Commit Graph

26772 Commits

Author SHA1 Message Date
Chris Lattner
e7bf6037e2 silence a warning when assertions are disabled.
llvm-svn: 62976
2009-01-25 23:08:00 +00:00
Torok Edwin
6f715ebe85 should have removed the + when manually applying a patch!
llvm-svn: 62973
2009-01-25 20:29:34 +00:00
Torok Edwin
3f54410405 revert this patch for now, because Codegen does still want to generate SSE code,
for example in the case of va-args. XFAIL associated tests.

llvm-svn: 62972
2009-01-25 20:21:24 +00:00
Torok Edwin
49b1d3e3cc If user explicitly asks not to use SSE, don't force it. This fixes LLVM part of PR3402.
llvm-svn: 62967
2009-01-25 17:58:56 +00:00
Dan Gohman
15770b411b Eliminate the loop that searches through each of the operands
of each use in the SelectionDAG ReplaceAllUses* functions. Thanks
to Chris for spotting this opportunity.

Also, factor out code from all 5 of the ReplaceAllUses* functions
into AddNonLeafNodeToCSEMaps, which is now renamed
AddModifiedNodeToCSEMaps to more accurately reflect its purpose.

llvm-svn: 62964
2009-01-25 16:29:12 +00:00
Dan Gohman
1302160ad4 Whitespace tidiments.
llvm-svn: 62963
2009-01-25 16:21:38 +00:00
Dan Gohman
1de7d569cf Move the N->use_empty() assert from DeleteNode to
DeleteNodeNotInCSEMaps, since DeleteNode just calls
DeleteNodeNotInCSEMaps.

llvm-svn: 62962
2009-01-25 16:20:37 +00:00
Nick Lewycky
4b9d56801b Start generating arbitrary precision integer SCEVs. This removes the temporary
code that rounded up and capped the size.

llvm-svn: 62958
2009-01-25 08:16:27 +00:00
Nick Lewycky
459667b48d The function that does nothing but call malloc is noalias return.
llvm-svn: 62956
2009-01-25 07:59:57 +00:00
Evan Cheng
71ca3e2bdb Private linkage support for PPC / Darwin.
llvm-svn: 62955
2009-01-25 06:32:01 +00:00
Evan Cheng
4ebe9b79fa Teach 2addr pass to be do more commuting. If both uses of a two-address instruction are killed, but the first operand has a use before and after the def, commute if the second operand does not suffer from the same issue.
%reg1028<def> = EXTRACT_SUBREG %reg1027<kill>, 1                                                                                                                                     
%reg1029<def> = MOV8rr %reg1028                                                                                                                                                      
%reg1029<def> = SHR8ri %reg1029, 7, %EFLAGS<imp-def,dead>                                                                                                                            
insert => %reg1030<def> = MOV8rr %reg1028                                                                                                                                            
%reg1030<def> = ADD8rr %reg1028<kill>, %reg1029<kill>, %EFLAGS<imp-def,dead>                                                                                                         

In this case, it might not be possible to coalesce the second MOV8rr                                                                                                                 
instruction if the first one is coalesced. So it would be profitable to                                                                                                              
commute it:                                                                                                                                                                          
%reg1028<def> = EXTRACT_SUBREG %reg1027<kill>, 1                                                                                                                                     
%reg1029<def> = MOV8rr %reg1028                                                                                                                                                      
%reg1029<def> = SHR8ri %reg1029, 7, %EFLAGS<imp-def,dead>                                                                                                                            
insert => %reg1030<def> = MOV8rr %reg1029                                                                                                                                            
%reg1030<def> = ADD8rr %reg1029<kill>, %reg1028<kill>, %EFLAGS<imp-def,dead>

llvm-svn: 62954
2009-01-25 03:53:59 +00:00
Nate Begeman
48f3fe9199 Fix an indent and a typo.
llvm-svn: 62940
2009-01-24 22:12:48 +00:00
Dale Johannesen
b818fa1bd3 Revert previous change; even this mild and clearly
more accurate change loses more than it gains on
benchmarks.

llvm-svn: 62938
2009-01-24 21:49:34 +00:00
Torok Edwin
6dd79be128 add note about possible GEP improvement with fields of size 0.
llvm-svn: 62925
2009-01-24 19:30:25 +00:00
Torok Edwin
2a7e7066b3 testcase for PR3381.
Also it was an empty struct, not a void after all.

llvm-svn: 62920
2009-01-24 17:16:04 +00:00
Torok Edwin
726354d4ce void* is represented as pointer to empty struct {}.
Thus we need to check whether the struct is empty before trying to index into
it. This fixes PR3381.

llvm-svn: 62918
2009-01-24 11:30:49 +00:00
Owen Anderson
78b6a961bb Some cleanups. No functional changes.
llvm-svn: 62917
2009-01-24 10:07:43 +00:00
Dale Johannesen
1867040e00 Improve the inlining cost function a bit.
Little practical effect.

llvm-svn: 62908
2009-01-24 01:27:33 +00:00
Chris Lattner
d386e82ec9 Make InstCombineStoreToCast handle aggregates more aggressively,
handling the case in Transforms/InstCombine/cast-store-gep.ll, which
is a heavily reduced testcase from Clang on x86-64.

llvm-svn: 62904
2009-01-24 01:00:13 +00:00
Evan Cheng
93b798e690 Refactor code. No functionality change.
llvm-svn: 62893
2009-01-23 23:27:33 +00:00
Devang Patel
d980abaae2 Introduce two DWARF attribute extentions DW_AT_APPLE_optimized, DW_AT_APPLE_flags.
DW_AT_APPLE_optimized flag is set when a compile_unit is optimized. The debugger takes advantage of this information some way.

DW_AT_APPLE_flags encodes command line options when certain env. variable is set. This is used by build engineers to track various gcc command lines used by by  a project, irrespective of whether the project used makefile, Xcode or something else.

llvm-gcc patch is next.

llvm-svn: 62888
2009-01-23 22:33:47 +00:00
Chris Lattner
97b6f6a674 hopefully address PR3379 by making the P modifier work in x86 inline asm.
llvm-svn: 62887
2009-01-23 22:33:40 +00:00
Gabor Greif
2d7bf2a76c use CallSite::isCalle instead of slow getOperandNo
llvm-svn: 62877
2009-01-23 21:17:04 +00:00
Gabor Greif
fbd40904c1 Simplify the logic of getting hold of a PHI predecessor block.
There is now a direct way from value-use-iterator to incoming block in PHINode's API.
This way we avoid the iterator->index->iterator trip, and especially the costly
getOperandNo() invocation. Additionally there is now an assertion that the iterator
really refers to one of the PHI's Uses.

llvm-svn: 62869
2009-01-23 19:40:15 +00:00
Nate Begeman
4de6ffb743 Add support for deleting a module provider from a JIT in such a way that it does not cause the owned module to be fully materialized.
llvm-svn: 62864
2009-01-23 19:27:28 +00:00
Devang Patel
37b2017172 Empty DIType represents void. In this case no need to construct any type DIE.
llvm-svn: 62861
2009-01-23 19:13:31 +00:00
Dan Gohman
c971f3cf5b Fold x-0 to x in unsafe-fp-math mode. This comes up in the
testcase from PR3376, and in fact is sufficient to completely
avoid the problem in that testcase.

There's an underlying problem though; TLI.isOperationLegal
considers Custom to be Legal, which might be ok in some
cases, but that's what DAGCombiner is using in many places
to test if something is legal when LegalOperations is true.
When DAGCombiner is running after legalize, this isn't
sufficient. I'll address this in a separate commit.

llvm-svn: 62860
2009-01-23 19:10:37 +00:00
Chris Lattner
b36503c31b fix two more cases where we could let the NLPDI cache get unsorted.
With this, sqlite3 now passes.

llvm-svn: 62839
2009-01-23 07:12:16 +00:00
Chris Lattner
1dc435cdc7 Unconditionally reset 'cache' to zero, even if we don't need to resort it.
This avoids using a dangling pointer.

Reset NumSortedEntries after restoring Cache to avoid extraneous sorts.
This fixes the reduced sqlite3 testcase, but apparently not the whole app.

llvm-svn: 62838
2009-01-23 06:48:41 +00:00
Evan Cheng
8ef6b2d75e Only check if coalescing is worthwhile when the result is targeting a more restrictive register class.
llvm-svn: 62837
2009-01-23 05:48:59 +00:00
Owen Anderson
8c41157459 Stage two of fixing pre-alloc-splitting's code size issues: filter out restores that are just
going to be re-spilled again.

This also helps performance.  Pre-alloc-splitting now seems to be an overall win on SPEC.

llvm-svn: 62834
2009-01-23 03:28:53 +00:00
Evan Cheng
7cdc6e4f8c Cross register class coalescing. Not yet enabled.
llvm-svn: 62832
2009-01-23 02:15:19 +00:00
Devang Patel
86e54f5636 Code did not follow associated comment. not a good idea.
llvm-svn: 62828
2009-01-23 01:25:58 +00:00
Devang Patel
b638679d9d Set appropriate tag for the composite type.
llvm-svn: 62827
2009-01-23 01:19:09 +00:00
Chris Lattner
f2adaaa0fc a minor tweak to my previous patch, handle the invalidation case
when there are multiple iterations of the loop.  This fixes PR3375.

llvm-svn: 62822
2009-01-23 00:27:03 +00:00
Owen Anderson
b9426381f3 Add an initial pass at dead spill/restore removal for pre alloc splitting.
llvm-svn: 62821
2009-01-23 00:23:32 +00:00
Chris Lattner
c843f0f484 do not sign extend characters input to isprint. This improves
compatibility with VC++.  Patch by Max Burke!

llvm-svn: 62813
2009-01-22 23:38:45 +00:00
Bob Wilson
d79e684d1e Fix a minor bug in DAGCombiner's folding of SELECT. Folding "select C, 0, 1"
to "C ^ 1" is only valid when C is known to be either 0 or 1.  Most of the
similar foldings in this function only handle "i1" types, but this one appears
intentionally written to handle larger integer types.  If C has an integer
type larger than "i1", this needs to check if the high bits of a boolean
are known to be zero.  I also changed the comment to describe this folding as
"C ^ 1" instead of "~C", since that is what the code does and since the latter
would only be valid for "i1" types.  The good news is that most LLVM targets
use TargetLowering::ZeroOrOneBooleanContent so this change will not disable
the optimization; the bad news is that I've been unable to come up with a
testcase to demonstrate the problem.

I have also removed a "FIXME" comment for folding "select C, X, 0" to "C & X",
since the code looks correct to me.  It could be made more aggressive by not
limiting the type to "i1", but that would then require checking for
TargetLowering::ZeroOrNegativeOneBooleanContent.  Similar changes could be
done for the other SELECT foldings, but it was decided to be not worth the
trouble and complexity (see e.g., r44663).

llvm-svn: 62790
2009-01-22 22:05:48 +00:00
Dan Gohman
a6e5948fce Don't create ISD::FNEG nodes after legalize if they aren't legal.
Simplify x+0 to x in unsafe-fp-math mode. This avoids a bunch of
redundant work in many cases, because in unsafe-fp-math mode,
ISD::FADD with a constant is considered free to negate, so the
DAGCombiner often negates x+0 to -0-x thinking it's free, when
in reality the end result is -x, which is more expensive than x.

Also, combine x*0 to 0.

This fixes PR3374.

llvm-svn: 62789
2009-01-22 21:58:43 +00:00
Gabor Greif
d56b0a8c03 introduce a useful abstraction to find out if a Use is in the call position of an instruction
llvm-svn: 62788
2009-01-22 21:35:57 +00:00
Tanya Lattner
33f3303401 Revert r62553 and r62616 due to issues with portability.
llvm-svn: 62777
2009-01-22 20:09:20 +00:00
Chris Lattner
da545d3918 rename methods in System/Host to be more consistent.
llvm-svn: 62776
2009-01-22 19:53:00 +00:00
Bob Wilson
186046e657 Add SelectionDAG::getNOT method to construct bitwise NOT operations,
corresponding to the "not" and "vnot" PatFrags.  Use the new method
in some places where it seems appropriate.

llvm-svn: 62768
2009-01-22 17:39:32 +00:00
Sanjiv Gupta
7173c05b38 Few targets like the tiny little PIC16 have only 16-bit pointers.
llvm-svn: 62763
2009-01-22 10:14:21 +00:00
Evan Cheng
c971801ae1 Eliminate a couple of fields from TargetRegisterClass: SubRegClasses and SuperRegClasses. These are not necessary. Also eliminate getSubRegisterRegClass and getSuperRegisterRegClass. These are slow and their results can change if register file names change. Just use TargetLowering::getRegClassFor() to get the right TargetRegisterClass instead.
llvm-svn: 62762
2009-01-22 09:10:11 +00:00
Chris Lattner
92f592ad8b fix a typo
llvm-svn: 62761
2009-01-22 07:21:55 +00:00
Chris Lattner
fcf56e7fbe add a note
llvm-svn: 62760
2009-01-22 07:16:03 +00:00
Chris Lattner
ddc8e78d54 Fix PR3358, a really nasty bug where recursive phi translated
analyses could be run without the caches properly sorted.  This
can fix all sorts of weirdness.  Many thanks to Bill for coming
up with the 'issorted' verification idea.

llvm-svn: 62757
2009-01-22 07:04:01 +00:00
Dan Gohman
29b575c6cd Recognize inline asm for bswap on x86-64 GLIBC. This allows it
to be supported in the JIT.

llvm-svn: 62730
2009-01-21 23:40:54 +00:00
Dale Johannesen
a5699a1e8b Do not use host floating point types when emitting
ASCII IR; loading and storing these can change the
bits of NaNs on some hosts.  Remove or add warnings
at a few other places using host floating point;
this is a bad thing to do in general.

llvm-svn: 62712
2009-01-21 20:32:55 +00:00