Commit Graph

151 Commits

Author SHA1 Message Date
whesse@chromium.org
20a6e7e5e9 Rewrite of VisitCountOperation that should speed it up
Review URL: http://codereview.chromium.org/56151

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1677 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-04-03 13:27:14 +00:00
lrn@chromium.org
5dc51480c5 Quick pointer comparison, removed undetectable tests.
Special case for NaN in equality test.


git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1676 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-04-03 12:44:45 +00:00
lrn@chromium.org
3e55df5a57 Tracks static type of strings in frame elements and results.
Uses static string type to optimize string additions.


git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1675 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-04-03 12:01:56 +00:00
kmillikin@chromium.org
06860982a9 Clean up return statements in the code generator by explicitly
counting the reference to the return value and passing it to the
return label.  This requires threading it through try/catch and
try/finally.  The return value is loaded into eax more lazily than
before.

Also, perform some related refactoring of jump targets.
Review URL: http://codereview.chromium.org/56172

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1669 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-04-02 08:58:18 +00:00
ager@chromium.org
deb0d4cc37 Revert change 1656 which was unsafe until it can be properly fixed.
TBR=whesse@chromium.org
Review URL: http://codereview.chromium.org/60010

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1666 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-04-01 10:35:20 +00:00
whesse@chromium.org
e2ed40a1eb Keep the result of postfix increment and decrement in a register.
Review URL: http://codereview.chromium.org/56106

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1656 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-31 15:03:02 +00:00
whesse@chromium.org
952179e323 Formatting changes and some slight improvements to revision 1621, issue 42006.
Review URL: http://codereview.chromium.org/51007

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1634 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-27 14:35:24 +00:00
kmillikin@chromium.org
b8af68b03f Make VirtualFrame::CallStub on IA32 responsible for moving arguments
into the appropriate registers for the stub call.  Since stubs that
take arguments in registers do not (currently) take any arguments on
the stack, the unused generality has been removed (it's easy to put
back in later).
Review URL: http://codereview.chromium.org/55003

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1629 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-27 10:29:37 +00:00
whesse@chromium.org
d970e04e04 Optimize binary operations in which one or both operands is a constant smi.
Review URL: http://codereview.chromium.org/42006

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1621 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-26 13:14:10 +00:00
kmillikin@chromium.org
f8bf152c89 Change the arguments access stub to take both arguments in registers
rathern than one in a register and one on the stack.

Fix a stack height bug in the case that the arguments access stub hit
its slow case.
Review URL: http://codereview.chromium.org/49037

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1620 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-26 13:02:00 +00:00
kmillikin@chromium.org
4dceb3c87f In the IA32 code genrator, handle call ICs and constructor calls the
same as load and store ICs.  Eliminate the general function used to
call any IC.
Review URL: http://codereview.chromium.org/42638

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1619 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-26 13:00:03 +00:00
kmillikin@chromium.org
03419c8369 Simplify the way the code generator handles calls to IC stubs. Before
we dispatched on the IC stub kind in a generic CallCodeObject
function.  Now, we have special functions for the load and store IC
stubs.

We also (for the load and store ICs) handle moving register arguments
into place only after the stack is prepared for the call.  This
replaces some memory-to-memory moves (for copies whose backing store
is occupied by a register needed for the arguments) with
memory-to-register moves.
Review URL: http://codereview.chromium.org/42602

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1613 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-25 14:08:41 +00:00
kmillikin@chromium.org
7c35a5d820 The code generator doesn't need to put the argument count in eax for
call ICs.  Refactor CallRuntime on ARM to match the structure of IA32.
Update some comments.
Review URL: http://codereview.chromium.org/42598

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1607 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-25 11:05:36 +00:00
whesse@chromium.org
36853c83d1 Remove register counts from VirtualFrame, use register indices instead.
Review URL: http://codereview.chromium.org/42565

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1605 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-25 10:00:52 +00:00
kmillikin@chromium.org
f38dc917b0 Make the "has function literal" default value for loops conservative.
It defaults to true so that if the analysis is not run we get a safe
analysis result.

Review URL: http://codereview.chromium.org/42562

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1600 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-24 14:25:22 +00:00
kasperl@chromium.org
70aa1fde83 Fix ARM performance regression on constructing array
and object literals by using the same logic as we use
on IA-32.
Review URL: http://codereview.chromium.org/53001

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1587 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-24 08:55:11 +00:00
kmillikin@chromium.org
5e2f3adfd0 Fix issue 284.
The problem was continuing out of the body of a for/in (where we do
register allocation) to the loop update (where we do not).  Variables
allocated to registers where not preserved.

Review URL: http://codereview.chromium.org/53002

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1585 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-24 08:29:24 +00:00
olehougaard
a8c50151fc Reapply revisions 1432, 1433, 1469 and 1472 while fixing issue 279.
Review URL: http://codereview.chromium.org/48006

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1571 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-23 07:27:47 +00:00
kmillikin@chromium.org
6565243504 Inline the Result::Unuse member function and change the Result
destructor to no longer invalidate the result.

Review URL: http://codereview.chromium.org/50051

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1560 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-20 12:58:14 +00:00
kmillikin@chromium.org
9752bf90b5 Remove a bunch of unnecessary includes from header files in favor of
forward declarations.

Review URL: http://codereview.chromium.org/42389

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1559 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-20 09:35:31 +00:00
erik.corry@gmail.com
3aa57f7f65 Revert 1432, 1433, 1469 and 1472 due to a bug with literal objects.
Review URL: http://codereview.chromium.org/46088

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1511 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-15 16:18:20 +00:00
kmillikin@chromium.org
30af089bc5 Fix a test in the IA32 code generator for whether loop conditions
contain function literals.  The test was not conservative enough.

Review URL: http://codereview.chromium.org/42138

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1504 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-12 19:49:19 +00:00
kmillikin@chromium.org
03d7c6bd1d Reenable the duplicated test at the bottom of for and while loops. It
is enabled for a loop by the absence of function literal
subexpressions in the condition expression.

Review URL: http://codereview.chromium.org/42127

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1503 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-12 15:44:05 +00:00
kmillikin@chromium.org
3c41aab15d Ensure deallocation of the dynamically allocated shadow targets used
for compilation of try/finally, even in the case where control cannot
reach the end of the finally block.

Review URL: http://codereview.chromium.org/43071

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1490 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-11 13:14:27 +00:00
kmillikin@chromium.org
1a7359fb9d Deallocate the dynamically-allocated shadow targets that we use for
compilation of try/catch and try/finally.
Review URL: http://codereview.chromium.org/42068

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1489 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-11 12:08:28 +00:00
kmillikin@chromium.org
82df32b78c Fix issue 268 by explicitly calling Unuse (to deallocate all contained
virtual frames) on zone-allocated jump targets.  These include jump
targets in AST nodes and the entry and exit targets of deferred code.

See http://code.google.com/p/v8/issues/detail?id=268

Review URL: http://codereview.chromium.org/42067

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1486 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-11 10:25:48 +00:00
kmillikin@chromium.org
0e60cf38dc Fix issue 265 by handling extra statement state on the frame based on
the expectation at the break, continue, and return labels (including
shadowed ones) instead of based on the AST nodes.

See http://code.google.com/p/v8/issues/detail?id=265

Review URL: http://codereview.chromium.org/42017

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1482 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-11 06:17:19 +00:00
lrn@chromium.org
34db0ff86c Issue 267: Calls to arguments in eval-tainted function scope uses global object as receiver.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1476 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-10 12:28:34 +00:00
kmillikin@chromium.org
773843b62e Introduce a BreakTarget subclass of JumpTarget used to represent the
blocks labeled by "break", "continue", and "return".  BreakTargets are
the only jump targets that appear in the AST, the only ones that can
be uninitialized, and the only ones that can be shadowed.
Review URL: http://codereview.chromium.org/42008

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1475 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-10 12:11:56 +00:00
olehougaard
860508f3e2 Optimizing generation of nested literals for both object and array literals.
Review URL: http://codereview.chromium.org/40295

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1469 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-10 08:59:00 +00:00
kmillikin@chromium.org
34af9f2ecf Work around issue 260 for now by disabling duplication of the loop
condition for while and for loops.

http://code.google.com/p/v8/issues/detail?id=260

Review URL: http://codereview.chromium.org/40294

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1453 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-09 14:12:20 +00:00
bak@chromium.org
9c96c68ab4 Reduced the code sequence for testing for object type.
We now use cmpb instead of movzx_b and cmp.



git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1452 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-09 14:00:51 +00:00
kmillikin@chromium.org
ece2c03160 Fix issue 263:
http://code.google.com/p/v8/issues/detail?id=263

Sharing the code to unlink the exception handler for a try/finally
causes us to try to merge virtual frames with different heights (due
to statements that keep state on the stack) at the entry to the unlink
code.

Avoid this by unlinking the handler separately for each exit from the
try block.
Review URL: http://codereview.chromium.org/39331

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1449 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-09 10:51:57 +00:00
kmillikin@chromium.org
182c3ebb2d Reapply r1434 and port to ARM.
Review URL: http://codereview.chromium.org/40220

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1441 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-06 14:18:03 +00:00
christian.plesner.hansen@gmail.com
2a2c9381ea Reverted r1434
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1437 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-06 11:59:53 +00:00
kmillikin@chromium.org
b638d5c613 Fix issue 259.
http://code.google.com/p/v8/issues/detail?id=259

Due to constant folding of loop conditions, it is possible to have a
try/finally with no exits from the try block (eg, an infinite loop in
the try block).  In that case the code to unlink the try handler is
dead and should not be emitted.

Review URL: http://codereview.chromium.org/39251

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1434 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-06 10:18:33 +00:00
olehougaard
d46a9900ee Optimizing nested, constant object literals (like JSON objects) by building one large object template for the entire object instead of one for each sub-object.
Review URL: http://codereview.chromium.org/39184

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1432 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-06 09:38:17 +00:00
kmillikin@chromium.org
870b6c46fa Fix a few indentation issues.
Review URL: http://codereview.chromium.org/40102

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1413 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-04 10:13:17 +00:00
kmillikin@chromium.org
fa5bc7ea0b Streamline CodeGenerator::Comparison in the IA32 code generator. When
doing constant folding, handle only the cases that actually arise.
This avoids some complicated logic, some extra comparisons, and fixes
issue 247:

http://code.google.com/p/v8/issues/detail?id=247

Review URL: http://codereview.chromium.org/40023

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1410 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-04 08:50:38 +00:00
kasperl@chromium.org
9f78714717 Disable the compilation caching for scripts to make it
possible to start generating code specific to a global
object. We should start caching scripts but make the
mechanism local to a specific global context inspired
by the way we handle eval-caching. 
Review URL: http://codereview.chromium.org/40007

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1406 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-03 13:35:05 +00:00
kmillikin@chromium.org
4678719d24 Simplify the way that we compile slow-case switch statements. Compile
all the reachable tests first, and then all the reachable bodies.
Review URL: http://codereview.chromium.org/28296

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1397 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-03-02 11:30:17 +00:00
kmillikin@chromium.org
f4735247cf Merge from experimental code generator branch to bleeding edge.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1389 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-27 13:00:32 +00:00
olehougaard
e3c0ae62af Go into slow case when encountering object initialization on the top level to optimize performance of code like
C.prototype.x = ...;
C.prototype.y = ...;
...
C.prototype.z = ...;
Review URL: http://codereview.chromium.org/27128

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1372 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-26 07:54:22 +00:00
kasperl@chromium.org
1e42e70021 Use SSE3 instructions - if available - for faster To(U)Int32
conversion in bit operation stubs. Disable serialization
support by default to allow us to use SSE3 instructions and
faster write barrier code when running without snapshot.
Review URL: http://codereview.chromium.org/27046

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1364 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-25 14:57:46 +00:00
ager@chromium.org
0c7af39737 Speed up access to global variables from eval scopes. Traverse the
surrounding context to figure out if the variable could be global.  If
the variable could be global we check context extension objects at
runtime and use a global LoadIC if no variables have been introduced
by eval.

Fix crash bug when loading function arguments from inside eval.  The 
shadowed variable in the DYNAMIC_LOCAL case does not rewrite to a slot in 
that case.

Review URL: http://codereview.chromium.org/28027

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1348 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-24 13:11:53 +00:00
kasperl@chromium.org
0e1819d0b8 Revert 1330:1331.
TBR=ager@chromium.org
Review URL: http://codereview.chromium.org/24026

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1332 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-20 17:42:04 +00:00
ager@chromium.org
b6a1cd6787 Simplify non-smi bit operation optimization. The explicit range
checks are not needed.  If we fail to store the float as a 64-bit
integer, we just go to the runtime system.  The only extra case in
which we enter the runtime system is for NaN which should not matter.

Review URL: http://codereview.chromium.org/21539

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1331 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-20 14:23:54 +00:00
kasperl@chromium.org
3afe470026 Improve generated code for non-smi bit operations by
converting the operands to 32-bit ints directly in the 
machine code stub instead of just checking if the heap 
numbers really just represent 32-bit ints and going
through the runtime system in that case.
Review URL: http://codereview.chromium.org/21536

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1330 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-20 13:30:32 +00:00
ager@chromium.org
195753ff7b Commit of eval changes left out because they were stuck in another changelist.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1307 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-19 07:22:53 +00:00
ager@chromium.org
cafe25f258 ARM side of load optimization in the presence of eval.
Review URL: http://codereview.chromium.org/20453

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1301 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-02-18 13:54:13 +00:00