Re: [PATCH] Constant fold VIEW_CONVERT_EXPR

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

Roger Sayle wrote:
> The following patch teaches the middle-end how to fold/evaluate
> VIEW_CONVERT_EXPR of INTEGER_CST, REAL_CST, COMPLEX_CST and VECTOR_CST
> at compile-time.  This type of binary reinterpretation can already
> be performed (for some of the above constant types) in the RTL optimizers,
> and this patch effectively duplicates the logic of simplify-rtx.c's
> simplify_immed_subreg in fold, but operating on trees.

Great, thank you very much!

> +
> + /* Subroutine of fold_view_convert_expr.  Interpet the contents of

Typo -- Interpret the contents of...

> + 	return NULL_TREE;

sizeof (tmp) * CHAR_BIT maybe?

> +       for (i = 0; i < bitsize; i += 8)

CHAR_BIT here as well?

> + 	  tmp[ibase / 32] |= (long) (*ptr++) << (i & 31);

What's happening on I32LP64 hosts?  (I just don't know the answer).

> + static tree
> + fold_view_convert_expr (tree type, tree expr)
> + {
> +   /* We support up to 512-bit values (for V8DFmode).  */
> +   enum {
> +     max_bitsize = 512
> +   };
> +   unsigned char buffer[max_bitsize / 8];

And similarly, you could hard code 512 here and later...

> +       if (isize > max_bitsize)

... use isize > sizeof (buffer) / CHAR_BIT.

Paolo