This CL runs a second round of constant propagation after range
analysis to eliminate additional unreachable code. Range analysis
is changed to mark branches as constant if the constraints they
generate are unsatisfiable.
The second pass of constant propagation only visits branches and
removes unreachable code, but does not do full constant propagation.
This proves useful when inlining array view operations where the
following pattern occurs:
for (i = 0; i < length; i++) {
if (i < 0 || i >= length) {
throw 123;
}
foo();
}
In this example the if-statement will be eliminated completely.
Also, fix a bug in range analyis where constraints of already
constrained values were missing.
Review URL: https://codereview.chromium.org//13469013
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20914 260f80e4-7a28-3924-810f-c04153c831b5
This CL removes optimized access for scalarlist, and only the new TypedData classes
are optimized. I changed the runtime libraries core and math to use typedData
instead of scalarlist (Uint16List is used in StringBuffer, Uint32List by Math.random).
Instead of using LoadIndexed for internal and external arrays,
split external loads into a load of the backing store and a load
of the element.
v3 <- LoadIndexed(v1, index)
becomes
v2 <- LoadUntagged(v1, ExternalTypedData::data_offset)
v3 <- LoadIndexed(v2, index);
For this I introduce two new representations in the IL:
kUntagged (for values that hold a untagged pointer) and
kNoRepresentation (for instructions accept any input
representation)
Deoptimization does not need to know about kUntagged
since these values can never occur in the environment.
Also with this change:
* fix COMPILE_ASSERT and use it in one place.
* Cleanup IL printer output of deopt ids.
Review URL: https://codereview.chromium.org//12871010
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20198 260f80e4-7a28-3924-810f-c04153c831b5
Branch optimization pushes some branches that test the value of a phi
to the predecessor blocks. This can avoid materializing a boolean
object solely for the purposes of branching on its boolean value.
The optimization is performed after inlinining which creates
opportunities, and before constant propagation, because it exposes
opportunities for unreachable code elimination.
R=vegorov@google.com
BUG=
Review URL: https://codereview.chromium.org//12540002
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@19682 260f80e4-7a28-3924-810f-c04153c831b5
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
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
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
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
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
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
Implement proposed new identity spec (flag --new_identity_spec). It will be enabled/committed once the new spec been confirmed.
The new spec says that identity of numbers is computed on their values instead of on the object reference. This provides deterministic behavior without having to disable optimizations in the VM. In some cases 'identical' is now slower than 'equality' as more checks need to be done. The VM compiler can optimize 'identical' in the future.
My benchmarks do not show any slow down.
Review URL: https://codereview.chromium.org//11414136
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@15332 260f80e4-7a28-3924-810f-c04153c831b5
This optimization does not bring any benefit since all functions
have at least a call in the slow-path of the stack-overflow check.
I'm suggesting to remove support for this optimization since it
does not get triggered currently, therefore is also not tested.
Plus we have inlining now, which already makes calls to (small
enough) leaf functions fast.
Review URL: https://codereview.chromium.org//11086044
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13555 260f80e4-7a28-3924-810f-c04153c831b5
In case of a mixed integer/double operation that has a mint
input, the corresponding conversion from unboxed mint to unboxed
double was missing.
This CL converts a unboxed mint to a double by first boxing,
and then unboxing. Currently, we deoptimize in case the result
does not fit into a smi because we can only optimize mixed
smi/double operations and not yet mint/double operations.
BUG=dart:5721
TEST=tests/language/mint_arithmetic.dart
Review URL: https://codereview.chromium.org//11017017
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13372 260f80e4-7a28-3924-810f-c04153c831b5
johnmccutchan@google.com