> > I'm wondering if this whole game is even necessary. Could we use the > > struct version in opt builds as well? The only reason I can think not > > to is if some compilers do not optimize a struct containing a single > > integer as well as they would a naked integer.
That's right, in particular for parameter passing, in many ABIs, structs always get passed on the stack. I tested this 3 years ago and it was definitely measurable on SS/V8. It probably matters less on these benchmarks nowadays b/c we spend less time in the VM, but, of course, benchmarks are not the only thing that matters. It'd be useful to measure if there was a difference on, e.g., the testcase in bug 907369, since proxy handlers shuffle around a lot of jsids as parameters. > At any rate, we're definitely already screwed on compilers than don't > properly optimized those kinds of structs. (See Value!) Value gets passed around by 'const Value &' since, on 32-bit, it gets passed on the stack anyways. Well, and HandleValue, but not by value (usually). _______________________________________________ dev-tech-js-engine-internals mailing list [email protected] https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
