aboutsummaryrefslogtreecommitdiff
path: root/lib/Transforms
AgeCommit message (Collapse)Author
2010-02-03Adjust the heuristics used to decide when SROA is likely to be profitable.Bob Wilson
The SRThreshold value makes perfect sense for checking if an entire aggregate should be promoted to a scalar integer, but it is not so good for splitting an aggregate into its separate elements. A struct may contain a large embedded array along with some scalar fields that would benefit from being split apart by SROA. Even if the total aggregate size is large, it may still be good to perform SROA. Thus, the most important piece of this patch is simply moving the aggregate size comparison vs. SRThreshold so that it guards only the aggregate promotion. We have also been checking the number of elements to decide if an aggregate should be split up. The limit of "SRThreshold/4" seemed rather arbitrary, and I don't think it's very useful to derive this limit from SRThreshold anyway. I've collected some data showing that the current default limit of 32 (since SRThreshold defaults to 128) is a reasonable cutoff for struct types. One thing suggested by the data is that distinguishing between structs and arrays might be useful. There are (obviously) a lot more large arrays than large structs (as measured by the number of elements and not the total size -- a large array inside a struct still counts as a single element given the way we do SROA right now). Out of 8377 arrays where we successfully performed SROA while compiling a large set of benchmarks, only 16 of them had more than 8 elements. And, for those 16 arrays, it's not at all clear that SROA was actually beneficial. So, to offset the compile time cost of investigating more large structs for SROA, the patch lowers the limit on array elements to 8. This fixes Apple Radar 7563690. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95224 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-03Revert 94937 and move the noreturn check to codegen.Evan Cheng
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95198 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-03Fix some comment typos.Bob Wilson
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95170 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-03Recommit this, looks like it wasn't the cause.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95165 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02Hopefully temporarily revert this.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95154 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02Reformat my last patch slightly.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95147 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02Re-add strcmp and known size object size checking optimization.Eric Christopher
Passed bootstrap and nightly test run here. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95145 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02don't turn (A & (C0?-1:0)) | (B & ~(C0?-1:0)) -> C0 ? A : BChris Lattner
for vectors. Codegen is generating awful code or segfaulting in various cases (e.g. PR6204). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95058 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02fix a crash in loop unswitch on a loop invariant vector condition.Chris Lattner
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95055 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02LangRef.html says that inttoptr and ptrtoint always use zero-extensionDan Gohman
when the cast is extending. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95046 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02Don't need to check the last argument since it'll always be bool. We alsoEric Christopher
don't use TargetData here. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95040 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02More indentation/tabification fixes.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95036 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-02Untabify previous commit.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95035 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01Formatting.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95027 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01Add an option to GVN to remove all partially redundant loads. This is currentlyBob Wilson
disabled by default. This divides the existing load PRE code into 2 phases: first it checks that it is safe to move the load to each of the predecessors where it is unavailable, and then if it is safe, the code is changed to move the load. Radar 7571861. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95007 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01cleanups.Chris Lattner
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94995 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01fix rdar://7590304, a miscompilation of objc apps on arm. The callerChris Lattner
of objc message send was getting marked arm_apcscc, but the prototype isn't. This is fine at runtime because objcmsgsend is implemented in assembly. Only turn a mismatched caller and callee into 'unreachable' if the callee is a definition. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94986 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01fix rdar://7590304, an infinite loop in instcombine. In the invokeChris Lattner
case, instcombine can't zap the invoke for fear of changing the CFG. However, we have to do something to prevent the next iteration of instcombine from inserting another store -> undef before the invoke thereby getting into infinite iteration between dead store elim and store insertion. Just zap the callee to null, which will prevent the next iteration from doing anything. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94985 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-01Fix pr6198 by moving the isSized() check to an outer conditional.Bob Wilson
The testcase from pr6198 does not crash for me -- I don't know what's up with that -- so I'm not adding it to the tests. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94984 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-31Simplify/generalize the xor+add->sign-extend instcombine.Eli Friedman
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94943 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-31Add a small transform: transform -(X<<Y) to (-X<<Y) when the shift has a singleEli Friedman
use and X is free to negate. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94941 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-31Do not mark no-return calls tail calls. It'll screw up special calls like ↵Evan Cheng
longjmp and it doesn't make much sense for performance reason. If my logic is faulty, please let me know. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94937 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-30Check alignment of loads when deciding whether it is safe to execute themBob Wilson
unconditionally. Besides checking the offset, also check that the underlying object is aligned as much as the load itself. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94875 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-30Use more specific types to avoid casts. No functionality change.Bob Wilson
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94863 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Keep iterating over all uses when meeting a phi node in ↵Jakob Stoklund Olesen
AllUsesOfValueWillTrapIfNull(). This bug was exposed by my inliner cost changes in r94615, and caused failures of lencod on most architectures when building with LTO. This patch fixes lencod and 464.h264ref on x86-64 (and likely others). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94858 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Preserve load alignment in instcombine transformations. I've been unable toBob Wilson
create a testcase where this matters. The select+load transformation only occurs when isSafeToLoadUnconditionally is true, and in those situations, instcombine also changes the underlying objects to be aligned. This seems like a good idea regardless, and I've verified that it doesn't pessimize the subsequent realignment. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94850 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Revert my last couple of patches. They appear to have broken bison.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94841 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Use uint64_t instead of unsigned for offsets and sizes.Bob Wilson
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94835 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Improve isSafeToLoadUnconditionally to recognize that GEPs with constantBob Wilson
indices are safe if the result is known to be within the bounds of the underlying object. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94829 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Having RHSKnownZero and RHSKnownOne be alternative names for KnownZero and ↵Duncan Sands
KnownOne (via APInt &RHSKnownZero = KnownZero, etc) seems dangerous and confusing to me: it is easy not to notice this, and then wonder why KnownZero/RHSKnownZero changed underneath you when you modified RHSKnownZero/KnownZero etc. So get rid of this. No intended functionality change (tested with "make check" + llvm-gcc bootstrap). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94802 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Make strcpy_chk lower to strcpy if we have a safe size.Eric Christopher
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94783 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Add constant support to object size handling and remove defaultEric Christopher
lowering. We'll either figure it out, or not and be lowered by SelectionDAGBuild. Add test. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94775 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Generic reformatting and comment fixing. No functionality change.Bill Wendling
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94771 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29Add newline to debugging output, and fix some grammar-os in comment.Bill Wendling
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94765 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-29mem2reg erases the dbg.declare intrinsics that it converts to dbg.val intrinsicsVictor Hernandez
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94763 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-28Fix PR6165. The bug was that LHSKnownZero was being and'd with DemandedMaskDuncan Sands
when it should have been and'd with LowBits. Fix that and while there beef up the logic in the case of a negative LHS. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94745 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27Avoid creating redundant PHIs in SSAUpdater::GetValueInMiddleOfBlock.Bob Wilson
This was already being done in SSAUpdater::GetValueAtEndOfBlock so I've just changed SSAUpdater to check for existing PHIs in both places. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94690 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27Kill ModuleProvider and ghost linkage by inverting the relationship betweenJeffrey Yasskin
Modules and ModuleProviders. Because the "ModuleProvider" simply materializes GlobalValues now, and doesn't provide modules, it's renamed to "GVMaterializer". Code that used to need a ModuleProvider to materialize Functions can now materialize the Functions directly. Functions no longer use a magic linkage to record that they're materializable; they simply ask the GVMaterializer. Because the C ABI must never change, we can't remove LLVMModuleProviderRef or the functions that refer to it. Instead, because Module now exposes the same functionality ModuleProvider used to, we store a Module* in any LLVMModuleProviderRef and translate in the wrapper methods. The bindings to other languages still use the ModuleProvider concept. It would probably be worth some time to update them to follow the C++ more closely, but I don't intend to do it. Fixes http://llvm.org/PR5737 and http://llvm.org/PR5735. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94686 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27Don't bother with sprintf, just pass the Twine through.Benjamin Kramer
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94684 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27Use the less expensive getName function instead of getNameStr.Benjamin Kramer
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94683 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27some cleanups.Chris Lattner
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94649 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27no need to check for nullChris Lattner
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94648 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-27When converting dbg.declare to dbg.value, attach promoted store's debug ↵Victor Hernandez
metadata to dbg.value git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94634 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-26Avoid extra calls to MD->getNumOperands()Victor Hernandez
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94618 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-26Switch AllocaDbgDeclares to SmallVector and don't leak DIFactoryVictor Hernandez
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94567 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-26In mem2reg, for all alloca/stores that get promoted where the alloca has an ↵Victor Hernandez
associated llvm.dbg.declare instrinsic, insert an llvm.dbg.var intrinsic before each store. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94493 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-25Remove check for an impossible condition: the condition of the while loop hasBob Wilson
already checked that TmpBB->getSinglePredecessor() is non-null. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94451 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-25Change Value::getUnderlyingObject to have the MaxLookup value specified as aBob Wilson
parameter with a default value, instead of just hardcoding it in the implementation. The limit of MaxLookup = 6 was introduced in r69151 to fix a performance problem with O(n^2) behavior in instcombine, but the scalarrepl pass is relying on getUnderlyingObject to go all the way back to an AllocaInst. Making the limit part of the method signature makes it clear that by default the result is limited and should help avoid similar problems in the future. This fixes pr6126. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94433 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-25Revert r94260 until findDbgDeclare() is made more efficientVictor Hernandez
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94432 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-24make -fno-rtti the default unless a directory builds with REQUIRES_RTTI.Chris Lattner
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94378 91177308-0d34-0410-b5e6-96231b3b80d8
generated by cgit v1.2.3-18-g5258 (git 2.32.0) at 2025-06-23 18:56:21 +0000