aboutsummaryrefslogtreecommitdiff
path: root/vm/compiler/codegen/arm
Commit message (Collapse)AuthorAgeFilesLines
...
* Merge "Defer marking of objects as finalizable" into dalvik-devAndy McFadden2011-02-181-3/+1
|\
| * Defer marking of objects as finalizableAndy McFadden2011-02-181-3/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This shifts responsibility for marking an object as "finalizable" from object creation to object initialization. We want to make the object finalizable when Object.<init> completes. For performance reasons we skip the call to the Object constructor (which doesn't do anything) and just take the opportunity to check the class flag. Handling of clone()d object isn't quite right yet. Also, fixed a minor glitch in stubdefs. Bug 3342343 Change-Id: I5b7b819079e5862dc9cbd1830bb445a852dc63bf
* | [JIT] Fix profile trace dumpbuzbee2011-02-181-3/+6
|/ | | | | | | | | | The debugging profile mode prints out a list of the top ten traces, followed by recompilations. In some cases, it is possible that a trace was requested, but did not finish compiling before the run ended. If so, that could cause the dump to fail. This CL adds a check for null codeAddress to detect those cases. Change-Id: I415fd94d8fa9e270f75d5114fa5cc5d993bd6997
* Merge "Ensure we always call inline natives." into dalvik-devElliott Hughes2011-02-161-17/+0
|\
| * Ensure we always call inline natives.Elliott Hughes2011-02-161-17/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | Even though execute-inline is now a mandatory optimization, you can't be sure the inline natives will be invoked that way. There's reflection and JNI, for example, and there's the special case of String.equals that might be invoked as Object.equals. This patch adds a regular native method corresponding to each inline native, so that a corresponding libcore patch can drop its implementations. (For example, despite the fact that we all believed last week that the Java implementation of String.equals is never used, that turned out not to be true: every HashMap lookup will have used it. This pair of patches brings reality in line with our existing belief.) Change-Id: I19e64c23bea83e91696206ca40ce4e3faf853040
* | Add support to do suspend polling on backward branches in JIT'ed code.Ben Cheng2011-02-165-565/+619
|/ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The polling is expensive for now as it is done through three instructions: ld/ld/branch. As a result, a bunch of bonus stuff has been worked on to mitigate the extra overhead: - Cleaned up resource flags for memory disambiguation. - Rewrote load/store elimination and scheduler routines to hide the ld/ld latency for GC flag. Seperate the dependency checking into memory disambiguation part and resource conflict part. - Allowed code motion for Dalvik/constant/non-aliasing loads to be hoisted above branches for null/range checks. - Created extended basic blocks following goto instructions so that longer instruction streams can be optimized as a whole. Without the bonus stuff, the performance dropped about ~5-10% on some benchmarks because of the lack of headroom to hide the polling latency in tight loops. With the bonus stuff, the performance delta is between +/-5% with polling code generated. With the bonus stuff but disabling polling, the new bonus stuff provides consistent performance improvements: CaffeineMark 3.6% Linpack 11.1% Scimark 9.7% Sieve 33.0% Checkers 6.0% As a result, GC polling is disabled by default but can be turned on through the -Xjitsuspendpoll flag for experimental purposes. Change-Id: Ia81fc85de3e2b70e6cc93bc37c2b845892003cdb
* Merge "Add scheduling barrier at the end of IT blocks for FP comparisons." ↵Ben Cheng2011-02-141-0/+5
|\ | | | | | | into dalvik-dev
| * Add scheduling barrier at the end of IT blocks for FP comparisons.Ben Cheng2011-02-111-0/+5
| | | | | | | | | | Bug: 3448446 Change-Id: I98e2bbc4886443ba3c27c2963d7540fcee5790bb
* | Rename invoke-direct-empty to invoke-object-initAndy McFadden2011-02-111-1/+1
|/ | | | | | | | | | | | | | The invoke-direct-empty instruction was introduced to remove the overhead of calling the empty Object constructor. We now need it to do some extra work on behalf of object construction, so it's appropriate to change the instruction name to match the role it fills rather than the more general role it was hoped to fill. No functional changes. Bug 3342343 Change-Id: I65dd6a2c00c99581c9a19b16fe193b70642c8fbb
* Misc goodies in the JIT in preparation for more aggressive code motion.Ben Cheng2011-02-089-84/+209
| | | | | | | | | - Set up resource masks correctly for Thumb push/pop when LR/PC are involved. - Preserve LR around simulated heap references under self-verification mode. - Compact a few simple flags in ArmLIR into bit fields. - Minor performance tuning in TEMPLATE_MEM_OP_DECODE Change-Id: Id73edac837c5bb37dfd21f372d6fa21c238cf42a
* Merge "[JIT] Fix for 3385583: Performance variance" into dalvik-devBill Buzbee2011-02-032-1/+7
|\
| * [JIT] Fix for 3385583: Performance varianceBill Buzbee2011-02-032-1/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Closes a window in which the "interpret-only" templace could get chained to an existing trace while the intended translation was under construction. Note that this CL also introduces some small, but fundamental changes in trace formation: 1. Previouosly, when an exception or other trace terminating event occurred during trace formation, the entire trace was abandoned. With this change, we instead end the trace at the last successful instruction. 2. We previously allowed multiple attempts (perhaps by multiple threads) to form a trace compilation request for a dalvik PC. This was done in an attempt to allow recovery from compiler failures. Now we enforce a new rule: only the thread that wins the race to allocate an entry in the JitTable will form the trace request. 3. In a (probably misguided) attempt avoid unnecessary contention, we previously allowed work order enqueue requests to be dropped if a requester did not aquire TableLock on first attempt (assuming that if the trace were hot, it would be requested again). Now we block on enqueue. Change-Id: I40ea4f1b012250219ca37d5c40c5f22cae2092f1
* | Remove the unused monitor tracking and deadlock prediction code.Carl Shapiro2011-02-012-12/+2
|/ | | | | | | This feature has been in the code base for several releases but has never been enabled. Change-Id: Ia770b03ebc90a3dc7851c0cd8ef301f9762f50db
* Add runtime support for method based compilation.Ben Cheng2011-01-2614-62/+420
| | | | | | | | | | Enhanced code cache management to accommodate both trace and method compilations. Also implemented a hacky dispatch routine for virtual leaf methods. Microbenchmark showed 3x speedup in leaf method invocation. Change-Id: I79d95b7300ba993667b3aa221c1df9c7b0583521
* Fix armv5te-vfp build.Ben Cheng2011-01-251-14/+14
| | | | | | c++ dislikes variables named template. Change-Id: I6aaf623b449bfdb0c88b9664c55824268992058d
* Fix generic-eng (ARMv5) build.Elliott Hughes2011-01-241-1/+3
| | | | Change-Id: I366df47ebb597a629cb50046320ee3a6230d1ed9
* Fix push/pop instruction selection and encoding.Ben Cheng2011-01-244-11/+35
| | | | | | | | | | 1) Thumb 'push' can handle lr and 'pop' can handle pc, so make use of them. 2) Thumb2 push was incorrectly encoded as stmia, which should be stmdb instead. None of the above affect the code that we currently ship. Change-Id: I89ab46b032a3d562355c2cc3bc05fe308ba40957
* Merge "Light refactoring of handleExecuteInline." into dalvik-devElliott Hughes2011-01-206-101/+100
|\
| * Light refactoring of handleExecuteInline.Elliott Hughes2011-01-206-101/+100
| | | | | | | | | | | | | | | | | | | | | | | | | | | | I wanted the code to JIT a call a C function extracted so I can potentially use it elsewhere. The functions that sometimes JIT instructions directly and other times bail out to C can now call this, simplifying the body of the switch. I think there's a behavioral change here with the ThumbVFP genInlineSqrt, which previously had the wrong return value. Tested on passion to ensure that the performance characteristics of assembler intrinsics, C intrinsics, and library native methods haven't changed (using the Math and Float classes). Change-Id: Id79771a31abe3a516f403486454e9c0d9793622a
* | Support traceview-style profiling in all buildsbuzbee2011-01-192-8/+44
|/ | | | | | | | | | | | | | | | | | | | | | | | This change builds on an earlier bccheng change that allowed JIT'd code to avoid reverting to the debug portable interpeter when doing traceview-style method profiling. That CL introduced a new traceview build (libdvm_traceview) because the performance delta was too great to enable the capability for all builds. In this CL, we remove the libdvm_traceview build and provide full-speed method tracing in all builds. This is done by introducing "_PROF" versions of invoke and return templates used by the JIT. Normally, these templates are not used, and performace in unaffected. However, when method profiling is enabled, all existing translation are purged and new translations are created using the _PROF templates. These templates introduce a smallish performance penalty above and beyond the actual tracing cost, but again are only used when tracing has been enabled. Strictly speaking, there is a slight burden that is placed on invokes and returns in the non-tracing case - on the order of an additional 3 or 4 cycles per invoke/return. Those operations are already heavyweight enough that I was unable to measure the added cost in benchmarks. Change-Id: Ic09baf4249f1e716e136a65458f4e06cea35fc18
* am 45e9a990: am 80211d2b: Only generate debugging LIRs in verbose mode.Ben Cheng2011-01-141-13/+22
|\ | | | | | | | | * commit '45e9a9908f8874b64294dbd3e4dcfb6b76c4b6e3': Only generate debugging LIRs in verbose mode.
| * Only generate debugging LIRs in verbose mode.Ben Cheng2011-01-141-13/+22
| | | | | | | | | | | | | | This should reduce memory usage and JIT time a bit. Affected opcodes: kArmPseudoSSARep and kArmPseudoDalvikByteCodeBoundary. Change-Id: I18ce9338b8d258270df51a66f9dc98cd2d9dd0e8
* | Added vm support for new jumbo opcodes.jeffhao2011-01-071-33/+94
| | | | | | | | | | | | | | | | | | This enables jumbo opcodes by default, and they will get used by the current build without modification. Support has been added for arm, x86, and the portable interpreter. x86-atom support is on the TODO list. This commit also includes a test for the new jumbo opcodes. Change-Id: Ic3f1b41b51645861c5196f76aaf0e96e727ea537
* | resolved conflicts for merge of 4fa9da7e to dalvik-devBen Cheng2011-01-054-12/+24
|\| | | | | | | Change-Id: I56b52104f50d2e67115227e61e4b250e1116135d
| * More Jit-to-Interp entry point cleanup.Ben Cheng2011-01-054-12/+24
| | | | | | | | | | | | | | | | | | | | | | Only register entry points dispatched through [r6+#offset] in JitToInterpEntries. For ARM targets check the size of JitToInterpEntries explicitly to make sure that its last entry is within 128 byte from InterpState due to the Thumb codegen constraint. Change-Id: I74184115cb3a3c89afc3a5fe53685671d9cb1027
* | am af5aa1f4: Don\'t treat dvmJitToPatchPredictedChain as a Jit-to-Interp ↵Ben Cheng2011-01-043-12/+12
|\| | | | | | | | | | | | | entry point. * commit 'af5aa1f4ce7eecc1b47a4c038cebb67d33f08f18': Don't treat dvmJitToPatchPredictedChain as a Jit-to-Interp entry point.
| * Don't treat dvmJitToPatchPredictedChain as a Jit-to-Interp entry point.Ben Cheng2011-01-043-12/+12
| | | | | | | | | | | | It is just a native callout helper function. Change-Id: I6398b6876f5ba579b76e732107157a4c99337796
* | am a8589335: [Jit] Fix for 3311468 Maps crashed at handleFmt...Bill Buzbee2010-12-271-2/+2
|\| | | | | | | | | * commit 'a85893356ac4d86ef7d7dd18807d7bef95d7dddb': [Jit] Fix for 3311468 Maps crashed at handleFmt...
| * [Jit] Fix for 3311468 Maps crashed at handleFmt...Bill Buzbee2010-12-271-2/+2
| | | | | | | | | | | | | | | | | | | | Change https://android-git.corp.google.com/g/#change,86452 eliminated unused chaining cells for direct JNI calls. However, a code path in CodegenDriver.c assumed all similar invokes would have such cells. Slightly re-arranged the to avoid relying on the existance of the cell in cases in which it isn't needed. Change-Id: Ifc28acf559455a292b4b915ef1302085557e1d81
* | am 0d1aac38: Revert "Remove inline natives for an unused performance test."Jesse Wilson2010-12-221-0/+1
|\| | | | | | | | | * commit '0d1aac383a4bdce9feaad2f614df42234c2dcced': Revert "Remove inline natives for an unused performance test."
| * Revert "Remove inline natives for an unused performance test."Jesse Wilson2010-12-221-0/+1
| | | | | | | | | | | | This reverts commit 7ecd89dc02ce00c425788bd4989bdb6cde9a618a. Change-Id: I427635b7e3f7be45cfde78b8046dab3b23b64562
* | am 7ecd89dc: Remove inline natives for an unused performance test.Jesse Wilson2010-12-221-1/+0
|\| | | | | | | | | * commit '7ecd89dc02ce00c425788bd4989bdb6cde9a618a': Remove inline natives for an unused performance test.
| * Remove inline natives for an unused performance test.Jesse Wilson2010-12-221-1/+0
| | | | | | | | Change-Id: I80cfb918bdf174aeb6de83909c840563f6b945dd
* | [JIT] Trace profiling supportbuzbee2010-12-175-86/+182
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | In preparation for method compilation, this CL causes all traces to include two entry points: profiling and non-profiling. For now, the profiling entry will only be used if dalvik is run with -Xjitprofile, and largely works like it did before. The difference is that profiling support no longer requires the "assert" build - it's always there now. This will enable us to do a form of sampling profiling of traces in order to identify hot methods or hot trace groups, while keeping the overhead low by only switching profiling on periodically. To turn the periodic profiling on and off, we simply unchain all existing translations and set the appropriate global profile state. The underlying translation lookup and chaining utilties will examine the profile state to determine which entry point to use (i.e. - profiling or non-profiling) while the traces naturally rechain during further execution. Change-Id: I9ee33e69e33869b9fab3a57e88f9bc524175172b
* | [JIT] Regalloc cleanupbuzbee2010-12-143-16/+0
| | | | | | | | | | | | | | | | Remove vestiges of code intended for linear scan register allocation in the trace compiler. New plan is to stick with local allocation for traces and build a new linear scan allocator for the method compiler. Change-Id: Ic265ab5a7936b144cbe7fa4dc667fa7aba579045
* | Merge "Stamp out some x86/host mode warnings" into dalvik-devbuzbee2010-12-142-7/+12
|\ \
| * | Stamp out some x86/host mode warningsbuzbee2010-12-142-7/+12
| | | | | | | | | | | | | | | | | | | | | Nuked a void* cast warnings and moved cacheflush into a target-specific utility wrapper. Change-Id: I36c841288b9ec7e03c0cb29b2e89db344f36fad1
* | | Add explicit casts from "void *" to destination types.Ben Cheng2010-12-142-4/+4
|/ / | | | | | | Change-Id: I8828bc628f110aaade578a197bf1f51b30bf1be7
* | Implement method parser and SSA transformation.Ben Cheng2010-12-132-46/+47
| | | | | | | | Change-Id: If3fb3a36f33aaee8e5fdded4e9fa607be54f0bfb
* | Fix implicit conversions, rename reserved works, enable -Wc++-compat.Carl Shapiro2010-12-077-41/+45
|/ | | | Change-Id: I06292964a6882ea2d0c17c5c962db95e46b01543
* Another clarifying rename and a bit more automation.Dan Bornstein2010-12-032-6/+6
| | | | | | | | kNumDalvikInstructions is now kNumPackedOpcodes, there is a new kMaxOpcodeValue, and both are generated by opcode-gen. Change-Id: Ic46f1f52d2d21382452c8e777024f4a985ad31d3 Bonus: Reworded the switch and array data comment for clarity.
* Make opcode 00ff be called "dispatch-ff".Dan Bornstein2010-12-021-1/+1
| | | | | | | With this change, it's still implemented as an unused opcode, but it's now ready for its new life! Change-Id: Ic70d311704925067e47d87b657d133a792144e65
* Rename some instruction/opcode types and utilities.Dan Bornstein2010-12-022-6/+6
| | | | | | | | | A lot of this is more about properties of opcodes as opposed to inspecting instructions per se, and the new naming attempts to make it clear what is being queried and what sort of data is being returned. Change-Id: Ice6f9f2ebf4f1cfa8c99597419aa13d1134a33b2
* It's "opcode" not "opCode".Dan Bornstein2010-12-0118-486/+486
| | | | | | | | | | Similarly "Opcode" not "OpCode". This appears to be the general worldwide consensus on the matter. Other residents of my office didn't seem to mind one way or the other how it's spelled in our code, but for whatever reason, it really bugged me. Change-Id: Ia0b73d19c54aefc0f543a9c9451dda22ee876a59
* Rename OpCode.h -> DexOpcodes.h.Dan Bornstein2010-12-016-11/+6
| | | | | | | | Also incorporate the former contents of OpCodeNames.h. This is a small attempt to increase naming consistency in libdex. There will be a bit more to come, in a follow-up. Change-Id: Ia7ab06042dde2e19eda02ef1fee72fb4260e899d
* Say kNumDalvikInstructions when appropriate.Dan Bornstein2010-11-192-6/+6
| | | | | | | In particular, use it instead of just saying 256, and similarly for 255. The number of opcodes will be changing soon. Change-Id: Icc77120c2673968dddd6b4003f717245d46e4159
* Get rid of the copies of the opcode table pointers.Dan Bornstein2010-11-192-7/+6
| | | | | | | | | | | | | | | This inclduded fixing all the accessor functions to refer to the global ones defined in InstrUtils.[ch] instead of taking separate "table pointer" arguments. This did end up adding a few more truly global references to some of the code paths, particularly when performing dex optimization, so I went ahead and measured the time to do a cold first-boot both before and after the change (on real hardware). The times were identical (to one-second granularity), so I'm reasonably comfortable making this change. Change-Id: I604d9f7882bad4245bb11371218d13b06c3a5375
* Stop reporting negative widths.Andy McFadden2010-11-122-7/+4
| | | | | | | | | At one point, returning a negative width for dexopt output was useful. That stopped being the case a long time ago. This also removes a bad assert that went into my previous checkin. Change-Id: I18880c2316f5499a09dc479d271ca70b2a5be259
* am 72ef412b: Fix uninitialized variable warning(error).Jing Yu2010-11-111-0/+1
|\ | | | | | | | | * commit '72ef412b56becfbdd54f239ea672a48b163ff1d2': Fix uninitialized variable warning(error).
| * Fix uninitialized variable warning(error).Jing Yu2010-11-111-0/+1
| | | | | | | | Change-Id: I10815b033199758976950b28af7cc412f093a7f5
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-22 01:45:42 +0000