<div dir="ltr">The compile time increase look small and acceptable given what it does and performance we saw.<div><br></div><div>On the other hand, I assume you will also be on the hook to solve the compile time problem longer term (i.e., helping newgvn deployment) :)</div><div><br></div><div>David</div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, May 9, 2017 at 3:59 PM, Wei Mi via Phabricator <span dir="ltr"><<a href="mailto:reviews@reviews.llvm.org" target="_blank">reviews@reviews.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">wmi updated this revision to Diff 98361.<br>
wmi added a comment.<br>
<br>
I change the DenseMap numberingExpression (mapping from value numbering to GVN::Expression) to std::vector. Although there is still some minor increase but better than before.<br>
<br>
400.perlbench   0.81%<br>
401.bzip2       -0.11%<br>
403.gcc -0.42%<br>
429.mcf 1.58%<br>
433.milc        -0.33%<br>
445.gobmk       0.85%<br>
456.hmmer       0.07%<br>
458.sjeng       -0.42%<br>
462.libquantum  -0.75%<br>
464.h264ref     -1.24%<br>
470.lbm -0.65%<br>
482.sphinx3     0.44%<br>
998.specrand    0.00%<br>
999.specrand    -1.74%<br>
444.namd        0.28%<br>
447.dealII      1.56%<br>
450.soplex      0.96%<br>
453.povray      -0.02%<br>
471.omnetpp     0.02%<br>
473.astar       0.18%<br>
483.xalancbmk   0.85%<br>
<div class="HOEnZb"><div class="h5"><br>
<br>
Repository:<br>
  rL LLVM<br>
<br>
<a href="https://reviews.llvm.org/D32252" rel="noreferrer" target="_blank">https://reviews.llvm.org/<wbr>D32252</a><br>
<br>
Files:<br>
  include/llvm/Transforms/<wbr>Scalar/GVN.h<br>
  lib/Transforms/Scalar/GVN.cpp<br>
  test/Transforms/GVN/PRE/phi-<wbr>translate-2.ll<br>
  test/Transforms/GVN/PRE/pre-<wbr>gep-load.ll<br>
  test/Transforms/GVN/PRE/pre-<wbr>load.ll<br>
<br>
</div></div></blockquote></div><br></div>