Commit Graph

5336 Commits

Author SHA1 Message Date
Owen Anderson 80baed63b4 Second batch of passes using LLVMContext.
llvm-svn: 74753
2009-07-03 00:54:20 +00:00
Owen Anderson b5618da226 Convert the first batch of passes to use LLVMContext.
llvm-svn: 74748
2009-07-03 00:17:18 +00:00
Chris Lattner f3f6aaa2c3 fix inverted logic pointed out by John McCall, noticed by inspection.
This was considering vector intrinsics to have cost 2, but non-vector
intrinsics to have cost 1, which is backward.

llvm-svn: 74698
2009-07-02 15:39:39 +00:00
Dan Gohman 43f33dd550 Fix a bunch of other places that used operator[] to test whether
a key is present in a std::map or DenseMap to use find instead.

llvm-svn: 74676
2009-07-02 00:17:47 +00:00
Dan Gohman cf092389a9 Request LCSSA after LoopSimplify. This fixes a problem in which the
PassManager was scheduling LCSSA before LoopSimplify, which does not
preserve LCSSA.

llvm-svn: 74661
2009-07-01 23:21:38 +00:00
Dan Gohman 83348f80b6 Fix an instcombine abort on a scalar-to-vector bitcast. This fixes PR4487.
llvm-svn: 74646
2009-07-01 21:38:46 +00:00
Owen Anderson 6773d388aa Add a pointer to the owning LLVMContext to Module. This requires threading LLVMContext through a lot
of the bitcode reader and ASM parser APIs, as well as supporting it in all of the tools.

Patches for Clang and LLVM-GCC to follow.

llvm-svn: 74614
2009-07-01 16:58:40 +00:00
Chris Lattner 96122debc1 improve the APIs for creating struct and function types with no arguments/elements
to not have to create a temporary vector (in the API at least).  Patch by Jay Foad!

llvm-svn: 74584
2009-07-01 04:13:31 +00:00
Dan Gohman 4dfc680059 Minor code simplification.
llvm-svn: 74491
2009-06-30 01:24:43 +00:00
Dan Gohman 317f054531 Don't try to split a loop when the controlling icmp instruction
doesn't have an IV-based operand. This fixes PR4471.

llvm-svn: 74399
2009-06-27 22:58:27 +00:00
Dan Gohman ffdcba3dbd Remove the block from the LoopInfo, rather than just the Loop.
LoopInfo will handle removing it from the Loop, as well as updating
its own tables.

llvm-svn: 74398
2009-06-27 22:32:36 +00:00
Dan Gohman c8ca49659a Teach LoopSimplify how to merge multiple loop exits into a single exit,
when one of them can be converted to a trivial icmp and conditional
branch.

This addresses what is essentially a phase ordering problem.
SimplifyCFG knows how to do this transformation, but it doesn't do so
if the primary block has any instructions in it other than an icmp and
a branch. In the given testcase, the block contains other instructions,
however they are loop-invariant and can be hoisted. SimplifyCFG doesn't
have LoopInfo though, so it can't hoist them. And, it's important that
the blocks be merged before LoopRotation, as it doesn't support
multiple-exit loops.

llvm-svn: 74396
2009-06-27 21:30:38 +00:00
Dan Gohman 8918b481bf More minor code simplifications.
llvm-svn: 74395
2009-06-27 21:23:40 +00:00
Dan Gohman fe174b6952 When a value is used multiple times within a single PHI, instructions
inserted to replace that value must dominate all of of the basic
blocks associated with the uses of the value in the PHI, not just
one of them.

llvm-svn: 74376
2009-06-27 05:16:57 +00:00
Dan Gohman daafbe6168 Incorporate the insertion point into the key of SCEVExpander's CSE map.
This helps it avoid reusing an instruction that doesn't dominate all
of the users, in cases where the original instruction was inserted
before all of the users were known.  This may result in redundant
expansions of sub-expressions that depend on loop-unpredictable values
in some cases, however this isn't very common, and it primarily impacts
IndVarSimplify, so GVN can be expected to clean these up.

This eliminates the need for IndVarSimplify's FixUsesBeforeDefs,
which fixes several bugs.

llvm-svn: 74352
2009-06-26 22:53:46 +00:00
Devang Patel 0f2eb5b9f7 Remove unused routines.
llvm-svn: 74351
2009-06-26 22:53:22 +00:00
Owen Anderson 01ad6605c0 Constify this value.
llvm-svn: 74330
2009-06-26 21:39:56 +00:00
Douglas Gregor 6d94e6a5f3 Fix linking of llvm-ld and lli with CMake, from Xerxes Rånby
llvm-svn: 74285
2009-06-26 15:37:00 +00:00
Devang Patel 0751a28888 Remove debug info anchors - llvm.dbg.compile_units, llvm.dbg.subprograms
and llvm.dbg.global_variables.

llvm-svn: 74251
2009-06-26 01:49:18 +00:00
Dan Gohman ac3b5382b8 Change this code to a form about which VC++ reportedly isn't unhappy.
llvm-svn: 74243
2009-06-26 00:35:12 +00:00
Dan Gohman 7eaf50ecac Fix LCSSA to avoid emitting a PHI node for the unwind destination of
an invoke instruction, since the value isn't really live across that
edge.

llvm-svn: 74242
2009-06-26 00:31:13 +00:00
Dan Gohman 31167c61d5 Minor code simplification.
llvm-svn: 74240
2009-06-26 00:26:03 +00:00
Dan Gohman 091e440568 Reword a few comments.
llvm-svn: 74146
2009-06-25 00:22:44 +00:00
Dan Gohman 929fa7b0f4 When inserting code into a loop preheader, insert it before the
terminator, instead of after the last phi. This fixes a bug
exposed by ScalarEvolution analyzing more kinds of loops.
This fixes PR4436.

llvm-svn: 74072
2009-06-24 14:31:06 +00:00
Dan Gohman f19aeec3f5 Extend ScalarEvolution's multiple-exit support to compute exact
trip counts in more cases.

Generalize ScalarEvolution's isLoopGuardedByCond code to recognize
And and Or conditions, splitting the code out into an
isNecessaryCond helper function so that it can evaluate Ands and Ors
recursively, and make SCEVExpander be much more aggressive about
hoisting instructions out of loops.

test/CodeGen/X86/pr3495.ll has an additional instruction now, but
it appears to be due to an arbitrary register allocation difference.

llvm-svn: 74048
2009-06-24 01:18:18 +00:00
Dan Gohman f522a4e034 Don't emit a redundant BitCastInst if the value to be defined in the
preheader is already an instruction.

llvm-svn: 74031
2009-06-24 00:28:59 +00:00
Dan Gohman fd76113e28 Fix a few minor issues that were exposed by the removal of SCEVHandle.
llvm-svn: 73910
2009-06-22 22:08:45 +00:00
Owen Anderson 65b6056e37 SCEVHandle is no more!
llvm-svn: 73906
2009-06-22 21:39:50 +00:00
Dan Gohman 78ea89e161 Fix this code to correctly handle loops with multiple exits. Until
now, this hasn't mattered, because ScalarEvolution hasn't been able
to compute trip counts for loops with multiple exits. But it will
soon.

llvm-svn: 73864
2009-06-22 00:15:15 +00:00
Dan Gohman 860379bcc2 Rename a variable for consistency with the ExitBlock vs ExitingBlock
terminology that LoopInfo uses.

llvm-svn: 73863
2009-06-21 23:48:38 +00:00
Dan Gohman 724f825f96 Fix a typo in a comment that Frits von Bommel noticed.
llvm-svn: 73796
2009-06-19 23:41:37 +00:00
Dan Gohman cc31110b95 Re-apply r73718, now that the fix in r73787 is in, and add a
hand-crafted testcase which demonstrates the bug that was exposed
in 254.gap.

llvm-svn: 73793
2009-06-19 23:23:27 +00:00
Dan Gohman 55e3dd9174 Fix LSR's OptimizeSMax to ignore max operators with more than 2 operands,
which it isn't prepared to handle.

llvm-svn: 73787
2009-06-19 23:03:46 +00:00
Evan Cheng 86076c9e30 Revert 73718. It's breaking 254.gap.
llvm-svn: 73783
2009-06-19 21:15:06 +00:00
Chris Lattner d0a363e03b make jump threading handle lexically identical compare instructions
as if they were multiple uses of the same instruction.  This interacts
well with the existing loadpre that j-t does to open up many new jump
threads earlier.

llvm-svn: 73768
2009-06-19 16:27:56 +00:00
Nick Lewycky 77585a24ac Teach jump threading to look at comparisons between phi nodes and non-constants.
llvm-svn: 73755
2009-06-19 04:56:29 +00:00
Chris Lattner 5ca4197829 Improve tail call elim to move loads above readonly calls
when it allows forming a tail call.  Patch by Frits van
Bommel.  This implements PR4323.

llvm-svn: 73752
2009-06-19 04:22:16 +00:00
Chris Lattner 87a222c5c8 part of PR4405: disable a contentious optimization for
strcmp -> memcmp when the lengths of the strings are unknown.

Patch by Nick Lewycky!

llvm-svn: 73751
2009-06-19 04:17:36 +00:00
Dan Gohman 8c9ac59455 Generalize LSR's OptimizeSMax to handle unsigned max tests as well
as signed max tests. Along with r73717, this helps CodeGen avoid
emitting code for a maximum operation for this class of loop.

llvm-svn: 73718
2009-06-18 20:23:18 +00:00
Anton Korobeynikov 6ee547bb1b Revert IRBuilder CC propagation. Fix SimplifyLibCalls instead.
llvm-svn: 73715
2009-06-18 20:05:31 +00:00
Dan Gohman a0348809b6 Remove the code from IVUsers that attempted to handle
casted induction variables in cases where the cast
isn't foldable. It ended up being a pessimization in
many cases. This could be fixed, but it would require
a bunch of complicated code in IVUsers' clients. The
advantages of this approach aren't visible enough to
justify it at this time.

llvm-svn: 73706
2009-06-18 16:54:06 +00:00
Dan Gohman 56bd02c55c Generalize the zext(trunc(t) & C) instcombine to work even with
C is not a low-bits mask, and add a similar instcombine for
zext((trunc(t) & C) ^ C).

llvm-svn: 73705
2009-06-18 16:30:21 +00:00
Dan Gohman 7f836c7c61 Instcombine zext(trunc(x) & mask) to x&mask, even if the trunc has
multiple users.

llvm-svn: 73656
2009-06-17 23:17:05 +00:00
Dale Johannesen 81b6463ed0 This fixes a bug introduced in 72661, which can
move loads back past a check that the load address
is valid, see new testcase.  The test that went
in with 72661 has exactly this case, except that
the conditional it's moving past is checking
something else; I've settled for changing that
test to reference a global, not a pointer.  It
may be possible to scan all the tests you pass and
make sure none of them are checking any component
of the address, but it's not trivial and I'm not
trying to do that here.

llvm-svn: 73632
2009-06-17 20:48:23 +00:00
Torok Edwin ba93ea7632 Add debug message about non-local loads being clobbered.
llvm-svn: 73625
2009-06-17 18:48:18 +00:00
Dan Gohman d8329e8378 Update comments to use doxygen syntax.
llvm-svn: 73621
2009-06-17 17:51:33 +00:00
Sanjiv Gupta 2f2b0a1985 >> What if my global variable was into a different address space than stack?
>>     
>
> It doesn't matter in terms of semantics: because AnalyzeGlobal
> returned false, we're guaranteed the address of the global is never
> taken.  I wouldn't be surprised if we end up generating invalid IR in
> some cases, though, because of the semantics of replaceAllUsesWith.
> Do you have a testcase that breaks?
>
>   
The problem is replaceAllUsesWith asserts for type mismatch here. Try attached .bc with llvm-ld.

assert(New->getType() == getType() &&
        "replaceAllUses of value with new value of different type!");

Since stack is always on address space zero, I don't think that type of GV in a different address space is ever going to match.
The other way is to allow replaceAllUsesWith to ignore address spaces while comparing types. (do we have  a way to do that ?).
But then such an optimization may fail the entire idea of user wanting to place a variable into different memory space. The original idea of user might be to save on the stack space (data memory) and hence he asked the variable to be placed into different memory space (program memory). So the best bet here is to deny this optimization by checking

GV->getType()->getAddressSpace() == 0. 

llvm-svn: 73605
2009-06-17 06:47:15 +00:00
Eli Friedman a0fba5319d PR3439: Correct a silly mistake in the SimplifyDemandedUseBits code for
SRem.

llvm-svn: 73598
2009-06-17 02:57:36 +00:00
Dan Gohman 0ed7756fbe Generalize a few more instcombines to be vector/scalar-independent.
llvm-svn: 73541
2009-06-16 19:55:29 +00:00
Chris Lattner 945d08d76f Generalize instcombine's isSafeToLoadUnconditionally() function
to ignore readonly calls, and factor it out of instcombine so
that it can be used by other passes.  Patch by Frits van Bommel!

llvm-svn: 73506
2009-06-16 17:23:12 +00:00