what FPU pipe does VBROADCASTSD go down?

[perfwise]

What pipe does VBROADCASTSD go down? I looked in the latency/throughput section of the intel optguide and it doesn't list this information. Thanks

perfwise

[perfwise]

I haven't heard which pipe it goes down but from my assembly performance it surely isn't pipe 0 or 1 where the 256-bit add and mul units are. Thought I'd let people know..

perfwise

[Bernard]

>>>I haven't heard which pipe it goes down but from my assembly performance it surely isn't pipe 0 or 1>>>

Do you mean execution unit's Port?

[perfwise]

Yes, but it's a moot point now.  I was tuning my dgemm for SB and IB and noticed some "replication" instructions utilized the same pipe as the + or *, can't remember off the top of my head.  vbroadcastsd doesn't and is preferrable for this purpose.  

perfwise

[Bernard]

I suppose that store/load Ports 2 and 3 are executing VBROADCASTSD instruction.