Re: auto predicating branches

On Tue, 22 Apr 2025 5:10:10 +0000, Anton Ertl wrote:

mitchalsup@aol.com (MitchAlsup1) writes:

--------------

>
My argument is that this is a SW decision (in the compiler) not a
HW decision (other than providing the PREDs).

>
That's a position, not an argument.  Do you have an argument for your
position?

Architecture defines the primitives, Hardware provides the primitives,
SW decides to use them (or not) on a case-by-case basis.

Since PREDs are not
predicted (unless you think they are predicted BOTH ways) they do
not diminish the performance of the branch predictors.

>
Nor increase it.

When there is an else-clause, the PRED uses 1 fewer instruction:
the branch that exits the then-clause and jumps over the else-
clause does not exist.

                  But it sounds like you think that the compiler
should choose predication when the condition is not particularly
predictable.  How should the compiler know that?

Compiler is setup to use PRED when the then-clause is 8 or fewer
instructions and when the else clause is 8 or fewer instructions,
otherwise it uses branches.
----------

>
- anton