aboutsummaryrefslogtreecommitdiff
path: root/test/Transforms
AgeCommit message (Collapse)Author
2013-06-03PNaCl: ExpandVarArgs: Use memcpy() instead of struct load+store for struct argsMark Seaborn
Although PNaCl doesn't fully support struct types as varargs arguments, there is a test in Spec2k (255.vortex) that passes a struct as a varargs argument but never reads the argument using va_arg (which is legal, but strange). ExpandVarArgs was handling the struct argument by copying it with a struct load+store. This is undesirable because currently SROA converts that into code that uses extractvalue, which was rejected by the PNaCl ABI checker. Struct load/store will soon be rejected by the ABI checker too. We could fix this by running ExpandStructRegs after ExpandVarArgs, but it's cleaner for ExpandVarArgs to use memcpy() instead of struct load+store. memcpy() is potentially more efficient because it avoids having a temporary copy of the struct, and using memcpy() avoids dependencies between IR passes. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3338 BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343 TEST=*.ll tests + built 255.vortex from Spec2k Review URL: https://codereview.chromium.org/16232021
2013-06-01Merge 183060: Prevent loop-unroll from making assumptions about undefined ↵Andrew Trick
behavior. Fixes rdar:14036816, PR16130. There is an opportunity to compute precise trip counts for 'or' expressions and multi-exit loops. rdar:14038809: Optimize trip count computation for multi-exit loops. To do this we need to record the fact that ExitLimit assumes NSW. When it does not we can safely assume that the loop trip count is the minimum ExitLimt across all subexpressions and loop exits. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183060 91177308-0d34-0410-b5e6-96231b3b80d8 git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@183066 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-31PNaCl: FlattenGlobals: Use ptrtoint+add instead of getelementptr+bitcastMark Seaborn
The constructor for getelementptr ConstantExprs does some constant folding which can add 1 or 2 more indexes to the getelementptr. This complicates checking for FlattenGlobals' normal form in the PNaCl ABI checker. Worse, the GCC torture tests turned up a pathological case where this constant folding adds 4 indexes to the getelementptr and leaves the original struct type behind: @q = global i8* getelementptr inbounds ( %union.u* bitcast ([260 x i8]* @v to %union.u*), i32 0, i32 0, i32 0, i32 0, i32 4) That comes from the following code in gcc/testsuite/gcc.c-torture/execute/pr43784.c: struct s { unsigned char a[256]; }; union u { struct { struct s b; int c; } d; struct { int c; struct s b; } e; }; static union u v; static struct s *q = &v.e.b; We can fix this by using ptrtoint+add instead of getelementptr+bitcast, because ConstantExpr won't automatically convert ptrtoint to something else. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3113 TEST=*.ll tests + PNaCl toolchain trybots Review URL: https://codereview.chromium.org/15647009
2013-05-31PNaCl: Remove and disallow llvm.invariant.start/end intrinsicsMark Seaborn
These markers work in a similar way to llvm.lifetime.start/end, so we should remove them for similar reasons: it's not very well defined how one marker cancels out the effects of the other. Arguably, invariant.start/end are less useful than lifetime.start/end. They are ignored by the backend. They are generated in fewer places: invariant.start is generated by Clang (at -O1 or higher) when a const global is initialised with a non-POD initialiser. invariant.end is apparently not generated at all. Do the stripping in ReplacePtrsWithInts for consistency with the existing lifetime.start/end stripping. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3443 TEST=PNaCl toolchain trybots Review URL: https://codereview.chromium.org/15995004
2013-05-30PNaCl: Add a pass to expand out Clang's use of registers of struct typeMark Seaborn
Clang's implementation of C++ method pointers generates IR that uses LLVM registers with struct type -- specifically, loads and stores of struct values, and extractvalue instructions. See lib/CodeGen/ItaniumCXXABI.cpp in Clang. Add a pass, ExpandStructRegs, which expands out those uses. Factor out a function from ExpandArithWithOverflow so that the two passes can share some code. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343 TEST=*.ll tests + trybots + GCC torture tests Review URL: https://codereview.chromium.org/15692014
2013-05-30Add switch support to integer promotion passDerek Schuff
Odd-sized switch statements can appear in the sandboxed translator build. R=mseaborn@chromium.org BUG= https://code.google.com/p/nativeclient/issues/detail?id=3360 Review URL: https://codereview.chromium.org/15894006
2013-05-29PNaCl: Add ExpandSmallArguments pass to widen parameters to 32 bitsMark Seaborn
This widens i1, i8 and i16 function arguments and return types. Factor out RecreateFunction() helper function from existing PNaCl passes since this is a reoccurring code fragment. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3342 TEST=*.ll tests + PNaCl toolchain trybots + GCC torture tests + LLVM test suite Review URL: https://codereview.chromium.org/15971007
2013-05-29Fix PromoteIntegers pass to handle undef constantsDerek Schuff
ConvertConstant now returns an undef constant of the appropriate type. This fixes the translator build failure caused by enabling the pass. R=mseaborn@chromium.org BUG= https://code.google.com/p/nativeclient/issues/detail?id=3360 Review URL: https://codereview.chromium.org/16086005
2013-05-28PNaCl: Disable parts of InstCombine that introduce *.with.overflow intrinsicsMark Seaborn
Change the PNaCl ABI checker to disallow these intrinsics. Note that I had originally intended to commit this before my earlier change (https://codereview.chromium.org/15688011) that enables the ExpandArithWithOverflow pass. Enabling ExpandArithWithOverflow without changing InstCombine causes ExpandArithWithOverflow to fail on some of the *.with.overflow intrinsic calls that InstCombine introduces. This change therefore fixes some breakage. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3434 TEST=PNaCl toolchain trybots + GCC torture tests + LLVM test suite Review URL: https://codereview.chromium.org/16042011
2013-05-28Fix integer promotion pass to handle casts to struct pointer types.Derek Schuff
Also fix the diagnostic asserts in getPromotedType. R=mseaborn@chromium.org BUG= https://code.google.com/p/nativeclient/issues/detail?id=3360 Review URL: https://codereview.chromium.org/16004003
2013-05-28PNaCl: Extend ExpandMulWithOverflow pass to handle uadd.with.overflow tooMark Seaborn
It turned out that umul.with.overflow wasn't the only *.with.overflow intrinsic usage introduced by Clang. I knew that Clang's CGExprCXX.cpp generates umul.with.overflow for an overflow check for C++'s "new Foo[]". The same code for handling "new Foo[]" also generates uadd.with.overflow in some cases. This happens if class Foo has a destructor or a delete[] operator that takes a size argument. In those cases, the C++ ABI adds a "cookie" to the allocation which contains the array's size. Rename the pass to "ExpandArithWithOverflow" and rename files accordingly. Also enable the pass. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3434 TEST=*.ll tests + trybots + GCC torture tests Review URL: https://codereview.chromium.org/15688011
2013-05-28Merging r182656:Bill Wendling
------------------------------------------------------------------------ r182656 | d0k | 2013-05-24 11:05:35 -0700 (Fri, 24 May 2013) | 3 lines LoopVectorize: LoopSimplify can't canonicalize loops with an indirectbr in it, don't assert on those cases. Fixes PR16139. ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@182785 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-28Add two passes that implement conversions from PNaCl's specific intrinsicsEli Bendersky
to external function calls during the translation stage (llc). One of the passes is a ModulePass that adds the appropriate function declarations to the module. The other is a FunctionPass that performs the actual call replacement. This split exists because of bitcode streaming. Initially the passes handle the llvm.nacl.{set|long}jmp intrinsics. In the future they may handle additional intrinsics that are part of the PNaCl stable bitcode ABI. This CL also removes the previous approach to handling this conversion (in SelectionDAGBuilder.cpp). That ended up not working - more details in issue 3429. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3429 R=mseaborn@chromium.org Review URL: https://codereview.chromium.org/16047002
2013-05-24PNaCl: Add "-pnacl-abi-simplify-{pre,post}opt" meta-passes to "opt"Mark Seaborn
These meta-passes will be used to replace the pass lists that are currently in the pnacl-ld.py driver script in the NaCl repo. I've moved the comments across from pnacl-ld.py and added a couple more comments for ExpandByVal and StripMetadata. Fix the declaration of createResolveAliasesPass(). BUG=https://code.google.com/p/nativeclient/issues/detail?id=3435 TEST=new *.ll tests + tested with change to pnacl-ld.py Review URL: https://codereview.chromium.org/15669002
2013-05-24PNaCl: Add pass to expand out Clang-generated umul.with.overflow callsMark Seaborn
This adds a pass, ExpandMulWithOverflow, to expand out the llvm.umul.with.overflow calls that Clang generates to implement an overflow check for C++'s new[] operator. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3434 TEST=expand-mul-with-overflow.ll Review URL: https://codereview.chromium.org/14649027
2013-05-24PNaCl: Fix ReplacePtrsWithInts to handle some corner cases correctlyMark Seaborn
Running the LLVM test suite with the ReplacePtrsWithInts pass enabled produced a single failure (in MultiSource/Applications/SPASS), revealing a corner case in which a mixture of forward and backward references plus a bitcast causes the pass to fail (see @forwards_reference() in the test). The problem was that we were doing replaceAllUsesWith() on a placeholder value too early. RewriteMap was mapping a bitcast to a placeholder P, but RewriteMap's reference to P didn't get updated by P->replaceAllUsesWith() and P became a dangling pointer. The fix is: * Change convert() to strip off casts first, so that RewriteMap isn't used for mapping casts to converted values. * Defer the replaceAllUsesWith() calls until after creating all the replacement instructions. This makes the pass more robust against instruction ordering in the input. This requires debug instrinsics to be updated in a separate pass, because replaceAllUsesWith() doesn't work for references by metadata nodes. This also fixes some pathological corner cases of cyclic references in unreachable blocks. Fix indentation in one place. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343 TEST=replace-ptrs-with-ints.ll + LLVM test suite Review URL: https://codereview.chromium.org/15761003
2013-05-22PNaCl: Add ReplacePtrsWithInts pass for stripping out pointer typesMark Seaborn
ReplacePtrsWithInts converts IR to a normal form in which functions don't reference any aggregate pointer types and pointer types only appear inside a few instructions. Change BlockAddress::replaceUsesOfWithOnConstant() to handle changing a function's type by replacing a function with a bitcast ConstantExpr of a new function. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343 TEST=replace-ptrs-with-ints.ll + PNaCl toolchain trybots, torture tests, etc. Review URL: https://codereview.chromium.org/14262011
2013-05-22Merging r182485:Bill Wendling
------------------------------------------------------------------------ r182485 | arnolds | 2013-05-22 09:54:56 -0700 (Wed, 22 May 2013) | 7 lines LoopVectorize: Make Value pointers that could be RAUW'ed a VH The Value pointers we store in the induction variable list can be RAUW'ed by a call to SCEVExpander::expandCodeFor, use a TrackingVH instead. Do the same thing in some other places where we store pointers that could potentially be RAUW'ed. Fixes PR16073. ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@182492 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-22PNaCl: Fix ExpandCtors to handle an empty list caseMark Seaborn
While writing a test, I noticed that ExpandCtors crashes if given the empty list "[]", because this gets converted into an UndefValue ConstantExpr by the LLVM assembly reader. Fix this by checking the array's size via its type. This replaces the isNullValue() check. Make error handling cleaner by splitting out a separate function. BUG=none TEST=test/Transforms/NaCl/expand-ctors-emptylist.ll Review URL: https://codereview.chromium.org/15659005
2013-05-20Fix stderr redirection in test to work on all platformsEli Bendersky
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3429 R=mseaborn@chromium.org Review URL: https://codereview.chromium.org/15470008
2013-05-20PNaCl: Fix FlattenGlobals to correctly handle implicitly-aligned variablesMark Seaborn
If a global variable has no "align" attribute, it must be aligned based on its type. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3437 TEST=flatten-globals.ll Review URL: https://codereview.chromium.org/15359006
2013-05-16Temporarily remove test that fails on MacEli Bendersky
BUG=None Review URL: https://codereview.chromium.org/15100013
2013-05-15Merging r181524:Bill Wendling
------------------------------------------------------------------------ r181524 | rafael | 2013-05-09 10:22:59 -0700 (Thu, 09 May 2013) | 4 lines Don't replace an alias in llvm.used with its target. When we replace an internal alias with its target, be careful not to replace the entry in llvm.used (and llvm.compiler_used). ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@181909 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-14Update tests to use FileCheck instead of grep (which is deprecated in the ↵Eli Bendersky
LLVM regression suite) BUG=None R=mseaborn@chromium.org Review URL: https://codereview.chromium.org/15042009
2013-05-14Merging r181586:Bill Wendling
------------------------------------------------------------------------ r181586 | d0k | 2013-05-10 02:16:52 -0700 (Fri, 10 May 2013) | 3 lines InstCombine: Verify the type before transforming uitofp into select. PR15952. ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@181813 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-14Adding a pass - RewritePNaClLibraryCalls, that replaces known library calls ↵Eli Bendersky
with stable bitcode intrinsics. Starting with setjmp and longjmp. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3429 R=jvoung@chromium.org, mseaborn@chromium.org Review URL: https://codereview.chromium.org/14617017
2013-05-10PNaCl ABI: Promote illegal integer typesDerek Schuff
This pass (mostly) legalizes integer types by promoting them. It has some limitations (e.g. it can't change function types) but it is more than sufficient for what clang and SROA generate. A more significant limitation of promotion is that packed bitfields of size > 64 bits are still not handled. There are none in our tests (other than callingconv_case_by_case which doesn't require a stable pexe) but we will want to either handle them by correctly expanding them, or find a better way to error out. BUG= https://code.google.com/p/nativeclient/issues/detail?id=3360 R=eliben@chromium.org, jvoung@chromium.org Review URL: https://codereview.chromium.org/14569012
2013-05-09PNaCl: Add NoAlias attributes in ExpandByVal and ExpandVarArgs passesMark Seaborn
This could help prevent these expansion passes from inhibiting optimisations than run after the expansion. e.g. It gives the optimiser more freedom to move around reads from the varargs buffer because they will not alias writes to other locations. BUG=https://code.google.com/p/nativeclient/issues/detail?id=3400 TEST=PNaCl toolchain trybots + GCC torture tests + LLVM test suite + Spec2k Review URL: https://codereview.chromium.org/14060026
2013-05-08Merging r181286:Bill Wendling
------------------------------------------------------------------------ r181286 | arnolds | 2013-05-06 21:37:05 -0700 (Mon, 06 May 2013) | 7 lines LoopVectorize: getConsecutiveVector must respect signed arithmetic We were passing an i32 to ConstantInt::get where an i64 was needed and we must also pass the sign if we pass negatives numbers. The start index passed to getConsecutiveVector must also be signed. Should fix PR15882. ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@181455 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06InstCombine: (X ^ signbit) + C -> X + (signbit ^ C)David Majnemer
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181249 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Test results verified using FileCheck rather than grep | countJean-Luc Duprat
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181234 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Rotate multi-exit loops even if the latch was simplified.Andrew Trick
Test case by Michele Scandale! Fixes PR10293: Load not hoisted out of loop with multiple exits. There are few regressions with this patch, now tracked by rdar:13817079, and a roughly equal number of improvements. The regressions are almost certainly back luck because LoopRotate has very little idea of whether rotation is profitable. Doing better requires a more comprehensive solution. This checkin is a quick fix that lacks generality (PR10293 has a counter-example). But it trivially fixes the case in PR10293 without interfering with other cases, and it does satify the criteria that LoopRotate is a loop canonicalization pass that should avoid heuristics and special cases. I can think of two approaches that would probably be better in the long run. Ultimately they may both make sense. (1) LoopRotate should check that the current header would make a good loop guard, and that the loop does not already has a sufficient guard. The artifical SimplifiedLoopLatch check would be unnecessary, and the design would be more general and canonical. Two difficulties: - We need a strong guarantee that we won't endlessly rotate, so the analysis would need to be precise in order to avoid the SimplifiedLoopLatch precondition. - Analysis like this are usually based on SCEV, which we don't want to rely on. (2) Rotate on-demand in late loop passes. This could even be done by shoving the loop back on the queue after the optimization that needs it. This could work well when we find LICM opportunities in multi-branch loops. This requires some work, and it doesn't really solve the problem of SCEV wanting a loop guard before the analysis. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181230 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Fix add4.ll test cmdline so that it passesJean-Luc Duprat
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181219 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Provide InstCombines for the following 3 cases:Jean-Luc Duprat
A * (1 - (uitofp i1 C)) -> select C, 0, A B * (uitofp i1 C) -> select C, B, 0 select C, 0, A + select C, B, 0 -> select C, B, A These come up in code that has been hand-optimized from a select to a linear blend, on platforms where that may have mattered. We want to undo such changes with the following transform: A*(1 - uitofp i1 C) + B*(uitofp i1 C) -> select C, A, B git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181216 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Revert r164763 because it introduces new shuffles.Nadav Rotem
Thanks Nick Lewycky for pointing this out. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181177 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Fix unchecked uses of DominatorTree in MemoryDependenceAnalysis.Matt Arsenault
Use unknown results for places where it would be needed git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181176 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-06Fix const merging when an alias of a const is llvm.used.Rafael Espindola
We used to disable constant merging not only if a constant is llvm.used, but also if an alias of a constant is llvm.used. This change fixes that. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181175 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-05LoopVectorize: Add support for floating point min/max reductionsArnold Schwaighofer
Add support for min/max reductions when "no-nans-float-math" is enabled. This allows us to assume we have ordered floating point math and treat ordered and unordered predicates equally. radar://13723044 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181144 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-05LoopVectorize: We don't need an identity element for min/max reductionsArnold Schwaighofer
We can just use the initial element that feeds the reduction. max(max(x, y), z) == max(max(x,y), max(x,z)) radar://13723044 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181141 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-03LoopVectorizer: Add support for if-conversion of PHINodes with 3+ incoming ↵Nadav Rotem
values. By supporting the vectorization of PHINodes with more than two incoming values we can increase the complexity of nested if statements. We can now vectorize this loop: int foo(int *A, int *B, int n) { for (int i=0; i < n; i++) { int x = 9; if (A[i] > B[i]) { if (A[i] > 19) { x = 3; } else if (B[i] < 4 ) { x = 4; } else { x = 5; } } A[i] = x; } } git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181037 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-02TBAA: remove !tbaa from testing cases if not used.Manman Ren
This will make it easier to turn on struct-path aware TBAA since the metadata format will change. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180935 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-02Add a test for the foldSelectICmpAndOr fix committed in r180779.David Majnemer
This tests a case where C1 and C2 were the same but X and Y were different widths. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180907 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-01SROA: Generate selects instead of shuffles when blending values because this ↵Nadav Rotem
is the cannonical form. Shuffles are more difficult to lower and we usually don't touch them, while we do optimize selects more often. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180875 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-01Revert "InstCombine: Fold more shuffles of shuffles."Jim Grosbach
This reverts commit r180802 There's ongoing discussion about whether this is the right place to make this transformation. Reverting for now while we figure it out. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180834 91177308-0d34-0410-b5e6-96231b3b80d8
2013-04-30InstCombine: Fold more shuffles of shuffles.Jim Grosbach
Always fold a shuffle-of-shuffle into a single shuffle when there's only one input vector in the first place. Continue to be more conservative when there's multiple inputs. rdar://13402653 PR15866 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180802 91177308-0d34-0410-b5e6-96231b3b80d8
2013-04-30TBAA: remove !tbaa from testing cases if not used.Manman Ren
This will make it easier to turn on struct-path aware TBAA since the metadata format will change. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180796 91177308-0d34-0410-b5e6-96231b3b80d8
2013-04-30Fix "Combine bit test + conditional or into simple math"David Majnemer
This fixes the optimization introduced in r179748 and reverted in r179750. While the optimization was sound, it did not properly respect differences in bit-width. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180777 91177308-0d34-0410-b5e6-96231b3b80d8
2013-04-29Use unique'ed types for varargs expansion instead of non-unique named types.Jan Voung
Saves a tiny bit of space for var-args heavy bitcode programs, since the anonymous types get unique'ed. E.g., saves 20KB out of 1.6MB in spec gcc when comparing the gzipped files, or about 100KB when not zipped. This is only a savings with the current wire format. If we change the alloca, etc. to only have sizes and not struct types then we would also not have this duplication. Just happened to notice while looking through code for what struct types remain used in the bitcode. random cleanup for: BUG= https://code.google.com/p/nativeclient/issues/detail?id=3338 R=mseaborn@chromium.org Review URL: https://codereview.chromium.org/14197004
2013-04-29SimplifyCFG: If convert single conditional storesArnold Schwaighofer
This resurrects r179957, but adds code that makes sure we don't touch atomic/volatile stores: This transformation will transform a conditional store with a preceeding uncondtional store to the same location: a[i] = may-alias with a[i] load if (cond) a[i] = Y into an unconditional store. a[i] = X may-alias with a[i] load tmp = cond ? Y : X; a[i] = tmp We assume that on average the cost of a mispredicted branch is going to be higher than the cost of a second store to the same location, and that the secondary benefits of creating a bigger basic block for other optimizations to work on outway the potential case where the branch would be correctly predicted and the cost of the executing the second store would be noticably reflected in performance. hmmer's execution time improves by 30% on an imac12,2 on ref data sets. With this change we are on par with gcc's performance (gcc also performs this transformation). There was a 1.2 % performance improvement on a ARM swift chip. Other tests in the test-suite+external seem to be mostly uninfluenced in my experiments: This optimization was triggered on 41 tests such that the executable was different before/after the patch. Only 1 out of the 40 tests (dealII) was reproducable below 100% (by about .4%). Given that hmmer benefits so much I believe this to be a fair trade off. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180731 91177308-0d34-0410-b5e6-96231b3b80d8
2013-04-29[objc-arc] Apply the RV optimization to retains next to calls in ↵Michael Gottesman
ObjCARCContract instead of ObjCARCOpts. Turning retains into retainRV calls disrupts the data flow analysis in ObjCARCOpts. Thus we move it as late as we can by moving it into ObjCARCContract. We leave in the conversion from retainRV -> retain in ObjCARCOpt since it enables the dataflow analysis. rdar://10813093 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180698 91177308-0d34-0410-b5e6-96231b3b80d8
generated by cgit v1.2.3-18-g5258 (git 2.32.0) at 2025-07-03 17:42:34 +0000