llvm-svn: 72411

low-level alias() method, allowing it to reason more aggressively
about pointers into constant memory.  PR4189

llvm-svn: 72403

or use labels as members of structures for example. Also included a couple of
whitespace fixes.

llvm-svn: 72402

even on Core2, not just AMD processors which was a surprise to me.

llvm-svn: 72396

llvm-svn: 72395

llvm-svn: 72387

can't just eliminate all global initializers.

llvm-svn: 72378

simplification. It's not clear to me whether this can replace the first of the
linear instruction simplification stages or not, so I left it in.

llvm-svn: 72377

llvm-svn: 72376

in the case where a loop exit value cannot be computed, instead of only in
some cases while using SCEVCouldNotCompute in others. This simplifies
getSCEVAtScope's callers.

llvm-svn: 72375

bit clearer.

llvm-svn: 72374

corresponding integer type is legal.

llvm-svn: 72373

one of the RecursivelyDeleteTriviallyDeadInstructions.
Add a comment explaining why the cache needs to be cleared.

llvm-svn: 72372

Also fix 80 column violation.

llvm-svn: 72371

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

Use &Ops[0] instead, which is safe since Ops will never be empty here.

llvm-svn: 72368

comparison, use takeName to give the integer comparison a name.

llvm-svn: 72367

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

of low bits of the other operand are preserved in the output.

llvm-svn: 72365

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

it's late, and I don't have any better ideas at the moment.  Fixes PR4257.

llvm-svn: 72363

can crash during instruction simplification (for example if it creates a
broken module).

llvm-svn: 72362

Sometimes when bugpointing a crash the bugpoint-reduced-simplified.bc reproduces
a totally different bug than the original one ("GV doesn't have initializer").
Although its useful to report that bug too, I need a way to reduce the original
bug, hence I introduced -disable-global-remove.

llvm-svn: 72361

llvm-svn: 72360

all results and all operands are legal, so this change shouldn't affect 
behavior at all.

llvm-svn: 72359

This leaves around 4000 lines of dead code; I'll clean that up 
in subsequent commits.

llvm-svn: 72358

llvm-svn: 72357

nodes it generates.

llvm-svn: 72356

ExpandExtractFromVectorThroughStack.

llvm-svn: 72351

moment.

llvm-svn: 72350

doesn't split legal vector operands.  This is necessary because the 
type legalization (and therefore, vector splitting) code will be going 
away soon.

llvm-svn: 72349

llvm-svn: 72343

llvm-svn: 72340

llvm-svn: 72339

llvm-svn: 72337

llvm-svn: 72336

llvm-svn: 72335

llvm-svn: 72332

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

llvm-svn: 72329