Description

200010036 – Om Patil
200010041 – Pranav Talegaonkar
Part A
The instruction mix of some benchmarks from the SPEC CPU2017 suite are as follows,
Total
Instructions Load (%) Store (%) Unconditional
Jump (%) Conditional Branch (%)
gcc 66461 6.9379 54.4831 0.371647 10.6363
Ibm 100103459318 29.4346 17.392 0.244648 1.07347
mcf 26714123511 34.4908 10.0128 1.86763 19.0455
namd 110783343 23.711 16.3114 3.7725 18.5401
xalancbmk 273980614900 32.8253 17.2457 2.21862 14.1084
These values were obtained using the Intel Pin Instrumentation Tool.
Part B
We have run the performance analysis of come benchmarks from the SPEC CPU2017 suite on the 11th Gen Intel i5-1135G7 processor. This processor is based on the Tiger Lake microarchitecture. Some details of the processor are as follows,
● Manufactured using Intel’s 10 nm SuperFin Process
● 4 Cores
● L1I Cache is 32 KiB/Core (8-way set associative)
● L1D Cache is 48 KiB/core (12-way set associative)
● L2 Cache is 1280 KiB/core (20-way set associative)
● L3 Cache 3 MiB/core (12-way set associative)
● 14-19 Pipeline Stages
The stats displayed below are the averages over 10 repetitions.

Observations:
● Increase in Branch Accuracy corresponds to increase in IPC. This makes sense as there is very few control penalties.
● There are certain points in the program where we notice very high memory activity. These spikes can be attributed to memory intensive work. The gcc benchmark repeatedly has such memory intensive segments. to intensive.

Observations:
● Increase in Branch Accuracy corresponds to increase in IPC. This makes sense as there is very few control penalties.
● There are certain points in the program where we notice very high memory activity. These spikes can be attributed to memory intensive work. The lbm benchmark has relatively fewer such memory intensive

Observations:
● Increase in Branch Accuracy corresponds to increase in IPC. This makes sense as there is very few control penalties.
● There are certain points in the program where we notice very high memory activity. These spikes can be attributed to memory intensive work. The mcfbenchmark has a moderate number of such memory intensive with one lengthy patch.

Observations:
● Increase in Branch Accuracy corresponds to increase in IPC. This makes sense as there is very few control penalties.
● There are certain points in the program where we notice very high memory activity. These spikes can be attributed to memory intensive work. The namd benchmark has relatively frequent such memory intensive patches.

Observations:
● Increase in Branch Accuracy corresponds to increase in IPC. This makes sense as there is very few control penalties.
● There are certain points in the program where we notice very high memory activity. These spikes can be attributed to memory intensive work. The xalancbmk benchmark has periodic short duration memory intensive patches.
Observations:
● Branch Accuracy and L1 Hit Rate together influence the IPC, higher they are higher the IPC. This can clearly be observed, benchmarks such as mcf which has low average branch accuracy and low average L1 Hit Rate has a low average IPC. Whereas xalancbmk which has high average branch accuracy and high average L1 Hit Rate has a high average IPC.
Average IPC Average
Branch
Accuracy (%) Branches per Second Average L1
Hit Rate
(%) Average
Power
(J)
gcc 1.4402847571
189268 97.450921273
03184 2,28,15,039 95.208174204
35503 0.1547403685
0921286
lbm 1.0507958921
694462 97.093421052
63157 38,91,472 79.632509627
72791 0.1429268292
6829306
mcf 0.9552622061
482827 95.439620253
16459 1,89,36,663 84.066871609
40324 0.1553164556
9620248
namd 2.3734111675
12689 96.010680203
04572 69,24,441 97.533461928
93405 0.1567817258
883247
xalancbmk 1.5651742993
848279 99.406032125
76909 3,32,56,573 97.847481203
00735 0.1653861927
5461316