summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Transforms
Commit message (Collapse)AuthorAgeFilesLines
* Print correct instruction in dump.Dale Johannesen2009-04-291-1/+1
| | | | llvm-svn: 70427
* Reword and tidy up some comments.Dan Gohman2009-04-291-10/+21
| | | | llvm-svn: 70416
* Remove an obsolete comment.Dan Gohman2009-04-271-5/+3
| | | | llvm-svn: 70262
* Fix PR 4086, a bug in FP IV elimination.Dale Johannesen2009-04-271-2/+4
| | | | llvm-svn: 70247
* Permit ChangeCompareStride to rewrite a comparison when the factorDan Gohman2009-04-271-1/+3
| | | | | | | between the comparison's iv stride and the candidate stride is exactly -1. llvm-svn: 70244
* Return null instead of false, as appropriate.Dan Gohman2009-04-251-3/+3
| | | | llvm-svn: 70054
* Add several more icmp simplifications. Transform signed comparisonsDan Gohman2009-04-251-82/+159
| | | | | | | into unsigned ones when the operands are known to have the same sign bit value. llvm-svn: 70053
* Allow i16 type indices to gep.Sanjiv Gupta2009-04-241-6/+1
| | | | llvm-svn: 69946
* Change SCEVExpander's expandCodeFor to provide more flexibilityDan Gohman2009-04-231-15/+12
| | | | | | | | | | | with the persistent insertion point, and change IndVars to make use of it. This fixes a bug where IndVars was holding on to a stale insertion point and forcing the SCEVExpander to continue to use it. This fixes PR4038. llvm-svn: 69892
* Make sure both operands have binary instructions have the same type.Evan Cheng2009-04-221-11/+23
| | | | llvm-svn: 69844
* A few more places where the check of use_empty is needed.Evan Cheng2009-04-221-0/+3
| | | | llvm-svn: 69842
* Avoid deferencing use_begin() if value does not have a use.Evan Cheng2009-04-221-0/+1
| | | | llvm-svn: 69836
* Real fix for PR3549, by using caching for predecessor counts in addition to ↵Owen Anderson2009-04-221-11/+13
| | | | | | | | the predecessors themselves. This halves the time to optimize the testcase, beyond what my previous patch did. llvm-svn: 69792
* Use PredIteratorCache in LCSSA, which gives a 37% overall speedup on the ↵Owen Anderson2009-04-221-2/+5
| | | | | | testcase from PR3549. More improvements to come. llvm-svn: 69788
* use predicate instead of hand-rolled loopChris Lattner2009-04-211-11/+4
| | | | llvm-svn: 69752
* fix a crash on a pointless but valid zero-length memset, rdar://6808691Chris Lattner2009-04-211-13/+15
| | | | llvm-svn: 69680
* Factor out a common base class from SCEVTruncateExpr, SCEVZeroExtendExpr,Dan Gohman2009-04-211-6/+2
| | | | | | and SCEVSignExtendExpr. llvm-svn: 69649
* Introduce encapsulation for ScalarEvolution's TargetData object, and refactorDan Gohman2009-04-212-61/+51
| | | | | | the code to minimize dependencies on TargetData. llvm-svn: 69644
* Adjust loop size estimate for full unrolling;Dale Johannesen2009-04-201-0/+2
| | | | | | GEP's don't usually become instructions. llvm-svn: 69631
* Before trying to introduce/eliminate cast/ext/trunc to make indices type asSanjiv Gupta2009-04-201-1/+6
| | | | | | pointer type, make sure that the pointer size is a valid sequential index type. llvm-svn: 69574
* Use more const qualifiers with SCEV interfaces.Dan Gohman2009-04-182-33/+34
| | | | llvm-svn: 69450
* remove trailing whitespaceJim Grosbach2009-04-171-50/+50
| | | | llvm-svn: 69402
* Use a safer iterator interface and get rid of std C++ library misuse.David Greene2009-04-171-3/+2
| | | | | | This fixes a --enable-expensive-checks problem. llvm-svn: 69353
* Don't create ConstantInts with pointer type. This fixes aDan Gohman2009-04-171-1/+1
| | | | | | | regression in 403.gcc in PIC_CODEGEN=1 and DISABLE_LTO=1 mode. llvm-svn: 69344
* Use TargetData::getTypeSizeInBits instead of getPrimitiveSizeInBits()Dan Gohman2009-04-161-5/+5
| | | | | | to get the correct answer for pointer types. llvm-svn: 69321
* Fix for PR3944: make mem2reg O(N) instead of O(N^2) in the number of Eli Friedman2009-04-161-19/+11
| | | | | | incoming edges for a block with many predecessors. llvm-svn: 69312
* Minor code simplifications. Don't attempt LSR on theoreticalDan Gohman2009-04-161-12/+6
| | | | | | | targets with pointers larger than 64 bits, due to the code not yet being APInt clean. llvm-svn: 69296
* LSR is no longer a GEP optimizer. It is now an IV expressionDan Gohman2009-04-161-2/+1
| | | | | | optimizer, which just happen to frequently involve optimizing GEPs. llvm-svn: 69295
* Use ConstantExpr::getIntToPtr instead of SCEVExpander::InsertCastOfTo,Dan Gohman2009-04-161-3/+2
| | | | | | since the operand is always a constant. llvm-svn: 69291
* Use a SCEV expression cast instead of immediately inserting aDan Gohman2009-04-161-8/+8
| | | | | | new instruction with SCEVExpander::InsertCastOfTo. llvm-svn: 69290
* Expand GEPs in ScalarEvolution expressions. SCEV expressions can nowDan Gohman2009-04-162-309/+103
| | | | | | | | | | | | | | | | | | | | | have pointer types, though in contrast to C pointer types, SCEV addition is never implicitly scaled. This not only eliminates the need for special code like IndVars' EliminatePointerRecurrence and LSR's own GEP expansion code, it also does a better job because it lets the normal optimizations handle pointer expressions just like integer expressions. Also, since LLVM IR GEPs can't directly index into multi-dimensional VLAs, moving the GEP analysis out of client code and into the SCEV framework makes it easier for clients to handle multi-dimensional VLAs the same way as other arrays. Some existing regression tests show improved optimization. test/CodeGen/ARM/2007-03-13-InstrSched.ll in particular improved to the point where if-conversion started kicking in; I turned it off for this test to preserve the intent of the test. llvm-svn: 69258
* Eliminate zext over (iv | const) or (signed iv),Dale Johannesen2009-04-151-22/+77
| | | | | | | | and sext over (iv | const), if a longer iv is available. Allow expressions to have more than one zext/sext parent. All from OpenSSL. llvm-svn: 69241
* Eliminate zext over (iv & const) or ((iv+const)&const)Dale Johannesen2009-04-151-31/+89
| | | | | | | if a longer iv is available. These subscript forms are not common; they're a bottleneck in OpenSSL. llvm-svn: 69215
* Enhance induction variable code to remove theDale Johannesen2009-04-151-40/+121
| | | | | | | | | sext around sext(shorter IV + constant), using a longer IV instead, when it can figure out the add can't overflow. This comes up a lot in subscripting; mainly affects 64 bit. llvm-svn: 69123
* Avoid making the transformation enabled by my last patch if the new ↵Evan Cheng2009-04-151-3/+13
| | | | | | destinations have phi nodes. llvm-svn: 69121
* While inlining, clone llvm.dbg.func.start intrinsic and adjust Devang Patel2009-04-152-3/+52
| | | | | | llvm.dbg.region.end instrinsic. This nested llvm.dbg.func.start/llvm.dbg.region.end pair now enables DW_TAG_inlined_subroutine support in code generator. llvm-svn: 69118
* Optimize conditional branch on i1 phis with non-constant inputs.Evan Cheng2009-04-141-14/+55
| | | | | | | | | | | | | | | | | | | | | | | | | | | | This turns: eq: %3 = icmp eq i32 %1, %2 br label %join ne: %4 = icmp ne i32 %1, %2 br label %join join: %5 = phi i1 [%3, %eq], [%4, %ne] br i1 %5, label %yes, label %no => eq: %3 = icmp eq i32 %1, %2 br i1 %3, label %yes, label %no ne: %4 = icmp ne i32 %1, %2 br i1 %4, label %yes, label %no llvm-svn: 69102
* LoopIndexSplit needs to inform the loop pass manager of the instructions it isOwen Anderson2009-04-141-0/+1
| | | | | | deleting, not just the basic block. llvm-svn: 69011
* eliminate unneeded parens.Chris Lattner2009-04-131-2/+2
| | | | llvm-svn: 68939
* "There was a typo in my previous patch which leads to miscompilation ofChris Lattner2009-04-121-1/+1
| | | | | | | | | | | | | | strncat :( strncat(foo, "bar", 99) would be optimized to memcpy(foo+strlen(foo), "bar", 100, 1) instead of memcpy(foo+strlen(foo), "bar", 4, 1)" Patch by Benjamin Kramer! llvm-svn: 68905
* add some optimizations for strncpy/strncat and factor someChris Lattner2009-04-121-22/+114
| | | | | | code. Patch by Benjamin Kramer! llvm-svn: 68885
* Instcombine should not promote whole computation trees to "strange"Chris Lattner2009-04-081-0/+20
| | | | | | | | | | | | | integer types, unless they are already strange. This prevents it from turning the code produced by SROA into crazy libcalls and stuff that the code generator can't handle. In the attached example, the result was an i96 multiply that caused the x86 backend to assert. Note that if TargetData had an idea of what the legal types are for a target that this could be used to stop instcombine from introducing i64 muls, as Scott wanted. llvm-svn: 68598
* fix rdar://6762290, a crash compiling cxx filt with clang.Chris Lattner2009-04-071-3/+4
| | | | llvm-svn: 68500
* remove empty sectionChris Lattner2009-04-071-4/+0
| | | | llvm-svn: 68485
* Let the strcat optimizer return the pointer to the start of the buffer,Ed Schouten2009-04-061-2/+2
| | | | | | | | | instead of the place where it started to perform the string copy. - PR3661 - Patch by Benjamin Kramer! llvm-svn: 68443
* Reapply r68211, with the miscompilations it caused fixed.Owen Anderson2009-04-011-9/+30
| | | | llvm-svn: 68262
* Revert r68172. It caused regressions inDan Gohman2009-04-011-38/+10
| | | | | | | | Applications/Burg/burg Applications/ClamAV/clamscan and many other tests. llvm-svn: 68211
* Enhance GVN to propagate simple conditionals. This fixes PR3921.Owen Anderson2009-04-011-10/+38
| | | | llvm-svn: 68172
* Make the key of ValueRankMap an AssertingVH, so that we die violentlyChris Lattner2009-03-311-6/+7
| | | | | | if it dangles. llvm-svn: 68150
* Throttle back "fold select into operand" transformation. InstCombine should ↵Evan Cheng2009-03-311-52/+81
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | not generate selects of two constants unless they are selects of 0 and 1. e.g. define i32 @t1(i32 %c, i32 %x) nounwind { %t1 = icmp eq i32 %c, 0 %t2 = lshr i32 %x, 18 %t3 = select i1 %t1, i32 %t2, i32 %x ret i32 %t3 } was turned into define i32 @t2(i32 %c, i32 %x) nounwind { %t1 = icmp eq i32 %c, 0 %t2 = select i1 %t1, i32 18, i32 0 %t3 = lshr i32 %x, %t2 ret i32 %t3 } For most targets, that means materializing two constants and then a select. e.g. On x86-64 movl %esi, %eax shrl $18, %eax testl %edi, %edi cmovne %esi, %eax ret => xorl %eax, %eax testl %edi, %edi movl $18, %ecx cmovne %eax, %ecx movl %esi, %eax shrl %cl, %eax ret Also, the optimizer and codegen can reason about shl / and / add, etc. by a constant. This optimization will hinder optimizations using ComputeMaskedBits. llvm-svn: 68142
OpenPOWER on IntegriCloud