AW: GNAT compiler switches and optimization

AW: GNAT compiler switches and optimization

[Grein, Christoph (Fa. ESG)]

Warner BRUNS claimed:

> A problem with Ada is that the memory layout of a twodimensional array
is
> not specified in the Ada Standard.

That's wrong:
  Ada is row-first order         (1, 1) (1, 2) etc.,
  Fortran specifies column-first (1, 2) (2, 1) etc.

This is what the pragma Convention is for.

Re: AW: GNAT compiler switches and optimization

[Warner BRUNS]

Grein, Christoph (Fa. ESG) wrote:
> Warner BRUNS claimed:
> 
>> A problem with Ada is that the memory layout of a twodimensional array
> is
>> not specified in the Ada Standard.
> 
> That's wrong:
>   Ada is row-first order         (1, 1) (1, 2) etc.,
>   Fortran specifies column-first (1, 2) (2, 1) etc.
> 
> This is what the pragma Convention is for.
I might be wrong, but my understanding is that the memory layout is not specified
in the standard. Can you point me to the section of the standard which specifies
the memory layout of a multidimensional array?
To me it seems as if you can only count on a specific memory layout if you
specify it to be the layout of a non-Ada languag, eg Fortran.
Strange indeed for a standard.

Warner

Re: AW: GNAT compiler switches and optimization

[tmoran]

>I might be wrong, but my understanding is that the memory layout is not specified
>in the standard. Can you point me to the section of the standard which specifies
>the memory layout of a multidimensional array?
  ARM 3.6.2(11) "An implementation should normally represent a
multidimensional array in row-major order, consistent with the notation
used for multidimensional array aggregates (see 4.3.3).  However, if a
Pragma Convention(Fortran, ...) applies to a multidimensional array type,
then column-major order should be used instead ..."
  It would be strange to include powerful rep spec capabilities to define
everything about layouts, but not how an array is stored.

Re: AW: GNAT compiler switches and optimization

[Robert A Duff]

tmoran@acm.org writes:

>   ARM 3.6.2(11) "An implementation should normally represent a
> multidimensional array in row-major order, ...

Ah, I had forgotten that.  Thanks.

When the RM says "should", you can usually count on it being true in all
serious Ada compilers.  So why doesn't it say "shall"?  Well, it's not
really a formal requirement -- you can't write a test case that can
tell for sure which way arrays are stored.  Nonetheless, I think you
can count on this rule being obeyed -- compiler writers don't
deliberately go out of their way to cause trouble for their customers.

I certainly do not think you should start putting Convention(Fortran) on
all your performance-critical arrays, just to make sure!

- Bob

Re: AW: GNAT compiler switches and optimization

[Jeffrey R. Carter]

Robert A Duff wrote:
> 
> I certainly do not think you should start putting Convention(Fortran) on
> all your performance-critical arrays, just to make sure!

In the case of records, the representation is up to the compiler unless 
there is a representation clause. I don't find it surprising that the 
same is true of arrays. However, as with records, there should be some 
way to specify the representation for arrays other than a pragma 
Convention for another language.

-- 
Jeff Carter
"Blessed are they who convert their neighbors'
oxen, for they shall inhibit their girth."
Monty Python's Life of Brian
83