Actions
Feature #12452
closed
Want support for AMD Zen 2 CPC Events
Start date:
Due date:
% Done:
100%
Estimated time:
Difficulty:
Medium
Tags:
Gerrit CR:
Description
We should update the AMD PMC data files to add support for Zen 2 based on the Rome and Matisse. Similar to the Zen 1 change in 10896, I have constructed the data files myself from the publicly available references.
Related issues
Updated by Robert Mustacchi over 1 year ago
- Related to Feature #10896: Want support for AMD Zen CPC events added
Updated by Robert Mustacchi over 1 year ago
I tested this in a few different ways that I've included below. First, without the change present I ran cpustat -h and verified that the system didn't think there were any CPU Performance counters. Then on one with it cpustat -h reports:
rm@elbereth ~ $ ssh rm@beowulf
The Illumos Project SunOS 5.11 /ws/rm/zen2-pmc Apr. 02, 2020
SunOS Internal Development: rm 2020-Apr-02 [zen2-pmc]
You have new mail.
rm@beowulf:~$ cpustat -h
cpustat: setrlimit failed - Not owner
Usage:
cpustat -c spec [-c spec]... [-p period] [-T u|d]
[-sntD] [interval [count]]
-c spec specify processor events to be monitored
-n suppress titles
-p period cycle through event list periodically
-s run user soaker thread for system-only events
-t include tsc register
-T d|u Display a timestamp in date (d) or unix time_t (u)
-D enable debug mode
-h print extended usage information
Use cputrack(1) to monitor per-process statistics.
CPU performance counter interface: AMD Family 17h
event specification syntax:
[picn=]<eventn>[,attr[n][=<val>]][,[picn=]<eventn>[,attr[n][=<val>]],...]
Generic Events:
event[0-5]: PAPI_br_cn PAPI_br_ins PAPI_tot_cyc PAPI_tot_ins
PAPI_tlb_dm PAPI_tlb_im PAPI_tot_cyc
See generic_events(3CPC) for descriptions of these events
Platform Specific Events:
event[0-5]: FpRetSseAvxOps FpRetSseAvxOps.MacFLOPs
FpRetSseAvxOps.DivFLOPs FpRetSseAvxOps.MultFLOPs
FpRetSseAvxOps.AddSubFLOPs FpRetiredSerOps
FpRetiredSerOps.SseBotRet FpRetiredSerOps.SseCtrlRet
FpRetiredSerOps.X87BotRet FpRetiredSerOps.X87CtrlRet
FpDispFaults FpDispFaults.YmmSpillFault
FpDispFaults.YmmFillFault FpDispFaults.XmmFillFault
FpDispFaults.x87FillFault LsBadStatus2
LsBadStatus2.StliOther LsLocks LsRetClClush LsRetCpuid
LsDispatch LsSmiRx LsIntTaken LsRdTsc LsSTLF
LsStCommitCancel2 LsStCommitCancel2.StCommitCancelWcbFull
LsDcAccesses LsMabAlloc LsMabAlloc.DcPrefetcher
LsMabAlloc.Stores LsMabAlloc.Loads LsRefillsFromSys
LsRefillsFromSys.LS_MABRESP_RMT_DRAM
LsRefillsFromSys.LS_MABRESP_RMT_CACHE
LsRefillsFromSys.LS_MABRESP_LCL_DRAM
LsRefillsFromSys.LS_MABRESP_LCL_CACHE
LsRefillsFromSys.MABRESP_LCL_L2 LsL1DTlbMiss
LsL1DTlbMiss.TlbReload1GL2Miss
LsL1DTlbMiss.TlbReload2ML2Miss
LsL1DTlbMiss.TlbReloadCoalescedPageMiss
LsL1DTlbMiss.TlbReload4KL2Miss
LsL1DTlbMiss.TlbReload1GL2Hit
LsL1DTlbMiss.TlbReload2ML2Hit
LsL1DTlbMiss.TlbReloadCoalescedPageHit
LsL1DTlbMiss.TlbReload4KL2Hit LsMisalAccesses
LsPrefInstrDisp LsPrefInstrDisp.PrefetchNTA
LsPrefInstrDisp.PrefetchW LsPrefInstrDisp.Prefetch
LsInefSwPref LsInefSwPref.MabMchCnt
LsInefSwPref.DataPipeSwPfDcHit LsSwPfDcFills
LsSwPfDcFills.LS_MABRESP_RMT_DRAM
LsSwPfDcFills.LS_MABRESP_RMT_CACHE
LsSwPfDcFills.LS_MABRESP_LCL_DRAM
LsSwPfDcFills.LS_MABRESP_LCL_CACHE
LsSwPfDcFills.MABRESP_LCL_L2 LsHwPfDcFills
LsHwPfDcFills.LS_MABRESP_RMT_DRAM
LsHwPfDcFills.LS_MABRESP_RMT_CACHE
LsHwPfDcFills.LS_MABRESP_LCL_DRAM
LsHwPfDcFills.LS_MABRESP_LCL_CACHE
LsHwPfDcFills.MABRESP_LCL_L2 LsNotHaltedCyc LsTlbFlush
IcCacheFillL2 IcCacheFillSys BpL1TlbMissL2TlbHit
BpL1TlbMissL2TlbMiss BpL1TlbMissL2TlbMiss.IF1G
BpL1TlbMissL2TlbMiss.IF2M BpL1TlbMissL2TlbMiss.IF4K
BpL1BTBCorrect BpL2BTBCorrect BpDynIndPred BpDeReDirect
BpL1TlbFetchHit BpL1TlbFetchHit.IF1G BpL1TlbFetchHit.IF2M
BpL1TlbFetchHit.IF4K DeDisUopQueueEmptyDi0
DeDisUopsFromDecoder
DeDisUopsFromDecoder.OpCacheDispatched
DeDisUopsFromDecoder.DecoderDispatched
DeDisDispatchTokenStalls1
DeDisDispatchTokenStalls1.FPMiscRsrcStall
DeDisDispatchTokenStalls1.FPSchRsrcStall
DeDisDispatchTokenStalls1.FpRegFileRsrcStall
DeDisDispatchTokenStalls1.TakenBrnchBufferRsrc
DeDisDispatchTokenStalls1.IntSchedulerMiscRsrcStall
DeDisDispatchTokenStalls1.StoreQueueRsrcStall
DeDisDispatchTokenStalls1.LoadQueueRsrcStall
DeDisDispatchTokenStalls1.IntPhyRegFileRsrcStall
DeDisDispatchTokenStalls0
DeDisDispatchTokenStalls0.ScAguDispatchStall
DeDisDispatchTokenStalls0.RetireTokenStall
DeDisDispatchTokenStalls0.AGSQTokenStall
DeDisDispatchTokenStalls0.ALUTokenStall
DeDisDispatchTokenStalls0.ALSQ3_0_TokenStall
DeDisDispatchTokenStalls0.ALSQ2RsrcStall
DeDisDispatchTokenStalls0.ALSQ1RsrcStall ExRetInstr
ExRetCops ExRetBrn ExRetBrnMisp ExRetBrnTkn
ExRetBrnTknMisp ExRetBrnFar ExRetNearRet
ExRetNearRetMispred ExRetBrnIndMisp ExRetMmxFpInstr
ExRetMmxFpInstr.SseInstr ExRetMmxFpInstr.MmxInstr
ExRetMmxFpInstr.X87Instr ExRetCond ExDivBusy ExDivCount
ExTaggedIbsOps ExTaggedIbsOps.IbsCountRollover
ExTaggedIbsOps.IbsTaggedOpsRet ExTaggedIbsOps.IbsTaggedOps
ExRetFusBrnchInst L2RequestG1 L2RequestG1.RdBlkL
L2RequestG1.RdBlkX L2RequestG1.LsRdBlkC_S
L2RequestG1.CacheableIcRead L2RequestG1.ChangeToX
L2RequestG1.PrefetchL2Cmd L2RequestG1.L2HwPf
L2RequestG1.Group2 L2RequestG2 L2RequestG2.Group1
L2RequestG2.LsRdSized L2RequestG2.LsRdSizedNC
L2RequestG2.IcRdSized L2RequestG2.IcRdSizedNC
L2RequestG2.SmcInval L2RequestG2.BusLocksOriginator
L2RequestG2.BusLocksResponses L2CacheReqStat
L2CacheReqStat.LsRdBlkCS L2CacheReqStat.LsRdBlkLHitX
L2CacheReqStat.LsRdBlkLHitS L2CacheReqStat.LsRdBlkX
L2CacheReqStat.LsRdBlkC L2CacheReqStat.IcFillHitX
L2CacheReqStat.IcFillHitS L2CacheReqStat.IcFillMiss
L2PfHitL2 L2PfMissL2HitL2 L2PfMissL2L3
attributes: edge pc inv cmask umask nouser sys
See "Preliminary Processor Programming Reference (PPR) for AMD
Family 17h Model 31h, Revision B0 Processors" (AMD publication
55803), "Processor Programming Reference (PPR) for AMD Family 17h
Model 71h, Revision B0 Processors" (AMD publication 56176), and
amd_f17h_zen2_events(3CPC)
This matches what we expect. I then tested one of the PAPI counters -- total cycles. To do this I bound a shell to cpu 0. I started up cputrack as I'll include below. A few seconds in, I lauched a bash while :; do :; done loop and watched as the cycle count went to the turbo boost limit of the processor (3.2 GHz):
rm@beowulf:~$ pfexec cpustat -c PAPI_tot_cyc 1 30 | grep ' 0 ' 1.000 0 tick 10413 2.000 0 tick 73507 3.000 0 tick 36979 4.000 0 tick 140720 5.000 0 tick 125885 6.000 0 tick 106126 7.000 0 tick 125494 8.000 0 tick 140529 9.000 0 tick 1636613247 10.000 0 tick 3177561332 11.000 0 tick 3177848817 12.000 0 tick 3178011789 13.000 0 tick 3178239888 14.000 0 tick 3178218914 15.000 0 tick 3178334000 16.000 0 tick 3177945700 17.000 0 tick 3178337902 18.000 0 tick 3178005176 19.000 0 tick 3178241498 20.000 0 tick 3178326925 21.000 0 tick 3178410671 22.000 0 tick 1479389657 23.000 0 tick 16441 24.000 0 tick 13102 25.000 0 tick 17766 26.000 0 tick 17350
Updated by Electric Monk over 1 year ago
- Status changed from New to Closed
git commit 31aa620247ae407b2bee2dccd71693d1938f54d6
commit 31aa620247ae407b2bee2dccd71693d1938f54d6
Author: Robert Mustacchi <rm@fingolfin.org>
Date: 2020-04-08T00:58:56.000Z
12452 Want support for AMD Zen 2 CPC Events
Reviewed by: Patrick Mooney <pmooney@pfmooney.com>
Approved by: Gordon Ross <gordon.w.ross@gmail.com>
Actions