rwessel (robertwessel@yahoo.com) on 7/23/10 wrote:
---------------------------
>? (0xe2.0x9a.0x9b@gmail.com) on 7/22/10 wrote:
>>[cut]
>
>Well, PPC already does that, and it does help code in some cases, but it doesn't
>seem like a major win, and no one is really rushing to copy that. Of course the
>RISC approach of not having a flags register at all, is another solution.
>
>Of more general use than a duplicate flags register would be more instructions
>that don't set the flags. And within the existing scope of x86, you can do that
>yourself (by replacing the add with a mov/lea sequence), or minimize the effect
>by unrolling. Of course those techniques might not apply to any particular case,
>but they do to this one, which goes to show that short code snippets are not worth
>too much when it comes to making architectural decisions.

(Sorry about the empty post, I accidentally pushed the "Post" button at the wrong moment.)

Well, I think what we should we mention here is that speculative execution of branches works only if the branch can be predicted with high probability (let's say, more than 95% correct predictions). However, if this is the case, then it also might be possible to make static assertions about the branch in question: you or the compiler will be able to speculate at compile-time or after you profile the code that the branch will be taken with 95% probability.

Now imagine that you have a branch instruction about which you can are predicting it will be 95% of the time. Does this affect your decisions when writing the code, and does it affect the decisions the compiler is making when generating code? It does.

If you have the code

likely_condition = (...)

if(likely_condition) {

    statement1;

    statement2;

    statement3;

    ...

}

you or the compiler can transform it into

likely_condition = (...)

speculative_execution(statement1)

if(likely_condition) {

    statement2;

    statement3;

    ...

}

else {

    undo(statement1)

}

So, you yourself can do speculative execution also, not just the CPU.

Notice that if you do not have multiple FLAGS registers (or some similar mechanism, like the ones you mentioned), you probably have to execute "speculative_execution(statement1)" *before* doing "likely_condition = (...)" - which is not only harder to do but also suboptimal in respect to pipelining.