Use a uniform way to emit code for all instructions.

Use a uniform way to emit code for all instructions.

Rather than having special cases for block entries and parallel moves,
use the normal code generation infrastructure.

Take advantage of this to avoid generating duplicate deoptimization entries
for some instructions (e.g., GotoInstr).

R=fschneider@google.com
BUG=

[Florian Schneider, 2013-05-22 15:56:15]

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language.cc (right):

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.cc:1428: return new LocationSummary(0, 0,
LocationSummary::kNoCall);
Is this ever called? If no, just return NULL. Otherwise, can you avoid
allocating an empty location summary for block entry instructions?

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language.h (right):

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.h:931: LocationSummary* locs_;
I'd prefer to have this member only on those that actually need a
LocationSummary.

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.h:1782: class PushArgumentInstr : public
Definition {
Can't PushArgument just inherit from TemplateDefinition<1>?

[Kevin Millikin (Google), 2013-05-23 09:09:49]

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language.cc (right):

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.cc:1428: return new LocationSummary(0, 0,
LocationSummary::kNoCall);
On 2013/05/22 15:56:15, Florian Schneider wrote:
> Is this ever called? If no, just return NULL. Otherwise, can you avoid
> allocating an empty location summary for block entry instructions?

Yes, it's called by the nonoptimizing compiler.  It's probably a wart that it's
not called by the register allocator.

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language.h (right):

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.h:931: LocationSummary* locs_;
On 2013/05/22 15:56:15, Florian Schneider wrote:
> I'd prefer to have this member only on those that actually need a
> LocationSummary.

Good: all instructions need a location summary --- to tell the compiler what
locations they use and whether they make a call.

I'd rather use a real location summary than have special case code to support
NULL as a redundant representation of new LocationSummary(0, 0,
LocationSummary::kNoCall).

[Florian Schneider, 2013-05-23 19:12:35]

lgtm

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language.h (right):

https://codereview.chromium.org/15741002/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language.h:931: LocationSummary* locs_;
On 2013/05/23 09:09:49, kmillikin wrote:
> On 2013/05/22 15:56:15, Florian Schneider wrote:
> > I'd prefer to have this member only on those that actually need a
> > LocationSummary.
> 
> Good: all instructions need a location summary --- to tell the compiler what
> locations they use and whether they make a call.
> 
> I'd rather use a real location summary than have special case code to support
> NULL as a redundant representation of new LocationSummary(0, 0,
> LocationSummary::kNoCall).

Fine with me, if there is no visible impact on compilation time (e.g. when
running dart2js).

If now all instructions have a location summary, shouldn't then locs_ be just an
embedded LocationSummary member instead of a pointer to a ZoneAllocated object?

[Kevin Millikin (Google), 2013-05-27 11:21:34]

Actually, I'll close this issue until we sort out what we want to do with
instructions/locations regarding the size of IL instructions.

I'll send you a change with the uncontroversial part.