Sort checks by counts before emitting test-and-call polymorphic instance calls.

Sort checks by counts before emitting test-and-call polymorphic instance calls.
Committed: https://code.google.com/p/dart/source/detail?r=16517

[regis, 2012-12-27 17:52:30]

LGTM

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
File runtime/vm/flow_graph_compiler.cc (right):

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
runtime/vm/flow_graph_compiler.cc:658: if ((*sorted)[largest_ix].count <
(*sorted)[k].count) {
You could assign (*sorted)[largest_ix].count to largest_count outside of the
loop for better readability and greater speed, in debug mode at least :-)

[srdjan, 2012-12-27 18:19:41]

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
File runtime/vm/flow_graph_compiler.cc (right):

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
runtime/vm/flow_graph_compiler.cc:658: if ((*sorted)[largest_ix].count <
(*sorted)[k].count) {
On 2012/12/27 17:52:30, regis wrote:
> You could assign (*sorted)[largest_ix].count to largest_count outside of the
> loop for better readability and greater speed, in debug mode at least :-)

largest_ix is not loop invariant. Maintaining the largest count in sync with
largest_ix makes the code less readable, IMHO. Let me know if you disagree.

[regis, 2012-12-27 18:26:34]

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
File runtime/vm/flow_graph_compiler.cc (right):

https://codereview.chromium.org/11602014/diff/3001/runtime/vm/flow_graph_comp...
runtime/vm/flow_graph_compiler.cc:658: if ((*sorted)[largest_ix].count <
(*sorted)[k].count) {
On 2012/12/27 18:19:42, srdjan wrote:
> On 2012/12/27 17:52:30, regis wrote:
> > You could assign (*sorted)[largest_ix].count to largest_count outside of the
> > loop for better readability and greater speed, in debug mode at least :-)
> 
> largest_ix is not loop invariant. Maintaining the largest count in sync with
> largest_ix makes the code less readable, IMHO. Let me know if you disagree.

I do not disagree :-) Slow start after the break...