I uploaded the changes (style and variable names) and ran new benchmarks here:
http://golem.spb:9013/golem/compare?vmname=v8-jschuh-const-cookies&vmrev=5720&ref_vmname=v8-jschuh-base&ref_vmrev=5720 <http://golem.spb:9013/golem/compare?vmname=v8-jschuh-const-cookies&vmrev=5720&ref_vmname=v8-jschuh-base&ref_vmrev=5720> -j On Mon, Oct 25, 2010 at 12:25 AM, <[email protected]> wrote: > > http://codereview.chromium.org/3973002/diff/1/3 > File src/ia32/codegen-ia32.cc (right): > > http://codereview.chromium.org/3973002/diff/1/3#newcode157 > src/ia32/codegen-ia32.cc:157: jit_cookie_((FLAG_disable_jit_cookie) ? 0 > : V8::Random()) { > > On 2010/10/21 18:08:08, iposva wrote: > >> On 2010/10/21 14:16:01, Søren Gjesse wrote: >> > Is it required to have a new cookie generated for each code >> > generator created? > >> > Maybe a global cookie will be sufficient to avoid calling >> > V8::Random() so many > >> > times. >> > > Premature optimization alert! Compared to the amount of work needed >> > for one > >> compilation unit, how many cycles does it add to calculate >> > V8::Random()? > > > At one time we talked about having a jit cookie per VM which could be > part of the crash dumps to make debugging easier. But that only makes > sense if we ever need to move to "real" code randomization. > > For this change a per code generator instance should be fine. > > > http://codereview.chromium.org/3973002/show > -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
