can be used in virtualized environment such as virtualbox or kvm ?

[baiwfg2]

I know PMU is not supported in virtualbox linux. I'm wondering whether such environment is suitable for using uarch-bench

[travisdowns]

@baiwfg2 - yes, you can, probably, but you would need to use a different "timer" which is the thing that times regions of code. Pass --list-timers to read about them and you'd want the "clock" timer:

./uarch-bench --timer=clock ...

This timer just uses standard wall-clock time methods to do the timing, and will still convert that to cycles using a calibration loop. Depending on how often frequency changes are in your environment, this can still produce very accurate results in cycles, e.g., on my system w/o any special adjustment:

$ ./uarch-bench --timer=clock
Welcome to uarch-bench (e4f54d5-dirty)
Supported CPU features: SSE3 PCLMULQDQ VMX SMX EST TM2 SSSE3 FMA CX16 SSE4_1 SSE4_2 MOVBE POPCNT AES AVX RDRND TSC_ADJ SGX BMI1 AVX2 BMI2 ERMS MPX RDSEED ADX CLFLUSHOPT INTEL_PT
Pinned to CPU 0
WARNING: couldn't determine hugepage info (it's OK, you are probably not running as root)
UARCH_BENCH_CLOCK_MHZ not set, running calibration
Median CPU speed: 4.693 GHz
Running benchmarks groups using timer clock

** Running group basic : Basic Benchmarks **
                               Benchmark    Cycles     Nanos
                     Dependent add chain      1.00      0.21
                   Independent add chain      0.25      0.05
                  Dependent imul 64->128      2.94      0.63
                   Dependent imul 64->64      2.94      0.63
                Independent imul 64->128      1.01      0.21
                 Independent imul 64->64      1.00      0.21
                    Same location stores      1.00      0.21
                Disjoint location stores      1.00      0.21
                Dependent push/pop chain      4.41      0.94
              Independent push/pop chain      1.00      0.21
     Back-to-back call of empty function      4.38      0.93
            64-bit dependent div 1/1 = 1     31.47      6.71
           64-bit dependent idiv 1/1 = 1     38.88      8.29
                 same location inc [mem]      5.38      1.15
              same location add [mem], 1      5.38      1.15
                   multiple add [mem], 1      3.92      0.84
                        Calibration subs      0.98      0.21
                    Calibration subs x10      0.98      0.21
                   Calibration subs x10b      0.98      0.21
            Calibration subs w/ register      0.98      0.21
                 Calibration subs w/ nop      0.98      0.21
             Calibration subs value 1000      0.98      0.21
              Calibration subs value 100      0.98      0.21
                Calibration subs value 2      0.98      0.21
                Calibration subs value 1      0.98      0.21
                   1 dependent add chain      0.98      0.21
                   2 dependent add chain      1.01      0.22

[travisdowns]

(I forget if uarch-bench will automatically fall back to "clock" timer if the others don't work)

[baiwfg2]

@travisdowns I see. The reason I mentioned PMU is that I see PMU counters output in this Denis's post. The results(including yours in the comment) output INSTRUCTIONS_RETIRED | UOPS_ISSUED | UOPS_RETIRED | LSD.UOPS. But when I run ./uarch-bench --test-name=dendibakh/unfused-original I only see Benchmark Cycles Nanos. Why is that ?