LRB and 32-bit compressed oops

[Aleksey Shipilev]

On 3/26/19 9:14 AM, Roman Kennke wrote:
> And BTW, does that phenomenon also show with current WB?

I think WB does it right:

               [Verified Entry Point]
  9.15%         0x00007f5a24490bd0: mov    %eax,-0x14000(%rsp)
  3.41%         0x00007f5a24490bd7: push   %rbp
  2.82%         0x00007f5a24490bd8: sub    $0x10,%rsp
  6.48%         0x00007f5a24490bdc: mov    -0x8(%rsi),%r10
  2.58%         0x00007f5a24490be0: mov    0xc(%r10),%r11d
  3.28%         0x00007f5a24490be4: testb  $0x1,0x20(%r15)      <--- WB fastpath check
  6.33%  ╭      0x00007f5a24490be9: jne    0x00007f5a24490bff
  0.99%  │      0x00007f5a24490beb: movl   $0x2a,0xc(%r11)      <--- 32-bit coops access
 11.81%  │      0x00007f5a24490bf3: add    $0x10,%rsp
  2.39%  │      0x00007f5a24490bf7: pop    %rbp
  2.65%  │      0x00007f5a24490bf8: test   %eax,0x17c35402(%rip)
  1.06%  │      0x00007f5a24490bfe: retq

-Aleksey