Poindexter (cherullo.delete@this.gmail.com) on November 2, 2015 5:39 am wrote:
> juanrga (nospam.delete@this.juanrga.com) on November 2, 2015 4:47 am wrote:
> > Poindexter (cherullo.delete@this.gmail.com) on November 1, 2015 9:25 am wrote:
> > > It may reduce the number of required uops, and thus change
> > > the optimal (whatever that means) pipelines ratio.
> >
> > Macro-op fusion doesn't reduce the required number of uops.
> > Macro-op fusion joins uops for simplifying the SS/OOOE
> > logic. E.g., instead a separate entry in the ROB for each uop, you have one entry for the macro-op.
> >
> > Once the macro-op is scheduled to the corresponding place, it
> > is broken into uops and the execution units execute each uop.
> >
> > The number of uops is not reduced.
>
> You're mixing up macro-op fusion with micro-op fusion. Go read:
> http://www.agner.org/optimize/microarchitecture.pdf
>

I am not. I know terminology is ambiguous and that macro-op for Intel doesn't mean the same than macro-op for AMD. Moreover I treat macro-op fusion as different from macro-ops fusion, whereas other don't.

But you cannot be confused by terminology because I explained exactly what I mean. Moreover my original question to you was clear as well:

> > > I would like to know how you think they affect the discussion.
> > > E.g., how read-modify or read-modify-write fusion reduce the number
> > > of loads and stores?