[bgl-discuss] PAPI events

["Alam, Sadaf R." <alamsr@xxxxxxxx>]

Hi

I ran the avail program by PAPI and it shows that the following events
are available on the system:

PAPI_L3_TCM     0x80000008      Yes     No      Level 3 cache misses
(BGL_UPC_L3_CACHE_MISS_DATA_WILL_BE_REQED_DDR)

PAPI_L3_LDM     0x8000000e      Yes     Yes     Level 3 load misses
(BGL_UPC_L3_MSHNDLR_TOOK_REQ_RDQ0, BGL_UPC_L3_MSHNDLR_TOOK_REQ_RDQ1)

PAPI_L3_STM     0x8000000f      Yes     No      Level 3 store misses
(BGL_UPC_L3_MSHNDLR_TOOK_REQ_WRBUF)

PAPI_FMA_INS    0x80000030      Yes     No      FMA instructions
completed
(BGL_FPU_ARITH_TRINARY_OP)

PAPI_TOT_CYC    0x8000003c      Yes     No      Total cycles (Timebase
register (null))

PAPI_L2_DCH     0x80000041      Yes     Yes     Level 2 data cache hits
(BGL_UPC_PU0_PREF_STREAM_HIT, BGL_UPC_PU1_PREF_STREAM_HIT)

PAPI_L2_DCA     0x80000043      Yes     Yes     Level 2 data cache
accesses (BGL_UPC_PU0_PREF_REQ_VALID, BGL_UPC_PU1_PREF_REQ_VALID)

PAPI_L3_TCH     0x80000059      Yes     No      Level 3 total cache hits
(BGL_UPC_L3_CACHE_HIT)

PAPI_FML_INS    0x80000063      Yes     No      Floating point multiply
instructions (BGL_FPU_ARITH_MULT_DIV)

PAPI_FAD_INS    0x80000064      Yes     No      Floating point add
instructions (BGL_FPU_ARITH_ADD_SUBTRACT)

PAPI_BGL_OED    0x80000068      Yes     No      Floating point Oedipus
operations (BGL_FPU_ARITH_OEDIPUS_OP)

PAPI_BGL_TS_32B 0x80000069      Yes     Yes     32B chunks sent in any
torus link (BGL_UPC_TS_XM_32B_CHUNKS, BGL_UPC_TS_XP_32B_CHUNKS, 
BGL_UPC_TS _YM_32B_CHUNKS, BGL_UPC_TS_YP_32B_CHUNKS, B...)

PAPI_BGL_TS_FULL        0x8000006a      Yes     Yes     CLOCKx2 cycles
with no torus token (accum) 
(BGL_UPC_TS_XM_LINK_AVAIL_NO_VCD0_VCD_VCBN_TOK
ENS, BGL_UPC_TS_XP_LINK_AVAIL_NO_VCD0_VCD_VCBN_TOKENS, BGL_U...)

PAPI_BGL_TR_DPKT        0x8000006b      Yes     Yes     Data packets
sent
on any tree channel (BGL_UPC_TR_SNDR_2_VC1_DPKTS_SENT, 
BGL_UPC_TR_SNDR_ 2_VC0_DPKTS_SENT, BGL_UPC_TR_SNDR_1_VC1_DPKTS_SENT,
BGL...)

PAPI_BGL_TR_FULL        0x8000006c      Yes     Yes     CLOCKx2 cycles
with tree receiver full (accum) (BGL_UPC_TR_RCV_0_VC0_FULL, 
BGL_UPC_TR_RCV _0_VC1_FULL, BGL_UPC_TR_RCV_1_VC0_FULL, 
BGL_UPC_TR_RCV_1_VC1_FUL...)

I am wondering if it is possible to calculate the number of load and
store operations and PAPI_FP_INS using a HW counter utility/library.
Does libhpm provide these counters?

Thanks

Sadaf Alam

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.