Re: [PATCH] Superscalar tweaks to synth_mult

[Richard Henderson <rth at redhat dot com>]

On Mon, Aug 30, 2004 at 02:09:21PM -0600, Roger Sayle wrote:
> >> alg_add_t_m2         total := total + multiplicand * coeff;
> >> alg_sub_t_m2         total := total - multiplicand * coeff;
...
> So for almost all algorithm steps, the critical path length
> increases by the same amount as the total rtx_cost.  The
> two exceptions, however, are the steps "alg_add_t_m2" and
> "alg_sub_t_m2", in which the evaluation of the multiplication
> can theoretically be overlapped with the evaluation of the
> sub total.

Ah, you actually mean

	total(i) := total(i-1) + multiplicand * coeff

with overlapping * with total(i-1).  Sure, ok.

> Hopefully, this explains/rationalizes things.  I'm happy to add
> some more comments to the code, if you think that'll help clarify
> further.  Having both this and my previous post searchable in the
> gcc-patches archive should certainly help future generations.

Some more commentary would be nice.


r~