Introduce Definition and Use types.

Introduce Definition and Use types.

Make Bind and PickTemp into definitions.  Allow values to be uses that
name a definition.  Begin converting all uses of (named) Temp into Use.

R=srdjan@google.com
BUG=
TEST=

Committed: https://code.google.com/p/dart/source/detail?r=6928

[Kevin Millikin (Google), 2012-04-24 12:56:35]

Named temporaries in the IL are being eliminated.  Instead, uses point to a
definition, and definitions will (eventually, but not with this change) have an
allocatable location.

This is a partial change that allows named temporaries and definition uses to
coexist.  More is coming tomorrow.

(An issue is how to handle uses that have more than one definition reaching
them.  I plan to introduce a temporary local variable.  There is only one live
at a time, so a single one per function will suffice.  This allows them to be
treated uniformly with variables when we translate to SSA).

[srdjan, 2012-04-24 22:02:18]

LGTM

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_builder.cc (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_builder.cc:1680: ValueGraphVisitor
for_instantiator(owner(), temp_index());
Why temp_index() instead of start_index, it seems wrong?  Also above.

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:450: if (temp == NULL && use == NULL)
return false;
Use parenthesis

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:454: if (current != previous + 1) return
false;
ditto

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/intermediat...
File runtime/vm/intermediate_language.h (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1295: intptr_t temp_index() const { return
temp_index_; }
Why don't you need DECLARE_INSTRUCTION(Definition) ?

[Kevin Millikin (Google), 2012-04-25 08:50:51]

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_builder.cc (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_builder.cc:1680: ValueGraphVisitor
for_instantiator(owner(), temp_index());
On 2012/04/24 22:02:18, srdjan wrote:
> Why temp_index() instead of start_index, it seems wrong?  Also above.

Oops, that's a stray change that crept in here.  I'll set it back to
start_index.

Intermediate-term (but real soon now) I'd like to get rid of the explicit
indexes in the graph builder.  It should be possible to teach the visitor to
track stack height and set it automatically on instructions that need it (e.g.,
Bind) by making adding a definition to the graph increment the stack height and
using a defintion decrement.

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:450: if (temp == NULL && use == NULL)
return false;
On 2012/04/24 22:02:18, srdjan wrote:
> Use parenthesis

Done.  Thanks for the reminder.

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:454: if (current != previous + 1) return
false;
On 2012/04/24 22:02:18, srdjan wrote:
> ditto

Done.

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/intermediat...
File runtime/vm/intermediate_language.h (right):

https://chromiumcodereview.appspot.com/10201017/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1295: intptr_t temp_index() const { return
temp_index_; }
On 2012/04/24 22:02:18, srdjan wrote:
> Why don't you need DECLARE_INSTRUCTION(Definition) ?

Because it's an abstract, not concrete, instruction class.  We don't want to
have to implement VisitXXX for those.  (It might be nice to get the predicates
etc. automatically, but in that case we might split into DECLARE_INSTRUCTION and
DECLARE_CONCRETE_INSTRUCTION).