Add rotate-right instruction to hydrogen and use it instead of bitwise operations

Add rotate-right instruction to hydrogen and use it instead of bitwise operations
of the form ((x >>> i) | (x << (32 - i))).

This CL is based on https://chromiumcodereview.appspot.com/10984057/
by Jay Conrod <dconrod@codeaurora.org>;.

R=danno@chromium.org,mstarzinger@chromium.org,dconrod@codeaurora.org

Committed: https://code.google.com/p/v8/source/detail?r=12855

[ulan, 2012-10-01 18:15:19]

Please take a look.

https://chromiumcodereview.appspot.com/11033005/diff/1/src/hydrogen-instructi...
File src/hydrogen-instructions.h (right):

https://chromiumcodereview.appspot.com/11033005/diff/1/src/hydrogen-instructi...
src/hydrogen-instructions.h:902: bool deleteable_operation =
IsBitwiseBinaryOperation();
danno, michael: any suggestions for other instruction types to add here?

https://chromiumcodereview.appspot.com/11033005/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

https://chromiumcodereview.appspot.com/11033005/diff/1/src/hydrogen.cc#newcod...
src/hydrogen.cc:3414: HPhase phase("H_EliminateUnusedInstructions", this);
Maybe I should just do one pass instead of this complex worklist algorithm?

[Jay Conrod, 2012-10-01 19:40:40]

LGTM. I was worried having pattern matching inside the graph building phase
would overcomplicate things, but it actually seems simpler since you don't have
to worry about representation or HChanges. I can't really comment on the
ia32/x64 parts. Does MIPS need support, too?

[Jay Conrod, 2012-10-01 19:41:17]

lgtm

http://codereview.chromium.org/11033005/diff/2001/src/arm/simulator-arm.cc
File src/arm/simulator-arm.cc (right):

http://codereview.chromium.org/11033005/diff/2001/src/arm/simulator-arm.cc#ne...
src/arm/simulator-arm.cc:1503: } else {
Erik Corry expressed some suspicion in my first rev about the shifts, and I
agree. "result" is signed (int32_t), so >> will sign extend. Also *carry_out
should contain the last bit that was shifted, so it can be
(static_cast<uint32_t>(result) >> 31) != 0, right?

[ulan, 2012-10-02 12:20:26]

I uploaded a new patch set that addresses the comments and fixes a
representation bug.

http://codereview.chromium.org/11033005/diff/2001/src/arm/simulator-arm.cc
File src/arm/simulator-arm.cc (right):

http://codereview.chromium.org/11033005/diff/2001/src/arm/simulator-arm.cc#ne...
src/arm/simulator-arm.cc:1503: } else {
On 2012/10/01 19:41:17, Jay Conrod wrote:
> Erik Corry expressed some suspicion in my first rev about the shifts, and I
> agree. "result" is signed (int32_t), so >> will sign extend. Also *carry_out
> should contain the last bit that was shifted, so it can be
> (static_cast<uint32_t>(result) >> 31) != 0, right?

Yes, you're right. I am taking the updated implementation that you uploaded in
the other CL.

[ulan, 2012-10-11 11:54:30]

I rebased to ToT and remove unused instruction elimination since it was landed
separately in r12697.

[danno, 2012-10-25 15:24:55]

lgtm

[ulan, 2012-11-05 09:49:37]

Rebased to ToT, landing.