--- [EMAIL PROTECTED] wrote: > There is a plot of runtime of CVSHEAD versus version 3.3.13 > at > > http://www.sqlite.org/relspeed-20070326-ephemeral.gif > > Any value less than > 1.0 means that performance has improved. The test script is > > http://www.sqlite.org/cvstrac/fileview?f=sqlite/test/speed2.test > > The orange bars show the relative performance of CVSHEAD with > no special build options. 1.0 is the performance of version 3.3.13 > so anything less than 1.0 is an improvement. The red bars show > CVSHEAD with all the source files concatenated into a single big > file and compiled as a single translation unit. Compiling this > way allows the compiler to do additional optimizations that result > in improved performance. The blue bars are the same sqlite-as-one- > great-big-source-file optimization but with -DSQLITE_OMIT_XFER_OPT=1 > defined.
Might I ask which compiler optimization flags were used for the 3 bars? In particular, was strict aliasing used for all of them and what was the CPU make/model or at least how much CPU cache. I wonder if the latter tests had working data sets small enough to run completely in the L2/L3 cache as result of being completely in the disk cache (i.e., copy3, random-del1/2). ____________________________________________________________________________________ Food fight? Enjoy some healthy debate in the Yahoo! Answers Food & Drink Q&A. http://answers.yahoo.com/dir/?link=list&sid=396545367 ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
