The object with strings that we create for tagged templates is a constant frozen array, so we should be able to optimize the loads from that array. I have a patch seems to work, but needs a few more comments and some of those trackOptimizationAttempt things.

Comment on attachment 8968482 [details] [diff] [review] Optimize loads from CallSiteObjects for tagged template literals Review of attachment 8968482 [details] [diff] [review]: ----------------------------------------------------------------- Sweet. ::: js/src/jit-test/tests/ion/template-tag-callsiteobject.js @@ +1,2 @@ > +function tagA(strings) { > + assertEq(strings.length, 2); Nit: the indentation in this test looks a bit off. ::: js/src/jit/IonBuilder.cpp @@ +8251,5 @@ > +IonBuilder::getElemTryCallSiteObject(bool* emitted, MDefinition* obj, MDefinition* index) > +{ > + MOZ_ASSERT(*emitted == false); > + > + if (!obj->isConstant() || obj->type() != MIRType::Object) { Nit: no {} here and ~4 times below.

Pushed by evilpies@gmail.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/0a19791ab773 Optimize loads from CallSiteObjects for tagged template literals. r=jandem

On AWFY this didn't help on the expected test templatestringtag-es6. Instead we seem to have a reversed a regression on map-set-es6. GC timing maybe?