RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-28 06:22:51 +01:00
Code Issues Projects Releases Wiki Activity
54,426 Commits 4 Branches 5 Tags 704 MiB
76fbcdaaa7
Commit Graph

3807 Commits

Author SHA1 Message Date
Chris Lattner
00c9eb665d fix a nasty iterator invalidation bug from my conversion from 
std::map to DenseMap, exposed on release llvm-gcc bootstrap.

llvm-svn: 85840
 2009-11-02 23:25:39 +00:00
Chris Lattner
6f515d4ba8 revert r8579[56], which are causing unhappiness in buildbot land. 
llvm-svn: 85818
 2009-11-02 19:31:10 +00:00
Chris Lattner
a1776913ab disable IPSCCP support for multiple return values, it is buggy, so just 
disable it until I can fix it.

llvm-svn: 85810
 2009-11-02 18:22:51 +00:00
Chris Lattner
f1afb57935 improve IPSCCP to be able to propagate the result of "!mayBeOverridden" 
function to calls of that function, regardless of whether it has local
linkage or has its address taken.  Not escaping should only affect 
whether we make an aggressive assumption about the arguments to a 
function, not whether we can track the result of it.

llvm-svn: 85795
 2009-11-02 07:33:59 +00:00
Chris Lattner
c9279a9b4d don't mark the arguments of prototype overdefined, they will never be queried. 
llvm-svn: 85793
 2009-11-02 06:34:04 +00:00
Chris Lattner
9c7e443b07 restore some code I removed in r85788, refactor it into 
a shared place instead of duplicating it 4 times.

llvm-svn: 85792
 2009-11-02 06:28:16 +00:00
Chris Lattner
23f5603692 remove some confused code that dates from when we had 
"multiple return values" but not "first class aggregates"

llvm-svn: 85791
 2009-11-02 06:17:06 +00:00
Chris Lattner
c670f866ce avoid redundant lookups in BBExecutable, and make it a SmallPtrSet. 
llvm-svn: 85790
 2009-11-02 06:11:23 +00:00
Chris Lattner
16b825e51f Use the libanalysis 'ConstantFoldLoadFromConstPtr' function 
instead of reinventing SCCP-specific logic.  This gives us
new powers.

llvm-svn: 85789
 2009-11-02 06:06:14 +00:00
Chris Lattner
32acebedf7 switch the main 'ValueState' map from being an std::map to being 
a DenseMap.  Doing this required being aware of subtle iterator
invalidation issues, but it provides a big speedup.  In a 
release-asserts build, this sped up optimizing 403.gcc from
1.34s -> 0.79s (IPSCCP) and 1.11s -> 0.44s (SCCP).

This commit also conflates in a bunch of general cleanups, sorry.

llvm-svn: 85788
 2009-11-02 05:55:40 +00:00
Chris Lattner
7e74831e52 only IPSCCP incoming arguments if the function is executable, this fixes 
an assertion on the buildbot.

llvm-svn: 85784
 2009-11-02 03:25:55 +00:00
Chris Lattner
0a8c553eb2 add a new ValueState::getConstantInt() helper, use it to 
simplify some code.

llvm-svn: 85783
 2009-11-02 03:21:36 +00:00
Chris Lattner
4372b58329 tidy up some more: remove some extraneous inline specifiers, return harder. 
llvm-svn: 85780
 2009-11-02 03:03:42 +00:00
Chris Lattner
1c8cb53667 eliminate the SCCPSolver::getValueMapping method. 
llvm-svn: 85778
 2009-11-02 02:54:24 +00:00
Chris Lattner
f02562df08 fix failures introduced in r85774 
llvm-svn: 85777
 2009-11-02 02:48:17 +00:00
Chris Lattner
d4f286cb0c factor duplicated code into a new DeleteInstructionInBlock 
function, eliminate temporary (and pointless) smallvector.

llvm-svn: 85776
 2009-11-02 02:47:51 +00:00
Chris Lattner
a806b18c99 Chris used to use '...' instead of proper grammar. 
llvm-svn: 85775
 2009-11-02 02:33:50 +00:00
Chris Lattner
37dc1cb0fb remove some extraneous llvmcontext stuff. 
llvm-svn: 85774
 2009-11-02 02:30:06 +00:00
Chris Lattner
28db47c45a change LatticeVal to use PointerIntPair to save some space. 
llvm-svn: 85773
 2009-11-02 02:20:32 +00:00
Chris Lattner
b94afb778c fix instcombine to only do store sinking when the alignments 
of the two loads agree.  Propagate that onto the new store.

llvm-svn: 85772
 2009-11-02 02:06:37 +00:00
Chris Lattner
f3c2039dde when merging two loads, make sure to take the min of their alignment, 
not the max.  This didn't matter until the previous patch because
instcombine would refuse to sink loads with differenting alignments.

llvm-svn: 85738
 2009-11-01 20:07:07 +00:00
Chris Lattner
e7b1e067e9 split load sinking out to its own function, like gep sinking. 
llvm-svn: 85737
 2009-11-01 20:04:24 +00:00
Chris Lattner
b3fe1009a5 fix a bug noticed by inspection: when instcombine sinks loads through 
phis, it didn't preserve the alignment of the load.  This is a missed
optimization of the alignment is high and a miscompilation when the
alignment is low.

llvm-svn: 85736
 2009-11-01 19:50:13 +00:00
Chris Lattner
ad2926bbb4 remove a bunch of locking from LLVMContextImpl. Since only one thread 
can be banging on a context at a time, this isn't needed.  Owen, please
review.

llvm-svn: 85728
 2009-11-01 18:42:03 +00:00
Chris Lattner
d8f9d01b76 teach ipsccp and ipconstprop that a blockaddress doesn't 'take the address' of a function 
in a way that should prevent ip constprop.  This allows clang/test/CodeGen/indirect-goto.c
to pass with the new indirect goto lowering.

llvm-svn: 85709
 2009-11-01 06:11:53 +00:00
Chris Lattner
746b4c0b20 strengthen an assumption: RevectorBlockTo knows that PredBB 
ended in an uncond branch because the pass requires BreakCriticalEdges.

However, BCE doesn't eliminate critical adges from indbrs.

llvm-svn: 85707
 2009-11-01 04:23:20 +00:00
Chris Lattner
0ee6bbbeb1 if CostMetrics says to never duplicate some code, don't unswitch a loop. 
This prevents unswitching from duplicating indbr's.

llvm-svn: 85705
 2009-11-01 03:42:55 +00:00
Chris Lattner
1f101c7a1a Make sure PRE doesn't split crit edges from indirectbr. 
llvm-svn: 85692
 2009-10-31 22:11:15 +00:00
Chris Lattner
43591a021a llvm::SplitEdge should refuse to split an edge from an indirectbr. 
Fix CodeGenPrepare to not try to split edges from indirectbr.

llvm-svn: 85690
 2009-10-31 22:04:43 +00:00
Chris Lattner
0be6fcef18 add a comment. 
llvm-svn: 85671
 2009-10-31 17:48:31 +00:00
Dan Gohman
0079f87732 Rename forgetLoopBackedgeTakenCount to forgetLoop, because it 
clears out more information than just the stored backedge taken count.

llvm-svn: 85664
 2009-10-31 15:04:55 +00:00
Dan Gohman
de593b49b6 Replace LoopUnrollPass.cpp's custom code-size estimation code using 
the new common CodeMetrics code.

llvm-svn: 85663
 2009-10-31 14:54:17 +00:00
Dan Gohman
0f8d034e2f Remove an unnecessary #include. 
llvm-svn: 85661
 2009-10-31 14:39:43 +00:00
Dan Gohman
793119ef88 Update CMakeLists for recent renames. 
llvm-svn: 85660
 2009-10-31 14:38:25 +00:00
Dan Gohman
0186fd092a Rename UnrollLoop.cpp to LoopUnroll.cpp, and LoopUnroll.cpp to 
LoopUnrollPass.cpp, for consistency with other passes which are
similarly split.

llvm-svn: 85659
 2009-10-31 14:37:31 +00:00
Dan Gohman
04c536dff6 Remove CodeGenLICM. It's largely obsoleted by MachineLICM's new ability 
to unfold loop-invariant loads.

llvm-svn: 85657
 2009-10-31 14:35:41 +00:00
Dan Gohman
25091aa227 Reapply r85634, with the bug fixed. 
llvm-svn: 85655
 2009-10-31 14:22:52 +00:00
Evan Cheng
a0457853b2 Revert 85634. It's breaking consumer-typeset (and others). 
llvm-svn: 85641
 2009-10-31 01:28:06 +00:00
Dan Gohman
a532f755f7 Optimize around the fact that pred_iterator is slow: instead of sorting 
PHI operands by the predecessor order, sort them by the order used by the
first PHI in the block. This is still suffucient to expose duplicates.

llvm-svn: 85634
 2009-10-30 23:15:21 +00:00
Dan Gohman
82139ffc65 Sort the incoming values in PHI nodes to match the predecessor order. 
This helps expose duplicate PHIs, which will make it easier for them
to be eliminated.

llvm-svn: 85623
 2009-10-30 22:22:22 +00:00
Evan Cheng
8db8f72185 Add option to createGVNPass to disable PRE. 
llvm-svn: 85609
 2009-10-30 20:12:24 +00:00
Nick Lewycky
5207d16c05 Apply some cleanups. No functionality changes. 
llvm-svn: 85498
 2009-10-29 07:35:15 +00:00
Chris Lattner
a0e35cae82 teach various passes about blockaddress. We no longer 
crash on any clang tests.

llvm-svn: 85465
 2009-10-29 01:21:20 +00:00
Edward O'Callaghan
fbe911a40e No newline at end of file. 
llvm-svn: 85390
 2009-10-28 15:04:53 +00:00
Benjamin Kramer
51f925548b Update CMake file. 
llvm-svn: 85389
 2009-10-28 13:29:18 +00:00
Owen Anderson
cfb2c9edeb Treat lifetime begin/end markers as allocations/frees respectively for the 
purposes for GVN/DSE.

llvm-svn: 85383
 2009-10-28 07:05:35 +00:00
Nick Lewycky
0376e34268 Add ABCD, a generalized implementation of the Elimination of Array Bounds 
Checks on Demand algorithm which looks at arbitrary branches instead of loop
iterations. This is GSoC work by Andre Tavares with only editorial changes
applied!

llvm-svn: 85382
 2009-10-28 07:03:15 +00:00
Devang Patel
b7572e3783 Factor out redundancy from clone() implementations. 
llvm-svn: 85327
 2009-10-27 22:16:29 +00:00
Victor Hernandez
0f662a7acf Rename MallocFreeHelper as MemoryBuiltins 
llvm-svn: 85286
 2009-10-27 20:05:49 +00:00
Mike Stump
5b347d4e9d VS build fix, patch by Marius Wachtler. 
llvm-svn: 85197
 2009-10-27 02:14:13 +00:00
Eric Christopher
89c2f934d6 Add objectsize intrinsic and hook it up through codegen. Doesn't 
do anything than return "I don't know" at the moment.

llvm-svn: 85189
 2009-10-27 00:52:25 +00:00
Dan Gohman
44189252b9 Add braces to avoid ambiguous else. 
llvm-svn: 85185
 2009-10-27 00:11:02 +00:00
Victor Hernandez
6c86b93663 Rename MallocHelper as MallocFreeHelper, since it now also identifies calls to free() 
llvm-svn: 85181
 2009-10-26 23:58:56 +00:00
Owen Anderson
a8d6745830 Add a straight-forward implementation of SCCVN for aggressively eliminating scalar redundancies. 
llvm-svn: 85179
 2009-10-26 23:55:47 +00:00
Victor Hernandez
673c036bc7 Remove FreeInst. 
Remove LowerAllocations pass.
Update some more passes to treate free calls just like they were treating FreeInst.

llvm-svn: 85176
 2009-10-26 23:43:48 +00:00
Dan Gohman
3f14cd3282 Simplify this code. LoopDeletion doesn't need to explicit check that 
the loop exiting block dominates the latch block; if ScalarEvolution
can prove that the trip-count is finite, that's sufficient.

llvm-svn: 85165
 2009-10-26 22:18:58 +00:00
Dan Gohman
795e3db7ce Code that checks WillNotOverflowSignedAdd before creating an Add 
can safely use the NSW bit on the Add.

llvm-svn: 85164
 2009-10-26 22:14:22 +00:00
Ted Kremenek
acb9fe806e Update CMake files. 
llvm-svn: 85161
 2009-10-26 22:06:01 +00:00
Dan Gohman
05a91cc09f Check in the experimental GEP splitter pass. This pass splits complex 
GEPs (more than one non-zero index) into simple GEPs (at most one
non-zero index).  In some simple experiments using this it's not
uncommon to see 3% overall code size wins, because it exposes
redundancies that can be eliminated, however it's tricky to use
because instcombine aggressively undoes the work that this pass does.

llvm-svn: 85144
 2009-10-26 19:12:14 +00:00
Dan Gohman
7e7f34ab55 Fix a typo in a comment. 
llvm-svn: 85120
 2009-10-26 15:55:24 +00:00
Chris Lattner
7f9ddaf7e3 reapply r85085 with a bugfix to avoid infinite looping. 
All of the 'demorgan' related xforms need to use 
dyn_castNotVal, not m_Not.

llvm-svn: 85119
 2009-10-26 15:40:07 +00:00
Dan Gohman
3d0c010ee4 Make LSR's OptimizeShadowIV ignore induction variables with negative 
strides for now, because it doesn't handle them correctly. This fixes a
miscompile of SingleSource/Benchmarks/Misc-C++/ray.

This problem was usually hidden because indvars transforms such induction
variables into negations of canonical induction variables.

llvm-svn: 85118
 2009-10-26 15:32:57 +00:00
Evan Cheng
7dad155ec7 Revert 85085. It causes infinite looping during llvm-gcc build. 
llvm-svn: 85090
 2009-10-26 03:51:32 +00:00
Chris Lattner
192549ef70 Implement PR3266 & PR5276, folding: 
not (or (icmp, icmp)) -> and(icmp, icmp)

llvm-svn: 85085
 2009-10-26 01:06:31 +00:00
Nick Lewycky
cacdc4d65a Remove ICmpInst::isSignedPredicate which was a reimplementation 
CmpInst::isSigned.

llvm-svn: 85037
 2009-10-25 05:20:17 +00:00
Dan Gohman
1e96c6acaf Rename isLoopExit to isLoopExiting, for consistency with the wording 
used elsewhere - an exit block is a block outside the loop branched to
from within the loop. An exiting block is a block inside the loop that
branches out.

llvm-svn: 85019
 2009-10-24 23:34:26 +00:00
Dan Gohman
477456910a Rewrite LoopRotation's SSA updating code using SSAUpdater. 
llvm-svn: 85016
 2009-10-24 23:19:52 +00:00
Victor Hernandez
e8998896f5 Auto-upgrade free instructions to calls to the builtin free function. 
Update all analysis passes and transforms to treat free calls just like FreeInst.
Remove RaiseAllocations and all its tests since FreeInst no longer needs to be raised.

llvm-svn: 84987
 2009-10-24 04:23:03 +00:00
Victor Hernandez
8428eb5720 Remove AllocationInst. Since MallocInst went away, AllocaInst is the only subclass of AllocationInst, so it no longer is necessary. 
llvm-svn: 84969
 2009-10-23 21:09:37 +00:00
Dan Gohman
ad2555c9a3 Make LoopDeletion check the maximum backedge taken count, rather than the 
exact backedge taken count, when checking for infinite loops. This allows
it to delete loops with multiple exit conditions.

llvm-svn: 84952
 2009-10-23 17:10:01 +00:00
Chris Lattner
cd80b1711b move another load optimization from instcombine -> libanalysis. 
llvm-svn: 84841
 2009-10-22 06:44:07 +00:00
Chris Lattner
3fb4c19b7d move 'loading i32 from string' optimization from instcombine 
to libanalysis.  Instcombine shrinking... does this even 
make sense???

llvm-svn: 84840
 2009-10-22 06:38:35 +00:00
Chris Lattner
2c8c8533b1 Move some constant folding logic for loads out of instcombine into 
Analysis/ConstantFolding.cpp.  This doesn't change the behavior of
instcombine but makes other clients of ConstantFoldInstruction
able to handle loads.  This was partially extracted from Eli's patch
in PR3152.

llvm-svn: 84836
 2009-10-22 06:25:11 +00:00
Chris Lattner
7c507ae857 fix PR5262. 
llvm-svn: 84810
 2009-10-22 00:17:26 +00:00
Chris Lattner
bf5a89d82b revert r84754, it isn't the right approach. Edwin, please propose 
patches for fixes like this instead of committing them directly.

llvm-svn: 84799
 2009-10-21 23:41:58 +00:00
Victor Hernandez
0996619220 Make changes to rev 84292 as requested by Chris Lattner. 
Most changes are cleanup, but there is 1 correctness fix:
I fixed InstCombine so that the icmp is removed only if the malloc call is removed (which requires explicit removal because the Worklist won't DCE any calls since they can have side-effects).

llvm-svn: 84772
 2009-10-21 19:11:40 +00:00
Torok Edwin
a8a12b7d98 Fix PR5262: when folding select into PHI, make sure all operands are available 
in the PHI's Basic Block. This uses a conservative approach, because we don't
have dominator info in instcombine.

llvm-svn: 84754
 2009-10-21 10:49:00 +00:00
Chris Lattner
82599d2b0e make GVN work better when TD is not around: 
"In the existing code, if the load and the value to replace it with are
of different types *and* target data is available, it tries to use the
target data to coerce the replacement value to the type of the load.
Otherwise, it skips all effort to handle the type mismatch and just
feeds the wrongly-typed replacement value to replaceAllUsesWith, which
triggers an assertion.

The patch replaces it with an outer if checking for type mismatch, and
an inner if-else that checks whether target data is available and, if
not, returns false rather than trying to replace the load."

Patch by Kenneth Uildriks!

llvm-svn: 84739
 2009-10-21 04:11:19 +00:00
Dan Gohman
ea9ddb0313 Restore LoopUnswitch's block-oriented threshold. LoopUnswitch now checks both 
the estimated code size and the number of blocks when deciding whether to
do a non-trivial unswitch. This protects it from some very undesirable
worst-case behavior on large numbers of loop-unswitchable conditions, such
as in the testcase in PR5259.

llvm-svn: 84661
 2009-10-20 20:06:09 +00:00
Torok Edwin
3801e755cc Fix PR4313: IPSCCP was not setting the lattice value for the invoke instruction 
when the invoke had multiple return values: it set the lattice value only on the
extractvalue.
This caused the invoke's lattice value to remain the default (undefined), and
later propagated to extractvalue's operand, which incorrectly introduces
undefined behavior.

llvm-svn: 84637
 2009-10-20 15:15:09 +00:00
Owen Anderson
06714a71a9 Refactor lookup_or_add to contain _MUCH_ less duplicated code. Add support for 
numbering first class aggregate instructions while we're at it.

llvm-svn: 84547
 2009-10-19 22:14:22 +00:00
Owen Anderson
4954f8ef02 Simplify some code. 
llvm-svn: 84533
 2009-10-19 21:14:57 +00:00
Victor Hernandez
a25a2890fa Remove MallocInst from LLVM Instructions. 
llvm-svn: 84299
 2009-10-17 01:18:07 +00:00
Victor Hernandez
4d7283f82c Autoupgrade malloc insts to malloc calls. 
Update testcases that rely on malloc insts being present.

Also prematurely remove MallocInst handling from IndMemRemoval and RaiseAllocations to help pass tests in this incremental step.

llvm-svn: 84292
 2009-10-17 00:00:19 +00:00
Dan Gohman
ef1f246d88 Move zext and sext casts fed by loads into the same block as the 
load, to help SelectionDAG fold them into the loads, unless
conditions are unfavorable.

llvm-svn: 84271
 2009-10-16 20:59:35 +00:00
Chris Lattner
389f4efae7 only try to fold constantexpr operands when the worklist is first populated, 
don't bother every time going around the main worklist.  This speeds up a 
release-asserts opt -std-compile-opts on 403.gcc by about 4% (1.5s).  It
seems to speed up the most expensive instances of instcombine by ~10%.

llvm-svn: 84171
 2009-10-15 04:59:28 +00:00
Chris Lattner
ddfa3af584 don't bother calling ConstantFoldInstruction unless there is a use of the 
instruction (which disqualifies stores, unreachable, etc) and at least the
first operand is a constant.  This filters out a lot of obvious cases that
can't be folded.  Also, switch the IRBuilder to a TargetFolder, which tries
harder.

llvm-svn: 84170
 2009-10-15 04:13:44 +00:00
Devang Patel
89ef6dcb54 Use isVoidTy() 
llvm-svn: 84118
 2009-10-14 17:29:00 +00:00
Chris Lattner
a1f5264dd2 make instcombine's instruction sinking more aggressive in the 
presence of PHI nodes.

llvm-svn: 84103
 2009-10-14 15:21:58 +00:00
Devang Patel
81485d992b Check void type before using RAUWd. 
llvm-svn: 84049
 2009-10-13 22:56:32 +00:00
Devang Patel
20d88f9e16 Do not check use_empty() before replaceAllUsesWith(). This gives ValueHandles a chance to get properly updated. 
llvm-svn: 84033
 2009-10-13 21:41:20 +00:00
Dan Gohman
c5a7668904 Use the new CodeMetrics class to compute code size instead of 
manually counting instructions.

llvm-svn: 84016
 2009-10-13 20:12:23 +00:00
Dan Gohman
d01c6196f4 Make LoopUnswitch's cost estimation count Instructions, rather than 
BasicBlocks, so that it doesn't blindly procede in the presence of
large individual BasicBlocks. This addresses a class of code-size
expansion problems.

llvm-svn: 83992
 2009-10-13 17:50:43 +00:00
Evan Cheng
3a513d33ed Make licm debug message readable. 
llvm-svn: 83908
 2009-10-12 22:25:23 +00:00
Dale Johannesen
c78690fcaa Fix warning. 
llvm-svn: 83870
 2009-10-12 18:45:32 +00:00
Chris Lattner
ff0d7e8c93 populate instcombine's initial worklist more carefully, causing 
it to visit instructions from the start of the function to the
end of the function in the first path.  This greatly speeds up
some pathological cases (e.g. PR5150).

Try #3, this time with some unneeded debug info stuff removed
which was causing dead pointers to be added to the worklist.

llvm-svn: 83818
 2009-10-12 03:58:40 +00:00
Chris Lattner
bc69f8b5bb revert r83814 for now, it is making the llvm-gcc bootstrap unhappy. 
llvm-svn: 83817
 2009-10-11 23:56:08 +00:00
Chris Lattner
1a6fbd067a populate instcombine's initial worklist more carefully, causing 
it to visit instructions from the start of the function to the
end of the function in the first path.  This greatly speeds up
some pathological cases (e.g. PR5150). 

llvm-svn: 83814
 2009-10-11 23:17:43 +00:00
Chris Lattner
85457e7c14 remove some harmful code that would turn an insertelement on an undef 
into a shuffle even if it was used by another insertelement.  If the
visitation order of instcombine was wrong, this would turn a chain of
insertelements into a chain of shufflevectors, which was quite painful.

Since CollectShuffleElements handles these cases, the code can just
be nuked.

llvm-svn: 83810
 2009-10-11 23:02:46 +00:00
Chris Lattner
b033a0ac47 teach instcombine to simplify xor's harder, catching the 
new testcase.

llvm-svn: 83799
 2009-10-11 22:22:13 +00:00
Chris Lattner
d2dc77f8c0 cleanups 
llvm-svn: 83797
 2009-10-11 22:00:32 +00:00
Chris Lattner
6cf4024732 cleanup, no functionality change. 
llvm-svn: 83795
 2009-10-11 21:36:10 +00:00
Chris Lattner
100f3ac565 generalize a transformation even more: we don't care whether the 
input the the mul is a zext from bool, just that it is all zeros
other than the low bit.  This fixes some phase ordering issues
that would cause us to miss some xforms in mul.ll when the worklist
is visited differently.

llvm-svn: 83794
 2009-10-11 21:29:45 +00:00
Chris Lattner
40d1618562 simplify a transformation by making it more general. 
llvm-svn: 83792
 2009-10-11 21:22:21 +00:00
Chris Lattner
0ad392bbe7 temporarily revert previous patch 
llvm-svn: 83791
 2009-10-11 21:05:34 +00:00
Chris Lattner
aafa929524 populate instcombine's initial worklist more carefully, causing 
it to visit instructions from the start of the function to the
end of the function in the first path.  This greatly speeds up
some pathological cases (e.g. PR5150). 

llvm-svn: 83790
 2009-10-11 21:04:37 +00:00
Torok Edwin
8525f5df9a Remove CleanupDbgInfo, instcombine does this and its not worth duplicating it 
here.

llvm-svn: 83789
 2009-10-11 19:58:35 +00:00
Torok Edwin
ed37f33b81 LICM shouldn't sink/delete debug information. Fix this and add a testcase. 
For now the metadata of sinked/hoisted instructions is still wrong, but that'll
be fixed when instructions will have debug metadata directly attached.

llvm-svn: 83786
 2009-10-11 19:15:54 +00:00
Chris Lattner
8ae883b5f0 when folding duplicate conditions, delete the 
now-probably-dead instruction tree feeding it.

llvm-svn: 83778
 2009-10-11 18:39:58 +00:00
Chris Lattner
601cc4c588 implement rdar://7293527, a trivial instcombine that llvm-gcc 
gets but clang doesn't, because it is implemented in GCC's
fold routine.

llvm-svn: 83761
 2009-10-11 07:53:15 +00:00
Chris Lattner
1832a32b78 implement a transformation in jump threading that is currently 
done by condprop, but do it in a much more general form.  The
basic idea is that we can do a limited form of tail duplication
in the case when we have a branch on a phi.  Moving the branch
up in to the predecessor block makes instruction selection
much easier and encourages chained jump threadings.

llvm-svn: 83759
 2009-10-11 07:24:57 +00:00
Chris Lattner
562cc63c29 restructure some code, no functionality change. 
llvm-svn: 83756
 2009-10-11 04:40:21 +00:00
Chris Lattner
b4e9fb6967 factor some code better and move a function, no functionality change. 
llvm-svn: 83755
 2009-10-11 04:33:43 +00:00
Chris Lattner
525c0272aa make jump threading on a phi with undef inputs happen. 
llvm-svn: 83754
 2009-10-11 04:18:15 +00:00
Chris Lattner
37ff3c0ed1 switch GVN to use SSAUpdater. Besides removing a lot of complexity 
from GVN, this also speeds it up, inserts fewer PHI nodes (see the
testcase) and allows it to remove more loads (due to fewer PHI nodes
standing in the way).

llvm-svn: 83746
 2009-10-10 23:50:30 +00:00
Chris Lattner
31d3137c01 remove dead code 
llvm-svn: 83742
 2009-10-10 23:04:12 +00:00
Chris Lattner
75e5de9243 Change jump threading to use the new SSAUpdater class instead of 
DemoteRegToStack.  This makes it more efficient (because it isn't
creating a ton of load/stores that are eventually removed by a later
mem2reg), and more slightly more effective (because those load/stores
don't get in the way of threading).

llvm-svn: 83706
 2009-10-10 09:05:58 +00:00
Chris Lattner
317eb8ff20 random tidying 
llvm-svn: 83701
 2009-10-10 06:22:45 +00:00
Dan Gohman
68e044e81a Add a form of addPreserved which takes a string argument, to allow passes 
to declare that they preserve other passes without needing to pull in
additional header file or library dependencies. Convert MachineFunctionPass
and CodeGenLICM to make use of this.

llvm-svn: 83555
 2009-10-08 17:00:02 +00:00
Jeffrey Yasskin
e36facef86 In instcombine's debug output, avoid printing ADD for instructions that are 
already on the worklist, and print Visited when an instruction is about to be
visited.  Net, on one input, this reduced the output size by at least 9x.

llvm-svn: 83510
 2009-10-08 00:12:24 +00:00
Eric Christopher
a42e55349d 80-column and whitespace fixes. 
llvm-svn: 83489
 2009-10-07 21:14:25 +00:00
Ted Kremenek
4b5e7b5661 Update CMake file. 
llvm-svn: 83404
 2009-10-06 19:45:38 +00:00
Chris Lattner
e87add4b88 remove predicate simplifier, it never got the last bugs beaten 
out of it, and jump threading, condprop and gvn are now getting
most of the benefit.  This was approved by Nicholas and Nicolas.

llvm-svn: 83390
 2009-10-06 16:59:46 +00:00
Duncan Sands
2400ad7236 Introduce and use convenience methods for getting pointer types 
where the element is of a basic builtin type.  For example, to get
an i8* use getInt8PtrTy.

llvm-svn: 83379
 2009-10-06 15:40:36 +00:00
Dan Gohman
176e47a413 Remove an unnnecessary LLVMContext argument in 
ConstantFoldLoadThroughGEPConstantExpr.

llvm-svn: 83311
 2009-10-05 16:36:26 +00:00
Dan Gohman
a926123a24 Use Use::operator= instead of Use::set, for consistency. 
llvm-svn: 83310
 2009-10-05 16:31:55 +00:00
Chris Lattner
a3cf123e86 strength reduce a ton of type equality tests to check the typeid (Through 
the new predicates I added) instead of going through a context and doing a
pointer comparison.  Besides being cheaper, this allows a smart compiler
to turn the if sequence into a switch.

llvm-svn: 83297
 2009-10-05 05:54:46 +00:00
Chris Lattner
513b66b237 instcombine shouldn't delete all null checks for mallocs. 
This fixes PR5130.

llvm-svn: 83290
 2009-10-05 02:47:47 +00:00
Douglas Gregor
b4e6fce681 Remove GVNPRE.cpp from the CMake makefile 
llvm-svn: 83194
 2009-10-01 05:30:05 +00:00
Chris Lattner
857ed445a8 remove the GVNPRE pass. It has been subsumed by the GVN pass. 
Ok'd by Owen.

llvm-svn: 83193
 2009-10-01 02:18:36 +00:00
Chris Lattner
d77fb17b28 The select instruction is not neccesarily in the same block as the 
phi nodes.  Make sure to phi translate from the right block. 

This fixes a llvm-building-llvm failure on GVN-PRE.cpp

llvm-svn: 82970
 2009-09-28 06:49:44 +00:00
Chris Lattner
2a85cfd972 simplify some code. 
llvm-svn: 82936
 2009-09-27 21:46:50 +00:00
Chris Lattner
75a5d76294 The bitcast case is not needed here: instcombine turns icmp(bitcast(x), null) -> icmp(x, null) already. 
llvm-svn: 82935
 2009-09-27 21:42:46 +00:00
Chris Lattner
e024fd6b47 calls are already unmovable, malloc doesn't need a special case. 
llvm-svn: 82933
 2009-09-27 21:36:19 +00:00
Chris Lattner
8a77e8b353 calls to external functions are already marked overdefined, special casing 
malloc isn't needed.

llvm-svn: 82932
 2009-09-27 21:35:11 +00:00
Chris Lattner
24af6de20a calls are rejected above, no need to special case malloc here. 
llvm-svn: 82929
 2009-09-27 21:31:39 +00:00
Chris Lattner
7a9e470ddc allow pushing icmps through phis with multiple uses and across critical edges. 
These are important to push up to encourage jump threading.  This shrinks 176.gcc a bit.

llvm-svn: 82923
 2009-09-27 20:46:36 +00:00
Chris Lattner
f7db1f1a3a Enhance the previous fix for PR4895 to allow more values than just 
simple constants for the true/false value of the select.  We now
do phi translation etc.  This really fixes PR4895 :)

llvm-svn: 82917
 2009-09-27 20:18:49 +00:00
Chris Lattner
0af7f0ceaf implement PR4895, by making FoldOpIntoPhi handle select conditions 
that are phi nodes.  Also tighten up FoldOpIntoPhi to treat constantexpr
operands to phis just like other variables, avoiding moving constantexpr
computations around.

Patch by Daniel Dunbar.

llvm-svn: 82913
 2009-09-27 19:57:57 +00:00
Dan Gohman
641ab5a61e Grab an LLVM Context from an instruction that exists rather than one 
that is deleted in some situations. This fixes a use-after-free.

llvm-svn: 82903
 2009-09-27 16:10:30 +00:00
Dan Gohman
fa43cefae4 Tell ScalarEvolution to forget everything it knows about a loop before 
rotating the loop, since loop rotation is a very significant change.

llvm-svn: 82901
 2009-09-27 15:37:03 +00:00
Nick Lewycky
d21c325f09 Instruction::clone does not need to take an LLVMContext&. Remove that and 
update all the callers.

llvm-svn: 82889
 2009-09-27 07:38:41 +00:00
Dan Gohman
0a396aa546 Fix SimplifyLibCalls to transfer attributes from callees rather than 
calls, since direct calls don't always reflect the attributes of their
callees.

llvm-svn: 82867
 2009-09-26 18:10:13 +00:00
Dan Gohman
d585364b0b Rename ConstantFP's getInf to getInfinity. 
llvm-svn: 82823
 2009-09-25 23:40:21 +00:00
Dan Gohman
d564cd42d9 Transform pow(x, 0.5) to (x == -inf ? inf : fabs(sqrt(x))), which is 
typically faster then doing a general pow.

llvm-svn: 82819
 2009-09-25 23:10:17 +00:00
Torok Edwin
193d22fa77 Constant propagating byval pointer is safe if function is readonly. 
llvm-svn: 82700
 2009-09-24 18:33:42 +00:00
Torok Edwin
81fe296f15 Don't constant propagate byval pointers, since they are not really pointers, but 
rather structs passed by value.
This fixes PR5038.

llvm-svn: 82689
 2009-09-24 09:47:18 +00:00
Chris Lattner
1d9ca20bf5 big endian systems shift by bits too, hopefully this will fix the ppc 
bootstrap problems.

llvm-svn: 82464
 2009-09-21 17:55:47 +00:00
Dan Gohman
3359b05815 Nick pointed out that DominanceFrontier and DominanceTree are preserved 
by setPreservesCFG().

llvm-svn: 82463
 2009-09-21 17:54:42 +00:00
Dan Gohman
ea9d53d563 Remove the special-case for constants in PHI nodes; it's not really 
helpful, and it didn't correctly handle the case of constants input
to PHIs for backedges.

llvm-svn: 82462
 2009-09-21 17:53:35 +00:00