Make the array bounds check elimination phase optional (and set the foundation for introducing SSI …

Make the array bounds check elimination phase optional (and set the foundation for introducing SSI in a simple way).

BUG=

Committed: https://code.google.com/p/v8/source/detail?r=13409

[Massi, 2013-01-10 10:42:45]

Addressed (offline) review comments.

[Jakob Kummerow, 2013-01-10 16:06:47]

LGTM with comments.

https://chromiumcodereview.appspot.com/11783055/diff/2002/src/hydrogen-instru...
File src/hydrogen-instructions.h (right):

https://chromiumcodereview.appspot.com/11783055/diff/2002/src/hydrogen-instru...
src/hydrogen-instructions.h:351: Representation KeyedAccessIndexRequirement() {
nit: I'm a bit unhappy about adding instruction-specific logic to the
Representation class. Three alternative suggestions, pick what you like:
1) Inline it. It's only one line, after all.
2) Move to a static method of ArrayInstructionInterface.
3) Move to a file-global static function "Representation
ArrayIndexRepresentation(Representation r) {...}" as close as possible to the
two call sites.

https://chromiumcodereview.appspot.com/11783055/diff/2002/src/hydrogen.cc
File src/hydrogen.cc (right):

https://chromiumcodereview.appspot.com/11783055/diff/2002/src/hydrogen.cc#new...
src/hydrogen.cc:3974: HPhase phase("H_Apply actual values", this);
It would be nice to be able to skip this completely when it is not needed (when
there is no HBoundsCheck around). Feel free to fix that in a follow-up CL.

https://chromiumcodereview.appspot.com/11783055/diff/2002/src/hydrogen.cc#new...
src/hydrogen.cc:3981: if (phi->ActualValue() != phi) {
This can never be true, right? Let's not do the work, then. If you want to be
extra careful, you can put in a guard:

#ifdef DEBUG
    for (...) {
      // A phi's actual value is currently always itself.
      HPhi* phi = block->phis()->at(i);
      ASSERT(phi->ActualValue() == phi);
    }
#endif

[Toon Verwaest, 2013-01-17 15:48:57]

This CL breaks crypto aes from the kraken-once benchmark suite.