In this design maps contain descriptor arrays, which in turn can contain transition arrays. If transitions are needed when no descriptor array is present, a descriptor array without real descriptors is inserted just so it can point at the transition array.
The transition array does not contain details about the field it transitions to. In order to weed out transitions to FIELDs from CONSTANT_FUNCTION (what used to be MAP_TRANSITION vs CONSTANT_TRANSITION), the transition needs to be followed and the details need to be looked up in the target map. CALLBACKS transitions are still easy to recognize since the transition targets are stored as an AccessorPair containing the maps, rather than the maps directly.
Currently AccessorPairs containing a transition and an accessor are shared between the descriptor array and the transition array. This simplifies lookup since we only have to look in one of both arrays. This will change in subsequent revisions, when descriptor arrays will become shared between multiple maps, since transitions cannot be shared.
Review URL: https://chromiumcodereview.appspot.com/10697015
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11994 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Only handles when x is int32 and y is int32 constant.
BUG=v8:2038
Currently implemented by imul (not fpmul).
x86 and x64 algorithm differs a bit.
x86 implementation is kind of cumbersome, but I couldn't think of better ways.
Review URL: https://chromiumcodereview.appspot.com/10382033
Patch from Zheng Liu <zheng.z.liu@intel.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11887 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
- Extend HValue interface to allow splitting observed input representation (comming from type feedback) from required input representation (dictated by instruction itself). Currently all instructions except for bitwise binary operations have this representations match. For bitwise binary operations hydrogen builder unconditionaly forces Integer32 representation for those operations that have Double type feedback. Thus causing representation inference to incorrectly count such uses as Integer32 instead of Double. This change also prepares for more fine grained type feedback for inputs of binary operations.
- For phies that are not convertable to Integer32 discard direct and indirect use count of Integer32 type to avoid propagation of these uses to connected phies.
R=jkummerow@chromium.org
BUG=v8:2096
Review URL: https://chromiumcodereview.appspot.com/10540049
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11737 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
tells us that a map can transition to another map when we are generating
code for load operations. This may cause us to deopt if the same routine
is seeing different maps caused by branching in constructors. If so, I
have a different change that is around 100 times more complicated that
lets us generated Crankshaft code for negative lookups.
Review URL: https://chromiumcodereview.appspot.com/10306010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11524 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
As a special case, for checks on index expressions with the form (expr + constant) if a smaller constant is checked later in the DT also eliminate the check.
Finally, if a larger constant is checked later in the same BB do the more general check (larger constant) earlier instead of the less general one.
This will not cause useless deoptimizations because, since we are in the same BB, all the checks would have been executed anyway.
BUG=
TEST=
Review URL: https://chromiumcodereview.appspot.com/10032029
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11437 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This change allows hydrogen instructions to keep track of instructions
that dominate certain side-effects (GVN flags) in the hydrogen graph. We
use the GVN pass to keep track of side-effects because accurate flags
are already in place.
It also adds a new side-effect (kChangesNewSpacePromotion) indicating
whether an instruction can cause a GC and have objects be promoted to
old-space. An object allocated in new-space is sure to stay on paths not
having said side-effect.
R=erik.corry@gmail.com
TEST=mjsunit/compiler/inline-construct
Review URL: https://chromiumcodereview.appspot.com/10031031
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11270 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This CL is a step towards removing ZoneObject's new operator without a Zone
parameter, which uses Isolate::Current. For e.g. the bulletben benchmark, this
CL reduces the number of calls to this new operator by roughly 120k, but we are
still left with 780k calls from other sites...
Review URL: https://chromiumcodereview.appspot.com/9487010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10860 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Only JSObject enumerables with enum cache (fast case properties, no interceptors, no enumerable properties on the prototype) are supported.
HLoadKeyedGeneric with keys produced by for-in enumeration are recognized and rewritten into direct property load by index. For this enum-cache was extended to store property indices in a separate array (see handles.cc).
New hydrogen instructions:
- HForInPrepareMap: checks for-in fast case preconditions and returns map that contains enum-cache;
- HForInCacheArray: extracts enum-cache array from the map;
- HCheckMapValue: map check with HValue map instead of immediate;
- HLoadFieldByIndex: load fast property by it's index, positive indexes denote in-object properties, negative - out of object properties;
Changed hydrogen instructions:
- HLoadKeyedFastElement: added hole check suppression for loads from internal FixedArrays that are knows to have no holes inside.
R=fschneider@chromium.org
BUG=
TEST=
Review URL: https://chromiumcodereview.appspot.com/9425045
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10794 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This extends the current support for nested object literals we already
have in Crankshaft, to also support nested array literals and mixed
nested literals containing arrays and objects. All three types are
generated by the unified HFastLiteral instruction.
All previous upper bounds on nested literal graphs remain unchanged,
keeping the size of generated code in check.
The main intention is to boost performance of two-dimensional array
literals containing constant elements (aka. matrices).
R=danno@chromium.org
TEST=mjsunit/compiler/literals-optimized
Review URL: https://chromiumcodereview.appspot.com/9403018
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10734 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Yak shaving for map sharing with accessor properties contd.: When CALLBACKS can
have map transitions, simply looking at the property type is not sufficient
anymore to decide if a property is there or not. One has to look at the actual
contents of the descriptor entry then, but this breaks down sometimes when the
lookup is being done with a NULL holder. Luckily enough, we can oftren replace
IsProperty by the simpler IsFound, because we inspect the type immediately
afterwards, anyway.
Review URL: https://chromiumcodereview.appspot.com/9280007
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10474 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This generates optimized code for deep-copying of nested object literal
boilerplates which are statically known. Most of the boilerplates have
already been generated at crankshaft time, so this optimization should
kick in for virtually every object literal. Only nested object literal
graphs up to a certain depth and containing up to a certain total number
of properties are considered for this optimization. This will prevent
explosion of code size due to large object literals (e.g. eval on JSON).
Improves splay performance because object literals are created often.
R=fschneider@chromium.org
Review URL: http://codereview.chromium.org/8640001
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10061 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This time, we initially leave the HTypeof instruction in the Hydrogen graph,
even for the special cases. We later try to remove this instruction (and any
HConstant) in the canonicalization pass, if possible. Always removing the
HTypeof during the initial graph construction is wrong if e.g. it is used in an
HSimulate.
The removals can be generalized a bit, but this will happen in a separate CL.
TEST=mjsunit/optimized-typeof.js
Review URL: http://codereview.chromium.org/8334021
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9688 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Currently we avoid checking for the hole value after array loads, if the
result is only used by instructions that definitely deoptimize in case
of the hole value (HChange instructions).
This change performs the same procedure for loading from deleteable/read-only
global variable where we can also avoid the check in the same cases.
Review URL: http://codereview.chromium.org/8054008
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9453 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This improves our static type information by calculating the result type
of conversions (HChange) during range analysis. It allows e.g. to eliminate
the write barrier in the following example where it was not possible before:
function f(x) {
var y = x + 1;
if (y > 0 && y < 100) {
a[0] = y;
}
}
* Fix bug in Range::Copy. The minus-zero flags has to be preserved by default.
Review URL: http://codereview.chromium.org/7634022
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8994 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Some AST nodes (Property, Call, etc.) store either a list of receiver
types or a monomorphic receiver type. This patch merges the two fields
using a small pointer list. GetMonomorphicReceiverType() is now a
purely convenience function returning the first and only recorded
type.
This saves about 500K (of about 39M) on average when compiling V8
benchmark as measured by a simple patch adding a zone allocation
counter (https://gist.github.com/1149397).
R=kmillikin@chromium.org
Review URL: http://codereview.chromium.org/7655017
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8993 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
It is available platforms that have SSE 4.1 and allows us to handle
negative numbers without deoptimization. Before we would deoptimize
on negative inputs to Math.floor. x64 already uses this instruction.
* Change Math.floor unit test to make sure every test case gets
optimized by changing the source code for each test case.
* Fix HIR debug printing for some instructions.
Review URL: http://codereview.chromium.org/7628017
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8921 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
* Bug fix for range analysis (contributed by Andy Wingo). Ranges of
double values have to include negative zero. Original code review:
http://codereview.chromium.org/7514040/
* Fix a bug in optimized Math.round on ARM. When emitting minus-zero checks
we previously return a wrong result because of incorrect register assignment.
* Fix performance problem in IA32 and x64. Refine the checks
for minus zero and avoid unnecessary deoptimizations on Math.floor.
* Improve mjsunit test for Math.round to make sure we also
get the optimized version of the code for each test case.
Review URL: http://codereview.chromium.org/7604028
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8877 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
* src/hydrogen-instructions.cc (HValue::InferRange): Only mark values
with int32 representation as never being -0. Always return a non-NULL
value; callers should check for representation().IsNone() if that's
their concern.
In practice these invariants were not violated by callers, but they
were sometimes two calls away, which seems brittle.
BUG=
TEST=tests pass, modulo http://code.google.com/p/v8/issues/detail?id=1572
Review URL: http://codereview.chromium.org/7514040
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8804 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
If the input range is positive and the shift count is constant
we can replace >>> with >> to compute the output range.
For negative inputs, we can only compute a range if the
result always fits into a signed int32.
BUG=v8:1510
Review URL: http://codereview.chromium.org/7489043
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8735 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Eliminates the enum flag RESTORE_CONTEXT and CONTEXT_ADJUSTED, and adds a context HValue and LOperand to many hydrogen and lithium instructions.
Context is still used from the stack from in CallKnownFunction (this seems safe), and in CallRuntimeFromDeferred in lithium-codegen-ia32.cc, which needs to be fixed.
BUG=
TEST=
Review URL: http://codereview.chromium.org/7132002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8529 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This change makes the number of successors of a control instruction
configurable with a template parameter and changes the existing instructions
to use it.
To iterate over all successors I added an iterator instead of always calling
First- and SecondSuccessor.
Review URL: http://codereview.chromium.org/7114004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8262 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
If type-feedback indicates that an expression was never executed in
the non-optimized code, we insert a forced deoptimization right away
to enable re-optimization if we ever hit this path.
With this change we still continue to build the graph. As a next step, we
should remove the dead code after the deoptimize.
I had to remove one assert about the optimization status in a test since
we now immediately deoptimize after exiting the loop that triggers OSR.
Also remove a restriction that control-flow from an inlined function in a
test context always reaches both true- and false-target.
Review URL: http://codereview.chromium.org/7105015
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8140 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
- Introduce a class JSReceiver, that is a common superclass of JSObject and
JSProxy. Use JSReceiver where appropriate (probably lots of places that we
still have to migrate, but we will find those later with proxy test suite).
- Move appropriate methods to JSReceiver class (SetProperty,
GetPropertyAttribute, Get/SetPrototype, Lookup, and so on).
- Introduce new JSFunctionProxy subclass of JSProxy. Currently only a stub.
- Overhaul enum InstanceType:
* Introduce FIRST/LAST_SPEC_OBJECT_TYPE that ranges over all types that
represent JS objects, and use that consistently to check language types.
* Rename FIRST/LAST_JS_OBJECT_TYPE and FIRST/LAST_FUNCTION_CLASS_TYPE
to FIRST/LAST_[NON]CALLABLE_SPEC_OBJECT_TYPE for clarity.
* Eliminate the overlap over JS_REGEXP_TYPE.
* Also replace FIRST_JS_OBJECT with FIRST_JS_RECEIVER, but only use it where
we exclusively talk about the internal representation type.
* Insert JS_PROXY and JS_FUNCTION_PROXY in the appropriate places.
- Fix all checks concerning classification, especially for functions, to
use the CALLABLE_SPEC_OBJECT range (that includes funciton proxies).
- Handle proxies in SetProperty (that was the easiest part :) ).
- A few simple test cases.
R=kmillikin@chromium.org
Review URL: http://codereview.chromium.org/6992072
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8126 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This allows us to get rid of totally fake LAST_STRING_TYPE and makes
it possible to test for symbols.
I considered splitting HCheckInstanceType into two instructions, but
it seems nice to be able to hide the instance type implementation
details from the hydrogen level.
Review URL: http://codereview.chromium.org/6964011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7840 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
If the result of an fast elements load is converted to an untagged
representation we can omit the hole check if the value is not used
anywhere else except for HChange instructions converting it to
an untagged representation since those will deoptimize for the hole
value anyway.
Review URL: http://codereview.chromium.org/6964012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7827 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Rather than representing a use as a pointer to an HValue and then searching
for the specific (ambiguous) operand, we now represent a use as a pair of an
HValue and the input operand index. Additionally, use a linked list instead
of a growable array list since we never use random access.
This allows us to remove a bunch of similarly named and subtly different
functions from the HValue API. The cost in extra zone allocation per use is
partially offset by reusing use list nodes when replacing a use of one value
with another.
R=danno@chromium.org,fschneider@chromium.org
Review URL: http://codereview.chromium.org/6881044
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7674 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Instead of using virtual type-tester functions we can just
generate non-virtual ones for all concrete IR classes.
This is changes reduces the V8 binary size by ~2%.
I also simplified the macros to declare new hydrogen instructions slightly.
The name used for debug output is no longer passed as a separate string.
Instead we just use the class name.
Review URL: http://codereview.chromium.org/6880014
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7659 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
In the cases where a global property cell cannot be used in the optimized code
use standard load ic to get the property instead of bailing out.
This is re-committing r7212 and r7215 which where reverted in r7239 with the addition of recoring the source position in the hydrogen code for the LoadGlobalCell instruction. To record that position an optional position field has been added to the variable proxy AST node.
Review URL: http://codereview.chromium.org/6758007
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7474 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This commit adds current working versions of assembler, macro-assembler,
disassembler, and simulator.
All other mips arch files are replaced with stubbed-out versions that
will build.
Arch independent files are updated as needed to support building and
running mips.
The only test is cctest/test-assembler-mips, and this passes on the
simulator and on mips hardware.
TEST=none
BUG=none
Patch by Paul Lind from MIPS.
Review URL: http://codereview.chromium.org/6730029/
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7388 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This change adds a new IR instruction for polymorphic loads. It performs
map compares and loads in one IR instruction instead of splitting each
load into a graph of map-compares and field loads.
The advantage is a smaller IR and less basic blocks, plus it allows to
do GVN on polymorphic loads.
Review URL: http://codereview.chromium.org/6708085
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7336 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Do not use subgraphs to implement the translation of simple branching
control flow, for the function body entry, or for labeled blocks.
Combine all the loop construction functions into a single one.
Resolve a possible problem with duplicate AST IDs used both for joined loop
break blocks, the normal loop exit, and for their common successor, by never
introducing the extra (successor) block and instead making the normal loop
exit a predecessor of the break join block. There is a similar issue with
joined continue blocks.
Remove a (never needed) two-element zone list per each time we replace one
hydrogen value with another.
Review URL: http://codereview.chromium.org/6604002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7027 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Rather than passing in a pair of unequal-length lists, pass the default
subgraph separately. Construct the typecase from the top down rather than
the bottom up, so it doesn't need an intermediate zone list.
Also, change a basic block's 'last' instruction field to really be its last
instruction by correctly updating it when inserting and removing
instructions.
Review URL: http://codereview.chromium.org/6516016
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6783 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Before, Hydrogen call instructions had uses of the PushArgument instructions
for their arguments. These operands were unneeded, bloated the IR, and
caused calls to be the only Hydrogen instructions with an unpredictable
number of operands.
Now, PushArgument is a pure side-effecting instruction that has no uses.
Review URL: http://codereview.chromium.org/6480030
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6749 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Make esi available to the register allocator rather than dedicating it
permanently to the context.
The context is still passed in register esi to JavaScript and to the runtime
as part of the calling convention. Because some stubs might end up calling
JS or the runtime, it is also conservatively passed to stubs.
Roughly half the calls have been modified to use the context as an input
value in fixed register esi. The other half are marked as calls or deferred
code so esi is spilled and can be explicitly set.
It is no longer necessary to restore the context to esi after a call that
might change it.
Review URL: http://codereview.chromium.org/6452001
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6713 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
When constructing stack traces we interpret the deoptimization data for
optimized frames to find the receiver value. This value could sometimes be
eliminated from the deoptimization data if we though it was unused.
BUG=v8:1118
Review URL: http://codereview.chromium.org/6465023
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6694 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
1. The placement of checks for negative zero has to be computed after
all conversion instructions have been inserted. I separated the code
into its own phase.
2. GVN need to take instruction flags into account when comparing
instructions for redundancy.
Review URL: http://codereview.chromium.org/6260035
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6534 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
It is a type error to treat a flag (an enum) as if it were a flag mask (an
int derived from shifting a bit by the enum value). It is error prone to
have functions that take flag mask arguments because they will silently
accept flags.
Review URL: http://codereview.chromium.org/6373011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6500 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The former default was true. The new default is false and the default
implementation is UNREACHABLE so it asserts in debug builds. The function
is overridden in all concrete instruction classes that might have the flag
kUseGVN set.
Review URL: http://codereview.chromium.org/6255013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6493 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Rename HBranch (the instruction that coerces an arbitrary HValue to
control flow) to HTest to free up the term Branch to refer to any
control instruction with two successors.
Change the virtual FirstSuccessor and SecondSuccessor functions on
control instructions to a pair of data members.
Review URL: http://codereview.chromium.org/6366002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6417 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This prevents code motion from hoisting map-checks across such stores
which may result in unnecessary deoptimizations.
In the following example program we would move a map-check from the inner loop out before the outer loop which is not desirable:
function f() {
var o = {};
var j = 0;
o.a = 1;
do {
o.b = 6; // Map transition
for (var i=0; i<10; i++) {
o.a = o.b + i;
}
} while(++j < 1) {}
}
for (var i = 0; i < 1000000; i++) f();
Review URL: http://codereview.chromium.org/5991001
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6071 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
