Metrics (Intel) - phoronix/povrayThis is a test of POV-Ray, the Persistence of Vision Raytracer. POV-Ray is used to create 3D graphics using ray-tracing.
sh - pid 4433 On_CPU 0.980 On_Core 7.841 IPC 1.299 Retire 0.543 (54.3%) FrontEnd 0.229 (22.9%) Spec 0.155 (15.5%) Backend 0.072 (7.2%) Elapsed 134.57 Procs 30 Minflt 10840 Majflt 0 Utime 1054.79 (100.0%) Stime 0.33 (0.0%) Start 67208.77 Finish 67343.34
This benchmark runs for approximately one minute. It has a relatively high proportion of both front-end and speculation times. Both are areas for further exploration. This is a much larger application than some of the small microbenchmarks, are icache and itlb factors here? Note: As described below I think the speculation is a tool problem as I investigate below.
prompt$ size povray text data bss dec hex filename 3308040 365648 140592 3814280 3a3388 povrayMetrics (AMD) - phoronix/povray
sh - pid 10133 On_CPU 0.945 On_Core 15.115 IPC 1.420 FrontCyc 0.056 (5.6%) BackCyc 0.062 (6.2%) Elapsed 67.23 Procs 46 Minflt 16165 Majflt 0 Utime 1016.11 (100.0%) Stime 0.09 (0.0%) Start 50130.55 Finish 50197.78
Process Tree - phoronix/povray
Process Tree
A total of 30 processes are run as part of the application, some run for very short amounts of time.
4433) sh elapsed=134.57 start=0.00 finish=134.57
4434) povray elapsed=134.57 start=0.00 finish=134.57
4435) povray elapsed=0.00 start=0.00 finish=0.00
4436) povray elapsed=134.56 start=0.00 finish=134.56
4437) povray elapsed=134.55 start=0.01 finish=134.56
4438) povray elapsed=134.55 start=0.01 finish=134.56
4439) povray elapsed=134.50 start=0.01 finish=134.51
4440) povray elapsed=134.05 start=0.26 finish=134.31
4441) povray elapsed=0.30 start=0.26 finish=0.56
4442) povray elapsed=0.00 start=0.57 finish=0.57
4443) povray elapsed=133.73 start=0.58 finish=134.31
4444) povray elapsed=0.00 start=0.58 finish=0.58
4445) povray elapsed=0.00 start=0.58 finish=0.58
4446) povray elapsed=1.30 start=0.63 finish=1.93
4447) povray elapsed=0.19 start=0.63 finish=0.82
4448) povray elapsed=0.00 start=0.63 finish=0.63
4449) povray elapsed=0.00 start=0.63 finish=0.63
4450) povray elapsed=0.00 start=0.63 finish=0.63
4451) povray elapsed=0.00 start=0.63 finish=0.63
4452) povray elapsed=0.00 start=0.63 finish=0.63
4453) povray elapsed=0.00 start=0.63 finish=0.63
4454) povray elapsed=0.02 start=1.93 finish=1.95
4455) povray elapsed=131.48 start=1.98 finish=133.46
4456) povray elapsed=131.83 start=1.98 finish=133.81
4457) povray elapsed=131.42 start=1.98 finish=133.40
4458) povray elapsed=132.22 start=1.98 finish=134.20
4459) povray elapsed=131.37 start=1.98 finish=133.35
4460) povray elapsed=132.05 start=1.98 finish=134.03
4461) povray elapsed=131.89 start=1.98 finish=133.87
4462) povray elapsed=131.43 start=1.98 finish=133.41
About this graph
CPU utilization is close to 100% across all cores.
The IPC is 1.30
About this graph
This shows results across three runs. The speculation numbers in this graph aren’t consistent with metric above, need to investigate further.
Next steps: Investigate integrity of the speculation numbers between the topdown and metrics. Investigate frontend events for this benchmark. There seem to be a few “spikes” of very high numbers in both groups and perf is more consistent with a lower number:
prompt% perf stat -a --topdown ./pbm.sh
Performance counter stats for 'system wide':
retiring bad speculation frontend bound backend bound
S0-C0 2 67.6% 2.0% 22.8% 7.7%
S0-C1 2 67.5% 2.0% 22.5% 8.1%
S0-C2 2 67.5% 2.0% 22.6% 7.9%
S0-C3 2 67.6% 1.9% 22.8% 7.7%
133.380714246 seconds time elapsed