[vsipl++] [patch] Fix stride for CBE float vmul

[Jules Bergmann]

Don McCoy wrote:
> Jules Bergmann wrote:
>> Stride is passed down in elements, not bytes!
>>
>>
> This looks good, but I had a couple of minor questions/comments.

Don,

Thanks for looking over this!


>>    Rand<T> gen(0, 0);
>> -  A = gen.randu(size);
>> -  B = gen.randu(size);
>> +//  A = gen.randu(size);
>> +//  B = gen.randu(size);
>>   
> Did you mean to leave this commented out?

Good catch!  No, I didn't mean to leave this in, but I made a second 
silent checkin afterwards that fixed it.

> 
>>  
>> @@ -61,7 +61,7 @@
>>        std::cout << "A(i) * B(i) = " << A(i) * B(i) << std::endl;
>>      }
>>  #endif
>> -    test_assert(almost_equal(Z(i), A(i) * B(i)));
>> +    test_assert(almost_equal(Z.get(i), A(i) * B(i)));
>>   
> For consistency, should we use .get for all accesses (or drop it)?  I 
> guess for tests we ought to always use it, in my opinion.

Using get() is probably best.

Here it is a correctness issue.  If A is split complex, A(i) returns an 
Lvalue_proxy, and no operator*() can be found to handle it.  Using 
A.get(i) returns a complex, and all is fine.

				-- Jules


-- 
Jules Bergmann
CodeSourcery
jules at codesourcery.com
(650) 331-3385 x705