Commit Graph

7181 Commits

Author SHA1 Message Date
Evan Cheng fa0ac19b82 RALinScan::attemptTrivialCoalescing() was returning a virtual register instead of the physical register it is allocated to. This resulted in virtual register(s) being added the live-in sets.
llvm-svn: 72890
2009-06-04 20:53:36 +00:00
Evan Cheng 60fdf787a7 A value defined by an implicit_def can be liven to a use BB. This is unfortunate. But register allocator still has to add it to the live-in set of the use BB.
llvm-svn: 72888
2009-06-04 20:25:48 +00:00
Dale Johannesen 12313e327f For XTARGET to work on targets not in the list,
there must also be an XFAIL line.  Fix a couple
examples of this.

llvm-svn: 72876
2009-06-04 18:27:43 +00:00
Dan Gohman 1aa86203f6 Check in test changes that I accidentally left out of r72872.
llvm-svn: 72875
2009-06-04 18:22:31 +00:00
Eli Friedman 63488f1fbf PR3739, part 2: Use an explicit store to spill XMM registers. (Previously,
the code tried to use "push", which doesn't exist for XMM registers.)

llvm-svn: 72836
2009-06-04 02:32:04 +00:00
Eli Friedman 0cb0c78a26 PR3739, part 1: Disable the red zone on Win64.
llvm-svn: 72830
2009-06-04 02:02:01 +00:00
Evan Cheng 7f5976e11b Re-apply 72756 with fixes. One of those was introduced by we changed MachineInstrBuilder::addReg() interface.
llvm-svn: 72826
2009-06-04 01:15:28 +00:00
Eli Friedman ee06b752f0 PR4317: Handle splits where the new block is unreachable correctly in
DominatorTreeBase::Split.

llvm-svn: 72810
2009-06-03 21:42:06 +00:00
Evan Cheng ad6f3ff2c0 For Darwin / x86_64, override -relocation-model=static to pic if the output is assembly since Darwin assembler does not really support -static codeine.
I view this as a temporary workaround until the assembler / linker changes.

llvm-svn: 72806
2009-06-03 21:13:54 +00:00
Dan Gohman c380cca7ae Don't attempt to simplify an non-affine IV expression if it can't
be simplified to a loop-invariant value. This fixes PR4315.

llvm-svn: 72798
2009-06-03 19:11:31 +00:00
Evan Cheng b39a6be77a Fix for PR4225: When rewriter reuse a value in a physical register , it clear the register kill operand marker and its kill ops information. However, the cleared operand may be a def of a super-register. Clear the kill ops info for the super-register's sub-registers as well.
llvm-svn: 72758
2009-06-03 09:00:27 +00:00
Evan Cheng ab0c710fae Temporarily revert 72756 for now.
llvm-svn: 72757
2009-06-03 07:40:47 +00:00
Evan Cheng dfe6e689fd Fold preceding / trailing base inc / dec into the single load / store as well.
llvm-svn: 72756
2009-06-03 06:14:58 +00:00
Dan Gohman fc262babc3 Revert r72734. The Darwin assembler doesn't support the static
relocation model on x86-64. Higher level logic should override
the relocation model to PIC on x86_64-apple-darwin.

llvm-svn: 72746
2009-06-03 00:37:20 +00:00
Dan Gohman 760377effc Fix CodeGenPrepare's address-mode sinking to handle unusual
addresses, involving Base values which do not have Pointer type.
This fixes PR4297.

llvm-svn: 72739
2009-06-02 21:29:13 +00:00
Evan Cheng 448641d87c On Darwin x86_64 small code model doesn't guarantee code address fits in 32-bit.
llvm-svn: 72734
2009-06-02 20:09:31 +00:00
Evan Cheng 836894405f Avoid infinite looping in AllGlobalLoadUsesSimpleEnoughForHeapSRA(). This can happen when PHI uses are recursively dependent on each other.
llvm-svn: 72710
2009-06-02 00:56:07 +00:00
Eli Friedman ee94e3cc9e PR4286: Make RewriteLoadUserOfWholeAlloca and
RewriteStoreUserOfWholeAlloca deal with tail padding because 
isSafeUseOfBitCastedAllocation expects them to.  Otherwise, we crash 
trying to erase the bitcast.

llvm-svn: 72688
2009-06-01 09:14:32 +00:00
Owen Anderson cc0c75c74d Be more aggressive in doing LoadPRE by tracing backwards when a block only has
a single predecessor.

Patch by Jakub Staszak.

llvm-svn: 72661
2009-05-31 09:03:40 +00:00
Chris Lattner 221895303c fix PR4284, a bug in simplifylibcalls handling memcmp. Patch by
Benjamin Kramer!

llvm-svn: 72625
2009-05-30 18:43:04 +00:00
Duncan Sands 24ad1feb77 Adjust these tests now that "extern inline"
functions are being output with bodies and
available_externally linkage.

llvm-svn: 72620
2009-05-30 13:57:05 +00:00
Evan Cheng 7142ad75a1 (i64 (zext (srl GR32 8))) -> movzbl AH is not safe since srl 8 only clear the top 8 bits.
llvm-svn: 72618
2009-05-30 08:43:27 +00:00
Nick Lewycky adbc284666 Give embedded metadata its own type instead of relying on EmptyStructTy.
llvm-svn: 72610
2009-05-30 05:06:04 +00:00
Duncan Sands 15de591890 Dan noticed that the verifier wasn't thoroughly checking uses of
invoke results (see the testcases).  Tighten up the checking.

llvm-svn: 72586
2009-05-29 19:39:36 +00:00
Evan Cheng 8e97b85cde Remove an accidental commit.
llvm-svn: 72560
2009-05-29 05:28:52 +00:00
Evan Cheng 716e688fca More h-registers tricks: folding zext nodes.
llvm-svn: 72558
2009-05-29 01:44:43 +00:00
Evan Cheng 86cdb4b345 Do not try to create a MVT type of width 0.
llvm-svn: 72557
2009-05-28 23:52:18 +00:00
Eli Friedman e4b43e60b3 Add explicit test for PR4280.
llvm-svn: 72539
2009-05-28 21:04:35 +00:00
Eli Friedman 1f906448bc Add a testcase which got fixed by recent legalization work.
llvm-svn: 72517
2009-05-28 05:10:20 +00:00
Nick Lewycky 206876e2da Use Operands.data() instead of &Operands[0] where Operands is a potentially
empty SmallVector.

llvm-svn: 72512
2009-05-28 04:08:10 +00:00
Evan Cheng a9cda8abf2 Added optimization that narrow load / op / store and the 'op' is a bit twiddling instruction and its second operand is an immediate. If bits that are touched by 'op' can be done with a narrower instruction, reduce the width of the load and store as well. This happens a lot with bitfield manipulation code.
e.g.
orl     $65536, 8(%rax)
=>
orb     $1, 10(%rax)

Since narrowing is not always a win, e.g. i32 -> i16 is a loss on x86, dag combiner consults with the target before performing the optimization.

llvm-svn: 72507
2009-05-28 00:35:15 +00:00
Dan Gohman 4d1823680d Revert 72493 and replace it with a more conservative fix, for now: don't
rewrite the comparison if there is any implicit extension or truncation
on the induction variable. I'm planning for IVUsers to eventually take
over some of the work of this code, and for it to be generalized.

llvm-svn: 72496
2009-05-27 21:10:47 +00:00
Dan Gohman f4d85325c0 In ChangeCompareStride, when the stride to be reused is truncated to
a smaller type, promoted its offset back up to the type of the new
comparison. This fixes PR4222.

llvm-svn: 72493
2009-05-27 20:00:18 +00:00
Bill Wendling 0a9ea80013 This looks like it passes now.
llvm-svn: 72485
2009-05-27 17:43:21 +00:00
Dan Gohman 1789362d70 Teach SCEVExpander to avoid creating over-indexed GEP indices when
possible. For example, it now emits

  %p.2.ip.1 = getelementptr [3 x [3 x double]]* %p, i64 2, i64 %tmp, i64 1

instead of the equivalent but less obvious

  %p.2.ip.1 = getelementptr [3 x [3 x double]]* %p, i64 0, i64 %tmp, i64 19

llvm-svn: 72452
2009-05-27 02:00:53 +00:00
Dan Gohman 776e4c8d35 Teach BasicAliasAnalysis to understand constant gep indices that fall
beyond their associated static array type.

I believe that this fixes a legitimate bug, because BasicAliasAnalysis
already has code to check for this condition that works for non-constant
indices, however it was missing the case of constant indices. With this
change, it checks for both.

This fixes PR4267, and miscompiles of SPEC 188.ammp and 464.h264.href.

llvm-svn: 72451
2009-05-27 01:48:27 +00:00
Dale Johannesen af4fa989e4 Testcase for (llvm-gcc-4.2) 72442 (PR 4242).
llvm-svn: 72443
2009-05-26 23:19:19 +00:00
Dan Gohman 9fe212806b In cases where a pointer value is an operand of a multiplication or
division operation, don't attempt to use the operation's value as
the base of a getelementptr. This fixes PR4271.

llvm-svn: 72422
2009-05-26 17:41:16 +00:00
Chris Lattner 02274a7171 make memdep use the getModRefInfo method for stores instead of the
low-level alias() method, allowing it to reason more aggressively
about pointers into constant memory.  PR4189

llvm-svn: 72403
2009-05-25 21:28:56 +00:00
Dan Gohman 4486da5b78 When rewriting the loop exit test with the canonical induction variable,
leave the original comparison in place if it has other uses, since the
other uses won't be dominated by the new comparison instruction.

llvm-svn: 72369
2009-05-24 19:11:38 +00:00
Dan Gohman 291c2e04bd Generalize SCEVExpander::visitAddRecExpr's GEP persuit, and avoid
sending SCEVUnknowns to expandAddToGEP. This avoids the need for
expandAddToGEP to bend the rules and peek into SCEVUnknown
expressions.

Factor out the code for testing whether a SCEV can be factored by
a constant for use in a GEP index. This allows it to handle
SCEVAddRecExprs, by recursing.

As a result, SCEVExpander can now put more things in GEP indices,
so it emits fewer explicit mul instructions.

llvm-svn: 72366
2009-05-24 18:06:31 +00:00
Torok Edwin d184bc209c The rewriter may hold references to instructions that are deleted because they are trivially dead.
Fix by clearing the rewriter cache before deleting the trivially dead
instructions.
Also make InsertedExpressions use an AssertingVH to catch these
bugs easier.

llvm-svn: 72364
2009-05-24 14:23:16 +00:00
Torok Edwin be6a9a151a Fix PR4254.
The DAGCombiner created a negative shiftamount, stored in an
unsigned variable. Later the optimizer eliminated the shift entirely as being
undefined.
Example: (srl (shl X, 56) 48). ShiftAmt is 4294967288.
Fix it by checking that the shiftamount is positive, and storing in a signed
variable.

llvm-svn: 72331
2009-05-23 17:29:48 +00:00
Torok Edwin 7996339dd8 available_externall linkage is not local, this was confusing the codegenerator,
and it wasn't generating calls through @PLT for these functions.
hasLocalLinkage() is now false for available_externally,
I attempted to fix the inliner and dce to handle available_externally properly.
It passed make check.

llvm-svn: 72328
2009-05-23 14:06:57 +00:00
Eli Friedman 53a71147ba Fix test to account for legalization changes; I think this ends up
running an extra DAGCombine pass which improves the code a bit.

llvm-svn: 72326
2009-05-23 13:15:11 +00:00
Evan Cheng a838a40bc4 Fix bug in FoldFCmp_IntToFP_Cst. If inttofp is a uintofp, use unsigned instead of signed integer constant.
llvm-svn: 72300
2009-05-22 23:10:53 +00:00
Duncan Sands d6fb6501e3 Add a new codegen pass that normalizes dwarf exception handling
code in preparation for code generation.  The main thing it does
is handle the case when eh.exception calls (and, in a future
patch, eh.selector calls) are far away from landing pads.  Right
now in practice you only find eh.exception calls close to landing
pads: either in a landing pad (the common case) or in a landing
pad successor, due to loop passes shifting them about.  However
future exception handling improvements will result in calls far
from landing pads:
(1) Inlining of rewinds.  Consider the following case:
In function @f:
...
  invoke @g to label %normal unwind label %unwinds
...
unwinds:
  %ex = call i8* @llvm.eh.exception()
...

In function @g:
...
  invoke @something to label %continue unwind label %handler
...
handler:
  %ex = call i8* @llvm.eh.exception()
... perform cleanups ...
  "rethrow exception"

Now inline @g into @f.  Currently this is turned into:
In function @f:
...
  invoke @something to label %continue unwind label %handler
...
handler:
  %ex = call i8* @llvm.eh.exception()
... perform cleanups ...
  invoke "rethrow exception" to label %normal unwind label %unwinds
unwinds:
  %ex = call i8* @llvm.eh.exception()
...

However we would like to simplify invoke of "rethrow exception" into
a branch to the %unwinds label.  Then %unwinds is no longer a landing
pad, and the eh.exception call there is then far away from any landing
pads.

(2) Using the unwind instruction for cleanups.
It would be nice to have codegen handle the following case:
  invoke @something to label %continue unwind label %run_cleanups
...
handler:
... perform cleanups ...
  unwind

This requires turning "unwind" into a library call, which
necessarily takes a pointer to the exception as an argument
(this patch also does this unwind lowering).  But that means
you are using eh.exception again far from a landing pad.

(3) Bugpoint simplifications.  When bugpoint is simplifying
exception handling code it often generates eh.exception calls
far from a landing pad, which then causes codegen to assert.
Bugpoint then latches on to this assertion and loses sight
of the original problem.

Note that it is currently rare for this pass to actually do
anything.  And in fact it normally shouldn't do anything at
all given the code coming out of llvm-gcc!  But it does fire
a few times in the testsuite.  As far as I can see this is
almost always due to the LoopStrengthReduce codegen pass
introducing pointless loop preheader blocks which are landing
pads and only contain a branch to another block.  This other
block contains an eh.exception call.  So probably by tweaking
LoopStrengthReduce a bit this can be avoided.

llvm-svn: 72276
2009-05-22 20:36:31 +00:00
Dan Gohman 781b75a7df Teach IndVarSimplify's FixUsesBeforeDefs to handle InvokeInsts by
assuming that the use of the value is in a block dominated by the
"normal" destination. LangRef.html and other documentation sources
don't explicitly guarantee this, but it seems to be assumed in
other places in LLVM at least.

This fixes an assertion failure on the included testcase, which
is derived from the Ada testsuite.

FixUsesBeforeDefs is a temporary measure which I'm looking to
replace with a more capable solution.

llvm-svn: 72266
2009-05-22 16:47:11 +00:00
Dan Gohman 88e06db116 Fix a thinko in the code that adapted SCEVMulExpr operands for
use in expanding SCEVAddExprs with GEPs. The operands of a
SCEVMulExpr need to be multiplied together, not added.

llvm-svn: 72250
2009-05-22 07:14:20 +00:00
Torok Edwin 9c4dcfb4a7 Revert this. There's no way to verifiy indirect calls, and an optimizer can turn
indirect call into direct call, thus the verifier would reject something it
previously accepted.

llvm-svn: 72249
2009-05-22 07:12:05 +00:00
Torok Edwin a39d69684a Verify that calling conventions match function prototype.
This only rejects mismatches between target specific calling convention
and C/LLVM specific calling convention.
There are too many fastcc/C, coldcc/cc42 mismatches in the testsuite, these are
not reject by the verifier.

llvm-svn: 72248
2009-05-22 06:41:43 +00:00
Eli Friedman 84295eb636 Fix broken logic in DominatorTreeBase::Split. Part of PR4238.
llvm-svn: 72231
2009-05-21 21:47:54 +00:00
Eli Friedman 0ac904455e Fix some incorrect logic in DominanceFrontier::splitBlock. Part of
PR4238.

llvm-svn: 72223
2009-05-21 20:40:30 +00:00
Dan Gohman bf0002e7c1 Teach ValueTracking a new way to analyze PHI nodes, and and teach
Instcombine to be more aggressive about using SimplifyDemandedBits
on shift nodes. This allows a shift to be simplified to zero in the
included test case.

llvm-svn: 72204
2009-05-21 02:28:33 +00:00
Eli Friedman 9030c35eb4 Fix for PR4235: to build a floating-point value from integer parts,
build an integer and cast that to a float.  This fixes a crash 
caused by trying to split an f32 into two f16's.

This changes the behavior in test/CodeGen/XCore/fneg.ll because that 
testcase now triggers a DAGCombine which converts the fneg into an integer
operation.  If someone is interested, it's probably possible to tweak 
the test to generate an actual fneg.

llvm-svn: 72162
2009-05-20 06:02:09 +00:00
Evan Cheng 1fbc2a4754 Fix test on non-darwin hosts.
llvm-svn: 72161
2009-05-20 05:45:36 +00:00
Evan Cheng 960983371c Try again. Allow call to immediate address for ELF or when in static relocation mode.
llvm-svn: 72160
2009-05-20 04:53:57 +00:00
Evan Cheng 61da18645b Cannot use immediate as call absolute target in PIC mode.
llvm-svn: 72154
2009-05-20 01:11:00 +00:00
Dan Gohman 7248923a5d Suppress the IV reversal transformation in the case that the RHS
of the comparison is defined inside the loop. This fixes a
use-before-def problem, because the transformation puts a use
of the RHS outside the loop.

llvm-svn: 72149
2009-05-20 00:34:08 +00:00
Bob Wilson e666cc5206 Fix pr4058 and pr4059. Do not split i64 or double arguments between r3 and
the stack.  Patch by Sandeep Patel.

llvm-svn: 72106
2009-05-19 10:02:36 +00:00
Bob Wilson a2c462bbe9 Fix pr4091: Add support for "m" constraint in ARM inline assembly.
llvm-svn: 72105
2009-05-19 05:53:42 +00:00
Dan Gohman 2649491f9c Teach SCEVExpander to expand arithmetic involving pointers into GEP
instructions. It attempts to create high-level multi-operand GEPs,
though in cases where this isn't possible it falls back to casting
the pointer to i8* and emitting a GEP with that. Using GEP instructions
instead of ptrtoint+arithmetic+inttoptr helps pointer analyses that
don't use ScalarEvolution, such as BasicAliasAnalysis.

Also, make the AddrModeMatcher more aggressive in handling GEPs.
Previously it assumed that operand 0 of a GEP would require a register
in almost all cases. It now does extra checking and can do more
matching if operand 0 of the GEP is foldable. This fixes a problem
that was exposed by SCEVExpander using GEPs.

llvm-svn: 72093
2009-05-19 02:15:55 +00:00
Bill Wendling b17990999f Commands beginning with '--' are converted to '-f' by gcc. Blech!
llvm-svn: 72023
2009-05-18 18:09:36 +00:00
Dan Gohman 6350296efc Teach ScalarEvolution to recognize x^-1 in the case where non-demanded
bits have been stripped out by instcombine.

llvm-svn: 72010
2009-05-18 16:29:04 +00:00
Dan Gohman 8c77f1a275 Make ScalarEvolution::isLoopGuardedByCond work even when the edge
entering a loop is a non-split critical edge.

llvm-svn: 72004
2009-05-18 15:36:09 +00:00
Dan Gohman b81dd48fd2 Add nounwind to a few tests.
llvm-svn: 72002
2009-05-18 15:16:49 +00:00
Duncan Sands c5161d4b83 Check that the gcc front-end is not doing inlining
when not doing unit-at-a-time.

llvm-svn: 71986
2009-05-17 19:37:02 +00:00
Anton Korobeynikov 6de08cd093 Mark rotl/rotr as expand. This generates pretty ugly code, but this is better than nothing.
llvm-svn: 71976
2009-05-17 10:16:28 +00:00
Anton Korobeynikov 6b5523aec2 Typo
llvm-svn: 71975
2009-05-17 10:15:22 +00:00
Jakob Stoklund Olesen 9d7fb58581 Help DejaGnu avoid pipe-jam by producing less output from certain test cases.
When a test fails with more than a pipeful of output on stdout AND stderr, one
of the DejaGnu programs blocks. The problem can be avoided by redirecting
stdout to a file.

llvm-svn: 71919
2009-05-16 00:34:42 +00:00
David Greene 3587eed2c4 Implement !if, analogous to $(if) in GNU make.
llvm-svn: 71815
2009-05-14 23:26:46 +00:00
David Greene 9bb2de17c2 Fix tests to not upset DejaGNU.
llvm-svn: 71811
2009-05-14 23:21:40 +00:00
David Greene d571b3c94b Graduate LLVM to the big leagues by embedding a LISP processor into TableGen.
Ok, not really, but do support some common LISP functions:

* car
* cdr
* null

llvm-svn: 71805
2009-05-14 22:38:31 +00:00
David Greene e917fff30f Implement a !foreach operator analogous to GNU make's $(foreach).
Use it on dags and lists like this:

class decls {
  string name;
}

def Decls : decls;

class B<list<string> names> : A<!foreach(Decls.name, names, !strconcat(Decls.name, ", Sr."))>;

llvm-svn: 71803
2009-05-14 22:23:47 +00:00
David Greene 98ed3c7273 Implement a !subst operation simmilar to $(subst) in GNU make to do
def/var/string substitution on generic pattern templates.  For example:

def Type;
def v4f32 : Type;
def TYPE : Type;

class GenType<Type t> {
  let type = !(subst TYPE, v4f32, t);
}

def TheType : GenType<TYPE>;

llvm-svn: 71801
2009-05-14 21:54:42 +00:00
David Greene e8f3b27e65 Implement !cast.
llvm-svn: 71794
2009-05-14 21:22:49 +00:00
Dan Gohman 6b43dd7664 Add nounwind to this test.
llvm-svn: 71734
2009-05-13 22:29:12 +00:00
Bill Wendling 7abc25ad68 Remove too large testcase.
llvm-svn: 71730
2009-05-13 21:51:26 +00:00
Bill Wendling 1d601cdaf4 Move the bookkeeping of the debug scopes back to the place where it
belonged. The variable declaration stuff wasn't happy with it where it
was. Sorry that the testcase is so big. Bugpoint wasn't able to reduce it
successfully.

llvm-svn: 71714
2009-05-13 20:33:33 +00:00
Dale Johannesen 9e26d3708a Testcase for 71688.
llvm-svn: 71691
2009-05-13 18:33:24 +00:00
Chris Lattner 149546a6a0 calls in nothrow functions can be marked nothrow even if the callee
is not known to be nothrow.  This allows readnone/readonly functions
to be deleted even if we don't know whether the callee can throw.

llvm-svn: 71676
2009-05-13 17:39:14 +00:00
Chris Lattner 7e335a763a Fix PR4206 - crash in simplify lib calls
llvm-svn: 71644
2009-05-13 06:26:11 +00:00
Evan Cheng 85cca64dd7 If header of inner loop is aligned, do not align the outer loop header. We don't want to add nops in the outer loop for the sake of aligning the inner loop.
llvm-svn: 71609
2009-05-12 23:58:14 +00:00
Evan Cheng df1aeeeb90 Teach TransferDeadness to delete truly dead instructions if they do not produce side effects.
llvm-svn: 71606
2009-05-12 23:07:00 +00:00
Evan Cheng 79afe74195 Add nounwind.
llvm-svn: 71575
2009-05-12 18:35:43 +00:00
Evan Cheng c94de92c54 Fixed a stack slot coloring with reg bug: do not update implicit use / def when doing forward / backward propagation.
llvm-svn: 71574
2009-05-12 18:31:57 +00:00
Bob Wilson 9e3d48f10d Fix pr4195: When iterating through predecessor blocks, break out of the loop
after finding the (unique) layout predecessor.  Sometimes a block may be listed
more than once, and processing it more than once in this loop can lead to
inconsistent values for FtTBB/FtFBB, since the AnalyzeBranch method does not
clear these values.  There's no point in continuing the loop regardless.
The testcase for this is reduced from the 2003-05-02-DependentPHI SingleSource
test.

llvm-svn: 71536
2009-05-12 03:48:10 +00:00
Dan Gohman d76d71a291 Factor the code for collecting IV users out of LSR into an IVUsers class,
and generalize it so that it can be used by IndVarSimplify. Implement the
base IndVarSimplify transformation code using IVUsers. This removes
TestOrigIVForWrap and associated code, as ScalarEvolution now has enough
builtin overflow detection and folding logic to handle all the same cases,
and more. Run "opt -iv-users -analyze -disable-output" on your favorite
loop for an example of what IVUsers does.

This lets IndVarSimplify eliminate IV casts and compute trip counts in
more cases. Also, this happens to finally fix the remaining testcases
in PR1301.

Now that IndVarSimplify is being more aggressive, it occasionally runs
into the problem where ScalarEvolutionExpander's code for avoiding
duplicate expansions makes it difficult to ensure that all expanded
instructions dominate all the instructions that will use them. As a
temporary measure, IndVarSimplify now uses a FixUsesBeforeDefs function
to fix up instructions inserted by SCEVExpander. Fortunately, this code
is contained, and can be easily removed once a more comprehensive
solution is available.

llvm-svn: 71535
2009-05-12 02:17:14 +00:00
Dan Gohman ae451600fa When forgetting SCEVs for loop PHIs, don't forget SCEVUnknown values.
These values aren't analyzable, so they don't care if more information
about the loop trip count can be had. Also, SCEVUnknown is used for
a PHI while the PHI itself is being analyzed, so it needs to be left
in the Scalars map. This fixes a variety of subtle issues.

llvm-svn: 71533
2009-05-12 01:27:58 +00:00
Evan Cheng 78a4eb844b Teach LSR to optimize more loop exit compares, i.e. change them to use postinc iv value. Previously LSR would only optimize those which are in the loop latch block. However, if LSR can prove it is safe (and profitable), it's now possible to change those not in the latch blocks to use postinc values.
Also, if the compare is the only use, LSR would place the iv increment instruction before the compare instead in the latch.

llvm-svn: 71485
2009-05-11 22:33:01 +00:00
Dale Johannesen b571463363 Fix PR4188. TailMerging can't tolerate inexact
sucessor info.

llvm-svn: 71478
2009-05-11 21:54:13 +00:00
Dan Gohman 5b3a56bfac Make this grep line a little more specific so that it doesn't
accidentally match something unrelated.

llvm-svn: 71458
2009-05-11 18:49:56 +00:00
Dan Gohman 9521cadff7 When scalarizing a vector BITCAST, check whether the operand has vector
type, rather than assume that it does. If the operand is not vector, it
shouldn't be run through ScalarizeVectorOp. This fixes one of the
testcases in PR3886.

llvm-svn: 71453
2009-05-11 18:30:42 +00:00
Dan Gohman faf75c8c9a Convert a subtract into a negate and an add when it helps x86
address folding.

llvm-svn: 71446
2009-05-11 18:02:53 +00:00
Dale Johannesen 02cb2bf2e3 Reverse a loop that is counting up to a maximum to
count down to 0 instead, under very restricted
circumstances.  Adjust 4 testcases in which this
optimization fires.

llvm-svn: 71439
2009-05-11 17:15:42 +00:00
Nick Lewycky b8f9b7a965 Make MDNode use CallbackVH. Also change MDNode to store Value* instead of
Constant* in preperation of a future change to support holding non-Constants
in an MDNode.

llvm-svn: 71407
2009-05-10 20:57:05 +00:00
Anton Korobeynikov 7706cb8d5d Add MSP430 test for PR4136
llvm-svn: 71392
2009-05-10 14:48:36 +00:00
Eli Friedman ebf98b0212 Allow scalar evolution to compute iteration counts for loops with a
pointer-based condition.  This fixes PR3171.

llvm-svn: 71354
2009-05-09 12:32:42 +00:00
Evan Cheng 51fa9c7138 Enable loop bb placement optimization.
llvm-svn: 71291
2009-05-08 23:35:49 +00:00
Dan Gohman 35dc9b65ed Fix bogus overflow checks by replacing them with actual
overflow checks.

llvm-svn: 71284
2009-05-08 23:11:16 +00:00
Dan Gohman 18f026afa6 Use .td for tablegen files, not .ll.
llvm-svn: 71277
2009-05-08 23:01:28 +00:00
Dan Gohman 2e55cc5a4a Fold trunc casts into add-recurrence expressions, allowing the
add-recurrence to be exposed. Add a new SCEV folding rule to
help simplify expressions in the presence of these extra truncs.

llvm-svn: 71264
2009-05-08 21:03:19 +00:00
Chris Lattner f1d9b91434 Fix PR4152: asm constraint validation happens before dag combine, so we
need to work a bit to combine things like (x+c1+c2) into x+c3.

llvm-svn: 71232
2009-05-08 18:23:14 +00:00
Chris Lattner c48091f141 fix RewriteStoreUserOfWholeAlloca to use the correct type size
method, fixing a crash on PR4146.  While the store will 
ultimately overwrite the "padded size" number of bits in memory,
the stored value may be a subset of this size.  This function
only wants to handle the case where all bits are stored.

llvm-svn: 71224
2009-05-08 15:54:41 +00:00
Evan Cheng 2fa281106a Optimize code placement in loop to eliminate unconditional branches or move unconditional branch to the outside of the loop. e.g.
///       A:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       <fallthrough to B>                                                                                                                                                 
///                                                                                                                                                                          
///       B:  --> loop header                                                                                                                                                
///       ...                                                                                                                                                                
///       jcc <cond> C, [exit]                                                                                                                                               
///                                                                                                                                                                          
///       C:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jmp B                                                                                                                                                              
///                                                                                                                                                                          
/// ==>                                                                                                                                                                      
///                                                                                                                                                                          
///       A:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jmp B                                                                                                                                                              
///                                                                                                                                                                          
///       C:  --> new loop header                                                                                                                                            
///       ...                                                                                                                                                                
///       <fallthough to B>                                                                                                                                                  
///                                                                                                                                                                          
///       B:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jcc <cond> C, [exit] 

llvm-svn: 71209
2009-05-08 06:34:09 +00:00
Eli Friedman 36b9026fa7 PR4123: don't crash when inlining a call which uses its own result.
llvm-svn: 71199
2009-05-08 00:22:04 +00:00
Bob Wilson e20be4183c Fix pr4100. Do not remove no-op copies when they are dead. The register
scavenger gets confused about register liveness if it doesn't see them.
I'm not thrilled with this solution, but it only comes up when there are dead
copies in the code, which is something that hopefully doesn't happen much.

Here is what happens in pr4100: As shown in the following excerpt from the
debug output of llc, the source of a move gets reloaded from the stack,
inserting a new load instruction before the move.  Since that source operand
is a kill, the physical register is free to be reused for the destination
of the move.  The move ends up being a no-op, copying R3 to R3, so it is
deleted.  But, it leaves behind the load to reload %reg1028 into R3, and
that load is not updated to show that it's destination operand (R3) is dead.
The scavenger gets confused by that load because it thinks that R3 is live.

Starting RegAlloc of: %reg1025<def,dead> = MOVr %reg1028<kill>, 14, %reg0, %reg0
  Regs have values: 
  Reloading %reg1028 into R3
  Last use of R3[%reg1028], removing it from live set
  Assigning R3 to %reg1025
  Register R3 [%reg1025] is never used, removing it from live set

Alternative solutions might be either marking the load as dead, or zapping
the load along with the no-op copy.  I couldn't see an easy way to do
either of those, though.

llvm-svn: 71196
2009-05-07 23:47:03 +00:00
Dan Gohman 4bb6fa23cb Revert 71165. It did more than just revert 71158 and it introduced
several regressions. The problem due to 71158 is now fixed.

llvm-svn: 71176
2009-05-07 19:46:24 +00:00
Duncan Sands 7a8cfafd2e Revert r70876 and add a testcase (@c7) showing the problem:
bits captured, but the pointer marked nocapture.  In fact
I now recall that this problem is why only readnone functions
returning void were considered before!  However keep a small
fix that was also in r70876: a readnone function returning
void can result in bits being captured if it unwinds, so
test for this.

llvm-svn: 71168
2009-05-07 18:08:34 +00:00
Bill Wendling 17f0f65499 Temporarily revert r71158. It was causing a failure during a full bootstrap:
checking for bcopy... no
checking for getc_unlocked... Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decUtility.c:360: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decUtility.o] Error 1
make[4]: *** Waiting for unfinished jobs....
Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decNumber.c:5591: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decNumber.o] Error 1
make[3]: *** [all-stage2-libdecnumber] Error 2
make[3]: *** Waiting for unfinished jobs....

llvm-svn: 71165
2009-05-07 17:26:14 +00:00
Dan Gohman e5d2a04ace Constant-fold ptrtoint+add+inttoptr to gep when the pointer is an
array and the add is within range. This helps simplify expressions
expanded by ScalarEvolutionExpander.

llvm-svn: 71158
2009-05-07 14:24:56 +00:00
Bill Wendling 759de47964 THis doesn't fail.
llvm-svn: 71142
2009-05-07 01:41:42 +00:00
Bill Wendling 6144f63bd5 Temporarily revert r71010. It was causing massive failures during self-hosting.
llvm-svn: 71138
2009-05-07 01:27:25 +00:00
Evan Cheng cfc0513080 Do not use register as base ptr of pre- and post- inc/dec load / store nodes.
llvm-svn: 71098
2009-05-06 18:25:01 +00:00
Duncan Sands 86f2bc596a Nounwind is not valid for function return values.
llvm-svn: 71082
2009-05-06 13:51:18 +00:00
Duncan Sands 7374a0118d OCaml parameter attribute bindings from PR2752.
Incomplete, but better than nothing.

llvm-svn: 71081
2009-05-06 12:21:17 +00:00
Duncan Sands 9759f2e063 Fix PR3754: don't mark functions that wrap MallocInst with
the readnone.  Since MallocInst is scheduled for deletion
it doesn't seem worth doing anything more subtle, such as
having mayWriteToMemory return true for MallocInst.

llvm-svn: 71077
2009-05-06 08:42:00 +00:00
Duncan Sands 1efabaaa2a Allow readonly functions to unwind exceptions. Teach
the optimizers about this.  For example, a readonly
function with no uses cannot be removed unless it is
also marked nounwind.

llvm-svn: 71071
2009-05-06 06:49:50 +00:00
Lang Hames ab68cf24bb Renamed Spiller classes (plus uses and related files) to VirtRegRewriter.
llvm-svn: 71057
2009-05-06 02:36:21 +00:00
Mikhail Glushenkov 6da4749ce6 The 'forward_as' property did not use its second argument.
See PR4159 for details. Patch by Martin Nowack!

llvm-svn: 71054
2009-05-06 01:41:19 +00:00
Evan Cheng cfdbfccd91 Quotes should be printed before private prefix; some code clean up.
llvm-svn: 71032
2009-05-05 22:50:29 +00:00
Dan Gohman bb2f10759e If a MachineBasicBlock has multiple ways of reaching another block,
allow it to have multiple CFG edges to that block. This is needed
to allow MachineBasicBlock::isOnlyReachableByFallthrough to work
correctly. This fixes PR4126.

llvm-svn: 71018
2009-05-05 21:10:19 +00:00
Bill Wendling 5e2ac0cd9c Temporarily reverting r71008. It was causing this failure:
Running /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.src/test/
CodeGen/X86/dg.exp ...
FAIL: /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.src/test/
CodeGen/X86/change-compare-stride-1.ll
Failed with exit(1) at line 2
while running: grep {cmpq       $-478,} change-compare-stride-1.ll.tmp
child process exited abnormally

llvm-svn: 71013
2009-05-05 20:49:46 +00:00
Evan Cheng 9bf9002161 Enable stack coloring with regs at -O3.
llvm-svn: 71010
2009-05-05 20:30:36 +00:00
David Greene 246a3dfb10 Handle overflow of 64-bit loop conditions.
llvm-svn: 71008
2009-05-05 20:22:36 +00:00
Chris Lattner be9fa506ad Add basic support for code generation of
addrspace(257) -> FS relative on x86.  Patch by Zoltan Varga!

llvm-svn: 70992
2009-05-05 18:52:19 +00:00
David Greene 44f9d7a8f0 Allow multiclass def names to contain "#NAME"" where TableGen replaces
#NAME# with the name of the defm instantiating the multiclass.  This is
useful for AVX instruction naming where a "V" prefix is standard
throughout the ISA.  For example:

multiclass SSE_AVX_Inst<...> {
   def SS : Instr<...>;
   def SD : Instr<...>;
   def PS : Instr<...>;
   def PD : Instr<...>;

   def V#NAME#SS : Instr<...>;
   def V#NAME#SD : Instr<...>;
   def V#NAME#PS : Instr<...>;
   def V#NAME#PD : Instr<...>;
}

defm ADD : SSE_AVX_Inst<...>;

Results in 

ADDSS
ADDSD
ADDPS
ADDPD

VADDSS
VADDSD
VADDPS
VADDPD

llvm-svn: 70979
2009-05-05 16:28:25 +00:00
Mikhail Glushenkov d953cd9401 Fix incorrect code generation with ENV.
See PR4157 for details. Patch by Martin Nowack!

llvm-svn: 70973
2009-05-05 12:34:34 +00:00
Dan Gohman bb525f7e02 X86FastISel doesn't support the -tailcallopt ABI.
llvm-svn: 70902
2009-05-04 19:50:33 +00:00
Anton Korobeynikov 2d1e7321f6 Fix code emission for conditional branches.
Patch by Collin Winter!

llvm-svn: 70898
2009-05-04 19:10:38 +00:00
Bill Wendling 357e03cede Use %llvmgcc instead of llvm-gcc.
llvm-svn: 70886
2009-05-04 18:00:27 +00:00
Duncan Sands 50c70336b0 Teach capture tracking that readonly functions can
only capture their arguments by returning them or
throwing an exception or not based on the argument
value.  Patch essentially by Frits van Bommel.

llvm-svn: 70876
2009-05-04 16:50:29 +00:00
Duncan Sands b88227ef4c Check that pure/const functions are marked nounwind.
llvm-svn: 70875
2009-05-04 16:47:11 +00:00
Argyrios Kyrtzidis 9ae29b2d8f -Remove the DwarfWriter::RecordSourceLine calls from the instruction selectors.
-Depend on DebugLocs for source line info.

(Comes with Regression-Be-Gone(tm))

llvm-svn: 70871
2009-05-04 16:23:49 +00:00
Duncan Sands d84881ec53 Testcase for PR3967.
llvm-svn: 70856
2009-05-04 12:54:02 +00:00
Chris Lattner d579cb1167 * Sink 4 duplicates of edge threading validity checks and DOUT prints into
ThreadEdge directly.  This shares the code, but is just a refactoring.
* Make JumpThreading compute the set of loop headers and avoid threading
  across them.  This prevents jump threading from forming irreducible 
  loops (goodness) but also prevents it from threading in other cases that
  are beneficial (see the comment above FindFunctionBackedges).

llvm-svn: 70820
2009-05-04 02:28:08 +00:00
Argyrios Kyrtzidis 79be34012f Revert r70803 for now, it causes a regression.
llvm-svn: 70811
2009-05-03 23:27:19 +00:00
Argyrios Kyrtzidis ce7196b903 -Remove the DwarfWriter::RecordSourceLine calls from the instruction selectors.
-Depend on DebugLocs for source line info.

llvm-svn: 70803
2009-05-03 22:03:35 +00:00
Dan Gohman ff08995589 Previously, RecursivelyDeleteDeadInstructions provided an option
of returning a list of pointers to Values that are deleted. This was
unsafe, because the pointers in the list are, by nature of what
RecursivelyDeleteDeadInstructions does, always dangling. Replace this
with a simple callback mechanism. This may eventually be removed if
all clients can reasonably be expected to use CallbackVH.

Use this to factor out the dead-phi-cycle-elimination code from LSR
utility function, and generalize it to use the
RecursivelyDeleteTriviallyDeadInstructions utility function.

This makes LSR more aggressive about eliminating dead PHI cycles;
adjust tests to either be less trivial or to simply expect fewer
instructions.

llvm-svn: 70636
2009-05-02 18:29:22 +00:00
Chris Lattner e01821edbd 'The attached patch fixes an issue where llc -march=cpp fails with
"Invalid primitive type" on input containing the x86_fp80 type.'
Patch by Collin Winter!

llvm-svn: 70610
2009-05-01 23:54:26 +00:00
Dan Gohman 7227bc88f0 When printing a SCEVUnknown with pointer type, don't print an
artificial "ptrtoint", as it tends to clutter up complicated
expressions. The cast operators now print both source and
destination types, which is usually sufficient.

llvm-svn: 70554
2009-05-01 17:02:22 +00:00
Dan Gohman 2b8da35f9d Extend ScalarEvolution's getBackedgeTakenCount to be able to
compute an upper-bound value for the trip count, in addition to
the actual trip count. Use this to allow getZeroExtendExpr and
getSignExtendExpr to fold casts in more cases.

This may eventually morph into a more general value-range
analysis capability; there are certainly plenty of places where
more complete value-range information would allow more folding.

llvm-svn: 70509
2009-04-30 20:47:05 +00:00
Dan Gohman 4bafc42185 Don't try to mix integers and pointers in an icmp instruction
in getSCEVAtScope.

llvm-svn: 70495
2009-04-30 16:40:30 +00:00
Evan Cheng 99578674fd Mark MOV8mr_NOREX and MOV8rm_NOREX as mayStore / mayLoad respectively.
llvm-svn: 70461
2009-04-30 00:58:57 +00:00
Chris Lattner 5ab42e93c4 fix a regression handling indirect results: these need to be considered
memory operands otherwise the writebacks get lost when the inline asm 
doesn't otherwise have side effects.  This fixes rdar://6839427, though
clang really shouldn't generate these anymore.

llvm-svn: 70455
2009-04-30 00:48:50 +00:00
Nate Begeman 7e6e352735 Fix infinite recursion in the C++ code which handles movddup by making it unnecessary.
llvm-svn: 70425
2009-04-29 22:47:44 +00:00
Dan Gohman 494dac3f84 Generalize the cast-of-addrec folding to handle folding of SCEVs like
(sext i8 {-128,+,1} to i64) to i64 {-128,+,1}, where the iteration
crosses from negative to positive, but is still safe if the trip
count is within range.

llvm-svn: 70421
2009-04-29 22:28:28 +00:00
Dan Gohman d9775a3be1 Fix this test to match the new output from scalar-evolution.
llvm-svn: 70410
2009-04-29 21:06:20 +00:00
Dan Gohman d9b11b2ef4 Include the source type in SCEV cast expression debug output, and
print sext, zext, and trunc, instead of signextend, zeroextend,
and truncate, respectively, for consistency with the main IR.

llvm-svn: 70405
2009-04-29 20:27:52 +00:00
Dale Johannesen b0bc5c2f81 Fix recent regression in gcc.dg/pr26719.c (6835035).
llvm-svn: 70386
2009-04-29 16:38:47 +00:00
Evan Cheng 9cce299cc8 spillPhysRegAroundRegDefsUses() may have invalidated iterators stored in fixed_ IntervalPtrs. Reset them.
llvm-svn: 70378
2009-04-29 07:16:34 +00:00
Chris Lattner e6d2e627da testcase for PR4082
llvm-svn: 70375
2009-04-29 06:46:27 +00:00
Chris Lattner 7d10386113 Disable the load-shrinking optimization from looking at
anything larger than 64-bits, avoiding a crash.  This should
really be fixed to use APInts, though type legalization happens
to help us out and we get good code on the attached testcase at
least.

This fixes rdar://6836460

llvm-svn: 70360
2009-04-29 03:45:07 +00:00
Bill Wendling 084669a1c9 Second attempt:
Massive check in. This changes the "-fast" flag to "-O#" in llc. If you want to
use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.

Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'll change the JIT with a follow-up patch.

llvm-svn: 70343
2009-04-29 00:15:41 +00:00
Dan Gohman ed7f3a0eec As with r70333, give the primary induction variable a use so
that it can't be trivially eliminated.

llvm-svn: 70334
2009-04-28 22:05:13 +00:00
Dan Gohman 2625f2cb80 Make this testcase slightly less trivial, so that it doesn't fail
if indvars happens to optimize away the unused primary induction
variable.

llvm-svn: 70333
2009-04-28 22:03:26 +00:00
Dan Gohman 807dff7486 Fix a grammaro in a comment.
llvm-svn: 70331
2009-04-28 21:54:23 +00:00
Anton Korobeynikov dac88bae4f Properly print 'P' modifier on inline asm memory operands.
This should fix PR3379 and PR4064.
Patch inspired by Edwin Török!

llvm-svn: 70328
2009-04-28 21:49:33 +00:00
Dale Johannesen 25148e11fb Test for llvm-gcc bug fixed by 70301.
llvm-svn: 70302
2009-04-28 17:16:30 +00:00
Evan Cheng 7e09994cb8 Fix PR4034. Bug in LiveInterval::join when it's compacting new valno's.
llvm-svn: 70291
2009-04-28 06:24:09 +00:00
Evan Cheng c3884be983 Fix for PR4051. When 2address pass delete an instruction, update kill info when necessary.
llvm-svn: 70279
2009-04-28 02:12:36 +00:00
Bill Wendling 56f2987a87 r70270 isn't ready yet. Back this out. Sorry for the noise.
llvm-svn: 70275
2009-04-28 01:04:53 +00:00
Bill Wendling d0ae15946c Massive check in. This changes the "-fast" flag to "-O#" in llc. If you want to
use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.

Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'm not 100% sure if it's necessary to change it there...

llvm-svn: 70270
2009-04-28 00:21:31 +00:00
Dale Johannesen 27b4f222cf Fix PR 4086, a bug in FP IV elimination.
llvm-svn: 70247
2009-04-27 21:03:15 +00:00
Evan Cheng 093e4c578d Fix PR4076. Correctly create live interval of physical register with two-address update.
llvm-svn: 70245
2009-04-27 20:42:46 +00:00
Dan Gohman e99f98262c Permit ChangeCompareStride to rewrite a comparison when the factor
between the comparison's iv stride and the candidate stride is
exactly -1.

llvm-svn: 70244
2009-04-27 20:35:32 +00:00
Dan Gohman 7646637379 Teach getZeroExtendExpr and getSignExtendExpr to use trip-count
information to simplify [sz]ext({a,+,b}) to {zext(a),+,[zs]ext(b)},
as appropriate.

These functions and the trip count code each call into the other, so
this requires careful handling to avoid infinite recursion. During
the initial trip count computation, conservative SCEVs are used,
which are subsequently discarded once the trip count is actually
known.

Among other benefits, this change lets LSR automatically eliminate
some unnecessary zext-inreg and sext-inreg operation where the
operand is an induction variable.

llvm-svn: 70241
2009-04-27 20:16:15 +00:00
Dale Johannesen 95743b2efd Test for (llvm-gcc) 70231.
llvm-svn: 70233
2009-04-27 19:15:09 +00:00
Nate Begeman c79f731531 Revert accidental testcase reduction
llvm-svn: 70226
2009-04-27 18:42:40 +00:00
Nate Begeman 8d6d4b9289 2nd attempt, fixing SSE4.1 issues and implementing feedback from duncan.
PR2957

ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask.  A value of -1 represents UNDEF.

In addition to eliminating the creation of illegal BUILD_VECTORS just to 
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.

llvm-svn: 70225
2009-04-27 18:41:29 +00:00
Evan Cheng 0f85bd368c Fix PR4056. It's possible a physical register def is dead if its implicit use is deleted by two-address pass.
llvm-svn: 70213
2009-04-27 17:36:47 +00:00
Dan Gohman 3266391e86 Fix the syntax for a PR number in a test.
llvm-svn: 70208
2009-04-27 15:08:34 +00:00
Dan Gohman 58440cf305 Make this test slightly more strict.
llvm-svn: 70180
2009-04-27 03:05:26 +00:00
Dan Gohman be36f5ccda When transforming sext(trunc(load(x))) into sext(smaller load(x)),
the trunc is directly replaced with the smaller load, so don't
try to create a new sext node. This fixes PR4050.

llvm-svn: 70179
2009-04-27 02:00:55 +00:00
Dan Gohman 05c1d372b7 Handle ands with ~0 correctly too. This fixes PR4052.
llvm-svn: 70176
2009-04-27 01:41:10 +00:00
Sanjiv Gupta 8d319047c5 Any size of integral indices are allowed in gep for indexing into sequential types. Also adding a test case to check the indices type allowed into struct.
llvm-svn: 70134
2009-04-26 17:14:35 +00:00
Chris Lattner 73b5b5adae add testcase for strange types of gep indices
llvm-svn: 70085
2009-04-25 22:20:49 +00:00
Chris Lattner 4c73d7a9b8 testcase and asmparser fix for PR4066
llvm-svn: 70080
2009-04-25 21:26:00 +00:00
Dan Gohman 5638e0d642 Add several more icmp simplifications. Transform signed comparisons
into unsigned ones when the operands are known to have the same
sign bit value.

llvm-svn: 70053
2009-04-25 17:12:48 +00:00
Dan Gohman df19948ea7 Handle ands with 0 and shifts by 0 correctly. These aren't
common, but indvars shouldn't crash on them. This fixes PR4054.

llvm-svn: 70051
2009-04-25 17:05:40 +00:00
Torok Edwin c8afb09a3b Fix g++-4.4.0 warning, it was causing llvm-nm to fail on wrapped BC files:
Path.cpp:59: warning: case label value exceeds maximum value for type
magic[0] is a (signed) char, but some case values are unsigned (e.g. 0xde).
When magic[0] was 0xde, the switch has taken the default branch instead of case
0xde branch.
Apparently this was the behaviour with older versions of gcc too, but not with g++.
Now g++-4.4 behaves as gcc, and ignores unsigned case values out of range signed
range.

llvm-svn: 70038
2009-04-25 10:25:12 +00:00
Evan Cheng 362acf8a56 Do not share a single unknown val# for all the live ranges merged into a physical sub-register live interval. When coalescer is merging in clobbered virtaul register live interval into a physical register live interval, give each virtual register val# a separate val# in the physical register live interval. Otherwise, the coalescer would have lost track of the definitions information it needs to make correct coalescing decisions.
llvm-svn: 70026
2009-04-25 09:25:19 +00:00
Dale Johannesen 56cb14c874 Fix PR 4057, a crash doing float->char const folding.
This particular one is undefined behavior (although this
isn't related to the crash), so it will no longer do it
at compile time, which seems better.

llvm-svn: 69990
2009-04-24 21:34:13 +00:00
David Greene 7049e79e45 Fix multiclass inheritance to limit value resolution to new defs added
by base multiclasses.  Do not attempt to alter defs from previous base
multiclasses.  This fixes multiple multiclass inheritance.

llvm-svn: 69974
2009-04-24 16:55:41 +00:00
Rafael Espindola c1396a2313 Fix PR 4004 by including the call to __tls_get_addr in X86tlsaddr. This is not
very elegant, but neither is the tls specification :-(

llvm-svn: 69968
2009-04-24 12:59:40 +00:00
Rafael Espindola b93db668b3 Revert 69952. Causes testsuite failures on linux x86-64.
llvm-svn: 69967
2009-04-24 12:40:33 +00:00
Nate Begeman bb881d66f4 PR2957
ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask.  A value of -1 represents UNDEF.

In addition to eliminating the creation of illegal BUILD_VECTORS just to 
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.

A clean up of x86 shuffle code, and some canonicalizing in DAGCombiner is next.

llvm-svn: 69952
2009-04-24 03:42:54 +00:00
David Greene 196ac3c69a Make BinOps typed and require a type specifier for !nameconcat. This
allows binops to be used in typed contexts such as when passing
arguments to classes.

llvm-svn: 69921
2009-04-23 21:25:15 +00:00
Dan Gohman 723f175bc9 Explicitly pass -tailcallopt=false to these tests so that they
work as intended no matter what the default setting of that
option is.

llvm-svn: 69911
2009-04-23 19:39:41 +00:00
Dale Johannesen 466af3eb46 Testcase for 69795.
llvm-svn: 69901
2009-04-23 18:04:04 +00:00
Dan Gohman d6f9713634 Fix an error in this test.
llvm-svn: 69893
2009-04-23 15:22:28 +00:00
Dan Gohman 86bcd97014 Change SCEVExpander's expandCodeFor to provide more flexibility
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
2009-04-23 15:16:49 +00:00
Nick Lewycky b4d9f7a9b3 Simplify trunc(extend(x)) in SCEVs, just for completeness. Also fix some odd
whitespace in the same file.

llvm-svn: 69870
2009-04-23 05:15:08 +00:00
Owen Anderson 7d82244be7 Testcase for PR3909.
llvm-svn: 69868
2009-04-23 04:33:42 +00:00
Owen Anderson b94ea3ebf1 Testcase for PR2639.
llvm-svn: 69867
2009-04-23 04:30:52 +00:00
Owen Anderson b211b5c478 Testcase for PR2537.
llvm-svn: 69866
2009-04-23 04:26:42 +00:00
Owen Anderson 437c6357e0 Fix typo.
llvm-svn: 69865
2009-04-23 04:24:19 +00:00
Owen Anderson 0e357e5fd3 Testcase for PR3085.
llvm-svn: 69863
2009-04-23 04:21:14 +00:00
Owen Anderson 7e96d34f2e Add testcase from PR3086.
llvm-svn: 69862
2009-04-23 04:14:03 +00:00
Dan Gohman 4cc0cae46c Add more ulimit limits, to catch more kinds of runaway behavior.
llvm-svn: 69847
2009-04-23 00:28:31 +00:00
Evan Cheng d8174d3d09 Make sure both operands have binary instructions have the same type.
llvm-svn: 69844
2009-04-22 23:39:28 +00:00
Evan Cheng cbfe9df096 Avoid deferencing use_begin() if value does not have a use.
llvm-svn: 69836
2009-04-22 22:45:37 +00:00
David Greene f00919a040 Allow defm to inherit from multiple multiclasses.
llvm-svn: 69832
2009-04-22 22:17:51 +00:00
David Greene a9c6c5d39b Implement !nameconcat to concatenate strings and look up the resulting
name in the symbol table, returning an object.

llvm-svn: 69822
2009-04-22 20:18:10 +00:00
Duncan Sands 7bccadac92 Testcase for PR2958.
llvm-svn: 69818
2009-04-22 18:55:17 +00:00
David Greene 753ed8fd9c Implement multiclass inheritance.
llvm-svn: 69810
2009-04-22 16:42:54 +00:00
Dan Gohman 54fd2869a3 SCEVExpander's InsertCastOfTo knows how to move existing cast
instructions in order to avoid inserting new ones. However, if
the cast instruction is the SCEVExpander's InsertPt, this
causes subsequently emitted instructions to be inserted near
the cast, and not at the location of the original insert point.
Fix this by adjusting the insert point in such cases.
This fixes PR4009.

llvm-svn: 69808
2009-04-22 16:11:16 +00:00
Duncan Sands ab4c600e8b These tests are x86 specific.
llvm-svn: 69798
2009-04-22 10:39:51 +00:00
Evan Cheng 1a99a5f501 It has finally happened. Spiller is now using live interval info.
This fixes a very subtle bug. vr defined by an implicit_def is allowed overlap with any register since it doesn't actually modify anything. However, if it's used as a two-address use, its live range can be extended and it can be spilled. The spiller must take care not to emit a reload for the vn number that's defined by the implicit_def. This is both a correctness and performance issue.

llvm-svn: 69743
2009-04-21 22:46:52 +00:00
Dan Gohman e14efcc9f4 When turning (ashr(shl(x, n), n)) into sext(trunc(x)), the width of the
type to truncate to should be the number of bits of the value that are
preserved, not the number that are clobbered with sign-extension.
This fixes regressions in ldecod.

llvm-svn: 69704
2009-04-21 20:18:36 +00:00
Devang Patel 55f84db7ea Test case for revision 69683.
llvm-svn: 69684
2009-04-21 17:21:01 +00:00
Chris Lattner 69223bb7f5 fix a crash on a pointless but valid zero-length memset, rdar://6808691
llvm-svn: 69680
2009-04-21 16:52:12 +00:00
Evan Cheng d67efaa847 Added a linearscan register allocation optimization. When the register allocator spill an interval with multiple uses in the same basic block, it creates a different virtual register for each of the reloads. e.g.
%reg1498<def> = MOV32rm %reg1024, 1, %reg0, 12, %reg0, Mem:LD(4,4) [sunkaddr39 + 0]
        %reg1506<def> = MOV32rm %reg1024, 1, %reg0, 8, %reg0, Mem:LD(4,4) [sunkaddr42 + 0]
        %reg1486<def> = MOV32rr %reg1506
        %reg1486<def> = XOR32rr %reg1486, %reg1498, %EFLAGS<imp-def,dead>
        %reg1510<def> = MOV32rm %reg1024, 1, %reg0, 4, %reg0, Mem:LD(4,4) [sunkaddr45 + 0]

=>

        %reg1498<def> = MOV32rm %reg2036, 1, %reg0, 12, %reg0, Mem:LD(4,4) [sunkaddr39 + 0]
        %reg1506<def> = MOV32rm %reg2037, 1, %reg0, 8, %reg0, Mem:LD(4,4) [sunkaddr42 + 0]
        %reg1486<def> = MOV32rr %reg1506
        %reg1486<def> = XOR32rr %reg1486, %reg1498, %EFLAGS<imp-def,dead>
        %reg1510<def> = MOV32rm %reg2038, 1, %reg0, 4, %reg0, Mem:LD(4,4) [sunkaddr45 + 0]

From linearscan's point of view, each of reg2036, 2037, and 2038 are separate registers, each is "killed" after a single use. The reloaded register is available and it's often clobbered right away. e.g. In thise case reg1498 is allocated EAX while reg2036 is allocated RAX. This means we end up with multiple reloads from the same stack slot in the same basic block.

Now linearscan recognize there are other reloads from same SS in the same BB. So it'll "downgrade" RAX (and its aliases) after reg2036 is allocated until the next reload (reg2037) is done. This greatly increase the likihood reloads from SS are reused.

This speeds up sha1 from OpenSSL by 5.8%. It is also an across the board win for SPEC2000 and 2006.

llvm-svn: 69585
2009-04-20 08:01:12 +00:00
Chris Lattner 1e7da23983 testcase for PR3898
llvm-svn: 69473
2009-04-18 20:49:22 +00:00
Duncan Sands e4ff21ba4b Don't try to make BUILD_VECTOR operands have the same
type as the vector element type: allow them to be of
a wider integer type than the element type all the way
through the system, and not just as far as LegalizeDAG.
This should be safe because it used to be this way
(the old type legalizer would produce such nodes), so
backends should be able to handle it.  In fact only
targets which have legal vector types with an illegal
promoted element type will ever see this (eg: <4 x i16>
on ppc).  This fixes a regression with the new type
legalizer (vec_splat.ll).  Also, treat SCALAR_TO_VECTOR
the same as BUILD_VECTOR.  After all, it is just a
special case of BUILD_VECTOR.

llvm-svn: 69467
2009-04-18 20:16:54 +00:00
Dale Johannesen 2f6263fea3 Adjust XFAIL syntax, maybe that will help. The other
way worked for me...

llvm-svn: 69414
2009-04-18 02:01:23 +00:00
Dale Johannesen e34fb6b5ce patch 69408 breaks this by removing the opportunity
for the optimization it's testing to kick in (although
it improves the code, getting rid of all spills).
I don't understand the optimization well enough to
rescue the test, so XFAILing.

llvm-svn: 69409
2009-04-18 00:11:50 +00:00
Bob Wilson 9c1ec76084 Rename file to have the correct suffix.
llvm-svn: 69380
2009-04-17 20:40:20 +00:00
Bob Wilson a4c2290e5f Use CallConvLower.h and TableGen descriptions of the calling conventions
for ARM.  Patch by Sandeep Patel.

llvm-svn: 69371
2009-04-17 19:07:39 +00:00
Rafael Espindola 355fe12c82 For general dynamic TLS access we must use
leaq	foo@TLSGD(%rip), %rdi

as part of the instruction sequence. Using a register other than %rdi and then
copying it to %rdi is not valid.

llvm-svn: 69350
2009-04-17 14:35:58 +00:00
Evan Cheng b96a1082a9 Teach spiller to unfold instructions which modref spill slot when a scratch
register is available and when it's profitable.

e.g.
     xorq  %r12<kill>, %r13
     addq  %rax, -184(%rbp)
     addq  %r13, -184(%rbp)
==>
     xorq  %r12<kill>, %r13
     movq  -184(%rbp), %r12
     addq  %rax, %r12
     addq  %r13, %r12
     movq  %r12, -184(%rbp)

Two more instructions, but fewer memory accesses. It can also open up
opportunities for more optimizations.

llvm-svn: 69341
2009-04-17 01:29:40 +00:00
Rafael Espindola 5e42177a0f fix PR3995. A scale must be 1, 2, 4 or 8.
llvm-svn: 69284
2009-04-16 12:34:53 +00:00
Dan Gohman 0a40ad93a9 Expand GEPs in ScalarEvolution expressions. SCEV expressions can now
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
2009-04-16 03:18:22 +00:00
Dale Johannesen 8958f4f30f Another testcase for IV shortening.
llvm-svn: 69247
2009-04-16 00:45:21 +00:00
Bill Wendling 066b8023a4 Check for alignment.
llvm-svn: 69140
2009-04-15 04:51:05 +00:00
Dale Johannesen 7ffb7d5728 Enhance induction variable code to remove the
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
2009-04-15 01:10:12 +00:00
Devang Patel 046bf624b9 While inlining, clone llvm.dbg.func.start intrinsic and adjust
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
2009-04-15 00:17:06 +00:00
Bill Wendling e665e3cf2f Testcase for r69104.
llvm-svn: 69110
2009-04-15 00:04:11 +00:00
Evan Cheng 5ebf2acd84 Optimize conditional branch on i1 phis with non-constant inputs.
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
2009-04-14 23:40:03 +00:00
Dan Gohman 86e43e7999 Fix the RUN lines so that this test actually tests.
llvm-svn: 69096
2009-04-14 22:50:17 +00:00
Dan Gohman 62f4498646 For the h-register addressing-mode trick, use the correct value for
any non-address uses of the address value. This fixes 186.crafty.

llvm-svn: 69094
2009-04-14 22:45:05 +00:00
Dan Gohman e5cd1fcdb9 When the result of an EXTRACT_SUBREG, INSERT_SUBREG, or SUBREG_TO_REG
operator is used by a CopyToReg to export the value to a different
block, don't reuse the CopyToReg's register for the subreg operation
result if the register isn't precisely the right class for the
subreg operation.

Also, rename the h-registers.ll test, now that there are more
than one.

llvm-svn: 69087
2009-04-14 22:17:14 +00:00
Evan Cheng dfbbf5c043 Some of GR8_NOREX registers are only available in 64-bit mode.
llvm-svn: 69049
2009-04-14 16:57:43 +00:00
Dale Johannesen b866ce73b2 Use the output of the asm so the optimizer won't
delete it.

llvm-svn: 69018
2009-04-14 01:51:40 +00:00
Evan Cheng 9787183b9b Fix PR3934 part 2. findOnlyInterestingUse() was not setting IsCopy and IsDstPhys which are returned by value and used by callee. This happened to work on the earlier test cases because of a logic error in the caller side.
llvm-svn: 69006
2009-04-14 00:32:25 +00:00
Evan Cheng f0843803a0 PR3934: Fix a bogus two-address pass assertion.
llvm-svn: 68979
2009-04-13 20:04:24 +00:00
Dan Gohman 57d6bd36b2 Implement x86 h-register extract support.
- Add patterns for h-register extract, which avoids a shift and mask,
   and in some cases a temporary register.
 - Add address-mode matching for turning (X>>(8-n))&(255<<n), where
   n is a valid address-mode scale value, into an h-register extract
   and a scaled-offset address.
 - Replace X86's MOV32to32_ and related instructions with the new
   target-independent COPY_TO_SUBREG instruction.

On x86-64 there are complicated constraints on h registers, and
CodeGen doesn't currently provide a high-level way to express all of them,
so they are handled with a bunch of special code. This code currently only
supports extracts where the result is used by a zero-extend or a store,
though these are fairly common.

These transformations are not always beneficial; since there are only
4 h registers, they sometimes require extra move instructions, and
this sometimes increases register pressure because it can force out
values that would otherwise be in one of those registers. However,
this appears to be relatively uncommon.

llvm-svn: 68962
2009-04-13 16:09:41 +00:00
Rafael Espindola 6d6c6043ea X86-64 TLS support for local exec and initial exec.
llvm-svn: 68947
2009-04-13 13:02:49 +00:00
Chris Lattner 184f1be4a8 Add a new "available_externally" linkage type. This is intended
to support C99 inline, GNU extern inline, etc.  Related bugzilla's
include PR3517, PR3100, & PR2933.  Nothing uses this yet, but it
appears to work.

llvm-svn: 68940
2009-04-13 05:44:34 +00:00
Rafael Espindola 7186f20a1b In X86DAGToDAGISel::MatchWrapper, if base or index are set, avoid matching
only if symbolic addresses are RIP relatives.

llvm-svn: 68924
2009-04-12 23:00:38 +00:00
Rafael Espindola e4bd8904f7 Add tests for the parts of X86-64 TLS that are already implemented.
llvm-svn: 68901
2009-04-12 10:43:41 +00:00
Chris Lattner ce6bcf0847 fix a cross-block fastisel crash handling overflow intrinsics.
See comment for details.  This fixes rdar://6772169

llvm-svn: 68890
2009-04-12 07:51:14 +00:00
Chris Lattner 91b6af24ac add some optimizations for strncpy/strncat and factor some
code.  Patch by Benjamin Kramer!

llvm-svn: 68885
2009-04-12 05:06:39 +00:00
Chris Lattner 4d59f88e60 move a target-specific test into its directory so it isn't run if you
don't configure the ARM target in.

llvm-svn: 68843
2009-04-10 23:58:38 +00:00
Chris Lattner 30c3de6461 fix two problems with machine sinking:
1. Sinking would crash when the first instruction of a block was
   sunk due to iterator problems.
2. Instructions could be sunk to their current block, causing an
   infinite loop.

This fixes PR3968

llvm-svn: 68787
2009-04-10 16:38:36 +00:00
Rafael Espindola bb834f0929 Don't fold a load if the other operand is a TLS address.
With this we generate

movl    %gs:0, %eax
leal    i@NTPOFF(%eax), %eax

instead of

movl    $i@NTPOFF, %eax
addl    %gs:0, %eax

llvm-svn: 68778
2009-04-10 10:09:34 +00:00
Bob Wilson 51856173c8 Fix pr3954. The register scavenger asserts for inline assembly with
register destinations that are tied to source operands.  The
TargetInstrDescr::findTiedToSrcOperand method silently fails for inline
assembly.  The existing MachineInstr::isRegReDefinedByTwoAddr was very
close to doing what is needed, so this revision makes a few changes to
that method and also renames it to isRegTiedToUseOperand (for consistency
with the very similar isRegTiedToDefOperand and because it handles both
two-address instructions and inline assembly with tied registers).

llvm-svn: 68714
2009-04-09 17:16:43 +00:00
Chris Lattner a725028d41 reg0 references are not real registers. This fixes a crash on the
attached testcase.

llvm-svn: 68712
2009-04-09 16:50:43 +00:00
Dan Gohman 0e8d199f91 Generalize ExtendUsesToFormExtLoad to be usable for ANY_EXTEND,
in addition to ZERO_EXTEND and SIGN_EXTEND. Fix a bug in the
way it checked for live-out values, and simplify the way it
find users by using SDNode::use_iterator's (relatively) new
features. Also, make it slightly more permissive on targets
with free truncates.

In SelectionDAGBuild, avoid creating ANY_EXTEND nodes that are
larger than necessary. If the target's SwitchAmountTy has
enough bits, use it. This exposes the truncate to optimization
early, enabling more optimizations.

llvm-svn: 68670
2009-04-09 03:51:29 +00:00
Rafael Espindola 3b2df10c9e Re-apply 68552.
Tested by bootstrapping llvm-gcc and using that to build llvm.

llvm-svn: 68645
2009-04-08 21:14:34 +00:00