Delta encode pc descriptors, and combine pc kind and try index into single field.

Delta encode pc descriptors, and combine pc kind and try index into single field.

328.3kB -> 126.8kB (-61.4%) mirrors_reader_test

Avoid quadratic var-int decoding in extracting edge counters by decoding once into a temporary list.

R=fschneider@google.com

Committed: https://github.com/dart-lang/sdk/commit/300a2f34cf3632892b36aaf149c9f4fd241e6324

[rmacnak, 2015-05-12 19:38:12]

rmacnak@google.com changed reviewers:
+ fschneider@google.com, iposva@google.com

[rmacnak, 2015-05-14 17:14:56]

PTAL.  Golem now reports +6.8% for CompileHello instead of -33.3%.

[Florian Schneider, 2015-05-18 11:33:28]

lgtm

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/block...
File runtime/vm/block_scheduler.cc (right):

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/block...
runtime/vm/block_scheduler.cc:29: while (i < n) {
I'd turn this into a for-loop.

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/block...
runtime/vm/block_scheduler.cc:88: ZoneGrowableArray<uword>* deopt_id_pc_pairs =
new ZoneGrowableArray<uword>();
I wonder if using DirectChainedHashMap from hash_map.h would speed up finding
the edge counters even more.

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/raw_o...
File runtime/vm/raw_object.h (right):

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/raw_o...
runtime/vm/raw_object.h:1033: kOther           = kOsrEntry << 1,
kLastKind = kOther,

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/raw_o...
runtime/vm/raw_object.h:1034: kAnyKind         = 0xFF
I think -1 would be a more useful value for kAnyKind - should also work if there
are more than 8 kinds.

https://chromiumcodereview.appspot.com/1128183007/diff/60001/runtime/vm/raw_o...
runtime/vm/raw_object.h:1060: static const intptr_t kKindShiftSize = 3;
Maybe assert that 3 bits is enough: kLastKind <= 128

sth like:

COMPILE_ASSERT(kLastKind <= 1 << ((1 << kKindShiftSize) - 1));

[rmacnak, 2015-05-18 18:03:55]

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/block_schedu...
File runtime/vm/block_scheduler.cc (right):

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/block_schedu...
runtime/vm/block_scheduler.cc:29: while (i < n) {
On 2015/05/18 11:33:28, Florian Schneider wrote:
> I'd turn this into a for-loop.

Done.

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/block_schedu...
runtime/vm/block_scheduler.cc:88: ZoneGrowableArray<uword>* deopt_id_pc_pairs =
new ZoneGrowableArray<uword>();
On 2015/05/18 11:33:28, Florian Schneider wrote:
> I wonder if using DirectChainedHashMap from hash_map.h would speed up finding
> the edge counters even more.

I'll take a look.

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/raw_object.h
File runtime/vm/raw_object.h (right):

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/raw_object.h...
runtime/vm/raw_object.h:1034: kAnyKind         = 0xFF
On 2015/05/18 11:33:28, Florian Schneider wrote:
> I think -1 would be a more useful value for kAnyKind - should also work if
there
> are more than 8 kinds.

sgtm

https://codereview.chromium.org/1128183007/diff/60001/runtime/vm/raw_object.h...
runtime/vm/raw_object.h:1060: static const intptr_t kKindShiftSize = 3;
On 2015/05/18 11:33:28, Florian Schneider wrote:
> Maybe assert that 3 bits is enough: kLastKind <= 128
> 
> sth like:
> 
> COMPILE_ASSERT(kLastKind <= 1 << ((1 << kKindShiftSize) - 1));

Done.

[rmacnak, 2015-05-18 18:04:51]

Committed patchset #5 (id:80001) manually as
300a2f34cf3632892b36aaf149c9f4fd241e6324 (presubmit successful).