Description - phoronix/asmfish

This is a test of asmFish, an advanced chess benchmark written in Assembly.

Metrics (Intel) - phoronix/asmfish

sh - pid 32504
	On_CPU   0.998
	On_Core  7.984
	IPC      0.962
	Retire   0.467	(46.7%)
	FrontEnd 0.290	(29.0%)
	Spec     0.170	(17.0%)
	Backend  0.074	(7.4%)
	Elapsed  220.05
	Procs    12
	Minflt   273834
	Majflt   0
	Utime    1756.59 	(100.0%)
	Stime    0.35    	(0.0%)
	Start    243559.28
	Finish   243779.33

The benchmark runs for almost four minutes. During that time it is scheduled on all CPUs 100% of the time.

Metrics (AMD) - phoronix/asmfish

sh - pid 12237
	On_CPU   0.997
	On_Core  15.952
	IPC      0.935
	FrontCyc 0.089	(8.9%)
	BackCyc  0.083	(8.3%)
	Elapsed  213.43
	Procs    20
	Minflt   279864
	Majflt   0
	Utime    3404.30 	(100.0%)
	Stime    0.29    	(0.0%)
	Start    236054.23
	Finish   236267.66

Process Tree - phoronix/asmfish
Process Tree

   32504) sh elapsed=220.05 start=0.00 finish=220.05
      32505) asmfish elapsed=220.05 start=0.00 finish=220.05
        32506) asmfish elapsed=0.00 start=0.00 finish=0.00
        32507) asmFishL_2018-0 elapsed=220.05 start=0.00 finish=220.05
        32508) asmFishL_2018-0 elapsed=219.96 start=0.05 finish=220.01
        32509) asmFishL_2018-0 elapsed=219.96 start=0.05 finish=220.01
        32510) asmFishL_2018-0 elapsed=219.96 start=0.05 finish=220.01
        32511) asmFishL_2018-0 elapsed=219.96 start=0.05 finish=220.01
        32512) asmFishL_2018-0 elapsed=219.96 start=0.05 finish=220.01
        32513) asmFishL_2018-0 elapsed=219.95 start=0.06 finish=220.01
        32514) asmFishL_2018-0 elapsed=219.95 start=0.06 finish=220.01
        32515) asmFishL_2018-0 elapsed=219.95 start=0.06 finish=220.01

The process tree is relatively simple and symmetric for all cores.

^{About this graph}
CPU utilization maxed out for all cores.

The IPC is just slightly less than one.

^{About this graph}
Reasonably high percentages of time speculating slots that are never retired as well a moderately high delays in the front end contribute to the lower IPC.

Next steps: Deeper investigation into causes for speculation e.g. branch misses and front end e.g. icache/itlb or other reasons.