summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Transforms
Commit message (Collapse)AuthorAgeFilesLines
* CodeGenPrepare: Don't crash when TLI is not available.Benjamin Kramer2012-06-291-1/+2
| | | | | | This happens when codegenprepare is invoked via opt. llvm-svn: 159457
* Rework this to clarify where the removal of nodes from the queue isDuncan Sands2012-06-291-8/+9
| | | | | | really happening. No intended functionality change. llvm-svn: 159451
* ignore 'invoke new' in isInstructionTriviallyDead, since most callers are ↵Nuno Lopes2012-06-291-1/+2
| | | | | | not ready to handle invokes. instcombine will take care of this. llvm-svn: 159440
* Fix a reassociate crash on sozefx when compiling with dragonegg+gcc-4.7 due toDuncan Sands2012-06-291-5/+13
| | | | | | | the optimizers producing a multiply expression with more multiplications than the original (!). llvm-svn: 159426
* Move llvm/Support/IRBuilder.h -> llvm/IRBuilder.hChandler Carruth2012-06-2919-110/+110
| | | | | | | | | | | | | | | | | This was always part of the VMCore library out of necessity -- it deals entirely in the IR. The .cpp file in fact was already part of the VMCore library. This is just a mechanical move. I've tried to go through and re-apply the coding standard's preferred header sort, but at 40-ish files, I may have gotten some wrong. Please let me know if so. I'll be committing the corresponding updates to Clang and Polly, and Duncan has DragonEgg. Thanks to Bill and Eric for giving the green light for this bit of cleanup. llvm-svn: 159421
* The DIBuilder class is just a wrapper around debug info creationBill Wendling2012-06-293-3/+3
| | | | | | | (a.k.a. MDNodes). The module doesn't belong in Analysis. Move it to the VMCore instead. llvm-svn: 159414
* make simplifyCFG erase invokes to readonly/readnone functionsNuno Lopes2012-06-281-6/+7
| | | | llvm-svn: 159385
* make instcombine produce calls to llvm.donothing instead of a random intrinsicNuno Lopes2012-06-281-7/+4
| | | | llvm-svn: 159384
* [asan] set a hard limit on the number of instructions instrumented pear each ↵Kostya Serebryany2012-06-281-0/+12
| | | | | | BB. This is (hopefully temporary) workaround for PR13225 llvm-svn: 159344
* Precompute SCEV pointer analysis prior to instruction fusion in BBVectorize.Hal Finkel2012-06-281-10/+51
| | | | | | | | | | | | | When both a load/store and its address computation are being vectorized, it can happen that the address-computation vectorization destroys SCEV's ability to analyize the relative pointer offsets. As a result (like with the aliasing analysis info), we need to precompute the necessary information prior to instruction fusing. This was found during stress testing (running through the test suite with a very low required chain length); unfortunately, I don't have a small test case. llvm-svn: 159332
* Remove a useless check in BBVectorize.Hal Finkel2012-06-281-5/+0
| | | | | | | A shuffle mask will always be a constant, but I did not realize that when I originally wrote the code. llvm-svn: 159331
* Allow BBVectorize to form non-2^n-length vectors.Hal Finkel2012-06-281-132/+532
| | | | | | | | | | | | | | | | | | | | The original algorithm only used recursive pair fusion of equal-length types. This is now extended to allow pairing of any types that share the same underlying scalar type. Because we would still generally prefer the 2^n-length types, those are formed first. Then a second set of iterations form the non-2^n-length types. Also, a call to SimplifyInstructionsInBlock has been added after each pairing iteration. This takes care of DCE (and a few other things) that make the following iterations execute somewhat faster. For the same reason, some of the simple shuffle-combination cases are now handled internally. There is some additional refactoring work to be done, but I've had many requests for this feature, so additional refactoring will come soon in future commits (as will additional test cases). llvm-svn: 159330
* Refactor operation equivalence checking in BBVectorize by extending ↵Hal Finkel2012-06-281-23/+2
| | | | | | | | | | | Instruction::isSameOperationAs. Maintaining this kind of checking in different places is dangerous, extending Instruction::isSameOperationAs consolidates this logic into one place. Here I've added an optional flags parameter and two flags that are important for vectorization: CompareIgnoringAlignment and CompareUsingScalarTypes. llvm-svn: 159329
* Move lib/Analysis/DebugInfo.cpp to lib/VMCore/DebugInfo.cpp andBill Wendling2012-06-287-8/+8
| | | | | | | | | include/llvm/Analysis/DebugInfo.h to include/llvm/DebugInfo.h. The reasoning is because the DebugInfo module is simply an interface to the debug info MDNodes and has nothing to do with analysis. llvm-svn: 159312
* Revert r159136 due to PR13124.Matt Beaumont-Gay2012-06-271-11/+0
| | | | | | | | | | | Original commit message: If a constant or a function has linkonce_odr linkage and unnamed_addr, mark it hidden. Being linkonce_odr guarantees that it is available in every dso that needs it. Being a constant/function with unnamed_addr guarantees that the copies don't have to be merged. llvm-svn: 159272
* Some reassociate optimizations create new instructions, which they insert justDuncan Sands2012-06-271-11/+7
| | | | | | | | | | | before the expression root. Any existing operators that are changed to use one of them needs to be moved between it and the expression root, and recursively for the operators using that one. When I rewrote RewriteExprTree I accidentally inverted the logic, resulting in the compacting going down from operators to operands rather than up from operands to the operators using them, oops. Fix this, resolving PR12963. llvm-svn: 159265
* Remove a instcombine transform that (no longer?) makes sense:Evan Cheng2012-06-261-5/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // C - zext(bool) -> bool ? C - 1 : C if (ZExtInst *ZI = dyn_cast<ZExtInst>(Op1)) if (ZI->getSrcTy()->isIntegerTy(1)) return SelectInst::Create(ZI->getOperand(0), SubOne(C), C); This ends up forming sext i1 instructions that codegen to terrible code. e.g. int blah(_Bool x, _Bool y) { return (x - y) + 1; } => movzbl %dil, %eax movzbl %sil, %ecx shll $31, %ecx sarl $31, %ecx leal 1(%rax,%rcx), %eax ret Without the rule, llvm now generates: movzbl %sil, %ecx movzbl %dil, %eax incl %eax subl %ecx, %eax ret It also helps with ARM (and pretty much any target that doesn't have a sext i1 :-). The transformation was done as part of Eli's r75531. He has given the ok to remove it. rdar://11748024 llvm-svn: 159230
* Replacing zero-sized alloca's with a null pointer is too aggressive, insteadDuncan Sands2012-06-261-8/+40
| | | | | | | | | | | | | | | | merge all zero-sized alloca's into one, fixing c43204g from the Ada ACATS conformance testsuite. What happened there was that a variable sized object was being allocated on the stack, "alloca i8, i32 %size". It was then being passed to another function, which tested that the address was not null (raising an exception if it was) then manipulated %size bytes in it (load and/or store). The optimizers cleverly managed to deduce that %size was zero (congratulations to them, as it isn't at all obvious), which made the alloca zero size, causing the optimizers to replace it with null, which then caused the check mentioned above to fail, and the exception to be raised, wrongly. Note that no loads and stores were actually being done to the alloca (the loop that does them is executed %size times, i.e. is not executed), only the not-null address check. llvm-svn: 159202
* revert my previous commit (r159173), since as Eli pointed out, it's ↵Nuno Lopes2012-06-251-0/+1
| | | | | | perfectly ok to mark realloc as noalias llvm-svn: 159175
* do not set realloc() as NotAlias, since it can return the same pointer. This ↵Nuno Lopes2012-06-251-1/+0
| | | | | | whole thing should be upgraded to use the MemoryBuiltin interface anyway.. llvm-svn: 159173
* Fix the objc_autoreleasedReturnValue optimization code to locateDan Gohman2012-06-251-2/+17
| | | | | | | the call correctly even in the case where it is an invoke. This fixes rdar://11714057. llvm-svn: 159157
* improve optimization of invoke instructions:Nuno Lopes2012-06-252-3/+15
| | | | | | | | - simplifycfg: invoke undef/null -> unreachable - instcombine: invoke new -> invoke expect(0, 0) (an arbitrary NOOP intrinsic; only done if the allocated memory is unused, of course) - verifier: allow invoke of intrinsics (to make the previous step work) llvm-svn: 159146
* If a constant or a function has linkonce_odr linkage and unnamed_addr, mark itRafael Espindola2012-06-251-0/+11
| | | | | | | | hidden. Being linkonce_odr guarantees that it is available in every dso that needs it. Being a constant/function with unnamed_addr guarantees that the copies don't have to be merged. llvm-svn: 159136
* The name (and comment describing) of llvm::GetFirstDebuigLocInBasicBlock no ↵Eli Bendersky2012-06-252-11/+5
| | | | | | longer represents what the function does. Therefore, the function is removed and its functionality is folded into the only place in the code-base where it was being used. llvm-svn: 159133
* llvm/lib: [CMake] Add explicit dependency to intrinsics_gen.NAKAMURA Takumi2012-06-246-0/+12
| | | | llvm-svn: 159112
* Allow controlling vectorization of boolean values separately from other ↵Hal Finkel2012-06-241-4/+14
| | | | | | | | integer types. These are used as the result of comparisons, and often handled differently from larger integer types. llvm-svn: 159111
* Remove dyn_cast + dereference pattern by replacing it with a cast and changingNick Lewycky2012-06-241-3/+3
| | | | | | | the safety check to look for the same type we're going to actually cast to. Fixes PR13180! llvm-svn: 159110
* Tab to spaces. No functionality change.Nick Lewycky2012-06-241-1/+1
| | | | llvm-svn: 159104
* Remove a dangling reference to a deleted instruction. Fixes PR13185!Nick Lewycky2012-06-241-0/+1
| | | | llvm-svn: 159096
* Allow BBVectorize to fuse compare instructions.Hal Finkel2012-06-231-0/+8
| | | | llvm-svn: 159088
* Extend the IL for selecting TLS models (PR9788)Hans Wennborg2012-06-234-10/+10
| | | | | | | | | | | | | | | This allows the user/front-end to specify a model that is better than what LLVM would choose by default. For example, a variable might be declared as @x = thread_local(initialexec) global i32 42 if it will not be used in a shared library that is dlopen'ed. If the specified model isn't supported by the target, or if LLVM can make a better choice, a different model may be used. llvm-svn: 159077
* Optimized usage of new SwitchInst case values (IntegersSubset type) in ↵Stepan Dyatkovskiy2012-06-231-4/+3
| | | | | | | | Local.cpp, Execution.cpp and BitcodeWriter.cpp. I got about 1% of compile-time improvement on my machines (Ubuntu 11.10 i386 and Ubuntu 12.04 x64). llvm-svn: 159076
* BoundsChecking: attach debug info to traps to make my life a bit more saneNuno Lopes2012-06-231-9/+10
| | | | llvm-svn: 159055
* Revert remaining part of r93200: "Disable folding sext(trunc(x)) -> x"Jakob Stoklund Olesen2012-06-221-9/+4
| | | | | | | | | | | This fixes PR5997. These transforms were disabled because codegen couldn't deal with other uses of trunc(x). This is now handled by the peephole pass. This causes no regressions on x86-64. llvm-svn: 159003
* Fixed r158979.Stepan Dyatkovskiy2012-06-221-1/+2
| | | | | | | | | | Original message: Performance optimizations: - SwitchInst: case values stored separately from Operands List. It allows to make faster access to individual case value numbers or ranges. - Optimized IntItem, added APInt value caching. - Optimized IntegersSubsetGeneric: added optimizations for cases when subset is single number or when subset consists from single numbers only. llvm-svn: 158997
* fix whitespace in my last commit.Nuno Lopes2012-06-221-1/+1
| | | | | | sorry for the churn :S enough for today; going to sleep. llvm-svn: 158953
* remove extractMallocCallFromBitCast, since it was tailor maded for its sole ↵Nuno Lopes2012-06-221-2/+4
| | | | | | user. Update GlobalOpt accordingly. llvm-svn: 158952
* instcombine: disable optimization of 'invoke null/undef'. I'll move this ↵Nuno Lopes2012-06-211-11/+11
| | | | | | | | functionality to SimplifyCFG (since we cannot make changes to the CFG here). Fixes the crashes with the attached test case llvm-svn: 158951
* Look pass zext to strength reduce an udiv. Patch by David Majnemer. ↵Evan Cheng2012-06-211-1/+4
| | | | | | rdar://11721329 llvm-svn: 158946
* Add support for invoke to the MemoryBuiltin analysid.Nuno Lopes2012-06-212-2/+7
| | | | | | | | Update comments accordingly. Make instcombine remove useless invokes to C++'s 'new' allocation function (test attached). llvm-svn: 158937
* port the BoundsChecking patch to the new MemoryBuiltin API (i.e., remove ↵Nuno Lopes2012-06-211-393/+30
| | | | | | | | most of the code from here). Remove the alloc_size.ll test until we settle on a metadata format that makes everyone happy.. llvm-svn: 158920
* refactor the MemoryBuiltin analysis:Nuno Lopes2012-06-215-136/+19
| | | | | | | | | | | | - provide more extensive set of functions to detect library allocation functions (e.g., malloc, calloc, strdup, etc) - provide an API to compute the size and offset of an object pointed by Move a few clients (GVN, AA, instcombine, ...) to the new API. This implementation is a lot more aggressive than each of the custom implementations being replaced. Patch reviewed by Nick Lewycky and Chandler Carruth, thanks. llvm-svn: 158919
* Add a number of threshold arguments to the SRA pass.Nadav Rotem2012-06-211-18/+60
| | | | | | A patch by Tom Stellard with minor changes. llvm-svn: 158918
* replace usage of EmitGEPOffset() with TargetData::getIndexedOffset() when ↵Nuno Lopes2012-06-202-8/+6
| | | | | | | | | | the GEP offset is known to be constant. With this change, we avoid relying on the IR Builder to constant fold the operations. No functionality change intended. llvm-svn: 158829
* Fix two rather subtle internal vs. external linker issues.Chandler Carruth2012-06-201-25/+20
| | | | | | | | | | | | | | | | | | | | | | I'll admit I'm not entirely satisfied with this change, but it seemed the cleanest option. Other suggestions quite welcome The issue is that the traits specializations have static methods which return the typedef'ed PHI_iterator type. In both the IR and MI layers this is typedef'ed to a custom iterator class defined in an anonymous namespace giving the types and the functions returning them internal linkage. However, because the traits specialization is defined in the 'llvm' namespace (where it has to be, specialized template lives there), and is in turn used in the templated implementation of the SSAUpdater. This led to the linkage conflict that Clang now warns about. The simplest solution to me was just to define the PHI_iterator as a nested class inside the trait specialization. That way it still doesn't get scoped widely, it can't be accidentally reused somewhere, etc. This is a little gross just because nested class definitions are a little gross, but the alternatives seem more ad-hoc. llvm-svn: 158799
* Now that SROA can form alloca's for dynamic vector accesses, further improve ↵Pete Cooper2012-06-171-28/+99
| | | | | | it to be able to replace operations on these vector alloca's with insert/extract element insts llvm-svn: 158623
* Teach BBVectorize to combine, when possible, or discard metadata when fusing ↵Hal Finkel2012-06-161-0/+30
| | | | | | | | | | | | instructions. The present implementation handles only TBAA and FP metadata, discarding everything else. For debug metadata, the current behavior is maintained (the debug metadata associated with one of the instructions will be kept, discarding that attached to the other). This should address PR 13040. llvm-svn: 158606
* Move the Metadata merging methods from GVN and make them public in MDNode.Hal Finkel2012-06-161-153/+3
| | | | | | | There are other passes, BBVectorize specifically, that also need some of this functionality. llvm-svn: 158605
* It's not deterministic to iterate over SmallPtrSet. Replace it with ↵Evan Cheng2012-06-161-14/+14
| | | | | | SmallSetVector. Patch by Daniel Reynaud. rdar://11671029 llvm-svn: 158594
* Fix crash from r158529 on Bullet.Pete Cooper2012-06-161-1/+10
| | | | | | | | Dynamic GEPs created by SROA needed to insert extra "i32 0" operands to index through structs and arrays to get to the vector being indexed. llvm-svn: 158590
OpenPOWER on IntegriCloud