<div dir="ltr"><div>Here is the result of running 20 threads on 20 physical cores (40 virtual cores).</div><div><br></div><div><font face="monospace, monospace">      19002.081139 task-clock (msec)         #    2.147 CPUs utilized            ( +-  2.88% )</font></div><div><font face="monospace, monospace">            23,006 context-switches          #    0.001 M/sec                    ( +-  2.24% )</font></div><div><font face="monospace, monospace">             1,491 cpu-migrations            #    0.078 K/sec                    ( +- 22.50% )</font></div><div><font face="monospace, monospace">         2,607,076 page-faults               #    0.137 M/sec                    ( +-  0.83% )</font></div><div><font face="monospace, monospace">    56,818,049,785 cycles                    #    2.990 GHz                      ( +-  2.54% )</font></div><div><font face="monospace, monospace">    41,072,435,357 stalled-cycles-frontend   #   72.29% frontend cycles idle     ( +-  3.36% )</font></div><div><font face="monospace, monospace">   <not supported> stalled-cycles-backend  </font></div><div><font face="monospace, monospace">    41,090,608,917 instructions              #    0.72  insns per cycle        </font></div><div><font face="monospace, monospace">                                             #    1.00  stalled cycles per insn  ( +-  0.46% )</font></div><div><font face="monospace, monospace">     7,621,825,115 branches                  #  401.105 M/sec                    ( +-  0.52% )</font></div><div><font face="monospace, monospace">       139,383,452 branch-misses             #    1.83% of all branches          ( +-  0.18% )</font></div><div><font face="monospace, monospace"><br></font></div><div><font face="monospace, monospace">       8.848611242 seconds time elapsed                                          ( +-  2.72% )</font></div><div><br></div><div>and this is the single-thread result.</div><div><br></div><div><font face="monospace, monospace">      12738.416627 task-clock (msec)         #    1.000 CPUs utilized            ( +-  5.04% )</font></div><div><font face="monospace, monospace">             1,283 context-switches          #    0.101 K/sec                    ( +-  5.49% )</font></div><div><font face="monospace, monospace">                 3 cpu-migrations            #    0.000 K/sec                    ( +- 55.20% )</font></div><div><font face="monospace, monospace">         2,614,435 page-faults               #    0.205 M/sec                    ( +-  2.52% )</font></div><div><font face="monospace, monospace">    41,732,843,312 cycles                    #    3.276 GHz                      ( +-  5.76% )</font></div><div><font face="monospace, monospace">    26,816,171,736 stalled-cycles-frontend   #   64.26% frontend cycles idle     ( +-  8.48% )</font></div><div><font face="monospace, monospace">   <not supported> stalled-cycles-backend  </font></div><div><font face="monospace, monospace">    39,776,444,917 instructions              #    0.95  insns per cycle        </font></div><div><font face="monospace, monospace">                                             #    0.67  stalled cycles per insn  ( +-  0.84% )</font></div><div><font face="monospace, monospace">     7,288,624,141 branches                  #  572.177 M/sec                    ( +-  1.02% )</font></div><div><font face="monospace, monospace">       135,684,171 branch-misses             #    1.86% of all branches          ( +-  0.12% )</font></div><div><font face="monospace, monospace"><br></font></div><div><font face="monospace, monospace">      12.734335840 seconds time elapsed                                          ( +-  5.03% )</font></div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 16, 2016 at 6:13 PM, Joerg Sonnenberger via llvm-dev <span dir="ltr"><<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">On Wed, Nov 16, 2016 at 05:26:23PM -0800, Rui Ueyama wrote:<br>
> Did you see this<br>
> <a href="http://llvm.org/viewvc/llvm-project?view=revision&revision=287140" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project?view=revision&<wbr>revision=287140</a> ?<br>
> Interpreting these numbers may be tricky because of hyper threading, though.<br>
<br>
</span>Can you try that with a CPU set that explicitly doesn't include the HT<br>
cores? That's more likely to give a reasonable answer for "what is the<br>
thread overhead".<br>
<div class="HOEnZb"><div class="h5"><br>
Joerg<br>
______________________________<wbr>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><br>
</div></div></blockquote></div><br></div>