Commit Graph

254 Commits

Author SHA1 Message Date
fschneider@google.com 0a1d6d15f9 Fix bug in optimizing string .length loads.
We're not allowed to hoist a .length load if the corresponding
class-check is not also hoisted out of a loop.

TEST=tests/language/optimized_string_charcodeat_test.dart
Review URL: https://codereview.chromium.org//12256037

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18519 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-14 15:59:18 +00:00
fschneider@google.com e2a0c8f7ca Optimize stores to ExternalUint8Array and ExternalUint8ClampedArray in the optimizer.
This CL adds support for these two array types to the StoreIndexed instruction.
Review URL: https://codereview.chromium.org//12263010

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18517 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-14 12:59:51 +00:00
srdjan@google.com 5694a140e3 Fix allocation of array tables (use store barrier if needed, store values directly instead of via stack).
Review URL: https://codereview.chromium.org//12212050

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18464 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-13 18:44:27 +00:00
vegorov@google.com baa0347253 Reapply r18377 it was reverted due to the unrelated bug it surfaced.
Remove SminessPropagator and FlowGraphTypePropagator and all associated infrastructure and fields.

Replace multiple fields (result_cid_, propagated_cid_, propagated_type_, reaching_cid_) with a single field of type CompileType which represents an element of type analysis lattice and incorporates information about: value's nullability, concrete class id and abstract super type. This ensures that propagated cid and type are always in sync and complement each other

Implement a new FlowGraphPropagator that propagates types over the CompileType-lattice.

R=fschneider@google.com

Review URL: https://codereview.chromium.org//12260008

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18456 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-13 17:16:35 +00:00
vegorov@google.com 487c73e01f Revert "Remove SminessPropagator and FlowGraphTypePropagator and all associated infrastructure and fields."
This reverts commit r12365.

Reason: failure on vm-mac-debug (debug_ia32 standalone/io/file_fuzz_test).

TBR=fschneider@google.com

Review URL: https://codereview.chromium.org//12221139

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18378 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-12 12:36:21 +00:00
vegorov@google.com f8d9a90b2c Remove SminessPropagator and FlowGraphTypePropagator and all associated infrastructure and fields.
Replace multiple fields (result_cid_, propagated_cid_, propagated_type_, reaching_cid_) with a single field of type CompileType which represents an element of type analysis lattice and incorporates information about: value's nullability, concrete class id and abstract super type. This ensures that propagated cid and type are always in sync and complement each other

Implement a new FlowGraphPropagator that propagates types over the CompileType-lattice.

BUG=

Review URL: https://codereview.chromium.org//12221119

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18377 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-12 12:06:47 +00:00
kmillikin@google.com ab3cbf5796 Change CSE, LICM, and range analysis to preserve use lists.
Change the common subexpression elimination, loop-invariant code
motion, and range analysis passes to maintain use lists.

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18209 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-07 09:42:36 +00:00
fschneider@google.com f0caa40576 Add optimized loads from ExternalUint8ClampedArrays.
The generated code is exactly the same as for ExternalUint8Array.

TEST=tests/standalone/byte_array_test.dart
Review URL: https://codereview.chromium.org//12225049

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18183 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-06 19:00:50 +00:00
fschneider@google.com 8200c55ed9 Flip test condition when testing for fixed-length array length load.
This addresses additional comments to my previous CL:
https://codereview.chromium.org/12088108/
Review URL: https://codereview.chromium.org//12208046

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18175 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-06 16:21:51 +00:00
fschneider@google.com 7fc8f72c25 Inline getters of byte array view in the optimized flow graph.
This CL provides inline IL code for the getters _getInt8, _getInt16, etc.
to speed up [] and byte array views.

The code uses the existing LoadIndexed instructions by passing a index
scale factor explicitly: For normal arrays loads, the scale factor is equal
to the element size. For byte array access, the scale factor is always 1.

I'm adding inlined setters in a separate CL.
Review URL: https://codereview.chromium.org//12218008

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18173 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-06 15:18:57 +00:00
srdjan@google.com eaa1c974a3 Use SAR for divident by a power-of two constant divisor.
Review URL: https://codereview.chromium.org//12091100

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18075 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-04 20:29:58 +00:00
kmillikin@google.com a518a0a36f Add a use list iterator that allows mutation of current.
Add Value::Iterator that supports removing the current use from the
underlying use list, including deleting it or moving it to another use
list.  Use it in range analysis and representation selection.

R=fschneider@google.com

Review URL: https://codereview.chromium.org//12178019

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18055 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-04 17:04:02 +00:00
kmillikin@google.com 361300cb71 Move recording of definition uses from the value to the definition.
It was weird that this method mutated an object that wasn't even
mentioned in the call.  Also, change Definition::ReplaceUsesWith so that
it appends all at once instead of consing each individual use.

Review URL: https://codereview.chromium.org//12091091

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18048 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-04 14:01:35 +00:00
fschneider@google.com 2718b54625 Separate the array/string .length load from the bounds check.
This is done in preparation to inline more array and array view
operations.

Since the length is immutable for fixed length arrays, it allows
hoisting the load out of loops.
Review URL: https://codereview.chromium.org//12088108

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17984 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-01 15:10:24 +00:00
kmillikin@google.com a2436038da Make use lists into doubly-linked lists.
At the cost of a word per use, we gain constant time removal from use
lists.

BUG=
Review URL: https://codereview.chromium.org//12079096

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17924 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-31 16:10:15 +00:00
srdjan@google.com 64282183fe Try to have constant on the right hand side for commutative Smi instructions.
Review URL: https://codereview.chromium.org//12096067

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17879 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-30 22:57:20 +00:00
srdjan@google.com 35ef4ddbd5 Enable correct optimized double modulo operation. (TODO: enable remainder optimization).
Review URL: https://codereview.chromium.org//12082063

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17859 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-30 17:40:28 +00:00
srdjan@google.com 6e26ea871d On Ia32 optimistically assume that results from int32 and uint32 array loads fit into Smi. Only if the instruction caused deoptimization will we assume mixed Smi/Mint operations. In absence of precise tracking of deoptimization history (location/reason) conservatively assume that a method that was deoptimized once may have been deoptimzied because of uint32/int32 loads.
Significant improvement for MeshDecompression on ia32.
Review URL: https://codereview.chromium.org//12086045

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17810 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-29 21:16:44 +00:00
srdjan@google.com 3d1a139c4d Fix double modulo bug by temporary disabling calling to C (fix and reenabling it will follow later). Added test.
Review URL: https://codereview.chromium.org//12077052

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17806 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-29 20:36:59 +00:00
fschneider@google.com 35bc3b674e Widen the type of constant values handled by constant propagation.
The constant object has to be a smi or an old space heap object: Right
now, we can only embed old space objects into code.
Review URL: https://codereview.chromium.org//12096039

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17770 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-29 16:02:15 +00:00
fschneider@google.com dfda7854a8 Fix source position for stack traces with optimized top function.
Fix decoding of the deoptimization info when constructing a stack trace.

Fix stack trace for checked mode exceptions from optimized code.

BUG=dart:8058
TEST=runtime/tests/vm/dart/optimized_stacktrace_test.dart,
     tests/language/stack_overflow_stacktrace_test.dart
Review URL: https://codereview.chromium.org//12049039

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17755 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-29 12:38:27 +00:00
srdjan@google.com d3e5631456 Use builtin fmod for double modulo operation.
Review URL: https://codereview.chromium.org//12049056

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17514 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-24 00:06:16 +00:00
srdjan@google.com 76dffe600c Removed loop depth info tracking at graph build time.
60% compilation speed improvement on pathological case. Improvements for dart2js

The loop depth signal was made unnecessary by the call frequency signal.
Review URL: https://codereview.chromium.org//11975061

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17408 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-22 16:15:54 +00:00
vegorov@google.com 932a05420a Reland r17365.
Introduce InvokeMathCFunction that can be used to directly invoke mathematical
function provided by runtime.

Use it to unconditionally inline _Double.pow.

Use it to inline floor, ceil, round, truncate, round when SSE4.1 is not
available.

Perform representation selection phase after constant propagation to minimize
boxing.

Add support for enter instruction in the x64 disassembler.

Add test for optimized pow and fix compilation on windows.

R=fschneider@google.com
BUG=dart:8002

Review URL: https://codereview.chromium.org//12038013

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17393 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-22 12:18:31 +00:00
vegorov@google.com ab91977173 Revert "Introduce InvokeMathCFunction that can be used to directly invoke mathematical function provided by runtime."
This reverts commit r17365.

TBR=fschneider@google.com

Review URL: https://codereview.chromium.org//12042014

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17367 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-21 17:16:16 +00:00
vegorov@google.com a20c8d28b8 Introduce InvokeMathCFunction that can be used to directly invoke mathematical function provided by runtime.
Use it to unconditionally inline _Double.pow.

Use it to inline floor, ceil, round, truncate, round when SSE4.1 is not available.

Perform representation selection phase after constant propagation to minimize boxing.

Add support for enter instruction in the x64 disassembler.

R=fschneider@google.com
BUG=dart:8002

Review URL: https://codereview.chromium.org//12050002

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17365 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-21 16:59:37 +00:00
fschneider@google.com 64039d8ec3 Optimize loads and stores to Int32Array and Uint32Array.
Review URL: https://codereview.chromium.org//12041005

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17362 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-21 16:19:42 +00:00
fschneider@google.com 3e94e4a2fa Fix bug in optimization in the presence of externalized strings.
The class-id of string objects can be changed via the Dart API. The optimizer
has to disable hoisting class-id checks across calls. For now, loop-invariant
code motion is disabled for checks of string class-ids.
Review URL: https://codereview.chromium.org//11867020

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17288 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 16:29:02 +00:00
vegorov@google.com f6f7f1579e Optimize double.floor and double.ceil down to roundsd when SSE4.1 is available.
R=fschneider@google.com
BUG=dart:7971

Review URL: https://codereview.chromium.org//11929021

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17278 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 14:28:25 +00:00
fschneider@google.com 657c4944c3 Remove StringCharCodeAtInstr and handle it as part of LoadIndexed.
Review URL: https://codereview.chromium.org//11970038

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17274 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 14:03:40 +00:00
vegorov@google.com a77fc9701c When requested to extract a method M from class C inject a method extractor (consisting of a single AST node CreateClosure) as a getter get:M into C.
This allows to cache and optimize method extraction requests as normal method invocations and at hot method extraction sites that significantly decreases overhead of method extraction which previously required two trips into runtime system and was not cached at all.

BUG=

Review URL: https://codereview.chromium.org//11642003

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17261 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 11:54:45 +00:00
regis@google.com 912194a756 Fix vm code base so that it can be built for --arch=simarm (no snapshot yet).
Review URL: https://codereview.chromium.org//11956004

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17246 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 00:34:20 +00:00
fschneider@google.com 4982f8ff33 Fix a bug with store-to-load forwarding for typed arrays.
Stores to typed arrays implicitly convert the value stored.
This CL disables store-to-load forwarding for these array stores.

Normal arrays and float64 arrays can be still optimized as before (no
conversion occurs on store)

TEST=tests/language/int_array_load_elimination_test.dart
Review URL: https://codereview.chromium.org//11962036

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17208 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-17 14:08:04 +00:00
fschneider@google.com 8c5b198d95 Optimized loads/stores for scalar list: Uint8Clamped, Int8, Int16, Uint16.
This CL adds optimized stores for Uint8ClampedList, and
loads+stores for Int8List, Int16List and Uint16List.


TEST=tests/standalone/byte_array_test.dart, tests/standalone/int_array_test.dart
Review URL: https://codereview.chromium.org//11967012

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17190 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-17 10:43:46 +00:00
vegorov@google.com d11d988998 Set statically known ResultCid for AllocateObject.
Specialize PolymorphicInstanceCall in ApplyClassIds if receiver cid
was infered: filter out all but one target for attached ICData and drop
checks.

Canonicalize away CheckClass if reciever cid matches at least one in the list of checked.

R=fschneider@google.com
BUG=

Review URL: https://codereview.chromium.org//11886044

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17079 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-15 15:51:24 +00:00
fschneider@google.com e0395cf4bb Optimize stores to Uint8List.
Review URL: https://codereview.chromium.org//11855007

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17014 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-14 10:27:28 +00:00
srdjan@google.com 40a7130741 Inline Doubles truncate and round.
Review URL: https://codereview.chromium.org//11573044

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16983 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-11 20:42:47 +00:00
kmillikin@google.com 0f65888e88 Change the inlining context from an enum to a class.
To support inlining in test contexts, the inlining context needs to have its
dispatched behavior and state.  This change introduces a context class
representing calls inlined for their value or solely for their effects.  The
intermediate array of exits is moved from the graph to the inlining context.
The implementation behavior is otherwise the same as before.

Review URL: https://codereview.chromium.org//11856010

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16965 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-11 11:43:48 +00:00
fschneider@google.com 15432b2fee Add canonicalize optimization for branches.
Comparisons where the result is only used in a branch on true
can be folded into the branch instruction.

Patterns like this can occur after inlining and constant propagation.
In normal code, we merge branches and comparisons already at graph
building time.

v3 <- (v1 == v2)
Branch if (v3 === true)

is optimized to

Branch if (v1 == v2)

Also: Remove outdated TODOs and rename the canonicalization pass to a 
better name.
Review URL: https://codereview.chromium.org//11819031

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16919 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-10 13:18:31 +00:00
vegorov@google.com f115552d29 Constant propagator should revisit phis when it visits predecessor block.
Phi's value depends both on its inputs and reachability of join's predecessors. Thus it is essential to revisit phi when predecessor becomes reachable.

R=kmillikin@google.com
BUG=

Review URL: https://codereview.chromium.org//11824024

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16847 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-09 12:27:46 +00:00
fschneider@google.com d5f2305515 Support scalar lists in array bounds check elimination.
This allows to eliminate bounds checks for scalar lists in loops like:

for (var i=0; i < a.length; i++) {
  a[i]
}
Review URL: https://codereview.chromium.org//11665005

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16716 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-07 15:43:35 +00:00
srdjan@google.com 0af16251b5 Intrinsify Uint8ClampedArray's store and load indexed, inline load indexed.
Review URL: https://codereview.chromium.org//11783006

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16666 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-05 01:07:32 +00:00
regis@google.com 47e08870ca Support Null type in Class::TypeTest and simplify Instance::IsInstanceOf.
Review URL: https://codereview.chromium.org//11785006

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16664 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-05 00:45:52 +00:00
srdjan@google.com 5915aff811 Optimize instanceof: if all results are true and tests can be done using class only replace instanceof with a Boolean constant.
Review URL: https://codereview.chromium.org//11746024

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16648 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-04 21:29:05 +00:00
asiva@google.com 4c069867e4 - Make Boolean 'true' and 'false' singleton VM isolate objects.
- Change all uses of it
Review URL: https://codereview.chromium.org//11745022

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16623 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-04 01:52:05 +00:00
srdjan@google.com 29e06db609 In unoptimized code use call for instanceof instead of inlined checks. This allows us to collect type feedback and to reduce the code size of unoptimized code. Next will be work on type tests as well.
Review URL: https://codereview.chromium.org//11694003

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16589 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-02 22:22:39 +00:00
asiva@google.com 7022b39e35 Convert all symbol accessors to return read only handles so that it is not necessary to create a new handle in the code that uses it.
Added a few more strings to the symbols list.
Review URL: https://codereview.chromium.org//11667012

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16584 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-02 19:31:59 +00:00
asiva@google.com 7ea520f10f Create read only handles for empty_array and sentinel objects
(trying out a basic framework and will extend it to others once this
works).
Review URL: https://codereview.chromium.org//11648006

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16416 260f80e4-7a28-3924-810f-c04153c831b5
2012-12-21 02:33:05 +00:00
srdjan@google.com 324f1715bc Mark final fields as immutable.
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16390 260f80e4-7a28-3924-810f-c04153c831b5
2012-12-20 19:29:52 +00:00
vegorov@google.com 9a45691c76 When replacing one value with another ensure that replacement has SSA index assigned.
R=srdjan@google.com
BUG=dart:7513

Review URL: https://codereview.chromium.org//11638030

git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@16352 260f80e4-7a28-3924-810f-c04153c831b5
2012-12-20 00:16:42 +00:00