Intel ITANIUM ARCHITECTURE - SOFTWARE DEVELOPERS MANUAL VOLUME 1 REV 2.3 Manual page 569

initiates a new performance_index count, which is reported when the next
PAL_GET_PSTATE procedure call is made. A call to PAL_GET_PSTATE with a type
operand of 1 resets the performance measurement logic.
SCDD: If the logical processor belongs to a software-coordinated dependency domain,
the performance index returned (for either type=0 or 3) corresponds to the target
P-state requested by the most recent successful PAL_SET_PSTATE procedure call. No
weighted average (type=1 or 2) is computed by PAL; calling PAL_GET_PSTATE with
type=1 or 2 on a SCDD logical processor is undefined.
HCDD: If the logical processor belongs to a hardware-coordinated dependency domain,
the performance index returned (type=1 or 2) will be a weighted-average sum of the
performance_index values corresponding to the different P-states that the logical
processor was operating in since performance measurement was last reset. Note that
this return value may not necessarily correspond to the performance index of the target
P-state requested by the most recent PAL_SET_PSTATE procedure call. For example,
let's assume that the previous PAL_GET_PSTATE procedure was called at time t
the processor was operating in state P0. The previous PAL_SET_PSTATE procedure
requested a transition from P0 to P3. The transition happened over a period of time,
such that the logical processor went through states P1 at time t
at time t
was called. The performance_index returned is calculated as:
performance_index =
((time spent in P0 after the previous PAL_GET_PSTATE) * (performance_index for P0)+
(time spent in P1) * (performance_index for P1) +
(time spent in P2) * (performance_index for P2) +
(time spent in P3 up to the current PAL_GET_PSTATE) * (performance_index for P3)) /
(time interval between previous and current PAL_GET_PSTATE) =
Figure 11-11. Computation of performance_index
pf
pf
pf
pf
Volume 2, Part 1: Processor Abstraction Layer
, and was in state P3 at time t
3
   
t – t pf + t – t
1 0 0 2 1
--------------------------------------------------------------------------------------------------------------------------------------------------------
Performance
(P0)
0
(P1)
1
(P2)
2
(P3)
3
t t
0 1
(Previous) GET SET(P3)
when the current PAL_GET_PSTATE procedure
4
    
pf + t – t pf +
1 3 2 2
t – t
4 0
t t
2 3
, P2 at time t
1
  pf 
t – t
4 3 3
Time
t
4
(Current) GET
, when
0
and P3
2
2:321