Tail calls are matched on the graph, with a dedicated tail call
optimization that is actually testable. The instruction selection can
still fall back to a regular if the platform constraints don't allow to
emit a tail call (i.e. the return locations of caller and callee differ
or the callee takes non-register parameters, which is a restriction that
will be removed in the future).
Also explicitly limit tail call optimization to stubs for now and drop
the global flag.
BUG=v8:4076
LOG=n
Review URL: https://codereview.chromium.org/1114163005
Cr-Commit-Position: refs/heads/master@{#28219}
This CL contains the first steps towards tail call optimization:
* Structurally detect tail calls during instruction selection,
looking for special return/call combinations.
* Added new architecture-specific instructions for tail calls which
jump instead of call and take care of frame adjustment.
* Moved some code around.
Currently we restrict tail calls to callees which only use registers
for arguments/return value and to call sites which are explicitly
marked as being OK for tail calls. This excludes, among other things,
call sites in sloppy JS functions and our IC machinery (both need in
general to be able to access the caller's frame).
All this is behind a flag --turbo-tail-calls, which is currently off
by default, so it can easily be toggled.
Review URL: https://codereview.chromium.org/1108563002
Cr-Commit-Position: refs/heads/master@{#28150}
- allows the optimization of emitted gap move code since the representation of the value in the register is known
- necessary preparation for vector register allocation
- prepare for slot sharing for any value of the same byte width
TBR=jarin@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1111323003
Cr-Commit-Position: refs/heads/master@{#28140}
- allows the optimization of emitted gap move code since the representation of the value in the register is known
- necessary preparation for vector register allocation
- prepare for slot sharing for any value of the same byte width
BUG=
Review URL: https://codereview.chromium.org/1087793002
Cr-Commit-Position: refs/heads/master@{#28137}
- ConstantOperand was using a too-small field too store its virtual register
- drop ConvertTo, replace it with simple copy
- split AllocatedOperand off from Immediate and Constant to make assignment clearer, also paving the way for small Immediates
- put zone first in *Operand::New
- driveby: drop delayed ssa deconstruction experiment
R=titzer@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1050803002
Cr-Commit-Position: refs/heads/master@{#27692}
With this change, we remember the types of frame state inputs (in a new
operator, called TypedStateValues). Instead of inferring the value types
when building translations, we used the recorded types.
The original approach was not reliable because the passes after
simplified lowering can change node types, and this in turn confuses
the translation builder.
BUG=chromium:468727
LOG=n
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/1015423002
Cr-Commit-Position: refs/heads/master@{#27310}
A CompilationInfo constructed from just an Isolate* and a Zone* is in
weird an inconsistent state (calling e.g. flags() on it will crash),
so we need to avoid them. This CL removes almost all of them, the
remaining 2 call sites in (for testing only) will be handled in a
separate CL. Things which have been changed:
* Linkage is basically a decorator for CallDescriptor now.
* ChangeLowering doesn't need Linkage at all.
* JSGenericLowering doesn't need a full CompilationInfo*, just a
single flag.
* JSContextSpecializer doesn't need the full CompilationInfo, just a
Context.
* Removed unused CompilationInfo from SimplifiedLoweringTester.
This nicely decouples things already a bit more, but there's still
work to do...
Review URL: https://codereview.chromium.org/899803003
Cr-Commit-Position: refs/heads/master@{#26580}
Along the way:
- Thread isolate parameter explicitly through code that used to
rely on getting it from the zone.
- Canonicalize the parameter position of isolate and zone for
affected code
- Change Hydrogen New<> instruction templates to automatically
pass isolate
R=mstarzinger@chromium.org
LOG=N
Review URL: https://codereview.chromium.org/868883002
Cr-Commit-Position: refs/heads/master@{#26252}
- Use C++11 range based for loops.
- Remove duplicated virtual register set in unittests.
- Don't expose implementation details of InstructionSelector.
TEST=unittests
R=dcarney@chromium.org
Review URL: https://codereview.chromium.org/837423002
Cr-Commit-Position: refs/heads/master@{#25997}
This change enables non-tagged representations in FrameStates.
That allows us to run zlib with deoptimization support and have almost the same performance of the generated code (as the code with no deoptimization). Unfortunately, the frame states seem to confuse typer. As a consequence, we generate more representation changes, which in turn causes the scheduler to take a lot more time and memory (>4x). The added compiler time makes zlib with deopt be about 50% slower.
BUG=
R=titzer@chromium.org
Review URL: https://codereview.chromium.org/614713002
git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24454 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
