Sysbench CPU is absolutley broken.

[NoNaeAbC]

In

int cpu_execute_event()
{
  unsigned long long c;
  unsigned long long l;
  double t;
  unsigned long long n=0;

  /* So far we're using very simple test prime number tests in 64bit */

  for(c=3; c < max_prime; c++)
  {
    t = sqrt((double)c);
    for(l = 2; l <= t; l++)
      if (c % l == 0)
        break;
    if (l > t )
      n++; 
  }

  return 0;
}

n is not used, therefore everything can be and is optimized away. https://godbolt.org/z/8qbGhe9ba For some reason outside of compiler explorer, GCC doesn't optimize it away:

ALIGN   16
cpu_execute_event:
        mov     edi, dword [rel max_prime]              ; 0001D250 _ 8B. 3D, 00296C0A(rel)
        cmp     edi, 3                                  ; 0001D256 _ 83. FF, 03
        jbe     ?_01481                                 ; 0001D259 _ 76, 75
        vxorps  xmm1, xmm1, xmm1                        ; 0001D25B _ C5 F0: 57. C9
        mov     esi, 3                                  ; 0001D25F _ BE, 00000003
        vmovsd  xmm2, qword [rel ?_12157]               ; 0001D264 _ C5 FB: 10. 15, 0006BB04(rel)
        vmovsd  xmm3, qword [rel ?_12158]               ; 0001D26C _ C5 FB: 10. 1D, 0006BB04(rel)
        vcvtsi2sd xmm0, xmm1, rsi                       ; 0001D274 _ C4 E1 F3: 2A. C6
; Filling space: 7H
; Filler type: Multi-byte NOP
;       db 0FH, 1FH, 80H, 00H, 00H, 00H, 00H

ALIGN   8
?_01477:vcomisd xmm0, xmm3                              ; 0001D280 _ C5 F9: 2F. C3
        jc      ?_01480                                 ; 0001D284 _ 72, 2E
        test    sil, 01H                                ; 0001D286 _ 40: F6. C6, 01
        jz      ?_01480                                 ; 0001D28A _ 74, 28
        mov     ecx, 2                                  ; 0001D28C _ B9, 00000002
        jmp     ?_01479                                 ; 0001D291 _ EB, 12

; Filling space: 5H
; Filler type: Multi-byte NOP
;       db 0FH, 1FH, 44H, 00H, 00H

ALIGN   8
?_01478:mov     rax, rsi                                ; 0001D298 _ 48: 89. F0
        xor     edx, edx                                ; 0001D29B _ 31. D2
        div     rcx                                     ; 0001D29D _ 48: F7. F1
        test    rdx, rdx                                ; 0001D2A0 _ 48: 85. D2
        jz      ?_01480                                 ; 0001D2A3 _ 74, 0F
?_01479:inc     rcx                                     ; 0001D2A5 _ 48: FF. C1
        vcvtusi2sd xmm0, xmm1, rcx                      ; 0001D2A8 _ 62 F1 F7 08: 7B. C1
        vcomisd xmm2, xmm0                              ; 0001D2AE _ C5 F9: 2F. D0
        jnc     ?_01478                                 ; 0001D2B2 _ 73, E4
?_01480:inc     rsi                                     ; 0001D2B4 _ 48: FF. C6
        cmp     rsi, rdi                                ; 0001D2B7 _ 48: 39. FE
        jz      ?_01481                                 ; 0001D2BA _ 74, 14
        vcvtsi2sd xmm0, xmm1, rsi                       ; 0001D2BC _ C4 E1 F3: 2A. C6
        vsqrtsd xmm2, xmm0, xmm0                        ; 0001D2C1 _ C5 FB: 51. D0
        jmp     ?_01477                                 ; 0001D2C5 _ EB, B9

; Filling space: 9H
; Filler type: Multi-byte NOP
;       db 66H, 0FH, 1FH, 84H, 00H, 00H, 00H, 00H
;       db 00H

ALIGN   16
?_01481:xor     eax, eax                                ; 0001D2D0 _ 31. C0
        ret                                             ; 0001D2D2 _ C3

The result is Clang:

Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed:
    events per second: 11638783.77

Throughput:
    events/s (eps):                      11638783.7653
    time elapsed:                        10.0001s
    total number of events:              116388756

Latency (ms):
         min:                                    0.00
         avg:                                    0.00
         max:                                    0.13
         95th percentile:                        0.00
         sum:                                 2072.88

Threads fairness:
    events (avg/stddev):           116388756.0000/0.00
    execution time (avg/stddev):   2.0729/0.00

CGG:

Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed:
    events per second:  3999.47

Throughput:
    events/s (eps):                      3999.4704
    time elapsed:                        10.0003s
    total number of events:              39996

Latency (ms):
         min:                                    0.25
         avg:                                    0.25
         max:                                    1.06
         95th percentile:                        0.26
         sum:                                 9996.91

Threads fairness:
    events (avg/stddev):           39996.0000/0.00
    execution time (avg/stddev):   9.9969/0.00

To fix this:

1. Add google benchmark dependency
2. Convert the file from C to C++
3. Add

   benchmark::DoNotOptimize(std::move(n));
   

I have no clue how to use autogen so I'm not goint to do that.

[vaintroub]

FYI, This part of a larger pull request implements it this way, via storing TLS variable. It was noticed, because MSVC compiler also optimized away the function.