llvm-svn: 22696

llvm-svn: 22694

isHighOnes, where it would consider 0 to have high ones.

llvm-svn: 22693

llvm-svn: 22692

llvm-svn: 22691

* Teach this code to move allocas out of the loop when tail call eliminating
  a call marked 'tail'.  This implements TailCallElim/move_alloca_for_tail_call.ll
* Do not perform this transformation if a call is marked 'tail' and if there
  are allocas that we cannot move out of the loop in #2.  Doing so would increase
  the stack usage of the function.  This implements fixes
  PR615 and TailCallElim/dont-tce-tail-marked-call.ll.

llvm-svn: 22690

llvm-svn: 22689

depending on the command line option.  Now the command line option just
sets the subtarget as appropriate.  G5 opts will now default to on on
G5-enabled nightly testers among other machines.

llvm-svn: 22688

llvm-svn: 22687

to pass it in.  Also, since it always returns a non-null pointer, make it
return a reference instead for easier use.

llvm-svn: 22686

is available, since the target triple doesn't specify whether to use gpopts
or not.

llvm-svn: 22685

llvm-svn: 22683

llvm-svn: 22682

llvm-svn: 22681

avoid revisiting nodes more than once.  This eliminates a source of
potentially exponential behavior.  For a small function in 191.fma3d
(hexah_stress_divergence_), this speeds up isel from taking > 20mins to
taking 0.07s.

llvm-svn: 22680

llvm-svn: 22679

yesterday.  This fixes whetstone and a bunch of programs in the External tests.

llvm-svn: 22678

a subtarget.

llvm-svn: 22677

PHI is its only operand.

llvm-svn: 22676

llvm-svn: 22675

This fixes LSR crashes on 301.apsi, 191.fma3d, and 189.lucas

llvm-svn: 22673

the PHI node, this ugly code can vanish.

llvm-svn: 22672

llvm-svn: 22671

dominate the PHI node, this code can go away.  This also makes passes more
aggressive, e.g. implementing Transforms/CondProp/phisimplify2.ll

llvm-svn: 22670

prepared to deal with return values that do not dominate the PHI.  If we
cannot prove that the result dominates the PHI node, do not return it if
the client can't cope.

llvm-svn: 22669

after threading the branch, because both operands are the same value.

llvm-svn: 22668

llvm-svn: 22667

llvm-svn: 22666

llvm-svn: 22665

BasicBlock's removePredecessor routine.  This requires shuffling around
the definition and implementation of hasContantValue from Utils.h,cpp into
Instructions.h,cpp

llvm-svn: 22664

llvm-svn: 22663

that the symbolic evaluator is not always able to use subtraction to remove
expressions.  This makes the code faster, and fixes the last crash on 178.galgel.
Finally, add a statistic to see how many phi nodes are inserted.

On 178.galgel, we get the follow stats:

2562 loop-reduce  - Number of PHIs inserted
3927 loop-reduce  - Number of GEPs strength reduced

llvm-svn: 22662

llvm-svn: 22661

know what The Right Thing To Do is.

llvm-svn: 22660

and asm printer for PowerPC if one is not specified.

llvm-svn: 22659

  method.
* Fix a crash on 178.galgel, where we would insert expressions before PHI
  nodes instead of into the PHI node predecessor blocks.

llvm-svn: 22657

llvm-svn: 22656

not execution tests.

llvm-svn: 22655

llvm-svn: 22653

  for (i = 0; i < N; ++i)
    A[i][foo()] = 0;

here we still want to strength reduce the A[i] part, even though foo() is
l-v.

This also simplifies some of the 'CanReduce' logic.

This implements Transforms/LoopStrengthReduce/ops_after_indvar.ll

llvm-svn: 22652