| c393d63aa2b8f6984672fdd4de631bbeff14b6a2 |
|
15-Apr-2016 |
Alexandre Rames <alexandre.rames@linaro.org> |
Fix: correctly destruct VIXL labels.
(cherry picked from commit c01a66465a398ad15da90ab2bdc35b7f4a609b17)
Bug: 27505766
Change-Id: I077465e3d308f4331e7a861902e05865f9d99835
|
| dee58d6bb6d567fcd0c4f39d8d690c3acaf0e432 |
|
07-Apr-2016 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals""
This patch merges the instruction-building phases from HGraphBuilder
and SsaBuilder into a single HInstructionBuilder class. As a result,
it is not necessary to generate HLocal, HLoadLocal and HStoreLocal
instructions any more, as the builder produces SSA form directly.
Saves 5-15% of arena-allocated memory (see bug for more data):
GMS 20.46MB => 19.26MB (-5.86%)
Maps 24.12MB => 21.47MB (-10.98%)
YouTube 28.60MB => 26.01MB (-9.05%)
This CL fixed an issue with parsing quickened instructions.
Bug: 27894376
Bug: 27998571
Bug: 27995065
Change-Id: I20dbe1bf2d0fe296377478db98cb86cba695e694
|
| 60328910cad396589474f8513391ba733d19390b |
|
04-Apr-2016 |
David Brazdil <dbrazdil@google.com> |
Revert "Refactor HGraphBuilder and SsaBuilder to remove HLocals"
Bug: 27995065
This reverts commit e3ff7b293be2a6791fe9d135d660c0cffe4bd73f.
Change-Id: I5363c7ce18f47fd422c15eed5423a345a57249d8
|
| e3ff7b293be2a6791fe9d135d660c0cffe4bd73f |
|
02-Mar-2016 |
David Brazdil <dbrazdil@google.com> |
Refactor HGraphBuilder and SsaBuilder to remove HLocals
This patch merges the instruction-building phases from HGraphBuilder
and SsaBuilder into a single HInstructionBuilder class. As a result,
it is not necessary to generate HLocal, HLoadLocal and HStoreLocal
instructions any more, as the builder produces SSA form directly.
Saves 5-15% of arena-allocated memory (see bug for more data):
GMS 20.46MB => 19.26MB (-5.86%)
Maps 24.12MB => 21.47MB (-10.98%)
YouTube 28.60MB => 26.01MB (-9.05%)
Bug: 27894376
Change-Id: Iefe28d40600c169c5d306fd2c77034ae19476d90
|
| cac5a7e871f1f346b317894359ad06fa7bd67fba |
|
22-Feb-2016 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Improve const-string code generation.
For strings in the boot image, use either direct pointers
or pc-relative addresses. For other strings, use PC-relative
access to the dex cache arrays for AOT and direct address of
the string's dex cache slot for JIT.
For aosp_flounder-userdebug:
- 32-bit boot.oat: -692KiB (-0.9%)
- 64-bit boot.oat: -948KiB (-1.1%)
- 32-bit dalvik cache total: -900KiB (-0.9%)
- 64-bit dalvik cache total: -3672KiB (-1.5%)
(contains more files than the 32-bit dalvik cache)
For aosp_flounder-userdebug forced to compile PIC:
- 32-bit boot.oat: -380KiB (-0.5%)
- 64-bit boot.oat: -928KiB (-1.0%)
- 32-bit dalvik cache total: -468KiB (-0.4%)
- 64-bit dalvik cache total: -1928KiB (-0.8%)
(contains more files than the 32-bit dalvik cache)
Bug: 26884697
Change-Id: Iec7266ce67e6fedc107be78fab2e742a8dab2696
|
| 2ae48182573da7087bffc2873730bc758ec29696 |
|
16-Mar-2016 |
Calin Juravle <calin@google.com> |
Clean up NullCheck generation and record stats about it.
This removes redundant code from the generators and allows for easier
stat recording.
Change-Id: Iccd4368f9e9d87a6fecb863dee4e2145c97851c4
|
| c7098ff991bb4e00a800d315d1c36f52a9cb0149 |
|
09-Feb-2016 |
David Srbecky <dsrbecky@google.com> |
Remove HNativeDebugInfo from start of basic blocks.
We do not require full environment at the start of basic block.
The dex pc contained in basic block is sufficient for line mapping.
Change-Id: I5ba9e5f5acbc4a783ad544769f9a73bb33e2bafa
|
| 0c5b18edd1308975804ccf29a02a130a7b6f7fa7 |
|
06-Feb-2016 |
Mark Mendell <mark.p.mendell@intel.com> |
Support CMOV for x86 Select
If possible, generate CMOV to implement HSelect. Tricky cases are a
long or FP condition (no single CC generated), FP inputs (no FP CMOV)
and when the condition is a boolean or not emitted at the use site.
In these cases, keep using the existing HSelect code.
Change-Id: I4ff1e152b8ef126fbbabeb3316e9e2b6a6b74aeb
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 6e332529c33be4d7dae5dad3609a839f4c0d3bfc |
|
02-Feb-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Remove HTemporary
Change-Id: I21b984224370a9ce7a4a13a9652503cfb03c5f03
|
| a19616e3363276e7f2c471eb2839fb16f1d43f27 |
|
02-Feb-2016 |
Aart Bik <ajcbik@google.com> |
Implemented compare/signum intrinsics as HCompare
(with all code generation for all)
Rationale:
At HIR level, many more optimizations are possible, while ultimately
generated code can take advantage of full semantics.
Change-Id: I6e2ee0311784e5e336847346f7f3c4faef4fd17e
|
| 2f10a5fb8c236a6786928f0323bd312c3ee9a4cc |
|
25-Jan-2016 |
Mark P Mendell <mark.p.mendell@intel.com> |
Revert "Revert "X86: Use the constant area for more operations.""
This reverts commit cf8d1bb97e193e02b430d707d3b669565fababb4.
Handle the case of an intrinsic where CurrentMethod is still an input.
This will be the case when there are unresolved classes in the
hierarchy.
Add a test case to confirm that we don't crash when handling Math.abs,
which wants to add a pointer to the constant area for the bitmask to be
used to remove the sign bit.
Enhance 565-checker-condition-liveness to check for the case of deeply
nested EmitAtUseSite chains.
Change-Id: I022e8b96a32f5bf464331d0c318c56b9d0ac3c9a
|
| cf8d1bb97e193e02b430d707d3b669565fababb4 |
|
25-Jan-2016 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "X86: Use the constant area for more operations."
Hits a DCHECK:
dex2oatd F 19461 20411 art/compiler/optimizing/pc_relative_fixups_x86.cc:196] Check failed: !invoke_static_or_direct->HasCurrentMethodInput()
This reverts commit dc00454f0b9a134f01f79b419200f4044c2af5c6.
Change-Id: Idfcacf12eb9e1dd7e68d95e880fda0f76f90e9ed
|
| dc00454f0b9a134f01f79b419200f4044c2af5c6 |
|
30-Oct-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
X86: Use the constant area for more operations.
Allow FP HNeg to use the constant area to hold the constant to flip the
sign bit.
Enhance some math intrinsics to allow the use of the constant
area: Abs{Float,Double}, {Min,Max}{FloatFloat,DoubleDouble}.
Allow compares of floats/doubles to constants using the constant area.
These eliminate almost all uses of loading constants from the stack.
Change-Id: Ic4b831565825cbe9f0801b1b53c1013be7c87ae4
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 95e7ffc28ea4d6deba356e636b16120ae49b62e2 |
|
22-Jan-2016 |
Roland Levillain <rpl@google.com> |
Improve documentation and assertions of read barrier instrumentation.
For ARM, x86, x86-64 back ends. The case of the ARM64 back
end is already handled in
https://android-review.googlesource.com/#/c/197870/.
Bug: 12687968
Change-Id: I6df1128cc100cbdb89020876e1a54de719508be3
|
| e3f43ac79e50a4693ea4d46acf5cffca64910cee |
|
19-Jan-2016 |
Roland Levillain <rpl@google.com> |
Some read barrier clean-up in Optimizing.
These changes make the read barrier compiler instrumentation
code more uniform among the ARM, ARM64, x86 and x86-64 back
ends.
Bug: 12687968
Change-Id: I6b1c0cf2bc22ed6cd6b14754136bef4a2a036ea5
|
| 58282f4510961317b8d5a364a6f740a78926716f |
|
14-Jan-2016 |
David Brazdil <dbrazdil@google.com> |
ART: Remove Baseline compiler
We don't need Baseline any more and it hasn't been maintained for
a while anyway. Let's remove it.
Change-Id: I442ed26855527be2df3c79935403a25b1ee55df6
|
| 42249c3602c3d0243396ee3627ffb5906aa77c1e |
|
08-Jan-2016 |
Aart Bik <ajcbik@google.com> |
Reduce code size by sharing slow paths.
Rationale:
Sharing identical slow path code reduces code size.
Background:
Currently, slow paths with the same dex-pc, same physical register
spilling code, and identical stack maps are shared (making this
only useful for deopt slow paths). The newly introduced mechanism
is sufficiently general to allow future improvements by e.g.
allowing different dex-pc (by passing this to runtime) or even
the kind of slow paths (by passing runtime addresses to the slowpath).
Change-Id: I819615c47b4fd98440a241f681f93e4fc22d12e0
|
| 152408f8c2188a7ed950cad04883b2f67dc74e84 |
|
31-Dec-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
X86: templatize GenerateTestAndBranch and friends
Allow the use of NearLabel as well as Label. This will be used by the
HSelect patch.
Replace a couple of Label(s) with NearLabel(s) as well.
Change-Id: I8e674c89e691bcdbccf4a5cdc07ad13b29ec21dd
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 5f7b58ea1adfc0639dd605b65f59198d3763f801 |
|
23-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Rewrite HInstruction::Is/As<type>().
Make Is<type>() and As<type>() non-virtual for concrete
instruction types, relying on GetKind(), and mark GetKind()
as PURE to improve optimization opportunities. This reduces
the number of relocations in libart-compiler.so's .rel.dyn
section by ~4K, or ~44%, and in .data.rel.ro by ~18K, or
~65%. The file is 96KiB smaller for Nexus 5, including 8KiB
reduction of the .text section.
Unfortunately, the g++/clang++ __attribute__((pure)) is not
strong enough to avoid duplicated virtual calls and we would
need the C++ [[pure]] attribute proposed in n3744 instead.
To work around this deficiency, we introduce an extra
non-virtual indirection for GetKind(), so that the compiler
can optimize common expressions such as
instruction->IsAdd() || instruction->IsSub()
or
instruction->IsAdd() && instruction->AsAdd()->...
which contain two virtual calls to GetKind() after inlining.
Change-Id: I83787de0671a5cb9f5b0a5f4a536cef239d5b401
|
| f3e0ee27f46aa6434b900ab33f12cd3157578234 |
|
17-Dec-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "Revert "ART: Reduce the instructions generated by packed switch.""
This reverts commit b4c137630fd2226ad07dfd178ab15725374220f1.
The underlying issue was fixed by https://android-review.googlesource.com/188271 .
Bug: 26121945
Change-Id: I58b08eb1a9f0a5c861f8cda93522af64bcf63920
|
| 17077d888a6752a2e5f8161eee1b2c3285783d12 |
|
16-Dec-2015 |
Mark P Mendell <mark.p.mendell@intel.com> |
Revert "Revert "X86: Use locked add rather than mfence""
This reverts commit 0da3b9117706760e8722029f407da6d0297cc943.
Fix a compilation failure that slipped in somehow.
Change-Id: Ide8681cdc921febb296ea47aa282cc195f154049
|
| 0da3b9117706760e8722029f407da6d0297cc943 |
|
16-Dec-2015 |
Aart Bik <ajcbik@google.com> |
Revert "X86: Use locked add rather than mfence"
This reverts commit 7b3e4f99b25c31048a33a08688557b133ad345ab.
Reason: build error on sdk (linux) in git_mirror-aosp-master-with-vendor , please fix first
art/compiler/optimizing/code_generator_x86_64.cc:4032:7: error: use of
undeclared identifier 'codegen_'
codegen_->MemoryFence();
Change-Id: I91f8542cfd944b7425d1981c35872dcdcb901e18
|
| b4c137630fd2226ad07dfd178ab15725374220f1 |
|
16-Dec-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "ART: Reduce the instructions generated by packed switch."
This reverts commit 59f054d98f519a3efa992b1c688eb97bdd8bbf55.
bug:26121945
Change-Id: I8a5ad7ef1f1de8d44787c27528fa3f7f5c2e9cd3
|
| 7b3e4f99b25c31048a33a08688557b133ad345ab |
|
19-Nov-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
X86: Use locked add rather than mfence
Java semantics for memory ordering can be satisfied using
lock addl $0,0(SP)
rather than mfence. The locked add synchronizes the memory caches, but
doesn't affect device memory.
Timing on a micro benchmark with a mfence or lock add $0,0(sp) in a loop
with 600000000 iterations:
time ./mfence
real 0m5.411s
user 0m5.408s
sys 0m0.000s
time ./locked_add
real 0m3.552s
user 0m3.550s
sys 0m0.000s
Implement this as an instruction-set-feature lock_add. This is off by
default (uses mfence), and enabled for atom & silvermont variants.
Generation of mfence can be forced by a parameter to MemoryFence.
Change-Id: I5cb4fded61f4cbbd7b7db42a1b6902e43e458911
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 7c1559a06041c9c299d5ab514d54b2102f204a84 |
|
15-Dec-2015 |
Roland Levillain <rpl@google.com> |
x86 Baker's read barrier fast path implementation.
Introduce an x86 fast path implementation in Optimizing for
Baker's read barriers (for both heap reference loads and GC
root loads). The marking phase of the read barrier is
performed by a slow path, invoking a new runtime entry point
(artReadBarrierMark).
Other read barrier algorithms continue to use the original
slow path based implementation, which has been renamed as
GenerateReadBarrierSlow/GenerateReadBarrierForRootSlow.
Bug: 12687968
Change-Id: Ie610c4befc19ff22378a8cba38b422dcacb54320
|
| 59f054d98f519a3efa992b1c688eb97bdd8bbf55 |
|
07-Dec-2015 |
Zheng Xu <zheng.xu@linaro.org> |
ART: Reduce the instructions generated by packed switch.
Implement Vladimir Marko's suggestion. The new compare/jump series
reduce the number of instructions from (2*n+1) to (1.5*n+3).
Generate normal compare/jump series when numEntries <= 3.
Generate optimal compare/jump series when numEntries <= threshold.
Generate jump tables otherwise.
Change-Id: I425547b6787057c7fa84e71f17c145b63b208633
|
| 0debae7bc89eb05f7a2bf7dccd223318fad7c88d |
|
12-Nov-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Refactor GenerateTestAndBranch
Each code generator implements a method for generating condition
evaluation and branching to arbitrary labels. This patch refactors
it for better clarity but also to generate fewer jumps when the true
branch is the fallthrough successor.
This is preliminary work for implementing HSelect.
Change-Id: Iaa545a5ecbacb761c5aa241fa69140cf6eb5952f
|
| 0d5a281c671444bfa75d63caf1427a8c0e6e1177 |
|
13-Nov-2015 |
Roland Levillain <rpl@google.com> |
x86/x86-64 read barrier support for concurrent GC in Optimizing.
This first implementation uses slow paths to instrument heap
reference loads and GC root loads for the concurrent copying
collector, respectively calling the artReadBarrierSlow and
artReadBarrierForRootSlow (new) runtime entry points.
Notes:
- This implementation does not instrument HInvokeVirtual
nor HInvokeInterface instructions (for class reference
loads), as the corresponding read barriers are not stricly
required with the current concurrent copying collector.
- Intrinsics which may eventually call (on slow path) are
disabled when read barriers are enabled, as the current
slow path infrastructure does not support this case.
- When read barriers are enabled, the code generated for a
HArraySet instruction always go into the array set slow
path for object arrays (delegating the operation to the
runtime), as we are lacking a mechanism to keep a
temporary register live accross a runtime call (needed for
the instrumentation of type checking code, which requires
two successive read barriers).
Bug: 12687968
Change-Id: I14cd6107233c326389120336f93955b28ffbb329
|
| 0f7dca4ca0be8d2f8776794d35edf8b51b5bc997 |
|
02-Nov-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing/X86: PC-relative dex cache array addressing.
Add PC-relative dex cache array addressing for X86 and use
it for better invoke-static/-direct dispatch. Also delay
the initialization to the PC-relative base until needed.
Change-Id: Ib8634d5edce4920cd70172fd13211809cf6948d1
|
| dc151b2346bb8a4fdeed0c06e54c2fca21d59b5d |
|
15-Oct-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Determine invoke-static/-direct dispatch early.
Determine the dispatch type of invoke-static/-direct in a
special pass right after the type inference. This allows the
inliner to pass the "needs dex cache" check and inline more.
It also allows the code generator to avoid requesting a
register location for the ArtMethod* for kDexCachePcRelative
and direct methods.
The supported dispatch check handles also situations that
the CompilerDriver currently doesn't allow. The cleanup of
the CompilerDriver and required changes to Quick will come
in a separate change.
Change-Id: I3f8e903a119949e95871d8ab0a995f4731a13a07
|
| 805b3b56c6eb542298db33e0181f135dc9fed3d9 |
|
18-Sep-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
X86 jump tables for PackedSwitch
Implement X86PackedSwitch using a jump table of offsets to blocks. The
X86PackedSwitch version just adds an input to address the constant area.
Change-Id: Id2752a1ee79222493040c6fd0e59aee9a544b76a
Bug: 21119474
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| e460d1df1f789c7c8bb97024a8efbd713ac175e9 |
|
29-Sep-2015 |
Calin Juravle <calin@google.com> |
Revert "Revert "Support unresolved fields in optimizing"
The CL also changes the calling convetion for 64bit static field set
to use kArg2 instead of kArg1. This allows optimizing to keep
the asumptions:
- arm pairs are always of form (even_reg, odd_reg)
- ecx_edx is not used as a register on x86.
This reverts commit e6f49b47b6a4dc9c7684e4483757872cfc7ff1a1.
Change-Id: I93159917565824084abc96775f31be1a4249f2f3
|
| 225b6464a58ebe11c156144653f11a1c6607f4eb |
|
28-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag arena allocations in code generators.
And completely remove the deprecated GrowableArray.
Replace GrowableArray with ArenaVector in code generators
and related classes and tag arena allocations.
Label arrays use direct allocations from ArenaAllocator
because Label is non-copyable and non-movable and as such
cannot be really held in a container. The GrowableArray
never actually constructed them, instead relying on the
zero-initialized storage from the arena allocator to be
correct. We now actually construct the labels.
Also avoid StackMapStream::ComputeDexRegisterMapSize() being
passed null references, even though unused.
Change-Id: I26a46fdd406b23a3969300a67739d55528df8bf4
|
| 85b62f23fc6dfffe2ddd3ddfa74611666c9ff41d |
|
09-Sep-2015 |
Andreas Gampe <agampe@google.com> |
ART: Refactor intrinsics slow-paths
Refactor slow paths so that there is a default implementation for
common cases (only arm64 with vixl is special). Write a generic
intrinsic slow-path that can be reused for the specific architectures.
Move helper functions into CodeGenerator so that they are accessible.
Change-Id: Ibd788dce432601c6a9f7e6f13eab31f28dcb8550
|
| e6f49b47b6a4dc9c7684e4483757872cfc7ff1a1 |
|
17-Sep-2015 |
Calin Juravle <calin@google.com> |
Revert "Support unresolved fields in optimizing"
breaks debuggable tests.
This reverts commit 23a8e35481face09183a24b9d11e505597c75ebb.
Change-Id: I8e60b5c8f48525975f25d19e5e8066c1c94bd2e5
|
| 23a8e35481face09183a24b9d11e505597c75ebb |
|
08-Sep-2015 |
Calin Juravle <calin@google.com> |
Support unresolved fields in optimizing
Change-Id: I9941fa5fcb6ef0a7a253c7a0b479a44a0210aad4
|
| 175dc732c80e6f2afd83209348124df349290ba8 |
|
25-Aug-2015 |
Calin Juravle <calin@google.com> |
Support unresolved methods in Optimizing
Change-Id: If2da02b50d2fa668cd58f134a005f1752e7746b1
|
| fa6b93c4b69e6d7ddfa2a4ed0aff01b0608c5a3a |
|
15-Sep-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Tag arena allocations in HGraph.
Replace GrowableArray with ArenaVector in HGraph and related
classes HEnvironment, HLoopInformation, HInvoke and HPhi,
and tag allocations with new arena allocation types.
Change-Id: I3d79897af405b9a1a5b98bfc372e70fe0b3bc40d
|
| bfb5ba90cd6425ce49c2125a87e3b12222cc2601 |
|
01-Sep-2015 |
Andreas Gampe <agampe@google.com> |
Revert "Revert "Do a second check for testing intrinsic types.""
This reverts commit a14b9fef395b94fa9a32147862c198fe7c22e3d7.
When an intrinsic with invoke-type virtual is recognized, replace
the instruction with a new HInvokeStaticOrDirect.
Minimal update for dex-cache rework. Fix includes.
Change-Id: I1c8e735a2fa7cda4419f76ca0717125ef236d332
|
| 0616ae081e648f4b9b64b33e2624a943c5fce977 |
|
17-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Add support for x86 constant area
Use the Quick trick of finding the address of the method by calling the
next instruction and popping the return address into a register. This
trick is used because of the lack of PC-relative addressing in 32 bit
mode on the X86.
Add a HX86ComputeBaseMethodAddress instruction to trigger generation
of the method address, which is referenced by instructions needing
access to the constant area.
Add a HX86LoadFromConstantTable instruction that takes a
HX86ComputeBaseMethodAddress and a HConstant that will be used to load
the value when needed.
Change Add/Sub/Mul/Div to detect a HX86LoadFromConstantTable right hand
side, and generate code that directly references the constant area.
Other uses will be added later.
Change the inputs to HReturn and HInvoke(s), replacing the FP constants
with HX86LoadFromConstantTable instead. This allows values to be
loaded from the constant area into the right location.
Port the X86_64 assembler constant area handling to the X86.
Use the new per-backend optimization framework to do this conversion.
Change-Id: I6d235a72238262e4f9ec0f3c88319a187f865932
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| ecc4366670e12b4812ef1653f7c8d52234ca1b1f |
|
13-Aug-2015 |
Serban Constantinescu <serban.constantinescu@linaro.org> |
Add OptimizingCompilerStats to the CodeGenerator class.
Just refactoring, not yet used, but will be used by the incoming patch
series and future CodeGen specific stats.
Change-Id: I7d20489907b82678120518a77bdab9c4cc58f937
Signed-off-by: Serban Constantinescu <serban.constantinescu@linaro.org>
|
| 581550137ee3a068a14224870e71aeee924a0646 |
|
19-Aug-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "Revert "Optimizing: Better invoke-static/-direct dispatch.""
Fixed kCallArtMethod to use correct callee location for
kRecursive. This combination is used when compiling with
debuggable flag set.
This reverts commit b2c431e80e92eb6437788cc544cee6c88c3156df.
Change-Id: Idee0f2a794199ebdf24892c60f8a5dcf057db01c
|
| b2c431e80e92eb6437788cc544cee6c88c3156df |
|
19-Aug-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "Optimizing: Better invoke-static/-direct dispatch."
Reverting due to failing ndebug tests.
This reverts commit 9b688a095afbae21112df5d495487ac5231b12d0.
Change-Id: Ie4f69da6609df3b7c8443412b6cf7f5c43c2c5d9
|
| 9b688a095afbae21112df5d495487ac5231b12d0 |
|
06-May-2015 |
Vladimir Marko <vmarko@google.com> |
Optimizing: Better invoke-static/-direct dispatch.
Add framework for different types of loading ArtMethod*
and code pointer retrieval. Implement invoke-static and
invoke-direct calls the same way as Quick. Document the
dispatch kinds in HInvokeStaticOrDirect's new enumerations
MethodLoadKind and CodePtrLocation.
PC-relative loads from dex cache arrays are used only for
x86-64 and arm64. The implementation for other architectures
will be done in separate CLs.
Change-Id: I468ca4d422dbd14748e1ba6b45289f0d31734d94
|
| 8158f28b6689314213eb4dbbe14166073be71f7e |
|
07-Aug-2015 |
Alexandre Rames <alexandre.rames@linaro.org> |
Ensure coherency of call kinds for LocationSummary.
The coherency is enforced with checks added in the `InvokeRuntime`
helper, that we now also use on x86 and x86_64.
Change-Id: I8cb92b042f25dc3c5fd390e9c61a45b477d081f4
|
| c470193cfc522fc818eb2eaab896aef9caf0c75a |
|
10-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
Fuse long and FP compare & condition on x86/x86-64 in Optimizing.
This is a preliminary implementation of fusing long/float/double
compares with conditions to avoid materializing the result from the
compare and condition.
The information from a HCompare is transferred to the HCondition if it
is legal. There must be only a single use of the HCompare, the HCompare
and HCondition must be in the same block, the HCondition must not need
materialization.
Added GetOppositeCondition() to HCondition to return the flipped
condition.
Bug: 21120453
Change-Id: I1f1db206e6dc336270cd71070ed3232dedc754d6
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| fc6a86ab2b70781e72b807c1798b83829ca7f931 |
|
26-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "Revert "ART: Implement try/catch blocks in Builder""
This patch enables the GraphBuilder to generate blocks and edges which
represent the exceptional control flow when try/catch blocks are
present in the code. Actual compilation is still delegated to Quick
and Baseline ignores the additional code.
To represent the relationship between try and catch blocks, Builder
splits the edges which enter/exit a try block and links the newly
created blocks to the corresponding exception handlers. This layout
will later enable the SsaBuilder to correctly infer the dominators of
the catch blocks and to produce the appropriate reverse post ordering.
It will not, however, allow for building the complete SSA form of the
catch blocks and consequently optimizing such blocks.
To this end, a new TryBoundary control-flow instruction is introduced.
Codegen treats it the same as a Goto but it allows for additional
successors (the handlers).
This reverts commit 3e18738bd338e9f8363b26bc895f38c0ec682824.
Change-Id: I4f5ea961848a0b83d8db3673763861633e9bfcfb
|
| 3e18738bd338e9f8363b26bc895f38c0ec682824 |
|
26-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
Revert "ART: Implement try/catch blocks in Builder"
Causes OutOfMemory issues, need to investigate.
This reverts commit 0b5c7d1994b76090afcc825e737f2b8c546da2f8.
Change-Id: I263e6cc4df5f9a56ad2ce44e18932ca51d7e349f
|
| 0b5c7d1994b76090afcc825e737f2b8c546da2f8 |
|
11-Jun-2015 |
David Brazdil <dbrazdil@google.com> |
ART: Implement try/catch blocks in Builder
This patch enables the GraphBuilder to generate blocks and edges which
represent the exceptional control flow when try/catch blocks are
present in the code. Actual compilation is still delegated to Quick
and Baseline ignores the additional code.
To represent the relationship between try and catch blocks, Builder
splits the edges which enter/exit a try block and links the newly
created blocks to the corresponding exception handlers. This layout
will later enable the SsaBuilder to correctly infer the dominators of
the catch blocks and to produce the appropriate reverse post ordering.
It will not, however, allow for building the complete SSA form of the
catch blocks and consequently optimizing such blocks.
To this end, a new TryBoundary control-flow instruction is introduced.
Codegen treats it the same as a Goto but it allows for additional
successors (the handlers).
Change-Id: I415b985596d5bebb7b1bb358a46e08b7b04bb53a
|
| eb7b7399dbdb5e471b8ae00a567bf4f19edd3907 |
|
19-Jun-2015 |
Alexandre Rames <alexandre.rames@linaro.org> |
Opt compiler: Add disassembly to the '.cfg' output.
This is automatically added to the '.cfg' output when using the usual
`--dump-cfg` option.
Change-Id: I864bfc3a8299c042e72e451cc7730ad8271e4deb
|
| ef20f71e16f035a39a329c8524d7e59ca6a11f04 |
|
09-Jun-2015 |
Alexandre Rames <alexandre.rames@linaro.org> |
Add boilerplate code for architecture-specific HInstructions.
Change-Id: I2723cd96e5f03012c840863dd38d7b2168117db8
|
| 69aa60163989c33a008115205d39732a76ecc1dc |
|
09-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Pass current method to HNewInstance and HNewArray.""
Problem exposed by this change was fixed in:
https://android-review.googlesource.com/#/c/154031/
This reverts commit 7b0e353b49ac3f464c662f20e20e240f0231afff.
Change-Id: I680c13dc9db9ba223ab11c7af255222860b4e6d2
|
| 7b0e353b49ac3f464c662f20e20e240f0231afff |
|
09-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Pass current method to HNewInstance and HNewArray."
082-inline-execute fails on x86.
This reverts commit e21aa42e1341d34250742abafdd83311ad9fa737.
Change-Id: Ib3fd25faee2e0128001e40d3d51a74f959bc4449
|
| 94015b939060f5041d408d48717f22443e55b6ad |
|
04-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "Use HCurrentMethod in HInvokeStaticOrDirect.""
Fix was to special case baseline for x86, which does not have enough
registers to allocate the current method.
This reverts commit c345f141f11faad177aa9635a78088d00cf66086.
Change-Id: I5997aa52f8d4df373ae5ff4d4150dac0c44c4c10
|
| e21aa42e1341d34250742abafdd83311ad9fa737 |
|
08-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Pass current method to HNewInstance and HNewArray.
Also remove unsed CodeGenerator::LoadCurrentMethod.
Change-Id: I4b8d3f2a30b8e2c76b6b329a72555483c993cb73
|
| c345f141f11faad177aa9635a78088d00cf66086 |
|
04-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Use HCurrentMethod in HInvokeStaticOrDirect."
Fails on baseline/x86.
This reverts commit 38207af82afb6f99c687f64b15601ed20d82220a.
Change-Id: Ib71018367eb7c6046965494a7e996c22af3de403
|
| 38207af82afb6f99c687f64b15601ed20d82220a |
|
01-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Use HCurrentMethod in HInvokeStaticOrDirect.
Change-Id: I0d15244b6b44c8b10079398c55da5071a3e3af66
|
| fd88f16100cceafbfde1b4f095f17e89444d6fa8 |
|
03-Jun-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Factorize code for common LocationSummary of HInvoke.
This is one step forward, we could factorize more, but
I wanted to get this out of the way first.
Change-Id: I6ae411a737eebaecb64974f47af507ce0cfbae85
|
| 3d21bdf8894e780d349c481e5c9e29fe1556051c |
|
22-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move mirror::ArtMethod to native
Optimizing + quick tests are passing, devices boot.
TODO: Test and fix bugs in mips64.
Saves 16 bytes per most ArtMethod, 7.5MB reduction in system PSS.
Some of the savings are from removal of virtual methods and direct
methods object arrays.
Bug: 19264997
(cherry picked from commit e401d146407d61eeb99f8d6176b2ac13c4df1e33)
Change-Id: I622469a0cfa0e7082a2119f3d6a9491eb61e3f3d
Fix some ArtMethod related bugs
Added root visiting for runtime methods, not currently required
since the GcRoots in these methods are null.
Added missing GetInterfaceMethodIfProxy in GetMethodLine, fixes
--trace run-tests 005, 044.
Fixed optimizing compiler bug where we used a normal stack location
instead of double on ARM64, this fixes the debuggable tests.
TODO: Fix JDWP tests.
Bug: 19264997
Change-Id: I7c55f69c61d1b45351fd0dc7185ffe5efad82bd3
ART: Fix casts for 64-bit pointers on 32-bit compiler.
Bug: 19264997
Change-Id: Ief45cdd4bae5a43fc8bfdfa7cf744e2c57529457
Fix JDWP tests after ArtMethod change
Fixes Throwable::GetStackDepth for exception event detection after
internal stack trace representation change.
Adds missing ArtMethod::GetInterfaceMethodIfProxy call in case of
proxy method.
Bug: 19264997
Change-Id: I363e293796848c3ec491c963813f62d868da44d2
Fix accidental IMT and root marking regression
Was always using the conflict trampoline. Also included fix for
regression in GC time caused by extra roots. Most of the regression
was IMT.
Fixed bug in DumpGcPerformanceInfo where we would get SIGABRT due to
detached thread.
EvaluateAndApplyChanges:
From ~2500 -> ~1980
GC time: 8.2s -> 7.2s due to 1s less of MarkConcurrentRoots
Bug: 19264997
Change-Id: I4333e80a8268c2ed1284f87f25b9f113d4f2c7e0
Fix bogus image test assert
Previously we were comparing the size of the non moving space to
size of the image file.
Now we properly compare the size of the image space against the size
of the image file.
Bug: 19264997
Change-Id: I7359f1f73ae3df60c5147245935a24431c04808a
[MIPS64] Fix art_quick_invoke_stub argument offsets.
ArtMethod reference's size got bigger, so we need to move other args
and leave enough space for ArtMethod* and 'this' pointer.
This fixes mips64 boot.
Bug: 19264997
Change-Id: I47198d5f39a4caab30b3b77479d5eedaad5006ab
|
| e401d146407d61eeb99f8d6176b2ac13c4df1e33 |
|
22-Apr-2015 |
Mathieu Chartier <mathieuc@google.com> |
Move mirror::ArtMethod to native
Optimizing + quick tests are passing, devices boot.
TODO: Test and fix bugs in mips64.
Saves 16 bytes per most ArtMethod, 7.5MB reduction in system PSS.
Some of the savings are from removal of virtual methods and direct
methods object arrays.
Bug: 19264997
Change-Id: I622469a0cfa0e7082a2119f3d6a9491eb61e3f3d
|
| 07276db28d654594e0e86e9e467cad393f752e6e |
|
18-May-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Don't do a null test in MarkGCCard if the value cannot be null.
Change-Id: I45687f6d3505178e2fc3689eac9cb6ab1b2c1e29
|
| 7394569c9252b277710b2d7d3fc35fb0dd48fc4b |
|
29-Apr-2015 |
Mark P Mendell <mark.p.mendell@intel.com> |
Revert "Revert "Revert "Revert "[optimizing] Improve x86 shifts""""
This reverts commit 2a7a1d7808f003bea908023ebd11eb442d2fca39.
Fix the problem that a long long >> 63 got the wrong answer. The
problem was that a shr was used instead of a sar.
Change-Id: I0327f79c718016ddec9272a605fc50ec15ec4566
|
| 2d27c8e338af7262dbd4aaa66127bb8fa1758b86 |
|
28-Apr-2015 |
Roland Levillain <rpl@google.com> |
Refactor InvokeDexCallingConventionVisitor in Optimizing.
Change-Id: I7ede0f59d5109644887bf5d39201d4e1bf043f34
|
| 2a7a1d7808f003bea908023ebd11eb442d2fca39 |
|
29-Apr-2015 |
Roland Levillain <rpl@google.com> |
Revert "Revert "Revert "[optimizing] Improve x86 shifts"""
This reverts commit 9b95a057ee20e4b1ca2e9c663726482172dc9ba3.
Reverting this CL as it breaks libcore tests:
org.apache.harmony.tests.java.lang.DoubleTest#test_compare
junit.framework.AssertionFailedError: compare() -0.0 should be less 0.0
at junit.framework.Assert.assertTrue(Assert.java:140)
at org.apache.harmony.tests.java.lang.DoubleTest.test_compare(DoubleTest.java:258)
org.apache.harmony.tests.java.lang.DoubleTest#test_compare FAIL (EXEC_FAILED)
org.apache.harmony.tests.java.lang.DoubleTest#test_compareToLjava_lang_Double
junit.framework.AssertionFailedError: Assert 2: compare() -0.0 should be less 0.0
at junit.framework.Assert.assertTrue(Assert.java:140)
at org.apache.harmony.tests.java.lang.DoubleTest.test_compareToLjava_lang_Double(DoubleTest.java:1320)
org.apache.harmony.tests.java.lang.DoubleTest#test_compareToLjava_lang_Double FAIL (EXEC_FAILED)
Change-Id: I10f0ec8cc9495cc225fef1940b3f1a9fe87d996f
|
| 9b95a057ee20e4b1ca2e9c663726482172dc9ba3 |
|
29-Apr-2015 |
Roland Levillain <rpl@google.com> |
Revert "Revert "[optimizing] Improve x86 shifts""
This reverts commit f9aac1e9f442c2486cd54f045d43e15791601205.
Don't use Location::Any() for the first input if the output is
Location::SameAsFirstInput().
Change-Id: I400834052b114abf0d616da1b4b6506f7bba10ab
|
| 232ade0b9401404ad4b61b1003551b58b96195a8 |
|
20-Apr-2015 |
Roland Levillain <rpl@google.com> |
Revert "Revert "Optimizing: Fix long-to-fp conversion on x86.""
This reverts commit 386ce406f150645158d6067c4e0a36565aefc44f.
Bug: 20413424
Change-Id: I6e93ff132907f2653f1ae12d6676ff2298f62ca1
|
| ad4450e5c3ffaa9566216cc6fafbf5c11186c467 |
|
17-Apr-2015 |
Zheng Xu <zheng.xu@arm.com> |
Opt compiler: Implement parallel move resolver without using swap.
The algorithm of ParallelMoveResolverNoSwap() is almost the same with
ParallelMoveResolverWithSwap(), except the way we resolve the circular
dependency. NoSwap() uses additional scratch register to resolve the
circular dependency. For example, (0->1) (1->2) (2->0) will be performed
as (2->scratch) (1->2) (0->1) (scratch->0).
On architectures without swap register support, NoSwap() can reduce the
number of moves from 3x(N-1) to (N+1) when there is circular dependency
with N moves.
And also, NoSwap() algorithm does not depend on architecture register
layout information, which means it can support register pairs on arm32
and X/W, D/S registers on arm64 without additional modification.
Change-Id: Idf56bd5469bb78c0e339e43ab16387428a082318
|
| e14590bdfed24df30e6b7545fc819ba03ff8bba1 |
|
15-Apr-2015 |
Guillaume Sanchez <guillaumesa@google.com> |
Revert "[optimizing] Improve x86 parallel moves/swaps"
This reverts commit a5c19ce8d200d68a528f2ce0ebff989106c4a933.
This commit introduces a performance regression on CaffeineLogic of 30%.
Change-Id: I917e206e249d44e1748537bc1b2d31054ea4959d
|
| 386ce406f150645158d6067c4e0a36565aefc44f |
|
13-Apr-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Optimizing: Fix long-to-fp conversion on x86."
Test fails on arm.
This reverts commit 2d45b4df3838d9c0e5a213305ccd1d7009e01437.
Change-Id: Id2864917b52f7ffba459680303a2d15b34f16a4e
|
| 2d45b4df3838d9c0e5a213305ccd1d7009e01437 |
|
07-Apr-2015 |
Serguei Katkov <serguei.i.katkov@intel.com> |
Optimizing: Fix long-to-fp conversion on x86.
long-to-fp conversion implemented using SSE loses the precision.
The test is included. CL uses FPU to provide the correct result.
Change-Id: I8eaf3c46819a8cb52642a7e7d7c4e3e0edbc88db
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
|
| f9aac1e9f442c2486cd54f045d43e15791601205 |
|
10-Apr-2015 |
Roland Levillain <rpl@google.com> |
Revert "[optimizing] Improve x86 shifts"
This reverts commit 222fcf96c9b73bbb739012575e7e413caf9348ec.
Reverting this CL as it is breaking a few tests (see http://build.chromium.org/p/client.art/builders/host-x86/builds/3251/steps/test%20optimizing/logs/stdio). Will investigate ASAP.
Change-Id: Iddd8363e83a24aa49fbdf0f0c9dc12e63b4848de
|
| a5c19ce8d200d68a528f2ce0ebff989106c4a933 |
|
01-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Improve x86 parallel moves/swaps
Add a new constructor to ScratchRegisterScope that will supply a
register if there is a free one, but not spill to force one. Use this
to generated alternate code that doesn't use a temporary, as the
spill/restore of a register generates extra instructions that aren't
necessary on x86.
Here is the benefit for a 32 bit memory-to-memory exchange with no
free registers:
< 50 push eax
< 53 push ebx
< 8B44244C mov eax, [esp + 76]
< 8B5C246C mov ebx, [esp + 108]
< 8944246C mov [esp + 108], eax
< 895C244C mov [esp + 76], ebx
< 5B pop ebx
< 58 pop eax
---
> FF742444 push [esp + 68]
> FF742468 push [esp + 104]
> 8F44244C pop [esp + 72]
> 8F442468 pop [esp + 100]
Avoid using xchg instruction, as it is slow on smaller processors.
Change-Id: Id29ee3abd998577baaee552d55d23e60ae0c7871
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 222fcf96c9b73bbb739012575e7e413caf9348ec |
|
30-Mar-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Improve x86 shifts
Support memory operands for integer shifts. Generate better code for
long shifts by constants.
Change-Id: Icc92fa1b59cc280d4894af6f054e19b01977d5ce
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| b19930c5cba3cf662dce5ee057fcc9829b4cbb9c |
|
09-Apr-2015 |
Guillaume Sanchez <guillaumesa@google.com> |
Follow up of "div/rem on x86 and x86_64", to tidy up the code a little.
Change-Id: Ibf39cbc8ac1d773599d70be2cb1e941674b60f1d
|
| 0f88e87085b7cf6544dadff3f555773966a6853e |
|
30-Mar-2015 |
Guillaume Sanchez <guillaumesa@google.com> |
Speedup div/rem by constants on x86 and x86_64
This is done using the algorithms in Hacker's Delight chapter 10.
Change-Id: I7bacefe10067569769ed31a1f7834f796fb41119
|
| d43b3ac88cd46b8815890188c9c2b9a3f1564648 |
|
01-Apr-2015 |
Mingyao Yang <mingyao@google.com> |
Revert "Revert "Deoptimization-based bce.""
This reverts commit 0ba627337274ccfb8c9cb9bf23fffb1e1b9d1430.
Change-Id: I1ca10d15bbb49897a0cf541ab160431ec180a006
|
| fb8d279bc011b31d0765dc7ca59afea324fd0d0c |
|
01-Apr-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Implement x86/x86_64 math intrinsics
Implement floor/ceil/round/RoundFloat on x86 and x86_64.
Implement RoundDouble on x86_64.
Add support for roundss and roundsd on both architectures. Support them
in the disassembler as well.
Add the instruction set features for x86, as the 'round' instruction is
only supported if SSE4.1 is supported.
Fix the tests to handle the addition of passing the instruction set
features to x86 and x86_64.
Add assembler tests for roundsd and roundss to x86_64 assembler tests.
Change-Id: I9742d5930befb0bbc23f3d6c83ce0183ed9fe04f
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| d75948ac93a4a317feaf136cae78823071234ba5 |
|
27-Mar-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Intrinsify String.compareTo.
Change-Id: Ia540df98755ac493fe61bd63f0bd94f6d97fbb57
|
| 09ed1a3125849ec6ac07cb886e3c502e1dcfada2 |
|
25-Mar-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing] Implement X86 intrinsic support
Implement the supported intrinsics for X86.
Enhance the graph visualizer to print <U> for unallocated locations, to
allow calling the graph dumper from within register allocation for
debugging purposes.
Change-Id: I3b0319eb70a9a4ea228f67065b4c52d13a1ae775
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 0ba627337274ccfb8c9cb9bf23fffb1e1b9d1430 |
|
24-Mar-2015 |
Andreas Gampe <agampe@google.com> |
Revert "Deoptimization-based bce."
This breaks compiling the core image:
Error after BCE: art::SSAChecker: Instruction 219 in block 1 does not dominate use 221 in block 1.
This reverts commit e295e6ec5beaea31be5d7d3c996cd8cfa2053129.
Change-Id: Ieeb48797d451836ed506ccb940872f1443942e4e
|
| e295e6ec5beaea31be5d7d3c996cd8cfa2053129 |
|
07-Mar-2015 |
Mingyao Yang <mingyao@google.com> |
Deoptimization-based bce.
A mechanism is introduced that a runtime method can be called
from code compiled with optimizing compiler to deoptimize into
interpreter. This can be used to establish invariants in the managed code
If the invariant does not hold at runtime, we will deoptimize and continue
execution in the interpreter. This allows to optimize the managed code as
if the invariant was proven during compile time. However, the exception
will be thrown according to the semantics demanded by the spec.
The invariant and optimization included in this patch are based on the
length of an array. Given a set of array accesses with constant indices
{c1, ..., cn}, we can optimize away all bounds checks iff all 0 <= min(ci) and
max(ci) < array-length. The first can be proven statically. The second can be
established with a deoptimization-based invariant. This replaces n bounds
checks with one invariant check (plus slow-path code).
Change-Id: I8c6e34b56c85d25b91074832d13dba1db0a81569
|
| 234d69d075d1608f80adb647f7935077b62b6376 |
|
09-Mar-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "Revert "[optimizing] Enable x86 long support.""
This reverts commit 154552e666347d41d95d7619c6ee56249ff4feca.
Change-Id: Idc726551c249a888b7ff5fde8508ae50e81b2e13
|
| 154552e666347d41d95d7619c6ee56249ff4feca |
|
06-Mar-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Revert "[optimizing] Enable x86 long support."
Few libcore failures.
This reverts commit b4ba354cf8d22b261205494875cc014f18587b50.
Change-Id: I4a28d853e730dff9b69aec9555505803cf2fcd63
|
| b4ba354cf8d22b261205494875cc014f18587b50 |
|
05-Mar-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
[optimizing] Enable x86 long support.
Change-Id: I9006972a65a1f191c45691104a960366747f9d16
|
| dc23d8318db08cb42e20f1d16dbc416798951a8b |
|
16-Feb-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Avoid generating jmp +0.
When a block branches to a non-following block, but blocks
in-between do branch to it, we can avoid doing the branch.
Change-Id: I9b343f662a4efc718cd4b58168f93162a24e1219
|
| 1cf95287364948689f6a1a320567acd7728e94a3 |
|
12-Dec-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Small optimization for recursive calls: avoid dex cache.
Change-Id: I044757a2f06e535cdc1480c4fc8182b89635baf6
|
| 7c8d009552545e6f1fd6036721e4e42e3fd14697 |
|
26-Jan-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing compiler] Support x86 hard float ABI
Add support for the new ABI passing FP parameters in XMM0-XMM3. This
allows us to optimize for x86 methods that don't use 'long'.
Change-Id: Ic79a24767173451e7d7095ccc2a00b307593a868
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 966c3ae95d3c699ee9fbdbccc1acdaaf02325faf |
|
27-Jan-2015 |
Mark P Mendell <mark.p.mendell@intel.com> |
Revert "Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)""
This reverts commit 949c91fb91f40a4a80b2b492913cf8541008975e.
This time, don't clobber EBX before saving it.
Redo some of the macros to make register usage explicit.
Change-Id: I8db8662877cd006816e16a28f42444ab7c36bfef
|
| 949c91fb91f40a4a80b2b492913cf8541008975e |
|
27-Jan-2015 |
Vladimir Marko <vmarko@google.com> |
Revert "ART: Implement X86 hard float (Quick/JNI/Baseline)"
And the 3 Mac build fixes. Fix conflicts in context_x86.* .
This reverts commits
3d2c8e74c27efee58e24ec31441124f3f21384b9 ,
34eda1dd66b92a361797c63d57fa19e83c08a1b4 ,
f601d1954348b71186fa160a0ae6a1f4f1c5aee6 ,
bc503348a1da573488503cc2819c9e30807bea31 .
Bug: 19150481
Change-Id: I6650ee30a7d261159380fe2119e14379e4dc9970
|
| 3d2c8e74c27efee58e24ec31441124f3f21384b9 |
|
13-Jan-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
ART: Implement X86 hard float (Quick/JNI/Baseline)
Use XMM0-XMM3 as parameter registers for float/double on X86. X86_64
already uses XMM0-XMM7 for parameters.
Change the 'hidden' argument register from XMM0 to XMM7 to avoid a
conflict.
Add support for FPR save/restore in runtime/arch/x86.
Minimal support for Optimizing baseline compiler.
Bump the version in runtime/oat.h because this is an ABI change.
Change-Id: Ia6fe150e8488b9e582b0178c0dda65fc81d5a8ba
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| d97dc40d186aec46bfd318b6a2026a98241d7e9c |
|
22-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Support callee save floating point registers on x64.
- Share the computation of core_spill_mask and fpu_spill_mask
between backends.
- Remove explicit stack overflow check support: we need to adjust
them and since they are not tested, they will easily bitrot.
Change-Id: I0b619b8de4e1bdb169ea1ae7c6ede8df0d65837a
|
| 988939683c26c0b1c8808fc206add6337319509a |
|
21-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Enable core callee-save on x64.
Will work on other architectures and FP support in other CLs.
Change-Id: I8cef0343eedc7202d206f5217fdf0349035f0e4d
|
| 24f2dfae084b2382c053f5d688fd6bb26cb8a328 |
|
15-Jan-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing compiler] Implement inline x86 FP '%'
Replace the calls to fmod/fmodf by inline code as is done in the Quick
compiler.
Remove the quick fmod/fmodf runtime entries, as they are no longer in
use.
64 bit code generator Move() routine needed to be enhanced to handle
constants, as Location::Any() allows them to be generated.
Change-Id: I6b6a42f6faeed4b0b3c940453e487daf5b25d184
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| cd6dffedf1bd8e6dfb3fb0c933551f9a90f7de3f |
|
08-Jan-2015 |
Calin Juravle <calin@google.com> |
Add implicit null checks for the optimizing compiler
- for backends: arm, arm64, x86, x86_64
- fixed parameter passing for CodeGenerator
- 003-omnibus-opcodes test verifies that NullPointerExceptions work as
expected
Change-Id: I1b302acd353342504716c9169a80706cf3aba2c8
|
| f85a9ca9859ad843dc03d3a2b600afbaf2e9bbdd |
|
13-Jan-2015 |
Mark Mendell <mark.p.mendell@intel.com> |
[optimizing compiler] Compute live spill size
The current stack frame calculation assumes that each live register to
be saved/restored has the word size of the machine. This fails for X86,
where a double in an XMM register takes up 8 bytes. Change the
calculation to keep track of the number of core registers and number of
fp registers to handle this distinction.
This is slightly pessimal, as the registers may not be active at the
same time, but the only way to handle this would be to allocate both
classes of registers simultaneously, or remember all the active
intervals, matching them up and compute the size of each safepoint
interval.
Change-Id: If7860aa319b625c214775347728cdf49a56946eb
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
|
| 840e5461a85f8908f51e7f6cd562a9129ff0e7ce |
|
07-Jan-2015 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement double and float support for arm in register allocator.
The basic approach is:
- An instruction that needs two registers gets two intervals.
- When allocating the low part, we also allocate the high part.
- When splitting a low (or high) interval, we also split the high
(or low) equivalent.
- Allocation follows the (S/D register) requirement that low
registers are always even and the high equivalent is low + 1.
Change-Id: I06a5148e05a2ffc7e7555d08e871ed007b4c2797
|
| 52c489645b6e9ae33623f1ec24143cde5444906e |
|
16-Dec-2014 |
Calin Juravle <calin@google.com> |
[optimizing compiler] Add support for volatile
- for backends: arm, x86, x86_64
- added necessary instructions to assemblies
- clean up code gen for field set/get
- fixed InstructionDataEquals for some instructions
- fixed comments in compiler_enums
* 003-opcode test verifies basic volatile functionality
Change-Id: I144393efa312dfb2c332cb84056b00edffee338a
|
| 9aec02fc5df5518c16f1e5a9b6cb198a192db973 |
|
19-Nov-2014 |
Calin Juravle <calin@google.com> |
[optimizing compiler] Add shifts
Added SHL, SHR, USHR for arm, x86, x86_64.
Change-Id: I971f594e270179457e6958acf1401ff7630df07e
|
| 86a8d7afc7f00ff0f5ea7b8aaf4d50514250a4e6 |
|
19-Nov-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Consistently use k{InstructionSet}WordSize.
These constants were defined prior to k{InstructionSet}PointerSize. So
use them consistently in optimizing as a first step. We can discuss
whether we should remove them in a second step.
Change-Id: If129de1a3bb8b65f8d9c816a8ad466815fb202e6
|
| bacfec30ee9f2f6fdfd190f11b105b609938efca |
|
14-Nov-2014 |
Calin Juravle <calin@google.com> |
[optimizing compiler] Add REM_INT, REM_LONG
- for arm, x86, x86_64
- minor cleanup/fix in div tests
Change-Id: I240874010206a5a9b3aaffbc81a885b94c248f93
|
| f0e3937b87453234d0d7970b8712082062709b8d |
|
12-Nov-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Do a parallel move in BoundsCheckSlowPath.
The two locations of the index and length could overlap,
so we need a parallel move. Also factorize the code for
doing a parallel move based on two locations.
Change-Id: Iee8b3459e2eed6704d45e9a564fb2cd050741ea4
|
| 9574c4b5f5ef039d694ac12c97e25ca02eca83c0 |
|
12-Nov-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement and/or/xor in optimizing.
Change-Id: I7cf6da1fd334a7177a5580931b8f174dd40b7cec
|
| de58ab2c03ff8112b07ab827c8fa38f670dfc656 |
|
05-Nov-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement try/catch/throw in optimizing.
- We currently don't run optimizations in the presence of a try/catch.
- We therefore implement Quick's mapping table.
- Also fix a missing null check on array-length.
Change-Id: I6917dfcb868e75c1cf6eff32b7cbb60b6cfbd68f
|
| 424f676379f2f872acd1478672022f19f3240fc1 |
|
03-Nov-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement CONST_CLASS in optimizing compiler.
Change-Id: Ia8c8dfbef87cb2f7893bfb6e178466154eec9efd
|
| 19a19cffd197a28ae4c9c3e59eff6352fd392241 |
|
22-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add support for static fields in optimizing compiler.
Change-Id: Id2f010589e2bd6faf42c05bb33abf6816ebe9fa9
|
| 102cbed1e52b7c5f09458b44903fe97bb3e14d5f |
|
15-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement register allocator for floating point registers.
Also:
- Fix misuses of emitting the rex prefix in the x86_64 assembler.
- Fix movaps code generation in the x86_64 assembler.
Change-Id: Ib6dcf6e7c4a9c43368cfc46b02ba50f69ae69cbe
|
| 34bacdf7eb46c0ffbf24ba7aa14a904bc9176fb2 |
|
07-Oct-2014 |
Calin Juravle <calin@google.com> |
Add multiplication for integral types
This also fixes an issue where we could allocate a pair register even if
one of its parts was already blocked.
Change-Id: I4869175933409add2a56f1ccfb369c3d3dd3cb01
|
| 92a73aef279be78e3c2b04db1713076183933436 |
|
16-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Don't use assembler classes in code_generator.h.
The arm64 backend uses its own assembler and does not share
the same classes as the other backends. To avoid conflicts
or unnecessary mappings, just don't use those classes in the
shared part of the code generator.
Change-Id: I9e5fa40c1021d2e83a4ef14c52cd1ccd03f2f73d
|
| 71175b7f19a4f6cf9cc264feafd820dbafa371fb |
|
09-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Cleanup baseline register allocator.
- Use three arrays for blocking regsters instead of
one and computing offsets in that array.]
- Don't pass blocked_registers_ to methods, just use the field.
Change-Id: Ib698564c31127c59b5a64c80f4262394b8394dc6
|
| 360231a056e796c36ffe62348507e904dc9efb9b |
|
08-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix code generation of materialized conditions.
Move the logic for knowing if a condition needs to be materialized
in an optimization pass (so that the information does not change
as a side effect of another optimization).
Also clean-up arm and x86_64 codegen:
- arm: ldr and str are for power-users when a constant is
in play. We should use LoadFromOffset and StoreToOffset.
- x86_64: fix misuses of movq instead of movl.
Change-Id: I01a03b91803624be2281a344a13ad5efbf4f3ef3
|
| 56b9ee6fe1d6880c5fca0e7feb28b25a1ded2e2f |
|
09-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Stop converting from Location to ManagedRegister.
Now the source of truth is the Location object that knows
which register (core, pair, fpu) it needs to refer to.
Change-Id: I62401343d7479ecfb24b5ed161ec7829cda5a0b1
|
| 7fb49da8ec62e8a10ed9419ade9f32c6b1174687 |
|
06-Oct-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add support for floats and doubles.
- Follows Quick conventions.
- Currently only works with baseline register allocator.
Change-Id: Ie4b8e298f4f5e1cd82364da83e4344d4fc3621a3
|
| 3c04974a90b0e03f4b509010bff49f0b2a3da57f |
|
24-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Optimize suspend checks in optimizing compiler.
- Remove the ones added during graph build (they were added
for the baseline code generator).
- Emit them at loop back edges after phi moves, so that the test
can directly jump to the loop header.
- Fix x86 and x86_64 suspend check by using cmpw instead of cmpl.
Change-Id: I6fad5795a55705d86c9e1cb85bf5d63dadfafa2a
|
| 3bca0df855f0e575c6ee020ed016999fc8f14122 |
|
19-Sep-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Support for saving and restoring live registers in a slow path.
And use it in suspend check slow paths.
Change-Id: I79caf28f334c145a36180c79a6e2fceae3990c31
|
| e982f0b8e809cece6f460fa2d8df25873aa69de4 |
|
13-Aug-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement invoke virtual in optimizing compiler.
Also refactor 004 tests to make them work with both Quick and
Optimizing.
Change-Id: I87e275cb0ae0258fc3bb32b612140000b1d2adf8
|
| 3c7bb98698f77af10372cf31824d3bb115d9bf0f |
|
23-Jul-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Implement array get and array put in optimizing.
Also fix a couple of assembler/disassembler issues.
Change-Id: I705c8572988c1a9c4df3172b304678529636d5f6
|
| f12feb8e0e857f2832545b3f28d31bad5a9d3903 |
|
17-Jul-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Stack overflow checks and NPE checks for optimizing.
Change-Id: I59e97448bf29778769b79b51ee4ea43f43493d96
|
| 96f89a290eb67d7bf4b1636798fa28df14309cc7 |
|
11-Jul-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add assembly operations with constants in optimizing compiler.
Change-Id: I5bcc35ab50d4457186effef5592a75d7f4e5b65f
|
| ab032bc1ff57831106fdac6a91a136293609401f |
|
15-Jul-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Fix a braino in the stack layout.
Also do some refactoring to have this code be just in CodeGenerator.
Change-Id: I88de109889138af8d60027973c12a64bee813cb7
|
| e50383288a75244255d3ecedcc79ffe9caf774cb |
|
04-Jul-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Support fields in optimizing compiler.
- Required support for temporaries, to be only used by baseline compiler.
- Also fixed a few invalid assumptions around locations and instructions
that don't need materialization. These instructions should not have an Out.
Change-Id: Idc4a30dd95dd18015137300d36bec55fc024cf62
|
| 412f10cfed002ab617c78f2621d68446ca4dd8bd |
|
19-Jun-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Support longs in the register allocator for x86_64.
Change-Id: I7fb6dfb761bc5cf9e5705682032855a0a70ca867
|
| 86dbb9a12119273039ce272b41c809fa548b37b6 |
|
04-Jun-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Final CL to enable register allocation on x86.
This CL implements:
1) Resolution after allocation: connecting the locations
allocated to an interval within a block and between blocks.
2) Handling of fixed registers: some instructions require
inputs/output to be at a specific location, and the allocator
needs to deal with them in a special way.
3) ParallelMoveResolver::EmitNativeCode for x86.
Change-Id: I0da6bd7eb66877987148b87c3be6a983b4e3f858
|
| ffddfdf6fec0b9d98a692e27242eecb15af5ead2 |
|
03-Jun-2014 |
Tim Murray <timmurray@google.com> |
DO NOT MERGE
Merge ART from AOSP to lmp-preview-dev.
Change-Id: I0f578733a4b8756fd780d4a052ad69b746f687a9
|
| a7062e05e6048c7f817d784a5b94e3122e25b1ec |
|
22-May-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add a linear scan register allocator to the optimizing compiler.
This is a "by-the-book" implementation. It currently only deals
with allocating registers, with no hint optimizations.
The changes remaining to make it functional are:
- Allocate spill slots.
- Resolution and placements of Move instructions.
- Connect it to the code generator.
Change-Id: Ie0b2f6ba1b98da85425be721ce4afecd6b4012a4
|
| a7aca370a7d62ca04a1e24423d90e8020d6f1a58 |
|
28-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Setup policies for register allocation.
Change-Id: I857e77530fca3e2fb872fc142a916af1b48400dc
|
| a747a392fb5f88d2ecc4c6021edf9f1f6615ba16 |
|
17-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Code cleanup in preparation for x64 backend.
- Use InvokeDexCallingConventionVisitor for setting
up HParameterValues
- Use kVregSize instead of kX86WordSize when dealing with
virtual registers.
Change-Id: Ia520223010194c70a3ff0ed659077f55cec4e7d8
|
| 01bc96d007b67fdb7fe349232a83e4b354ce3d08 |
|
11-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Long support in optimizing compiler.
- Add stack locations to the Location class.
- Change logic of parameter passing/setup by setting the
location of such instructions the ones for the calling
convention.
Change-Id: I4730ad58732813dcb9c238f44f55dfc0baa18799
|
| 707c809f661554713edfacf338365adca8dfd3a3 |
|
04-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Use target-specific word instead of runtime word.
Change-Id: Ia11dc3cc520a1a5c7bd017013e5699af9570ce91
|
| 4a34a428c6a2588e0857ef6baf88f1b73ce65958 |
|
03-Apr-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Support passing arguments to invoke-static* instructions.
- Stop using the frame pointer for accessing locals.
- Stop emulating a stack when doing code generation. Instead,
rely on dex register model, where instructions only reference
registers.
Change-Id: Id51bd7d33ac430cb87a53c9f4b0c864eeb1006f9
|
| 8ccc3f5d06fd217cdaabd37e743adab2031d3720 |
|
19-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add support for invoke-static in optimizing compiler.
Support is limited to calls without parameters and returning
void. For simplicity, we currently follow the Quick ABI.
Change-Id: I54805161141b7eac5959f1cae0dc138dd0b2e8a5
|
| 787c3076635cf117eb646c5a89a9014b2072fb44 |
|
17-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Plug new optimizing compiler in compilation pipeline.
Also rename accessors to ART's conventions.
Change-Id: I344807055b98aa4b27215704ec362191464acecc
|
| bab4ed7057799a4fadc6283108ab56f389d117d4 |
|
11-Mar-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
More code generation for the optimizing compiler.
- Add HReturn instruction
- Generate code for locals/if/return
- Setup infrastructure for register allocation. Currently
emulate a stack.
Change-Id: Ib28c2dba80f6c526177ed9a7b09c0689ac8122fb
|
| d4dd255db1d110ceb5551f6d95ff31fb57420994 |
|
28-Feb-2014 |
Nicolas Geoffray <ngeoffray@google.com> |
Add codegen support to the optimizing compiler.
Change-Id: I9aae76908ff1d6e64fb71a6718fc1426b67a5c28
|