Re: [PATCH]: Matrix and vector math required for linear loop transforms

[Mark Mitchell <mark at codesourcery dot com>]

Daniel Berlin wrote:

> Whee.
> Bootstrapped and regtested on i686-pc-linux-gnu and powerpc-darwin

This is OK, with a couple of minor changes.

> +/* An integer vector, in the math sense. */
>  typedef int *lambda_vector;
> +/* An integer matrix, in the math sense.  */
> +typedef lambda_vector *lambda_matrix;

You need to say more than "in the math sense".  Give a complete 
definition here, or, at the very least, a reference to one.

>  /* Clear out vector VEC1 of length SIZE.  */
Say "Zero" rather than "Clear out".

> +  mat = ggc_alloc (m * sizeof (lambda_vector));

Do these really have to be ggc_alloc'd, or could they just be xmalloc'd 
or obstack_alloc'd?  If these are only going to live during loop 
optimization, we might be able to avoid putting them in GC'd memory.

> +void
> +lambda_matrix_id (lambda_matrix mat, int size)
> +{
> +  int i, j;
> +
> +  for (i = 0; i < size; i++)
> +    for (j = 0; j < size; j++)
> +      mat[i][j] = (i == j) ? 1 : 0;
> +}

I'd guess that memset, followed by initializing the diagonal to 1, would 
be a little quicker.  I'm not sure if that matters.

Thanks,

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com