tjbench is a JPEG decompression/compression benchmark part of libjpeg-turbo.
This code is single-threaded and was run pinned to core 1.
Metrics (Intel) - phoronix/tjbenchsh - pid 7457 On_CPU 0.125 On_Core 1.000 IPC 2.557 Retire 0.534 (53.4%) FrontEnd 0.044 (4.4%) Spec 0.236 (23.6%) Backend 0.186 (18.6%) Elapsed 6.18 Procs 16 Maxrss 82K Minflt 22471 Majflt 0 Inblock 0 Oublock 16 Msgsnd 0 Msgrcv 0 Nsignals 0 Nvcsw 129 (67.5%) Nivcsw 62 Utime 6.155447 Stime 0.024053 Start 692342.22 Finish 692348.40
The On_Core is 100%. The IPC is high with a higher than average percent of bad speculation. As seen below, this seems to be branch mis-prediction. The runtime is only 6 seconds.
Metrics (AMD) - phoronix/tjbenchsh - pid 11480 On_CPU 0.063 On_Core 1.001 IPC 2.752 FrontCyc 0.000 (0.0%) BackCyc 0.000 (0.0%) Elapsed 6.10 Procs 16 Maxrss 82K Minflt 22501 Majflt 0 Inblock 0 Oublock 16 Msgsnd 0 Msgrcv 0 Nsignals 0 Nvcsw 129 (17.0%) Nivcsw 629 Utime 6.080309 Stime 0.024571 Start 575740.65 Finish 575746.75
AMD IPC is also similarly high.
Process Tree - phoronix/tjbench
Process Tree
The process tree is simple:
7457) sh
7458) tjbench
7459) tjbench
7460) tjbench
7461) tjbench
7462) tjbench
7463) tjbench
7464) tjbench
7465) sed
7466) tjbench
7467) ls
7468) sed
7469) tjbench
7470) tjbench
7471) tjbench
7472) sed
On_Core is 100%.
IPC is consistently high.
Bad speculation is higher than average.
on_cpu 0.103 elapsed 23.483 utime 18.953 stime 0.341 nvcsw 1041 (95.59%) nivcsw 48 (4.41%) inblock 0 onblock 696 retire 0.647 ms_uops 0.004 speculation 0.115 branch_misses 92.79% machine_clears 7.21% frontend 0.054 idq_uops_delivered_0 0.012 icache_stall 0.002 itlb_misses 0.001 idq_uops_delivered_1 0.016 idq_uops_delivered_2 0.029 idq_uops_delivered_3 0.053 dsb_ops 82.93% backend 0.184 resource_stalls.sb 0.003 stalls_ldm_pending 0.545 l2_refs 0.003 l2_misses 0.001 l2_miss_ratio 27.44% l3_refs 0.000 l3_misses 0.000 l3_miss_ratio 18.53%
Speculation misses are mostly branch prediction related.
Next steps: None