Mark Haywood wrote: > Bill Holler wrote: >> Mark Haywood wrote: >>> Eric Saxe wrote: >>> >>>> Mark Haywood wrote: >>>> >>>>> Li, Aubrey wrote: >>>>> >>>>>> I guess Eric means P-state and C-state. >>>>>> If so, _PSD describes P-State Dependency and _CSD describes >>>>>> C-state Dependency. >>>>>> >>>>> Ah. Sorry. I'm familiar with the _CSD (and of course the _PSD), >>>>> but didn't recognize what Eric was referring to. So, I assume that >>>>> means _PSD defines the CPUs that share frequency change and >>>>> C-state dependency defines the CPUs that share a voltage change? >>>> Right, this is what I'm wondering as well (and I didn't know that >>>> _CSD and _PSD described those two domains...is that really so?). >>>> We can have the dispatcher consider thread placement across both >>>> levels if that makes sense. >>>> >>> >>> The _PSD domains define the CPUs that share P-state dependencies. >>> P-states are really a combination of frequency and voltage scaling. >>> >>> The _CSD domains define the CPUs that share C-state dependencies. Do >>> C-states really just equate to voltage scaling? I don't know. >>> Aubrey or Bill might know. >>> >> >> Are _CSD/c-state domains "flat", or can these be in a tree? >> For example if there is a 4-core chip with two shared caches? >> >> I am having a hard time envisioning how the PAD will place >> threads across both _PSD and _CSD defined domains. >> Would the PAD prefer high p-state domains, and then >> lower p-state domains, and then lower c-state domains last? > > My guess is that the _CSD and _PSD domains will be identical. I think > Aubrey will have a better idea.
>From what I understanding on Nehalem: As for _PSD domain, all logical processors in a specific processor package must be the same. A system with one physical package will have one _PSD domain. the coordination type could be SW_ALL, SW_ANY or HW_ALL. As for _CSD domain, all logical processors in a specific processor core must be the same. A system with one physical package will have four _CSD domains. And on Nehalem processors, the coordination type must be HW_ALL. Thanks, -Aubrey
