
         } 
     } 
  
     return fail; 
 } 

[Johannes Demel]

We could use long long int cf. Int types.

However, the fundamental int types just feel broken.
I assume we should #include <cstdint> and use fixed width integer types.

In this specific case int64_t would be correct. Though, this type is marked optional.

[Johannes Demel]

If we change this line:

volk/lib/qa_utils.cc

Line 493 in d5b317c

 if (((unsigned int)abs(int(((t*)(in1))[i]) - int(((t*)(in2))[i]))) > tol) { 

to

if (((unsigned int)abs(int64_t(((t*)(in1))[i]) - int64_t(((t*)(in2))[i]))) > tol) { 

we should be good. Technically, a #include <cstdint> line is required as well.