background image

19-154 Vol. 3B

PERFORMANCE-MONITORING EVENTS

19.13  PERFORMANCE MONITORING EVENTS FOR 45 NM AND 32 NM 

INTEL

®

ATOM

 PROCESSORS

45 nm and 32 nm processors based on the Intel

®

 Atom™ microarchitecture support the architectural performance-

monitoring events listed in Table 19-1 and fixed-function performance events using fixed counter listed in Table 
19-22
. In addition, they also support the following non-architectural performance-monitoring events listed in Table 
19-26

Table 19-26.  Non-Architectural Performance Events for 45 nm, 32 nm Intel® Atom™ Processors

Event

Num.

Umask

Value

Event Name

Definition

Description and Comment

02

H

81

H

STORe_FORWARDS.GO

OD

Good store forwards.

This event counts the number of times store data was 

forwarded directly to a load.

06

H

00

H

SEGMENT_REG_

LOADS.ANY

Number of segment 

register loads.

This event counts the number of segment register load 

operations. Instructions that load new values into segment 

registers cause a penalty. This event indicates performance 

issues in 16-bit code. If this event occurs frequently, it may be 

useful to calculate the number of instructions retired per 

segment register load. If the resulting calculation is low (on 

average a small number of instructions are executed between 

segment register loads), then the code’s segment register 

usage should be optimized. 
As a result of branch misprediction, this event is speculative and 

may include segment register loads that do not actually occur. 

However, most segment register loads are internally serialized 

and such speculative effects are minimized. 

07

H

01

H

PREFETCH.PREFETCHT

0

Streaming SIMD 

Extensions (SSE) 

PrefetchT0 

instructions executed.

This event counts the number of times the SSE instruction 

prefetchT0 is executed. This instruction prefetches the data to 

the L1 data cache and L2 cache.

07

H

06

H

PREFETCH.SW_L2

Streaming SIMD 

Extensions (SSE) 

PrefetchT1 and 

PrefetchT2 

instructions executed.

This event counts the number of times the SSE instructions 

prefetchT1 and prefetchT2 are executed. These instructions 

prefetch the data to the L2 cache.

07

H

08

H

PREFETCH.PREFETCHN

TA

Streaming SIMD 

Extensions (SSE) 

Prefetch NTA 

instructions executed.

This event counts the number of times the SSE instruction 

prefetchNTA is executed. This instruction prefetches the data 

to the L1 data cache. 

08H

07H

DATA_TLB_MISSES.DT

LB_MISS

Memory accesses that 

missed the DTLB.

This event counts the number of Data Table Lookaside Buffer 

(DTLB) misses. The count includes misses detected as a result 

of speculative accesses. Typically a high count for this event 

indicates that the code accesses a large number of data pages.

08H

05H

DATA_TLB_MISSES.DT

LB_MISS_LD

DTLB misses due to 

load operations.

This event counts the number of Data Table Lookaside Buffer 

(DTLB) misses due to load operations. This count includes 

misses detected as a result of speculative accesses.

08H

09H

DATA_TLB_MISSES.L0

_DTLB_MISS_LD

L0_DTLB misses due to 

load operations.

This event counts the number of L0_DTLB misses due to load 

operations. This count includes misses detected as a result of 

speculative accesses.

08H

06H

DATA_TLB_MISSES.DT

LB_MISS_ST

DTLB misses due to 

store operations.

This event counts the number of Data Table Lookaside Buffer 

(DTLB) misses due to store operations. This count includes 

misses detected as a result of speculative accesses.