- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Below is a list of all the event counter I have available on my Ivy Bridge CPU. It contains "L1D" cache event counters, but there are no "L2D" counters- only "L2". I was under the impression the L2 cache is split in to data and instruction caches.
(I have read the descriptions of the below event counters and the "L1D" says "L1 data cache" and the "L2" says "L2 cache", implying it is not specific to the data cache).
What event counters should I use to measure data cache misses for L1/L2/L3, ignoring instruction cache misses?
ARITH.FPU_DIV_ACTIVE |
BACLEARS.ANY |
BR_MISP_RETIRED.ALL_BRANCHES_PS |
CPU_CLK_UNHALTED.REF_TSC |
CPU_CLK_UNHALTED.THREAD |
CYCLE_ACTIVITY.CYCLES_NO_EXECUTE |
CYCLE_ACTIVITY.STALLS_L1D_PENDING |
CYCLE_ACTIVITY.STALLS_L2_PENDING |
CYCLE_ACTIVITY.STALLS_LDM_PENDING |
DSB2MITE_SWITCHES.PENALTY_CYCLES |
DTLB_LOAD_MISSES.STLB_HIT |
DTLB_LOAD_MISSES.WALK_DURATION |
DTLB_STORE_MISSES.STLB_HIT |
DTLB_STORE_MISSES.WALK_DURATION |
ICACHE.IFETCH_STALL |
ICACHE.MISSES |
IDQ.ALL_DSB_CYCLES_4_UOPS |
IDQ.ALL_DSB_CYCLES_ANY_UOPS |
IDQ.ALL_MITE_CYCLES_4_UOPS |
IDQ.ALL_MITE_CYCLES_ANY_UOPS |
IDQ.MS_CYCLES |
IDQ.MS_UOPS |
IDQ_UOPS_NOT_DELIVERED.CORE |
IDQ_UOPS_NOT_DELIVERED.CYCLES_0_UOPS_DELIV.CORE |
ILD_STALL.LCP |
INST_RETIRED.ANY |
INST_RETIRED.PREC_DIST |
INT_MISC.RECOVERY_CYCLES |
ITLB_MISSES.WALK_DURATION |
L1D.REPLACEMENT |
L1D_PEND_MISS.PENDING |
L2_LINES_IN.ALL |
LD_BLOCKS.NO_SR |
LD_BLOCKS.STORE_FORWARD |
LD_BLOCKS_PARTIAL.ADDRESS_ALIAS |
MACHINE_CLEARS.COUNT |
MACHINE_CLEARS.MASKMOV |
MACHINE_CLEARS.MEMORY_ORDERING |
MACHINE_CLEARS.SMC |
MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_HIT_PS |
MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_HITM_PS |
MEM_LOAD_UOPS_LLC_MISS_RETIRED.LOCAL_DRAM |
MEM_LOAD_UOPS_RETIRED.L1_MISS |
MEM_LOAD_UOPS_RETIRED.LLC_HIT_PS |
MEM_LOAD_UOPS_RETIRED.LLC_MISS_PS |
MEM_UOPS_RETIRED.ALL_STORES_PS |
MEM_UOPS_RETIRED.SPLIT_LOADS_PS |
MEM_UOPS_RETIRED.SPLIT_STORES_PS |
OFFCORE_RESPONSE.DEMAND_DATA_RD.LLC_MISS.ANY_DRAM_0 |
OFFCORE_RESPONSE.DEMAND_RFO.LLC_HIT.HITM_OTHER_CORE_0 |
RESOURCE_STALLS.SB |
RS_EVENTS.EMPTY_CYCLES |
RS_EVENTS.EMPTY_END |
UOPS_EXECUTED.CYCLES_GE_1_UOP_EXEC |
UOPS_EXECUTED.CYCLES_GE_2_UOPS_EXEC |
UOPS_EXECUTED.CYCLES_GE_3_UOPS_EXEC |
UOPS_ISSUED.ANY |
UOPS_RETIRED.RETIRE_SLOTS |
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
T C wrote:
Below is a list of all the event counter I have available on my Ivy Bridge CPU. It contains "L1D" cache event counters, but there are no "L2D" counters- only "L2". I was under the impression the L2 cache is split in to data and instruction caches.
(I have read the descriptions of the below event counters and the "L1D" says "L1 data cache" and the "L2" says "L2 cache", implying it is not specific to the data cache).
What event counters should I use to measure data cache misses for L1/L2/L3, ignoring instruction cache misses?
Have attached the list of hardware event counters in this message.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
T C:
please check out this thread: http://software.intel.com/en-us/forums/topic/506796
yes, L2 cache is "unified" wrt the CPU. Are you doing any GPU code excution?
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page