AuroraMiddleware/v8
1
0
Fork 0
Code Issues 2 Pull Requests Releases Wiki Activity
19,285 Commits 1 Branch 0 Tags 839 MiB
d7b5cd0817
Commit Graph

27 Commits

Author SHA1 Message Date
sigurds@chromium.org
4dc4b40925 Reland "Fix scheduler to correctly schedule nested diamonds". 
Reland fix: Consume less memory.

R=mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/636233006

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24597 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-10-14 12:08:55 +00:00
sigurds@chromium.org
aa67d12f09 Revert "Fix scheduler to correctly schedule nested diamonds." 
This reverts commit 0ab306041a6eba99fe5e2ef78251de137321f559.

"Fix" make scheduler run out of memory.

TBR=mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/656473002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24566 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-10-13 16:08:29 +00:00
sigurds@chromium.org
0f5949d605 Fix scheduler to correctly schedule nested diamonds. 
The scheduler rewires control based on the last *control*
node that appears in the schedule of a block. This is not
sufficient to account for dependencies.

This patch adds additional dependencies to floating control
nodes. Given a floating control node A, every non-control
dependency of every node B that depends on A is introduces
as an additional dependency of A.

This allows the scheduler to correctly schedule two
diamonds A, B, if their only correct schedule is to
schedule B into the ifTrue successor in A.

TEST=cctest/test-scheduler/NestedFloatingDiamonds
R=mstarzinger@chromium.org, titzer@chromium.org

Review URL: https://codereview.chromium.org/602083003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24561 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-10-13 13:07:49 +00:00
bmeurer@chromium.org
a17289f4bc [turbofan] Drop broken StaticParameterTraits. 
The StaticParameterTraits are broken by design, and cause way too much
trouble. The compilers usually pick the wrong specialization (i.e. the
default specialization is picked for Load and Phi even tho there is a
specialization for MachineType), which is not only the reason why GVN is
ineffective and slow, but can also lead to correctness issues in some
rare cases.

Also clean up some minor bugs/inconsistencies on the way.

TEST=cctest,unittests
R=svenpanne@chromium.org

Review URL: https://codereview.chromium.org/636893002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24437 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-10-07 13:30:28 +00:00
bmeurer@chromium.org
3eebdc3264 Replace OStream with std::ostream. 
Review URL: https://codereview.chromium.org/618643002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24319 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-30 10:29:32 +00:00
jarin@chromium.org
6a07181cde Refactor BasicBlock to not use GenericNode. 
To manage BasicBlock's predecessors and successors we now use plain
std::vector.

The change also moves bunch of method definitions from header files
to implementation files.

In zlib, the change brings 3x improvement in the scheduler's memory
consumption. The --turbo-stats flag says we go 169MB -> 55MB in
the scheduler, 383MB -> 268MB overall.

BUG=
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/606403003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@24308 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-30 08:23:20 +00:00
bmeurer@chromium.org
bfa3884ae0 [turbofan] Machine operators are globally shared singletons. 
TEST=compiler-unittests,cctest
R=svenpanne@chromium.org

Review URL: https://codereview.chromium.org/547233003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23864 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-11 10:37:49 +00:00
bmeurer@chromium.org
6bb08db27e [turbofan] Next step towards shared operators. 
TEST=compiler-unittests,cctest
R=svenpanne@chromium.org

Review URL: https://codereview.chromium.org/555283004

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23837 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-10 12:23:45 +00:00
bmeurer@chromium.org
99ccab81e2 [turbofan] Add MachineType to Phi. 
TEST=cctest,compiler-unittests,mjsunit
R=mstarzinger@chromium.org, titzer@chromium.org

Review URL: https://codereview.chromium.org/545153002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23728 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-05 11:44:31 +00:00
mstarzinger@chromium.org
4539e9b936 Remove deprecated PrintableUnique. 
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/543743002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23697 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-04 13:45:05 +00:00
jarin@chromium.org
a668cd6fc8 Context deoptimization and removal of the deoptimization block in Turbofan 
This adds context deoptimization to Turbofan and Crankshaft (also submitted separately as https://codereview.chromium.org/515723004/).

The second patchset removes the deoptimization/continuation block from calls.

BUG=
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/522873002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23547 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-09-01 09:31:14 +00:00
titzer@chromium.org
6bbf6c5cb0 Schedule floating control. 
This CL makes several changes to the scheduling algorithm to handle control
flow that is not connected to End. Such control nodes constitute "floating
control islands" that must be linearized by the schedule. This is done
by considering such nodes to be schedulable, and then editing the control
dependencies after a first pass of scheduling. Then a subsequent pass of
scheduling will place all nodes correctly into the fully connected graph.

R=mstarzinger@chromium.org, rossberg@chromium.org
BUG=

Review URL: https://codereview.chromium.org/499363002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23411 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-26 15:25:07 +00:00
titzer@chromium.org
5281cbebca Finish TODO in Schedule. s/entry/start/g and s/exit/end/g to be more regular. 
R=jarin@chromium.org
BUG=

Review URL: https://codereview.chromium.org/484653002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23161 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-18 13:28:10 +00:00
titzer@chromium.org
9803a72417 Unify MachineType and RepType. 
MachineType now tracks both the representation and the value type of machine quantities and is used uniformly throughout TurboFan.

These types can now express uint8, int8, uint16, and int16, i.e. signed and unsigned smallish integers. Note that currently only uint8 and uint16 are implemented in the TF backends.

R=bmeurer@chromium.org, mstarzinger@chromium.org
BUG=

Review URL: https://codereview.chromium.org/470593002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23122 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-14 09:19:54 +00:00
titzer@chromium.org
6603cbb319 Remove duplication in Scheduler and simplify interface. Make ComputeSchedule() and ComputeSpecialRPO() the only interface used by clients. 
R=bmeurer@chromium.org
BUG=

Review URL: https://codereview.chromium.org/460633002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23039 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 13:07:32 +00:00
jarin@chromium.org
1f776a07a6 Reland "More lazy deoptimization in Turbofan (binops, loads/stores)" 
This relands commit r23029 + adds missing lazy deopts for JSAdd operator inside VisitForInStatement. The added lazy deopt still refers to a bogus bailout id, but at least the graphs can be scheduled.

BUG=
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/450103004

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 12:26:17 +00:00
jarin@chromium.org
2558423c85 Revert "More lazy deoptimization in Turbofan (binops, loads/stores)" 
This reverts commit r23029.

BUG=
TBR=eisinger@chromium.org

Review URL: https://codereview.chromium.org/460623002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23031 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 11:21:51 +00:00
jarin@chromium.org
d04a7d6d91 More lazy deoptimization in Turbofan (binops, loads/stores) 
Deoptimizing binory operations, (Load|Store)(Property|Named),
constructors.

This also fixes safepoint lookup to account for lazily deoptimized code.

BUG=
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/453383002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23029 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 10:42:01 +00:00
jarin@chromium.org
e1deee4181 Revert "Initial shot at deoptimizing JSCallFunction in Turbofan." 
This reverts commit r23020.

BUG=
TBR=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/453363003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23022 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 07:33:58 +00:00
jarin@chromium.org
5c2e968bdb Initial shot at deoptimizing JSCallFunction in Turbofan. 
BUG=
R=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/444883006

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@23020 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-11 06:55:16 +00:00
jarin@chromium.org
117945ec6e Add deoptimization translations. 
BUG=
R=bmeurer@chromium.org, mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/442253002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22924 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-06 11:49:02 +00:00
bmeurer@chromium.org
70e7a119e6 Cleanup V8_TURBOFAN_TARGET definition. 
Move to globals.h so we do no longer skip any number of
tests. Introduce V8_TURBOFAN_BACKEND, which is 1 if we
actually have a backend and want to compile it, but don't
run the tests.

R=titzer@chromium.org

Review URL: https://codereview.chromium.org/444503003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22863 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-05 11:53:32 +00:00
sigurds@chromium.org
8322577eeb Make start node a value input to parameter nodes. 
BUG=
R=mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/437183002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22851 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-08-05 08:47:39 +00:00
danno@chromium.org
218395ebaa Fix MIPS build: use stubbed-out TF implementation 
R=titzer@chromium.org

Review URL: https://codereview.chromium.org/426833005

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22753 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-07-31 11:59:49 +00:00
svenpanne@chromium.org
337a7f61b2 Fixed a few space leaks. 
TBR=bmeurer@chromium.org

Review URL: https://codereview.chromium.org/430133002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22745 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-07-31 10:17:25 +00:00
danno@chromium.org
252cb8ba6e Fix 64-bit VS2010 build 
R=mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/420033003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22722 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-07-30 16:21:36 +00:00
danno@chromium.org
a1383e2250 Land the Fan (disabled) 
R=mstarzinger@chromium.org

Review URL: https://codereview.chromium.org/426233002

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22709 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
 2014-07-30 13:54:45 +00:00