Issue 10010029: When checking against non-parametrized types use a cache to hold result tuples (class, result). Tha…

Issue 10010029: When checking against non-parametrized types use a cache to hold result tuples (class, result). Tha… (Closed)

Created:
8 years, 8 months ago by srdjan

Modified:
8 years, 8 months ago

Reviewers:
regis

CC:
reviews_dartlang.org, vm-dev_dartlang.org

Base URL:
http://dart.googlecode.com/svn/branches/bleeding_edge/dart/

Visibility:
Public.

More Reviews

Description

When checking against non-parametrized types use a cache to hold result tuples (class, result). That cache is stored in the instruction stream. Currently implemented for instanceof, ia32 only. Should migrate to other type tests and architectures. 70x improvement on a benchmark (similar to what is seen in html _unwrap). Committed: https://code.google.com/p/dart/source/detail?r=6281

Patch Set 1 #

Patch Set 2 : #

Patch Set 3 : #

Total comments: 8

Patch Set 4 : #

Created: 8 years, 8 months ago

Download [raw] [tar.bz2]

	Unified diffs	Side-by-side diffs	Delta from patch set	Stats (+175 lines, -19 lines)			Patch
M	runtime/vm/class_finalizer.cc	View	1 2 3	1 chunk	+17 lines, -6 lines	0 comments	Download
M	runtime/vm/code_generator.cc	View	1 2 3	2 chunks	+43 lines, -7 lines	0 comments	Download
M	runtime/vm/code_generator_ia32.cc	View	1 2 3	5 chunks	+48 lines, -4 lines	0 comments	Download
M	runtime/vm/code_generator_x64.cc	View	1 2 3	2 chunks	+4 lines, -1 line	0 comments	Download
M	runtime/vm/code_patcher.h	View	1 2 3	1 chunk	+3 lines, -0 lines	0 comments	Download
M	runtime/vm/code_patcher_arm.cc	View	1 2 3	1 chunk	+12 lines, -0 lines	0 comments	Download
M	runtime/vm/code_patcher_ia32.cc	View	1 2 3	1 chunk	+16 lines, -0 lines	0 comments	Download
M	runtime/vm/code_patcher_x64.cc	View	1 2 3	1 chunk	+13 lines, -0 lines	0 comments	Download
M	runtime/vm/flow_graph_compiler_x64.cc	View	1 2 3	2 chunks	+4 lines, -1 line	0 comments	Download
M	runtime/vm/object.h	View	1 2 3	2 chunks	+2 lines, -0 lines	0 comments	Download
M	runtime/vm/object.cc	View	1 2 3	2 chunks	+13 lines, -0 lines	0 comments	Download

Messages

Total messages: 3 (0 generated)

Expand Messages | Collapse Messages

regis

LGTM Nice! Are you planning on caching the type for implicit type checks in checked ...

8 years, 8 months ago (2012-04-06 00:44:01 UTC) #2

srdjan

8 years, 8 months ago (2012-04-06 15:58:34 UTC) #3

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/class_fi...
File runtime/vm/class_finalizer.cc (right):

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/class_fi...
runtime/vm/class_finalizer.cc:1227: const String& super_name =
String::Handle(super_type.TypeClassName());
On 2012/04/06 00:44:01, regis wrote:
> Why not print the super type rather than just the super type class?
> TypeClassName() -> Name()

Done.

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/code_pat...
File runtime/vm/code_patcher_arm.cc (right):

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/code_pat...
runtime/vm/code_patcher_arm.cc:73: UNIMPLEMENTED();  // Untested.
On 2012/04/06 00:44:01, regis wrote:
> I hope the UNIMPLEMENTED() is tested by now ;-)

:-). Removed comment.

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/code_pat...
File runtime/vm/code_patcher_x64.cc (right):

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/code_pat...
runtime/vm/code_patcher_x64.cc:248: uint64_t* target_addr =
reinterpret_cast<uint64_t*>(instruction_address + 1);
On 2012/04/06 00:44:01, regis wrote:
> I do not think offset +1 is correct.
> You always told me not to submit untested code :-)

Deleting the code in both, leaving UNIMPLEMENTED.

https://chromiumcodereview.appspot.com/10010029/diff/5004/runtime/vm/code_pat...
runtime/vm/code_patcher_x64.cc:257: uint64_t* target_addr =
reinterpret_cast<uint64_t*>(instruction_address + 1);
On 2012/04/06 00:44:01, regis wrote:
> ditto

ditto

Expand Messages | Collapse Messages