RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 12:12:47 +01:00
Code Issues Projects Releases Wiki Activity
37,002 Commits 4 Branches 5 Tags 704 MiB
de35f122f2
Commit Graph

21785 Commits

Author SHA1 Message Date
Owen Anderson
3efff61819 In addition to arguments passed to it, memcpy (and all other calls) can ModRef pointers that alias their arguments as well. This fixes PR2057. 
llvm-svn: 47317
 2008-02-19 06:47:18 +00:00
Owen Anderson
dbc264003e Major improvements to yesterday's return slot optimization. Remove some unneccessary constraints, 
and add some others that should have been in from the first place.  Document the whole thing better.

llvm-svn: 47315
 2008-02-19 06:35:43 +00:00
Chris Lattner
85482657eb Fix some minor issues folding undef, PR2052 
llvm-svn: 47314
 2008-02-19 06:22:12 +00:00
Chris Lattner
6bb889cf84 fdiv/frem of undef can produce undef, because the undef operand 
can be a SNaN.  We could be more aggressive and turn this into 
unreachable, but that is less nice, and not really worth it.

llvm-svn: 47313
 2008-02-19 06:12:18 +00:00
Chris Lattner
851ebb75de regenerate 
llvm-svn: 47312
 2008-02-19 04:36:25 +00:00
Chris Lattner
5400cc3907 Fix PR2060 by rejecting invalid types for integer constants. 
llvm-svn: 47311
 2008-02-19 04:36:07 +00:00
Owen Anderson
4e6f18d5bf Factor the profitability check for return slot optimization out into a static function. 
At some point in the future, this check will become smarter.

llvm-svn: 47310
 2008-02-19 03:27:34 +00:00
Owen Anderson
3782cd74d1 An sret parameter is required to be the first parameter, so there's no need to loop 
over all the parameters of the callee looking for it.

llvm-svn: 47309
 2008-02-19 03:15:29 +00:00
Owen Anderson
ea5cdf1a83 Cleanup some of my patches from yesterday. Refactor the check for which xform 
to apply to a memcpy into processInstruction.  Also, fix a bug in the check due to
missing braces.

llvm-svn: 47307
 2008-02-19 03:09:45 +00:00
Owen Anderson
5c258ed93d Fix Transforms/GVN/memcpy.ll, which Chris broke in r47275 by reordering the branches. memcpy's are a kind of CallInst. 
llvm-svn: 47305
 2008-02-19 02:53:23 +00:00
Evan Cheng
e18e5201fc PR1909: Tail merging pass ran wild. It makes no sense to merge blocks in order to save a single instruction since a branch will be inserted for each BB. 
llvm-svn: 47301
 2008-02-19 02:09:37 +00:00
Evan Cheng
ece0db124f Me not like duplicated comments. 
llvm-svn: 47300
 2008-02-19 02:05:16 +00:00
Evan Cheng
bb577266bf - When DAG combiner is folding a bit convert into a BUILD_VECTOR, it should check if it's essentially a SCALAR_TO_VECTOR. Avoid turning (v8i16) <10, u, u, u> to <10, 0, u, u, u, u, u, u>. Instead, simply convert it to a SCALAR_TO_VECTOR of the proper type. 
- X86 now normalize SCALAR_TO_VECTOR to (BIT_CONVERT (v4i32 SCALAR_TO_VECTOR)). Get rid of X86ISD::S2VEC.

llvm-svn: 47290
 2008-02-18 23:04:32 +00:00
Nick Lewycky
1f3c58df08 Correctly fold divide-by-constant, even when faced with overflow. 
llvm-svn: 47287
 2008-02-18 22:48:05 +00:00
Dan Gohman
0d0f16ca85 Chris pointed out that it's not necessary to set i64 MUL to Expand 
on x86-32 since i64 itself is not a Legal type. And, update some
comments.

llvm-svn: 47282
 2008-02-18 19:34:53 +00:00
Evan Cheng
03fd8ea753 - Remove the previous check which broke coalescer-commute3.ll 
- For now, conservatively ignore copy MI whose source is a physical register. Commuting its def MI can cause a physical register live interval to be live through a loop (since we know it's live coming into the def MI).

llvm-svn: 47281
 2008-02-18 18:56:31 +00:00
Chris Lattner
79ecc053ca upgrade some tests. 
llvm-svn: 47280
 2008-02-18 18:46:39 +00:00
Nate Begeman
66df9740df Add a note 
llvm-svn: 47279
 2008-02-18 18:39:23 +00:00
Chris Lattner
6ab061dd2d Add a note about sext from i1 plus flags use. 
llvm-svn: 47278
 2008-02-18 18:30:13 +00:00
Dan Gohman
70b9b2f77f Don't mark scalar integer multiplication as Expand on x86, since x86 
has plain one-result scalar integer multiplication instructions.
This avoids expanding such instructions into MUL_LOHI sequences that
must be special-cased at isel time, and avoids the problem with that
code that provented memory operands from being folded.

This fixes PR1874, addressesing the most common case. The uncommon
cases of optimizing multiply-high operations will require work
in DAGCombiner.

llvm-svn: 47277
 2008-02-18 17:55:26 +00:00
Chris Lattner
a378b03483 Transforming -A + -B --> -(A + B) isn't safe for FP, thanks 
to Dale for noticing this!

llvm-svn: 47276
 2008-02-18 17:50:16 +00:00
Chris Lattner
a3318d17d4 minor code simplification, no functionality change. 
llvm-svn: 47275
 2008-02-18 17:47:29 +00:00
Duncan Sands
a85448506d Simplify caller updating using a CallSite, as 
requested by Chris.  While there, do the same
for an existing function committed by someone
called "lattner" :)

llvm-svn: 47273
 2008-02-18 17:32:13 +00:00
Chris Lattner
57eff5b5f5 don't bother calling getUnderlyingObject for non-pointers. 
llvm-svn: 47272
 2008-02-18 17:28:21 +00:00
Owen Anderson
f3107c8546 Since we're not checking for the more general AllocationInst first, we need to explicitly check 
that Object is an Argument before casting it to one.

llvm-svn: 47268
 2008-02-18 10:11:00 +00:00
Roman Levenstein
a7d4ae1435 New helper function getMBBFromIndex() that given an index in any instruction of an MBB returns a pointer the MBB. Reviewed by Evan. 
llvm-svn: 47267
 2008-02-18 09:35:30 +00:00
Owen Anderson
7b092ea631 Add support to GVN for performing sret return slot optimization. This means that, if an sret function tail calls 
another sret function, it should pass its own sret parameter to the tail callee, allowing it to fill in the correct
return value.  llvm-gcc does not emit this by default.  Instead, it allocates space in the caller for the sret of
the tail call and then uses memcpy to copy the result into the caller's sret parameter.  This optimization detects
and optimizes that case.

llvm-svn: 47265
 2008-02-18 09:24:53 +00:00
Owen Anderson
e4e1706f95 I got the predicate backwards in my last patch. The comment is correct, the code was not. 
llvm-svn: 47264
 2008-02-18 09:22:21 +00:00
Owen Anderson
1b743c32fd This check is not correct for mallocs, so exclude them earlier. 
llvm-svn: 47263
 2008-02-18 09:11:02 +00:00
Evan Cheng
6eeae5bd62 For now, avoid commuting def MI for copy MI's whose source is not killed. That simply trade a live interval for another and because only the non-two-address operands can be folded into loads, may end up pessimising code. 
llvm-svn: 47262
 2008-02-18 08:40:53 +00:00
Chris Lattner
b22dbdf4fc switch simplifycfg from using vectors for most things to smallvectors, 
this speeds it up 2.3% on eon.

llvm-svn: 47261
 2008-02-18 07:42:56 +00:00
Chris Lattner
9851db050b optimize away stackrestore calls that have no intervening alloca or call. 
llvm-svn: 47258
 2008-02-18 06:12:38 +00:00
Owen Anderson
18a7d290cb Duncan pointed out that we can fast fail here, because the sret parameter of 
a function must be the first parameter.

llvm-svn: 47254
 2008-02-18 04:06:26 +00:00
Owen Anderson
b0891af18b Fix a comment, and a bug where we weren't applying the tail call logic in cases that failed the first test. 
llvm-svn: 47253
 2008-02-18 03:52:21 +00:00
Owen Anderson
0ef73a5d06 Fix bugs that Chris noticed in my last patch. 
llvm-svn: 47252
 2008-02-18 02:31:23 +00:00
Chris Lattner
0b6ec4a6b2 simplify some code, BreakUpSubtract always returns nonnull now. 
llvm-svn: 47251
 2008-02-18 02:18:25 +00:00
Chris Lattner
d5885d43ef bitcasts of pointers are always pointers. 
If we see a memcpy of a pointer, make sure to check later
uses of the pointer as well.

llvm-svn: 47250
 2008-02-18 02:11:28 +00:00
Owen Anderson
48dd6a0f96 Add a predicate to Argument to check for the StructRet attribute. 
llvm-svn: 47248
 2008-02-17 23:22:28 +00:00
Owen Anderson
884f6801df Teach getModRefInfo that memcpy, memmove, and memset don't "capture" memory addresses. 
Also, noalias arguments are be considered "like" stack allocated ones for this purpose, because
the only way they can be modref'ed is if they escape somewhere in the current function.

llvm-svn: 47247
 2008-02-17 21:29:08 +00:00
Chris Lattner
2fa904b3af Fold (-x + -y) -> -(x+y) which promotes better association, fixing 
the second half of PR2047

llvm-svn: 47244
 2008-02-17 21:03:36 +00:00
Chris Lattner
51a663a8fe fix pasto 
llvm-svn: 47242
 2008-02-17 20:54:40 +00:00
Chris Lattner
4a34461d64 Split up subtracts into add+negate if they have a reassociable use or operand 
that is also a subtract.  This implements PR2047 and Transforms/Reassociate/subtest2.ll 

llvm-svn: 47241
 2008-02-17 20:51:26 +00:00
Chris Lattner
cd6f9cea98 make the logic for breaking up subtracts more explicit, no 
functionality change.

llvm-svn: 47239
 2008-02-17 20:44:51 +00:00
Chris Lattner
3e886fa85a move PR2053 to here. 
llvm-svn: 47237
 2008-02-17 19:43:57 +00:00
Duncan Sands
cd0325cce1 Remove any 'nest' parameter attributes if the function 
is not passed as an argument to a trampoline intrinsic.

llvm-svn: 47220
 2008-02-16 20:56:04 +00:00
Duncan Sands
753597d1cf Some micro-optimizations. 
llvm-svn: 47219
 2008-02-16 20:53:06 +00:00
Andrew Lenharth
da54523742 I cannot find a libgcc function for this builtin. Therefor expanding it to a noop (which is how it use to be treated). If someone who knows the x86 backend better than me could tell me how to get a lock prefix on an instruction, that would be nice to complete x86 support. 
llvm-svn: 47213
 2008-02-16 14:46:26 +00:00
Duncan Sands
22ebf8b70f Teach LegalizeTypes how to expand the operands of 
br_cc.  This fixes 5 "make check" failures.

llvm-svn: 47212
 2008-02-16 10:29:26 +00:00
Evan Cheng
dd13a96a60 Refactor some code; check if commuteInstruction is able to commute the instruction. 
llvm-svn: 47208
 2008-02-16 02:32:17 +00:00
Andrew Lenharth
c178981b85 llvm.memory.barrier, and impl for x86 and alpha 
llvm-svn: 47204
 2008-02-16 01:24:58 +00:00
Bill Wendling
34dd48a10c Fix typos. 
llvm-svn: 47200
 2008-02-16 01:09:25 +00:00
Dan Gohman
a9386d8501 Rename CountMemOperands to ComputeMemOperandsEnd to reflect what 
it actually does. Simplify CountOperands a little by reusing
ComputeMemOperandsEnd. And reword some comments for both.

llvm-svn: 47198
 2008-02-16 00:36:48 +00:00
Dan Gohman
9ce252af04 Revert 47177, which was incorrect. 
llvm-svn: 47196
 2008-02-16 00:25:40 +00:00
Scott Michel
3ad0263173 Make tblgen a little smarter about constants smaller than i32. Currently, 
tblgen will complain if a sign-extended constant does not fit into a
data type smaller than i32, e.g., i16. This causes a problem when certain
hex constants are used, such as 0xff for byte masks or immediate xor
values.

tblgen will try the sign-extended value first and, if the sign extended
value would overflow, it tries to see if the unsigned value will fit.
Consequently, a software developer can now safely incant:

	(XORHIr16 R16C:$rA, 0xffff)

which is somewhat clearer and more informative than incanting:

	(XORHIr16 R16C:$rA, (i16 -1))

even if the two are bitwise equivalent.

Tblgen also outputs the 64-bit unsigned constant in the generated ISel code
when getTargetConstant() is invoked.

llvm-svn: 47188
 2008-02-15 23:05:48 +00:00
Evan Cheng
09fee0e727 The copy instruction being coalesced will be removed, it is not a kill. 
llvm-svn: 47179
 2008-02-15 21:36:51 +00:00
Ted Kremenek
83e9c3ba5c Fixed bug in FoldingSetIteratorImpl where we did not correctly check if 
we had reached the "fake bucket" after the last bucket, allowing the iterator
in some cases to run off the end of the hashtable.

llvm-svn: 47178
 2008-02-15 21:12:46 +00:00
Dan Gohman
547cd33955 Skip over the defs and start at the uses when looking for operands 
with the TIED_TO attribute.

llvm-svn: 47177
 2008-02-15 20:59:17 +00:00
Dan Gohman
06ac7aca3e Use the TargetInstrDescr to determine the number of operands 
that should be checked for the TIED_TO attribute instead of
using CountOperands.

llvm-svn: 47176
 2008-02-15 20:50:13 +00:00
Duncan Sands
bf7d428157 Teach LegalizeTypes how to promote the flags 
in a ret node.  These are created as i32 constants
but on some platforms i32 is not legal.  This
fixes 26 "make check" failures, for example
Alpha/2005-07-12-TwoMallocCalls.ll.

llvm-svn: 47172
 2008-02-15 19:34:17 +00:00
Chris Lattner
d55c26a77d Handle \n's in value names for more targets. The asm printers 
really really really need refactoring :(

llvm-svn: 47171
 2008-02-15 19:04:54 +00:00
Chris Lattner
cf98f7291b If the llvm name contains an unprintable character, don't print it in 
the global comment.  This prevents printing things like:

...  # foo
bar

when the name is "foo\nbar".

llvm-svn: 47170
 2008-02-15 18:56:05 +00:00
Chris Lattner
bab03a1b50 targets that support quotes for mangled names still need to escape newlines 
when they occur in the name, just like " is escaped.

llvm-svn: 47169
 2008-02-15 18:54:56 +00:00
Dale Johannesen
b9e1a37497 Cosmetics. 
llvm-svn: 47168
 2008-02-15 18:40:53 +00:00
Evan Cheng
bb357cd8a6 - Removing the infamous r2rMap_ and rep() method. Now the coalescer will update 
register defs and uses after each successful coalescing.
- Also removed a number of hacks and fixed some subtle kill information bugs.

llvm-svn: 47167
 2008-02-15 18:24:29 +00:00
Evan Cheng
2cb3fd8f72 Added CommuteChangesDestination(). This returns true if commuting the specified 
machine instr will change its definition register.

llvm-svn: 47166
 2008-02-15 18:21:33 +00:00
Evan Cheng
fff5254adf Remove unnecessary #include. 
llvm-svn: 47164
 2008-02-15 18:12:09 +00:00
Dan Gohman
8ff2d5e0e3 Use StoreSDNode::getValue instead of calling getOperand directly 
with a hard-coded operand number.

llvm-svn: 47163
 2008-02-15 18:11:59 +00:00
Dale Johannesen
de53aaec39 Remove warning about 64-bit code on processor 
that doesn't support it.  Per Chris.

llvm-svn: 47162
 2008-02-15 18:09:51 +00:00
Lauro Ramos Venancio
4fcf9513f1 It is not safe to call fork in PrintStackTrace. Sometimes it freezes the program. 
llvm-svn: 47161
 2008-02-15 18:05:54 +00:00
Devang Patel
01f9252f73 Fix PR2028 
llvm-svn: 47150
 2008-02-15 01:24:49 +00:00
Dale Johannesen
da9de4b3f0 nocona, core2 and penryn support 64 bit. 
llvm-svn: 47149
 2008-02-15 01:22:41 +00:00
Chris Lattner
3bac9de032 Support vector constant zeros, thanks to Zack Rusin for the testcase. 
llvm-svn: 47148
 2008-02-15 00:57:28 +00:00
Dale Johannesen
6cb8a628a2 Rewrite tblgen handling of subtarget features so 
it follows the order of the enum, not alphabetical.
The motivation is to make -mattr=+ssse3,+sse41
select SSE41 as it ought to.  Added "ignored"
enum values of 0 to PPC and SPU to avoid compiler
warnings.

llvm-svn: 47143
 2008-02-14 23:35:16 +00:00
Devang Patel
8af92b942c If loop header is also loop exiting block then OrigPN is incoming value for B loop header. 
Fixes PR 2030.

llvm-svn: 47141
 2008-02-14 23:18:47 +00:00
Dan Gohman
9ca30afc39 Fix a warning about comparison between signed and unsigned, 
being consistent with the rest of the APInt implementation.

llvm-svn: 47138
 2008-02-14 22:38:45 +00:00
Chris Lattner
b8bf200b77 Fix PR2029 
llvm-svn: 47129
 2008-02-14 19:18:13 +00:00
Chris Lattner
9c24f3ec37 Fix a miscompilation from Dan's recent apintification. 
llvm-svn: 47128
 2008-02-14 18:48:56 +00:00
Nate Begeman
9deedb0114 Fix single precision FP constants on SPU. They are actually legal, 
which allows us to kill a target-specific node.

llvm-svn: 47127
 2008-02-14 18:43:04 +00:00
Duncan Sands
0056f1e823 In TargetLowering::LowerCallTo, don't assert that 
the return value is zero-extended if it isn't
sign-extended.  It may also be any-extended.
Also, if a floating point value was returned
in a larger floating point type, pass 1 as the
second operand to FP_ROUND, which tells it
that all the precision is in the original type.
I think this is right but I could be wrong.
Finally, when doing libcalls, set isZExt on
a parameter if it is "unsigned".  Currently
isSExt is set when signed, and nothing is
set otherwise.  This should be right for all
calls to standard library routines.

llvm-svn: 47122
 2008-02-14 17:28:50 +00:00
Nate Begeman
1ef1013b6c Change how FP immediates are handled. 
1) ConstantFP is now expand by default
2) ConstantFP is not turned into TargetConstantFP during Legalize
   if it is legal.

This allows ConstantFP to be handled like Constant, allowing for 
targets that can encode FP immediates as MachineOperands.

As a bonus, fix up Itanium FP constants, which now correctly match,
and match more constants!  Hooray.

llvm-svn: 47121
 2008-02-14 08:57:00 +00:00
Nate Begeman
8352abdd98 Support a new type of MachineOperand, MO_FPImmediate, used for holding 
FP Immediates, crazily enough

llvm-svn: 47117
 2008-02-14 07:39:30 +00:00
Chris Lattner
5005b4d0d8 simplify code, no functionality change. 
llvm-svn: 47116
 2008-02-14 07:39:01 +00:00
Nate Begeman
5d61361bb9 Move some useful operands up into the all-targets .td 
llvm-svn: 47115
 2008-02-14 07:25:46 +00:00
Nick Lewycky
0dd6ce5d3a Fix PR2032. Inform the alias analysis of changes to the underlying program. 
llvm-svn: 47111
 2008-02-14 07:11:24 +00:00
Chris Lattner
b714906acf upgrade some entries, remove stuff that is done. 
llvm-svn: 47109
 2008-02-14 06:19:02 +00:00
Chris Lattner
037aa64987 the mid-level optimizer removes this stuff. 
llvm-svn: 47108
 2008-02-14 05:43:18 +00:00
Chris Lattner
10dc770a36 this one is easy. 
llvm-svn: 47107
 2008-02-14 05:41:38 +00:00
Chris Lattner
d696c25db5 This readme entry is done, testcase here: CodeGen/X86/zero-remat.ll 
llvm-svn: 47106
 2008-02-14 05:39:46 +00:00
Dan Gohman
83374c9be5 Allow the APInt form of ComputeMaskedBits to operate on i128 types. 
llvm-svn: 47101
 2008-02-13 23:13:32 +00:00
Dan Gohman
737856bd0d Assigning an APInt to 0 with plain assignment gives it a one-bit 
size. Initialize these APInts to properly-sized zero values.

llvm-svn: 47099
 2008-02-13 23:07:24 +00:00
Dan Gohman
a5ad7ff23b Avoid setting bits that aren't demanded. 
llvm-svn: 47098
 2008-02-13 22:43:25 +00:00
Dan Gohman
99b38405e3 Simplify some logic in ComputeMaskedBits. And change ComputeMaskedBits 
to pass the mask APInt by value, not by reference. 

llvm-svn: 47096
 2008-02-13 22:28:48 +00:00
Devang Patel
38de2d11a5 A loop latch phi node may have uses inside loop, not just in loop header. 
llvm-svn: 47093
 2008-02-13 22:23:07 +00:00
Dan Gohman
588498082a Rename APInt's isPositive to isNonNegative, to reflect what it 
actually does.

llvm-svn: 47090
 2008-02-13 22:09:18 +00:00
Devang Patel
1e71afe2df While moving exit condition, do not drop loop latch on the floor. 
llvm-svn: 47089
 2008-02-13 22:06:36 +00:00
Dan Gohman
d22101a970 Add countTrailingOnes member functions to APInt. 
llvm-svn: 47086
 2008-02-13 21:11:05 +00:00
Devang Patel
d48bbbf07b Keep track of exit value operand number when operands are swapped. 
llvm-svn: 47082
 2008-02-13 19:48:48 +00:00
Nicolas Geoffray
72fa78e195 Enable exception handling int JIT 
llvm-svn: 47079
 2008-02-13 18:39:37 +00:00
Duncan Sands
2e9661573f Teach LegalizeTypes how to expand and promote CTLZ, 
CTTZ and CTPOP.  The expansion code differs from
that in LegalizeDAG in that it chooses to take the
CTLZ/CTTZ count from the Hi/Lo part depending on
whether the Hi/Lo value is zero, not on whether
CTLZ/CTTZ of Hi/Lo returned 32 (or whatever the
width of the type is) for it.  I made this change
because the optimizers may well know that Hi/Lo
is zero and exploit it.  The promotion code for
CTTZ also differs from that in LegalizeDAG: it
uses an "or" to get the right result when the
original value is zero, rather than using a compare
and select.  This also means the value doesn't
need to be zero extended.

llvm-svn: 47075
 2008-02-13 18:01:53 +00:00
Chris Lattner
80b3a56774 Fix the PPC JIT regressions by encoding zeroreg as 0 for BLR. 
llvm-svn: 47067
 2008-02-13 17:24:14 +00:00