SHORT UPI data traffic

EVENTSET
FIXC0 INSTR_RETIRED_ANY
FIXC1 CPU_CLK_UNHALTED_CORE
FIXC2 CPU_CLK_UNHALTED_REF
SBOX0C0 TXL_FLITS_ALL_DATA
SBOX0C1 RXL_FLITS_ALL_DATA
SBOX1C0 TXL_FLITS_ALL_DATA
SBOX1C1 RXL_FLITS_ALL_DATA
SBOX2C0 TXL_FLITS_ALL_DATA
SBOX2C1 RXL_FLITS_ALL_DATA


METRICS
Runtime (RDTSC) [s] time
Runtime unhalted [s] FIXC1*inverseClock
Clock [MHz]  1.E-06*(FIXC1/FIXC2)/inverseClock
CPI  FIXC1/FIXC0
Received data bandwidth [MByte/s] 1.0E-06*((SBOX0C1+SBOX1C1+SBOX2C1)/9.0)*64.0/time
Received data volume [GByte] 1.0E-09*((SBOX0C1+SBOX1C1+SBOX2C1)/9.0)*64.0
Sent data bandwidth [MByte/s] 1.0E-06*((SBOX0C0+SBOX1C0+SBOX2C0)/9.0)*64.0/time
Sent data volume [GByte] 1.0E-09*((SBOX0C0+SBOX1C0+SBOX2C0)/9.0)*64.0
Total data bandwidth [MByte/s] 1.0E-06*((SBOX0C0+SBOX1C0+SBOX2C0+SBOX0C1+SBOX1C1+SBOX2C1)/9.0)*64.0/time
Total data volume [GByte] 1.0E-09*((SBOX0C0+SBOX1C0+SBOX2C0+SBOX0C1+SBOX1C1+SBOX2C1)/9.0)*64.0


LONG
Formulas:
Received data bandwidth [MByte/s] = 1.0E-06*(SUM(RXL_FLITS_ALL_DATA)/9.0)*64.0/runtime
Received data volume [GByte] = 1.0E-09*(SUM(RXL_FLITS_ALL_DATA)/9.0)*64.0
Sent data bandwidth [MByte/s] = 1.0E-06*(SUM(TXL_FLITS_ALL_DATA)/9.0)*64.0/time
Sent data volume [GByte] = 1.0E-09*(SUM(TXL_FLITS_ALL_DATA)/9.0)*64.0
Total data bandwidth [MByte/s] = 1.0E-06*((SUM(RXL_FLITS_ALL_DATA)+SUM(TXL_FLITS_ALL_DATA))/9.0)*64.0/time
Total data volume [GByte] = 1.0E-09*((SUM(RXL_FLITS_ALL_DATA)+SUM(TXL_FLITS_ALL_DATA))/9.0)*64.0
--
This group measures the data traffic on the UPI (socket interconnect). The group
measures all filled data slots (9 slots per 64 Byte data transfer), that's why
the count needs to be divided by 9. These 9 data chunks are not transferred in
a single flit but there is one flit for the header and three flits for the data.
The metrics show higher values as expected because the events count also
different transfers which include data.