We have to reland these two commits at once, because the first breaks
some asm.js benchmarks without the second. The change was reverted
because of bogus checks in the verifier, which will not work in the
presence of OSR (and where hidden because of the type back propagation
hack in OSR so far). Original messages are below:
[turbofan] Add new JSFrameSpecialization reducer.
The JSFrameSpecialization specializes an OSR graph to the current
unoptimized frame on which we will perform the on-stack replacement.
This is used for asm.js functions, where we cannot reuse the OSR
code object anyway because of context specialization, and so we could as
well specialize to the max instead.
It works by replacing all OsrValues in the graph with their values
in the JavaScriptFrame.
The idea is that using this trick we get better performance without
doing the unsound backpropagation of types to OsrValues later. This
is the first step towards fixing OSR for TurboFan.
[turbofan] Perform OSR deconstruction early and remove type propagation.
This way we don't have to deal with dead pre-OSR code in the graph
and risk optimizing the wrong code, especially we don't make
optimistic assumptions in the dead code that leaks into the OSR code
(i.e. deopt guards are in dead code, but the types propagate to OSR
code via the OsrValue type back propagation).
BUG=v8:4273
LOG=n
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1226673005
Cr-Commit-Position: refs/heads/master@{#29486}
`WriteUtf16Slow` should not assume that the output buffer has enough
bytes to hold both words of surrogate pair. It should pass the number of
remaining bytes to the `Utf8::ValueOf` instead, just as we already do in
`Utf8DecoderBase::Reset`. Otherwise it will attempt to write the trail
uint16_t past the buffer boundary, leading to memory corruption and
possible crash.
Originally reported by: Kris Reeves <kris.re@bbhmedia.com>
BUG=v8:4274
R=danno
R=svenpanne
LOG=y
Review URL: https://codereview.chromium.org/1226493003
Cr-Commit-Position: refs/heads/master@{#29485}
This way we don't have to deal with dead pre-OSR code in the graph and
risk optimizing the wrong code, especially we don't make optimistic
assumptions in the dead code that leaks into the OSR code (i.e. deopt
guards are in dead code, but the types propagate to OSR code via the
OsrValue type back propagation).
BUG=v8:4273
LOG=n
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1215333005
Cr-Commit-Position: refs/heads/master@{#29478}
The JSFrameSpecialization specializes an OSR graph to the current
unoptimized frame on which we will perform the on-stack replacement.
This is used for asm.js functions, where we cannot reuse the OSR code
object anyway because of context specialization, and so we could as well
specialize to the max instead.
It works by replacing all OsrValues in the graph with their values in
the JavaScriptFrame.
The idea is that using this trick we get better performance without
doing the unsound backpropagation of types to OsrValues later. This is
the first step towards fixing OSR for TurboFan.
R=jarin@chromium.org
BUG=v8:4273
LOG=n
Review URL: https://codereview.chromium.org/1225683004
Cr-Commit-Position: refs/heads/master@{#29476}
The context constant cannot be materialized from the frame when we are
compiling for OSR, because the context spill slot contains the current
instead of the outermost context in full-codegen.
R=bmeurer@chromium.org
Review URL: https://codereview.chromium.org/1220013003
Cr-Commit-Position: refs/heads/master@{#29472}
This changes the OsrValue insertion in the AstGraphBuilder to emit a
proper OsrValue instead of a special Parameter for the inner context
value at the OSR entry point.
R=titzer@chromium.org
Review URL: https://codereview.chromium.org/1213043005
Cr-Commit-Position: refs/heads/master@{#29471}
Additionally speed up instantiation of ObjectTemplates by preallocating enough space in the descriptor arrays
BUG=v8:4184
LOG=n
Review URL: https://codereview.chromium.org/1218403002
Cr-Commit-Position: refs/heads/master@{#29468}
Currently we lower shifts directly to machine operators, and add an
appropriate Word32And to implement the & 0x1F operation on the right
hand side required by the specification. However for Word32And we assume
Int32 in simplified lowering, which is basically changes the right hand
side bit interpretation for the shifts from Uint32 to Int32, which is
obviously wrong. So now we represent that explicitly by proper
simplified operators for the shifts, which are lowered to machine in
simplified lowering.
R=jarin@chromium.org
Review URL: https://codereview.chromium.org/1213803008
Cr-Commit-Position: refs/heads/master@{#29465}
This makes sure that the GC fully flushes the optimized code map when
the serializer is preparing a snapshot. Otherwise closures and contexts
could end up in the startup snapshot.
R=hpayer@chromium.org
TEST=cctest/test-serialize/SerializeInternalReference
Review URL: https://codereview.chromium.org/1215063007
Cr-Commit-Position: refs/heads/master@{#29461}
Keeping this CL separate in case there are more GC-stress problems.
BUG=v8:3956
LOG=N
Review URL: https://codereview.chromium.org/1217543006
Cr-Commit-Position: refs/heads/master@{#29449}
