partially unroll a loop when fully unrolling would not fit under the threshold.

Patch by Mikael Lepistö.

llvm-svn: 54160

llvm-svn: 54144

This fixes PR2599.

llvm-svn: 54133

llvm-svn: 54128

command-line option, and disable it by default. It introduced performance
regressions because CodeGen is currently not able to remat such loads.

llvm-svn: 53997

case for this.

This allows instructions like loads from global variables declared to
be constant to be moved out of loops."

Patch by Stefanus Du Toit!

llvm-svn: 53945

Remove the GetResultInst instruction. It is still accepted in LLVM assembly
and bitcode, where it is now auto-upgraded to ExtractValueInst. Also, remove
support for return instructions with multiple values. These are auto-upgraded
to use InsertValueInst instructions.

The IRBuilder still accepts multiple-value returns, and auto-upgrades them
to InsertValueInst instructions.

llvm-svn: 53941

leads into a cycle involving a different PHI, LSR got stuck running
around that cycle looking for the original PHI. To avoid this, keep
track of visited PHIs and stop searching if we see one more than once.
This fixes PR2570.

llvm-svn: 53879

llvm-svn: 53771

Make PRE actually handle critical edges (by splitting them).  Confirmed that bootstrap passes with this change.

llvm-svn: 53762

llvm-svn: 53730

llvm-svn: 53715

llvm-svn: 53705

could cause problems for memdep when it breaks critical edges.

llvm-svn: 53691

FindInsertedValue, it now performs a number of simple transformations that
should result in the same effect when applied iteratively.

llvm-svn: 53673

llvm-svn: 53666

llvm-svn: 53627

llvm-svn: 53616

Have GVN do a pre-pass over the CFG that folds away unconditional branches where possible.  This allows local PRE to be more aggressive.

llvm-svn: 53615

llvm-svn: 53564

disproving a condition.  This actually compiles the existing testcase
(udiv_select_to_select_shift) to:

define i64 @test(i64 %X, i1 %Cond) {
entry:
	%divisor1.t = lshr i64 %X, 3		; <i64> [#uses=1]
	%quotient2 = lshr i64 %X, 3		; <i64> [#uses=1]
	%sum = add i64 %divisor1.t, %quotient2		; <i64> [#uses=1]
	ret i64 %sum
}

instead of:

define i64 @test(i64 %X, i1 %Cond) {
entry:
	%quotient1.v = select i1 %Cond, i64 3, i64 4		; <i64> [#uses=1]
	%quotient1 = lshr i64 %X, %quotient1.v		; <i64> [#uses=1]
	%quotient2 = lshr i64 %X, 3		; <i64> [#uses=1]
	%sum = add i64 %quotient1, %quotient2		; <i64> [#uses=1]
	ret i64 %sum
}

llvm-svn: 53534

Remove dead code analyzing urem. 'urem' of power-of-2 is canonicalized to an
'and' instruction.

llvm-svn: 53506

llvm-svn: 53470

llvm-svn: 53454

llvm-svn: 53452

similar.

llvm-svn: 53451

llvm-svn: 53449

llvm-svn: 53448

llvm-svn: 53447

the min/max values for an integer type, compare against the min/max
values we can prove contain the input.  This might be a tighter bound,
so this is general goodness.

llvm-svn: 53446

handle them in some code.

llvm-svn: 53445

was using the algorithm for folding unsigned comparisons which is
completely wrong.  This has been broken since the signless types change.

llvm-svn: 53444

This cause a regression in InstCombine/JavaCompare, which was doing the right
thing on accident.  To handle the missed case, generalize the comparisons based
on masked bits a little bit to handle comparisons against the max value. For 
example, we can now xform (slt i32 (and X, 4), 4) -> (setne i32 (and X, 4), 4)

llvm-svn: 53443

llvm-svn: 53393

llvm-svn: 53283

llvm-svn: 53282

llvm-svn: 53276

llvm-svn: 53271

llvm-svn: 53265

into phis.  This is actually the same bug as PR2262 /
2008-04-29-VolatileLoadDontMerge.ll, but I missed checking the first 
predecessor for multiple successors.  Testcase here:
InstCombine/2008-07-08-VolatileLoadMerge.ll

llvm-svn: 53240