<div dir="ltr">Hi Henry,<br><div class="gmail_extra"><br><div class="gmail_quote">On Wed, May 20, 2015 at 5:41 AM, Henrik Johansen <span dir="ltr"><<a href="mailto:henrik.s.johansen@veloxit.no" target="_blank">henrik.s.johansen@veloxit.no</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <br><div style="word-wrap:break-word">Could it be related to the change in 1166 to use XCHG between register/mem locations instead of a MOV to a (now unavailable) temp register in genPushRegisterArgsForNumArgs:?<div>According to <a href="http://www.agner.org/optimize/instruction_tables.pdf" target="_blank">http://www.agner.org/optimize/instruction_tables.pdf</a> , XCHG with m/r operands has major penalties on most platforms compared to m/r MOV, which would be a problem if it's used for a large percent of sends...</div></div></blockquote><div><br></div><div>I think you're right, but I'll check. The XCHG is only needed on certain two-argument sends, where we're short a register (I can't remember the details right now). I'll see if I can use XCHG less.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word"><div>Cheers,<br></div><div>Henry</div><div><br><div><blockquote type="cite"><div>On 20 May 2015, at 3:07 , David T. Lewis <<a href="mailto:lewis@mail.msen.com" target="_blank">lewis@mail.msen.com</a>> wrote:</div><br><div><br>The generated source files have a comment block that indicate the VMMaker<br>sources from which they were generated. The SVN 3306 src files came from<br>VMMaker.oscog-eem.1155, and the SVN 3308 src files were generated from<br>VMMaker.oscog-eem.1178. The changes that produce the performances difference<br>will be found somewhere in the VMMaker commits it that range.<br><br>Dave<br><br><br>On Tue, May 19, 2015 at 10:40:15PM +0200, Levente Uzonyi wrote:<br><blockquote type="cite"><br>And after that I checked out 3308 of src, and rebuilt the VM again:<br><br>'934306569 bytecodes/sec; 46368169 sends/sec'<br><br>The files which were changed from 3306 to 3308 are:<br><br>U src/vm/cointerp.c<br>U src/vm/cointerpmt.c<br>U src/vm/cogit.c<br>U src/vm/cointerp.h<br>U src/vm/gcc3x-cointerp.c<br>U src/vm/cointerpmt.h<br>U src/vm/gcc3x-cointerpmt.c<br>U src/vm/cogit.h<br>U src/plugins/QuicktimePlugin/QuicktimePlugin.c<br><br>Levente<br><br>On Tue, 19 May 2015, Levente Uzonyi wrote:<br><br><blockquote type="cite"><br>The difference is somewhere in the sources. I just checked out revision <br>3306 of the src directory, and rebuilt a CogVM (ht one, and kept the <br>platform files at 3345. I used some extra CFLAGS which are responsible for <br>the higher bytecodes/sec, ignore that.). It gives the following:<br><br>'941176470 bytecodes/sec; 122927085 sends/sec'<br><br>While 3343 gives:<br><br>'896672504 bytecodes/sec; 46102703 sends/sec'<br><br>Levente<br><br></blockquote></blockquote></div></blockquote></div><br></div></div><br></blockquote></div><br><br clear="all"><div><br></div>-- <br><div class="gmail_signature">best,<div>Eliot</div></div>
</div></div>