RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-26 14:33:02 +02:00
Code Issues Projects Releases Wiki Activity
34,948 Commits 4 Branches 5 Tags 704 MiB
70abd7943f
Commit Graph

4145 Commits

Author SHA1 Message Date
Evan Cheng
1f3724ee2d Recover most of the compile time regression due to recent live interval changes. 
1. Eliminate the costly live interval "swapping".
2. Change ValueNumberInfo container from SmallVector to std::vector. The former
   performs slowly when the vector size is very large.

llvm-svn: 41536
 2007-08-28 08:28:51 +00:00
Dan Gohman
123b8effaa Make DAGCombiner's global alias analysis query more precise in the case 
where both pointers have non-zero offsets.

llvm-svn: 41491
 2007-08-27 16:32:11 +00:00
Dan Gohman
2e7e251f24 If the source and destination pointers in an llvm.memmove are known 
to not alias each other, it can be translated as an llvm.memcpy.

llvm-svn: 41489
 2007-08-27 16:26:13 +00:00
Duncan Sands
883740b39f There is an impedance matching problem between LLVM and 
gcc exception handling: if an exception unwinds through
an invoke, then execution must branch to the invoke's
unwind target.  We previously tried to enforce this by
appending a cleanup action to every selector, however
this does not always work correctly due to an optimization
in the C++ unwinding runtime: if only cleanups would be
run while unwinding an exception, then the program just
terminates without actually executing the cleanups, as
invoke semantics would require.  I was hoping this
wouldn't be a problem, but in fact it turns out to be the
cause of all the remaining failures in the LLVM testsuite
(these also fail with -enable-correct-eh-support, so turning
on -enable-eh didn't make things worse!).  Instead we need
to append a full-blown catch-all to the end of each
selector.  The correct way of doing this depends on the
personality function, i.e. it is language dependent, so
can only be done by gcc.  Thus this patch which generalizes
the eh.selector intrinsic so that it can handle all possible
kinds of action table entries (before it didn't accomodate
cleanups): now 0 indicates a cleanup, and filters have to be
specified using the number of type infos plus one rather than
the number of type infos.  Related gcc patches will cause
Ada to pass a cleanup (0) to force the selector to always
fire, while C++ will use a C++ catch-all (null).

llvm-svn: 41484
 2007-08-27 15:47:50 +00:00
Dan Gohman
e2c92effea Add explicit keywords and remove spurious trailing semicolons. 
llvm-svn: 41482
 2007-08-27 14:50:10 +00:00
Dale Johannesen
2ceade197b Revise per review comments. 
llvm-svn: 41409
 2007-08-26 01:18:27 +00:00
Dale Johannesen
b52093236e Add APFloat interface to ConstantFPSDNode. Change 
over uses in DAGCombiner.  Fix interfaces to work
with APFloats.

llvm-svn: 41407
 2007-08-25 22:10:57 +00:00
Chris Lattner
093144e147 Allow target constants to be illegal types. The target should 
know how to handle them.  This fixes 
test/CodeGen/Generic/asm-large-immediate.ll

llvm-svn: 41388
 2007-08-25 01:00:22 +00:00
Chris Lattner
666f15ac0a Teach the dag scheduler to handle inline asm nodes with multi-value immediate operands. 
llvm-svn: 41386
 2007-08-25 00:53:07 +00:00
Bill Wendling
da2e7749bb The personality function might need to be declared as: 
.set  Lset0,___gxx_personality_v0-.
    .long Lset0

on some targets. Make it so!

llvm-svn: 41385
 2007-08-25 00:51:55 +00:00
Chris Lattner
1e089aac3a rename isOperandValidForConstraint to LowerAsmOperandForConstraint, 
changing the interface to allow for future changes.

llvm-svn: 41384
 2007-08-25 00:47:38 +00:00
Dale Johannesen
9ec1e28e79 Poison APFloat::operator==. Replace existing uses with bitwiseIsEqual. 
This means backing out the preceding change to Constants.cpp, alas.

llvm-svn: 41378
 2007-08-24 22:09:56 +00:00
Dale Johannesen
5c0b0ec1d6 Use APFloat internally for ConstantFPSDNode. 
llvm-svn: 41372
 2007-08-24 20:59:15 +00:00
Anton Korobeynikov
9451c871c5 Perform correct codegen for eh_dwarf_cfa intrinsic. 
llvm-svn: 41316
 2007-08-23 07:21:06 +00:00
Andrew Lenharth
3902197dc3 move this check. ppc outputs .no_dead_strip properly 
llvm-svn: 41286
 2007-08-22 19:33:11 +00:00
Dan Gohman
23db01efea Minor cleanups to reduce some spurious differences between different 
scheduler implementations.

llvm-svn: 41191
 2007-08-20 19:28:38 +00:00
Rafael Espindola
68d95ff2b1 Partial implementation of calling functions with byval arguments: 
*) The needed information is propagated to the DAG
 *) The X86-64 backend detects it and aborts

llvm-svn: 41179
 2007-08-20 15:18:24 +00:00
Evan Cheng
a65c956119 Fold C ? 0 : 1 to ~C or zext(~C) or trunc(~C) depending the types. 
llvm-svn: 41163
 2007-08-18 05:57:05 +00:00
Evan Cheng
930acafa5b Avoid issue on 64-bit hosts. 
llvm-svn: 41143
 2007-08-17 18:02:22 +00:00
David Greene
1d1b7f793d Fix GLIBCXX_DEBUG error of comparing two singular iterators 
llvm-svn: 41139
 2007-08-17 15:13:55 +00:00
Evan Cheng
11fa8d0f62 If dynamic_stackalloc alignment is > stack alignment, first issue an instruction to align the stack ptr before the decrement. 
llvm-svn: 41133
 2007-08-16 23:50:06 +00:00
Evan Cheng
9a05381a81 - If a dynamic_stackalloc alignment requirement is <= stack alignment, then the alignment argument is ignored. 
- *Always* round up the size of the allocation to multiples of stack
alignment to ensure the stack ptr is never left in an invalid state after a dynamic_stackalloc.

llvm-svn: 41132
 2007-08-16 23:46:29 +00:00
Evan Cheng
ee277f940b Fix some kill info update bugs; add hidden option -disable-rematerialization to turn off remat for debugging. 
llvm-svn: 41118
 2007-08-16 07:24:22 +00:00
Lauro Ramos Venancio
9f9e5b3971 Implement FPOWI ExpandOp. 
Fix PR1287.

llvm-svn: 41112
 2007-08-15 22:13:27 +00:00
Evan Cheng
f758fa5c35 If the source of a move is in spill slot, the reload may be folded to essentially a load from stack slot. It's ok to mark the stack slot value as available for reuse. But it should not be clobbered since the destination of the move is live. 
llvm-svn: 41109
 2007-08-15 20:20:34 +00:00
Evan Cheng
70879e8dae - If a def is dead, do not spill it. 
- If the defs of a spilled rematerializable MI are dead after the spill store is deleted, delete
  the def MI as well.

llvm-svn: 41086
 2007-08-14 23:25:37 +00:00
Evan Cheng
f901915207 Fix for PR1596: AdjustCopiesBackFrom() should conservatively check if any of its sub-registers may overlap with the interval of the copy that's being coalesced. 
llvm-svn: 41084
 2007-08-14 23:19:28 +00:00
Evan Cheng
0fbe2a0ec4 If a MI's def is remat as well as spilled, and the store is later deemed dead, mark the def operand as isDead. 
llvm-svn: 41083
 2007-08-14 20:23:13 +00:00
Evan Cheng
de24ad8897 If a spilled value is being reused and the use is a kill, that means there are 
no more uses within the MBB and the spilled value isn't live out of the MBB.
Then it's safe to delete the spill store.

llvm-svn: 41069
 2007-08-14 09:11:18 +00:00
Evan Cheng
886c1fe427 If a rematerializable def is not deleted, i.e. it is also spilled, check if the 
spilled value is available for reuse.

llvm-svn: 41067
 2007-08-14 05:42:54 +00:00
Evan Cheng
5b64e19e06 Kill info update bugs. 
llvm-svn: 41064
 2007-08-14 01:56:58 +00:00
Evan Cheng
215f802b75 Re-implement trivial rematerialization. This allows def MIs whose live intervals that are coalesced to be rematerialized. 
llvm-svn: 41060
 2007-08-13 23:45:17 +00:00
Evan Cheng
54b799dc76 Kill info update bugs. 
llvm-svn: 41043
 2007-08-13 07:12:23 +00:00
Evan Cheng
3b211635e8 No need to remove dead range from soon-to-be-dead live interval. Its val# may be out of whack. 
llvm-svn: 41024
 2007-08-12 01:26:19 +00:00
Evan Cheng
8968affb6d Code to maintain kill information during register coalescing. 
llvm-svn: 41016
 2007-08-11 00:59:19 +00:00
Christopher Lamb
e0c9bd8d2e Move isSubRegOf into MRegisterInfo. Fix a missed move elimination in LowerSubregs and add more debugging output there. 
llvm-svn: 41005
 2007-08-10 21:11:55 +00:00
Dan Gohman
f18e94535f Fix EXTRACT_ELEMENT, EXTRACT_SUBVECTOR, and EXTRACT_VECTOR_ELT to 
use an intptr ValueType instead of i32 for the index operand in
getCopyToParts.

llvm-svn: 40987
 2007-08-10 14:59:38 +00:00
Rafael Espindola
b20b9e985a propagate struct size and alignment of byval arguments to the DAG 
llvm-svn: 40986
 2007-08-10 14:44:42 +00:00
Chris Lattner
202e7f84e9 unbreak the build 
llvm-svn: 40976
 2007-08-09 23:55:17 +00:00
Evan Cheng
0293fe7483 Bug fix. ~1U marks the val# dead. 
llvm-svn: 40975
 2007-08-09 23:14:39 +00:00
Dale Johannesen
a1340d5916 Update per review comments. 
llvm-svn: 40965
 2007-08-09 17:27:48 +00:00
Dale Johannesen
79551baaad long double 9 of N. This finishes up the X86-32 bits 
(constants are still not handled).  Adds ConvertActions
to control fp-to-fp conversions (these are currently
defaulted for all other targets, so no changes there).

llvm-svn: 40958
 2007-08-09 01:04:01 +00:00
Scott Michel
86052f49c0 If a target really needs to custom lower constants, it should be allowed 
to do so.

llvm-svn: 40955
 2007-08-08 23:23:31 +00:00
Evan Cheng
5227e36428 Adding kill info to val#. 
llvm-svn: 40925
 2007-08-08 07:03:29 +00:00
Evan Cheng
02e7fbd7c8 Clean up and bug fix. 
llvm-svn: 40921
 2007-08-08 05:56:18 +00:00
Evan Cheng
12d72cc1a0 - Each val# can have multiple kills. 
- Fix some minor bugs related to special markers on val# def. ~0U means
  undefined, ~1U means dead val#.

llvm-svn: 40916
 2007-08-08 03:00:28 +00:00
Evan Cheng
b4135e3825 Remove a dead assertion. 
llvm-svn: 40914
 2007-08-08 01:00:21 +00:00
Evan Cheng
0182b495fd - LiveInterval value#'s now have 3 components: def instruction #, 
kill instruction #, and source register number (iff the value# is defined by a
copy).
- Now def instruction # is set for every value#, not just for copy defined ones.
- Update some outdated code related inactive live ranges.
- Kill info not yet set. That's next patch.

llvm-svn: 40913
 2007-08-07 23:49:57 +00:00
David Greene
f0a40d3dc2 Add a missing forward declaration. 
llvm-svn: 40896
 2007-08-07 16:34:05 +00:00
Chris Lattner
a374edf4a6 Fix a regression compiling 2005-05-11-Popcount-ffs-fls with the CBE, 
introduced by chandler's patch.

llvm-svn: 40864
 2007-08-06 16:36:18 +00:00
Christopher Lamb
8875f43912 Implement review feedback. No functionality change. 
llvm-svn: 40863
 2007-08-06 16:33:56 +00:00
Reid Spencer
3e54e72ef1 Fix minor doxygen nits. 
llvm-svn: 40854
 2007-08-05 20:06:04 +00:00
Chris Lattner
bc90e7255f shorten this name 
llvm-svn: 40843
 2007-08-05 18:45:33 +00:00
Chandler Carruth
00e56b0e81 This is the patch to provide clean intrinsic function overloading support in LLVM. It cleans up the intrinsic definitions and generally smooths the process for more complicated intrinsic writing. It will be used by the upcoming atomic intrinsics as well as vector and float intrinsics in the future. 
This also changes the syntax for llvm.bswap, llvm.part.set, llvm.part.select, and llvm.ct* intrinsics. They are automatically upgraded by both the LLVM ASM reader and the bitcode reader. The test cases have been updated, with special tests added to ensure the automatic upgrading is supported.

llvm-svn: 40807
 2007-08-04 01:51:18 +00:00
Dan Gohman
352de5aeb9 More explicit keywords. 
llvm-svn: 40757
 2007-08-02 21:21:54 +00:00
Chris Lattner
9319dfc93a don't redefine a parameter 
llvm-svn: 40748
 2007-08-02 18:08:16 +00:00
Evan Cheng
a06b9cf09f Do not emit copies for physical register output if it's not used. 
llvm-svn: 40722
 2007-08-02 05:29:38 +00:00
Scott Michel
199544367f Style police: Expand the tabs to spaces! 
llvm-svn: 40712
 2007-08-02 02:22:46 +00:00
Evan Cheng
14bd520741 Instead of adding copyfromreg's to handle physical definitions. Now isel can 
simply specify them as results and let scheduledag handle them. That
is, instead of
SDOperand Flag = DAG.getTargetNode(Opc, MVT::i32, MVT::Flag, ...)
SDOperand Result = DAG.getCopyFromReg(Chain, X86::EAX, MVT::i32, Flag)

Just write:
SDOperand Result = DAG.getTargetNode(Opc, MVT::i32, MVT::i32, ...)

And let scheduledag emit the move from X86::EAX to a virtual register.

llvm-svn: 40710
 2007-08-02 00:28:15 +00:00
Evan Cheng
d6d66f0cf4 Bugs: missing partial uses and redundant partial defs. 
llvm-svn: 40688
 2007-08-01 20:18:21 +00:00
Lauro Ramos Venancio
1ef30713de Expand unaligned loads/stores when the target doesn't support them. (PR1548) 
llvm-svn: 40682
 2007-08-01 19:34:21 +00:00
David Greene
f06a395bb9 New CallInst interface to address GLIBCXX_DEBUG errors caused by 
indexing an empty std::vector.

Updates to all clients.

llvm-svn: 40660
 2007-08-01 03:43:44 +00:00
Evan Cheng
8889a56947 simpleregistercoalescing -> regcoalescing. It's too long for me to handle. 
llvm-svn: 40654
 2007-07-31 22:37:44 +00:00
Scott Michel
c65b4c8eeb - Allow custom lowering for CTPOP, CTTZ, CTLZ. 
- Fixed an existing unexpanded tab.

llvm-svn: 40605
 2007-07-30 21:00:31 +00:00
Dan Gohman
375d541183 Fix a bug in getCopyFromParts turned up in the testcase for PR1132. 
llvm-svn: 40598
 2007-07-30 19:09:17 +00:00
Dan Gohman
555f611994 Print a space between the comment character and the basic block name, 
for prettiness.

llvm-svn: 40593
 2007-07-30 15:06:25 +00:00
Duncan Sands
e8bb2c6d32 Support for trampolines, except for X86 codegen which is 
still under discussion.

llvm-svn: 40549
 2007-07-27 12:58:54 +00:00
Christopher Lamb
55a1e36013 Move subreg lowering pass to be right after regalloc, per feedback. 
llvm-svn: 40548
 2007-07-27 07:36:14 +00:00
Dan Gohman
249090568d Fix the alias analysis query in DAGCombiner to not add in two 
offsets. The SrcValueOffset values are the real offsets from the
SrcValue base pointers.

llvm-svn: 40534
 2007-07-26 16:14:06 +00:00
Christopher Lamb
258dab5389 Add a MachineFunction pass, which runs post register allocation, that turns subreg insert/extract instruction into register copies. This ensures correct code gen if the coalescer isn't able to remove all subreg instructions. 
llvm-svn: 40521
 2007-07-26 08:18:32 +00:00
Christopher Lamb
08bcd5c0f8 Teach DAG scheduling how to properly emit subreg insert/extract machine instructions. PR1350 
llvm-svn: 40520
 2007-07-26 08:12:07 +00:00
Christopher Lamb
73c3cd47cc Add selection DAG nodes for subreg insert/extract. PR1350 
llvm-svn: 40516
 2007-07-26 07:34:40 +00:00
Christopher Lamb
cb63a77a51 Remove subreg index from MachineInstr's and also keep vregs as unsigned when adding operands. 
llvm-svn: 40514
 2007-07-26 07:00:46 +00:00
Christopher Lamb
b178ecbbc0 Fix infinite recursion for when extract_vector_elt is legal. Unfortunately no public targets use this code-path, so no test. 
llvm-svn: 40510
 2007-07-26 03:33:13 +00:00
Evan Cheng
4b747bb6a0 EmitAlignment() also emits optional fill value. 
llvm-svn: 40500
 2007-07-25 23:35:07 +00:00
Anton Korobeynikov
e0344393bb Minor cleanup: 
- Split EH and debug infiormation
 - Make DwarfWriter more verbose in some cases

llvm-svn: 40481
 2007-07-25 00:06:28 +00:00
Dan Gohman
1444c5840b Add const to CanBeFoldedBy, CheckAndMask, and CheckOrMask. 
llvm-svn: 40480
 2007-07-24 23:00:27 +00:00
Dan Gohman
54b8032d64 Fix some uses of dyn_cast to be uses of cast. 
llvm-svn: 40443
 2007-07-23 20:24:29 +00:00
Dan Gohman
ac8ccab48f Don't assume that only Uses can be kills. Defs are marked as kills initially 
when there are no uses. This fixes a dangling-pointer bug, where pointers to
deleted instructions were not removed from kills lists. More info here:
http://lists.cs.uiuc.edu/pipermail/llvmdev/2007-July/009749.html

llvm-svn: 40131
 2007-07-20 23:17:34 +00:00
Evan Cheng
cdb4e7949f Added -print-emitted-asm to print out JIT generated asm to cerr. 
llvm-svn: 40123
 2007-07-20 21:56:13 +00:00
Duncan Sands
840594467f As pointed out by g++-4.2, the original code didn't do 
what it thought it was doing.

llvm-svn: 40044
 2007-07-19 07:31:58 +00:00
Evan Cheng
5dd17373c5 Add comment. 
llvm-svn: 40022
 2007-07-18 23:34:48 +00:00
Dan Gohman
4c140b7128 It's not necessary to do rounding for alloca operations when the requested 
alignment is equal to the stack alignment.

llvm-svn: 40004
 2007-07-18 16:29:46 +00:00
Evan Cheng
6ad7064bdf Dead code. 
llvm-svn: 39979
 2007-07-17 20:01:19 +00:00
Dan Gohman
0ba554c0c8 Fix comments about vectors to use the current wording. 
llvm-svn: 39921
 2007-07-16 14:29:03 +00:00
Nick Lewycky
05e4170839 Fix the build. Patch from Holger Schurig. 
llvm-svn: 39856
 2007-07-14 15:11:14 +00:00
Anton Korobeynikov
5635277c36 Long live the exception handling! 
This patch fills the last necessary bits to enable exceptions
handling in LLVM. Currently only on x86-32/linux.

In fact, this patch adds necessary intrinsics (and their lowering) which
represent really weird target-specific gcc builtins used inside unwinder.

After corresponding llvm-gcc patch will land (easy) exceptions should be
more or less workable. However, exceptions handling support should not be 
thought as 'finished': I expect many small and not so small glitches
everywhere.

llvm-svn: 39855
 2007-07-14 14:06:15 +00:00
Dan Gohman
44771cf8e8 Fix the comment for LegalizeOp to more accurately reflect what it does. 
llvm-svn: 39827
 2007-07-13 20:14:11 +00:00
Dan Gohman
1f4580ceb4 Don't call SimplifyVBinOp for non-vector operations, following earlier review 
feedback. This theoretically makes the common (scalar) case more efficient.

llvm-svn: 39823
 2007-07-13 20:03:40 +00:00
Dale Johannesen
1a9d304c35 Modify previous patch per review comments. 
llvm-svn: 39817
 2007-07-13 17:31:29 +00:00
Dale Johannesen
469ed8e17e Skeleton of post-RA scheduler; doesn't do anything yet. 
Change name of -sched option and DEBUG_TYPE to
pre-RA-sched; adjust testcases.

llvm-svn: 39816
 2007-07-13 17:13:54 +00:00
Duncan Sands
b2a18dc10b The type ids making up a filter are unsigned, and 
should be output as unsigned values.  Checked against
gcc.

llvm-svn: 39775
 2007-07-12 13:51:39 +00:00
Evan Cheng
c7216b3a86 Missed a couple of places where new instructions are added due to spill / restore. 
llvm-svn: 39748
 2007-07-11 19:17:18 +00:00
Duncan Sands
bfbe683782 Exception handling has been implemented. 
llvm-svn: 39732
 2007-07-11 16:59:20 +00:00
Duncan Sands
aa83dc9037 If assertions are not enabled, we should return False here. 
llvm-svn: 38535
 2007-07-11 08:47:55 +00:00
Evan Cheng
2791a7ea6d No longer need to track last def / use. 
llvm-svn: 38534
 2007-07-11 08:47:44 +00:00
Evan Cheng
c42a2e03f1 Fix for PR1545: Revamp code that update kill information due to register reuse. 
llvm-svn: 38525
 2007-07-11 05:28:39 +00:00
David Greene
8d77dcca00 Make this work with GLIBCXX_DEBUG. 
llvm-svn: 38516
 2007-07-10 22:00:30 +00:00
Dan Gohman
629311ca52 Change the peep for EXTRACT_VECTOR_ELT of BUILD_PAIR to look for 
the new CONCAT_VECTORS node type instead, as that's what legalize
uses now. And add a peep for EXTRACT_VECTOR_ELT of INSERT_VECTOR_ELT.

llvm-svn: 38503
 2007-07-10 18:20:44 +00:00
Evan Cheng
d80c2821af If the operand is marked M_OPTIONAL_DEF_OPERAND, then it's a def. 
llvm-svn: 38496
 2007-07-10 17:52:20 +00:00
Evan Cheng
87e735c4be Somehow this wasn't committed last time. M_CLOBBERS_PRED is gone. 
llvm-svn: 38495
 2007-07-10 17:50:43 +00:00
Dan Gohman
bd0ff28e1c Fix a bug in the folding of binary operators to undef. 
Thanks to Lauro for spotting this!

llvm-svn: 38491
 2007-07-10 15:19:29 +00:00
Dan Gohman
6b21e11f1c Fix the folding of undef in several binary operators to recognize 
undef in either the left or right operand.

llvm-svn: 38489
 2007-07-10 14:20:37 +00:00
Evan Cheng
cf3b1c89a9 When a node value is only used by a CopyToReg, use the user's dest. This should not be restricted to nodes that produce only a single value. 
llvm-svn: 38485
 2007-07-10 07:08:32 +00:00
Evan Cheng
23dc96f640 Move DenseMapKeyInfo<SDOperand> from LegalizeDAG.cpp to SelectionDAGNodes.h 
llvm-svn: 38484
 2007-07-10 06:59:55 +00:00
Dan Gohman
150487b166 Preserve volatililty and alignment information when lowering or 
simplifying loads and stores.

llvm-svn: 38473
 2007-07-09 22:18:38 +00:00
Dan Gohman
81cfdc2f19 Change getCopyToParts and getCopyFromParts to always use target-endian 
register ordering, for both physical and virtual registers. Update the PPC
target lowering for calls to expect registers for the call result to
already be in target order.

llvm-svn: 38471
 2007-07-09 20:59:04 +00:00
Dan Gohman
be352027e6 Initialize the IndexedModeActions array with memset before 
updating it with calls to setIndexedLoadAction/setIndexedStoreAction,
which only update a few bits at a time. This avoids ostensible
undefined behavior of operationg on values which may be
trap-representations, and as a practical matter fixes errors from
valgrind, which doesn't track uninitialized memory with bit
granularity.

llvm-svn: 38468
 2007-07-09 20:49:44 +00:00
Chris Lattner
aff9a84b7a Fix this warning: 
DAGCombiner.cpp: In member function 'llvm::SDOperand<unnamed>::DAGCombiner::visitOR(llvm::SDNode*)':
DAGCombiner.cpp:1608: warning: passing negative value '-0x00000000000000001' for argument 1 to 'llvm::SDOperand llvm::SelectionDAG::getConstant(uint64_t, llvm::MVT::ValueType, bool)'

oiy.

llvm-svn: 38458
 2007-07-09 16:16:34 +00:00
Gabor Greif
89919e17b4 fix typos 
llvm-svn: 38453
 2007-07-09 12:00:59 +00:00
Evan Cheng
4c7b04f24d Teach if-conversion about instructions that were already predicated, e.g. conditional move. 
llvm-svn: 37964
 2007-07-06 23:24:39 +00:00
Duncan Sands
f926a3080c The exception handling intrinsics return values, 
so must be lowered to a value, not nothing at all.
Subtle point: I made eh_selector return 0 and
eh_typeid_for return 1.  This means that only
cleanups (destructors) will be run as the exception
unwinds [if eh_typeid_for returned 0 then it would
be as if the first catch always matched, and the
corresponding handler would be run], which is
probably want you want in the CBE.

llvm-svn: 37947
 2007-07-06 14:46:23 +00:00
Duncan Sands
404336eb55 Indexes into the list of filter ids cannot be output 
directly: they need to be turned into byte offsets
(often the same, but may not be if there are many
type infos).

llvm-svn: 37942
 2007-07-06 12:46:24 +00:00
Rafael Espindola
7b3de98989 Add the byval attribute 
llvm-svn: 37940
 2007-07-06 10:57:03 +00:00
Duncan Sands
7e50c11edd Remove propagateEHRegister in favour of a more limited 
fix, that is adequate while PR1508 remains unresolved.

llvm-svn: 37938
 2007-07-06 09:18:59 +00:00
Duncan Sands
e7650d2b1e Remove ExtractGlobalVariable - use StripPointerCasts 
instead.

llvm-svn: 37937
 2007-07-06 09:10:03 +00:00
Evan Cheng
bae19254f0 Workaround of getCopyToRegs and getCopyFromRegs bugs for big-endian machines. 
llvm-svn: 37935
 2007-07-06 01:47:35 +00:00
Evan Cheng
22957164aa Change CalculateHeights and CalculateDepths to be non-recursive. 
llvm-svn: 37934
 2007-07-06 01:37:28 +00:00
Dan Gohman
cd6f6a0a58 Make the debug string for ISD::MERGE_VALUES consistent with the others. 
llvm-svn: 37922
 2007-07-05 20:15:43 +00:00
Dan Gohman
90c6b87b31 Add a parameter to getCopyToParts and getCopyFromParts to specify whether 
endian swapping should be done, and update the code to use it. This fixes
some register ordering issues on big-endian systems, such as PowerPC,
introduced by the recent illegal by-val arguments changes.

llvm-svn: 37921
 2007-07-05 20:12:34 +00:00
Gabor Greif
5f705671e4 Here is the bulk of the sanitizing. 
Almost all occurrences of "bytecode" in the sources have been eliminated.

llvm-svn: 37913
 2007-07-05 17:07:56 +00:00
Duncan Sands
eca2fa295b Make sure only one copy of a filter is placed in the 
exception handling table if we encounter it multiple
times.  Filters could be folded harder than this, but
that would mean a lot more work for not much gain.

llvm-svn: 37908
 2007-07-05 15:15:01 +00:00
Evan Cheng
0c037f4ee0 Better assertion messages. 
llvm-svn: 37890
 2007-07-05 07:05:38 +00:00
Duncan Sands
4441eff1ac Extend eh.selector to support both catches and filters. 
Drop the eh.filter intrinsic.

llvm-svn: 37875
 2007-07-04 20:52:51 +00:00
Dan Gohman
d2d18267e6 Fix several over-aggressive folds for undef nodes in dagcombine, to 
follow the rules for undef used in instcombine.

llvm-svn: 37851
 2007-07-03 14:03:57 +00:00
Dale Johannesen
7af19491d3 Fix for PR 1505 (and 1489). Rewrite X87 register 
model to include f32 variants.  Some factoring
improvments forthcoming.

llvm-svn: 37847
 2007-07-03 00:53:03 +00:00
Dan Gohman
68f0cbccfb Replace ExpandScalarFormalArgs and ExpandScalarCallArgs with the newly 
refactored getCopyFromParts and getCopyToParts, which are more general.
This effectively adds support for lowering illegal by-val vector call
arguments.

llvm-svn: 37843
 2007-07-02 16:18:06 +00:00
Dan Gohman
5fe8266d04 Teach GetNegatedExpression to negate 0-B to B in UnsafeFPMath mode, and 
visitFSUB to fold 0-B to -B in UnsafeFPMath mode. Also change visitFNEG
to use isNegatibleForFree/GetNegatedExpression instead of doing a subset
of the same thing manually.

This fixes test/CodeGen/X86/negative-sin.ll.

llvm-svn: 37842
 2007-07-02 15:48:56 +00:00
Evan Cheng
a1a06d1763 Only do FNEG xform when the vector type is a floating point type. 
llvm-svn: 37818
 2007-06-29 21:44:35 +00:00
David Greene
75e58e2207 Remove unused variables. 
llvm-svn: 37816
 2007-06-29 21:42:03 +00:00
Evan Cheng
a6add260b3 Fix a vector FP constant CSE bug. 
llvm-svn: 37814
 2007-06-29 21:36:04 +00:00
David Greene
0942edd414 Remove unnecessary attributions in comments. 
llvm-svn: 37799
 2007-06-29 03:42:23 +00:00
David Greene
df6a87ea1c Fix reference to cached end iterator invalidated by an erase operation. 
Uncovered by _GLIBCXX_DEBUG.

llvm-svn: 37795
 2007-06-29 02:49:11 +00:00
David Greene
244fb44f6b Remove the "special tie breaker" because it resulted in inconsistent 
ordering and thus violated the strict weak ordering requirement of
priority_queue.  Uncovered by _GLIBCXX_DEBUG.

llvm-svn: 37794
 2007-06-29 02:48:09 +00:00
David Greene
ad230f223f Fix misue of iterator pointing to erased object. Uncovered by 
_GLIBCXX_DEBUG.

llvm-svn: 37793
 2007-06-29 02:45:24 +00:00
Dan Gohman
ad9cbaba40 Fix an assertion failure in legalizing bitcast operators on targets where 
vectors are split down to single elements as part of legalization.

llvm-svn: 37785
 2007-06-29 00:09:08 +00:00
Dan Gohman
c6bdcfa8c0 Add new TargetLowering code to provide the final register type that an 
illegal value type will be transformed to, for code that needs the
register type after all transformations instead of just after the first
transformation.

Factor out the code that uses this information to do copy-from-regs and
copy-to-regs for various purposes into separate functions so that they
are done consistently.

llvm-svn: 37781
 2007-06-28 23:29:44 +00:00
Evan Cheng
e67385d712 If a livein is not used in the block. It's live through. 
llvm-svn: 37764
 2007-06-27 18:47:28 +00:00
Evan Cheng
26542e347a Partial fix for PR1502: If a EH register is needed in a successor of landing pad, add it as livein to all the blocks in the paths between the landing pad and the specified block. 
llvm-svn: 37763
 2007-06-27 18:45:32 +00:00
Dan Gohman
cb89e19a6d Rename ("shrinkify") MVT::isExtendedValueType to MVT::isExtendedVT. 
llvm-svn: 37758
 2007-06-27 16:08:04 +00:00
Dan Gohman
c1c4b0972f Use getVectorTypeBreakdown in FunctionLoweringInfo::CreateRegForValue 
to compute the number and type of registers needed for vector values
instead of computing it manually. This fixes PR1529.

llvm-svn: 37755
 2007-06-27 14:34:07 +00:00
Dan Gohman
69a6154359 Make the comment for ScalarizeVectorOp mention that it is only for use 
with single-element vectors.

llvm-svn: 37752
 2007-06-27 14:06:22 +00:00
Duraid Madina
d32d8b9726 ok, this is something of a dirty hack, but it seems to work. (fixes e.g. 
the SPASS miscompilation)

llvm-svn: 37750
 2007-06-27 09:01:14 +00:00
Duraid Madina
73b0d188c8 ok, this much doesn't seem to bork anything 
llvm-svn: 37749
 2007-06-27 08:31:07 +00:00
Duraid Madina
94cd67290b revert evan's fixes (and my doofusness) since they had a huge code 
quality hit. will look at this soon.

llvm-svn: 37748
 2007-06-27 08:11:59 +00:00
Duraid Madina
c0d3c96333 pull evan's fixes - should help the nightly tester (but there are still 
some issues)

llvm-svn: 37747
 2007-06-27 07:07:13 +00:00
Evan Cheng
480cd89fa1 Replace std::set with SmallPtrSet. 
llvm-svn: 37746
 2007-06-27 05:23:00 +00:00
Evan Cheng
f74e333161 Fix an obvious bug. Old code only worked for the entry block. 
llvm-svn: 37743
 2007-06-27 01:16:36 +00:00
Evan Cheng
5c0f03bce4 Correctly handle implcit def / use operands. 
llvm-svn: 37740
 2007-06-26 21:05:13 +00:00
Evan Cheng
0eaa723739 Properly handle kills of a physical register which has sub-registers that are read by later instructions. 
llvm-svn: 37739
 2007-06-26 21:03:35 +00:00
Duraid Madina
c4b17f9a08 tidy this file up a bit 
llvm-svn: 37725
 2007-06-26 00:21:58 +00:00
Duraid Madina
1d8c70bf83 A bunch of fixes to the BigBlock allocator improve compile-time by ~20% 
and code quality by ~2% on my tests.

A big thank you to Roman Levenstein for this patch! See
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070618/050717.html
for more details.

llvm-svn: 37724
 2007-06-25 23:46:54 +00:00
Dan Gohman
354f02e03d Generalize MVT::ValueType and associated functions to be able to represent 
extended vector types. Remove the special SDNode opcodes used for pre-legalize
vector operations, and the special MVT::Vector type used with them. Adjust
lowering and legalize to work with the normal SDNode kinds instead, and to
use the normal MVT functions to work with vector types instead of using the
two special operands that the pre-legalize nodes held.

This allows pre-legalize and post-legalize DAGs, and the code that operates
on them, to be more consistent. Pre-legalize vector operators can be handled
more consistently with scalar operators. And, -view-dag-combine1-dags and
-view-legalize-dags now look prettier for vector code.

llvm-svn: 37719
 2007-06-25 16:23:39 +00:00
Dan Gohman
a62327ea40 Move ComputeMaskedBits, MaskedValueIsZero, and ComputeNumSignBits from 
TargetLowering to SelectionDAG so that they have more convenient
access to the current DAG, in preparation for the ValueType routines
being changed from standalone functions to members of SelectionDAG for
the pre-legalize vector type changes.

llvm-svn: 37704
 2007-06-22 14:59:07 +00:00
Duraid Madina
fe6e41b4f9 check in the BigBlock local register allocator 
llvm-svn: 37703
 2007-06-22 08:27:12 +00:00
Evan Cheng
85169b9b40 std::set is really really terrible. Switch to SmallPtrSet to reduce compile time. For Duraid's example. The overall isel time is reduced from 0.6255 sec to 0.1876 sec. 
llvm-svn: 37701
 2007-06-22 01:35:51 +00:00
Dan Gohman
72af1305fb Tidy up ValueType names in comments. 
llvm-svn: 37688
 2007-06-21 14:48:26 +00:00
Dan Gohman
1815e9bdb3 Rename TargetLowering::getNumElements and friends to 
TargetLowering::getNumRegisters and similar, to avoid confusion with
the actual number of elements for vector types.

llvm-svn: 37687
 2007-06-21 14:42:22 +00:00
Evan Cheng
f26fc091ac Xforms: 
(add (select cc, 0, c), x) -> (select cc, x, (add, x, c))
(sub x, (select cc, 0, c)) -> (select cc, x, (sub, x, c))

llvm-svn: 37685
 2007-06-21 07:39:16 +00:00
Evan Cheng
5e2aabd3ad Avoid if-converting simple block that ends with unconditional branch or fallthrough unless it branches / falls to the 'false' block. Not profitable, may end up increasing code size. 
llvm-svn: 37660
 2007-06-19 21:45:13 +00:00
Dan Gohman
485cb57eab Pass a SelectionDAG into SDNode::dump everywhere it's used, in prepration 
for needing the DAG node to print pre-legalize extended value types, and
to get better debug messages with target-specific nodes.

llvm-svn: 37656
 2007-06-19 14:13:56 +00:00
Chris Lattner
653c0ccfbe document and hide two options. 
llvm-svn: 37651
 2007-06-19 05:47:49 +00:00
Dan Gohman
b60d8a92c9 Replace M_REMATERIALIZIBLE and the newly-added isOtherReMaterializableLoad 
with a general target hook to identify rematerializable instructions. Some
instructions are only rematerializable with specific operands, such as loads
from constant pools, while others are always rematerializable. This hook
allows both to be identified as being rematerializable with the same
mechanism.

llvm-svn: 37644
 2007-06-19 01:48:05 +00:00
Evan Cheng
12b3002673 Replace TargetInstrInfo::CanBeDuplicated() with a M_NOT_DUPLICABLE bit. 
llvm-svn: 37643
 2007-06-19 01:26:51 +00:00
Chris Lattner
68f9f9fa3b add isVarArg to CCState 
llvm-svn: 37640
 2007-06-19 00:11:09 +00:00
Evan Cheng
b752a87db1 Fix some fragile code wrt CFG edge updating. 
llvm-svn: 37634
 2007-06-18 22:44:57 +00:00
Evan Cheng
978058bcd7 Move CorrectExtraCFGEdges() from BranchFolding.cpp to a MachineBasicBlock method. 
llvm-svn: 37633
 2007-06-18 22:43:58 +00:00
Chris Lattner
5b93ff84b9 make ComputeTopDownOrdering significantly faster and use less stack space 
by making it non-recursive

llvm-svn: 37629
 2007-06-18 21:28:10 +00:00
Dan Gohman
46f036ccb0 Make chain dependencies blue, in addition to being dashed. 
llvm-svn: 37626
 2007-06-18 15:30:16 +00:00
Evan Cheng
b3cefc0824 Properly remove duplicate instructions as result of diamond if-conversion. Other bug fixes. 
llvm-svn: 37623
 2007-06-18 08:37:25 +00:00
Evan Cheng
1732a41371 Really turn if-converter loose: 
1. Consider all possible ifcvt cases at once. No longer restricted to bottom
   up iterative approach.
2. Sort all possible cases based on a cost function. Perform the most profitable
   ones first invalidate others that target the same blocks.
3. Fixed a number of bugs related to block duplication.

llvm-svn: 37613
 2007-06-16 09:34:52 +00:00
Tanya Lattner
4c078c1ace Codegen support (stripped out) for the annotate attribute. 
llvm-svn: 37608
 2007-06-15 22:26:58 +00:00
Evan Cheng
759b3cca3c Not every predicable block can be safely duplicated. 
llvm-svn: 37607
 2007-06-15 21:18:05 +00:00
Chris Lattner
c08e8abb53 Fix CodeGen/X86/inline-asm-x-scalar.ll:test4, by retaining regclass info 
for tied register constraints.

llvm-svn: 37601
 2007-06-15 19:11:01 +00:00
Evan Cheng
88c463eb8c MachineInstr::isPredicable() is no longer needed. 
llvm-svn: 37599
 2007-06-15 19:06:07 +00:00
Duncan Sands
deef6fe78b Workaround for PR1508. 
llvm-svn: 37597
 2007-06-15 19:04:19 +00:00
Evan Cheng
a6504bc2e0 Extra edges are deleted later if needed. 
llvm-svn: 37593
 2007-06-15 17:34:48 +00:00
Evan Cheng
f4d3d695b2 Allow small blocks to be duplicated to enable if-conversion. 
llvm-svn: 37590
 2007-06-15 07:36:12 +00:00
Evan Cheng
6f348e1c75 No really, clear predcessors states. 
llvm-svn: 37581
 2007-06-14 23:34:09 +00:00
Evan Cheng
b1b1914582 If BB is predicated, invalidate its predecessor(s) which would if-convert it. It needs to be re-analyzed. 
llvm-svn: 37580
 2007-06-14 23:13:19 +00:00
Dan Gohman
2fd7d26df8 Rename MVT::getVectorBaseType to MVT::getVectorElementType. 
llvm-svn: 37579
 2007-06-14 22:58:02 +00:00
Evan Cheng
26d68261ae Fix typo. 
llvm-svn: 37577
 2007-06-14 21:26:08 +00:00
Dan Gohman
35f2b4d716 Add a target hook to allow loads from constant pools to be rematerialized, and an 
implementation for x86.

llvm-svn: 37576
 2007-06-14 20:50:44 +00:00
Evan Cheng
a2c2479e22 Fix some stupid bugs that have effectively disabled if-conversion. 
llvm-svn: 37575
 2007-06-14 20:28:52 +00:00
Duncan Sands
51e6294637 Only correctly lower exception handing intrinsics if exception handling is 
turned on.  Likewise for scanning of invokes to mark landing pads.

llvm-svn: 37570
 2007-06-13 16:53:21 +00:00
Dan Gohman
6415b2548e Introduce new SelectionDAG node opcodes VEXTRACT_SUBVECTOR and 
VCONCAT_VECTORS. Use these for CopyToReg and CopyFromReg legalizing in
the case that the full register is to be split into subvectors instead
of scalars. This replaces uses of VBIT_CONVERT to present values as
vector-of-vector types in order to make whole subvectors accessible via
BUILD_VECTOR and EXTRACT_VECTOR_ELT.

This is in preparation for adding extended ValueType values, where
having vector-of-vector types is undesirable.

llvm-svn: 37569
 2007-06-13 15:12:02 +00:00
Dan Gohman
3bc1455d49 When creating CopyFromReg nodes, always use legal types. And use the 
correct types for the result vector, even though it is currently bitcasted
to a different type immediately.

llvm-svn: 37568
 2007-06-13 14:55:16 +00:00
Duncan Sands
32eaa9b30a The fix that was applied for PR1224 stops the compiler 
crashing but breaks exception handling.  The problem
described in PR1224 is that invoke is a terminator that
can produce a value.  The value may be needed in other
blocks.  The code that writes to registers values needed
in other blocks runs before terminators are lowered (in
this case invoke) so asserted because the value was not
yet available.  The fix that was applied was to do invoke
lowering earlier, before writing values to registers.

The problem this causes is that the code to copy values
to registers can be output after the invoke call.  If
an exception is raised and control is passed to the
landing pad then this copy-code will never execute.  If
the value is needed in some code path reached via the
landing pad then that code will get something bogus.

So revert the original fix and simply skip invoke values
in the general copying to registers code.  Instead copy
the invoke value to a register in the invoke lowering code.

llvm-svn: 37567
 2007-06-13 05:51:31 +00:00
Evan Cheng
d4e9443a78 Typo 
llvm-svn: 37566
 2007-06-13 00:04:00 +00:00
Evan Cheng
d5aaab210c Now if-converting all 4 variants of triangles. 
llvm-svn: 37565
 2007-06-12 23:54:05 +00:00
Lauro Ramos Venancio
c9a156f041 Fix a typo in bswap lowering. 
llvm-svn: 37544
 2007-06-11 23:16:16 +00:00
Evan Cheng
d6740b478b Restructure code to reduce ifcvt compile time cost. 
llvm-svn: 37543
 2007-06-11 22:26:22 +00:00
Reid Spencer
0c11f4ebd8 Fix the build. 
llvm-svn: 37537
 2007-06-10 00:19:17 +00:00
Evan Cheng
859dab4bda Don't change CFG during analysis stage. Do so during ifcvt and invalidate predecessors accordingly. 
llvm-svn: 37531
 2007-06-09 01:03:43 +00:00
Evan Cheng
7996fc5d23 Carefully remove extraneous CFG edges after each ifcvt. 
llvm-svn: 37529
 2007-06-08 22:01:07 +00:00
Evan Cheng
515e2537d7 Correct transfer predicate information. 
llvm-svn: 37524
 2007-06-08 19:17:12 +00:00
Evan Cheng
49720a7312 Hidden options to help debugging ifcvt issues. 
llvm-svn: 37523
 2007-06-08 19:10:51 +00:00
David Greene
6b14050d1e Factor live variable analysis so it does not do register coalescing 
simultaneously.  Move that pass to SimpleRegisterCoalescing.

This makes it easier to implement alternative register allocation and
coalescing strategies while maintaining reuse of the existing live
interval analysis.

llvm-svn: 37520
 2007-06-08 17:18:56 +00:00
Evan Cheng
a41764f141 Allow more cmp / bcc to be predicated; clean up triangle ifcvt checking code. 
llvm-svn: 37518
 2007-06-08 09:36:04 +00:00
Duncan Sands
f31e17574a Use more realistically sized vectors. Reserve capacity if we know in advance 
how much will be used.

llvm-svn: 37515
 2007-06-08 08:59:11 +00:00
Dale Johannesen
74ffb6ffb3 Make throttle a hidden parameter, per review. 
llvm-svn: 37511
 2007-06-08 01:08:52 +00:00
Dale Johannesen
0903dd4e52 Throttle tail merging; handling blocks with large numbers of predecessors 
is too slow.

llvm-svn: 37509
 2007-06-08 00:34:27 +00:00
Evan Cheng
2b9e9abed1 Only remove the edge from entry to false if false block is merged. 
llvm-svn: 37503
 2007-06-07 22:31:28 +00:00
Dale Johannesen
0f3c011fd5 Do not change the size of function arguments. PR 1489. 
llvm-svn: 37496
 2007-06-07 21:07:15 +00:00
Evan Cheng
4f73706838 ifcvt a triangle: don't merge ifcvt block with rejoin block if it can fall through to it. If merged, the resulting block is not a candidate for iterative ifcvting since it contains both predicated and non-predicated code. 
llvm-svn: 37487
 2007-06-07 08:13:00 +00:00
Evan Cheng
d440a235a1 Lots of bug fixes. Now finally in a reasonable state. 
llvm-svn: 37485
 2007-06-07 02:12:15 +00:00
Owen Anderson
b333dd064e Quick patch to fix the build, based on what it appears Evan meant to write. 
Evan, please check that this is in fact correct.

llvm-svn: 37471
 2007-06-06 16:22:00 +00:00
Duncan Sands
9bef60ea1a Fold the exception actions table harder: if two typeid lists start the 
same, only output one copy of the common part.

llvm-svn: 37470
 2007-06-06 15:37:31 +00:00
Evan Cheng
aeccfbef8b Lots of bug fixes. 
llvm-svn: 37467
 2007-06-06 10:16:17 +00:00
Duncan Sands
484bdb927b Additional fix for PR1422: make sure the landing pad label is placed in the 
correct machine basic block - do not rely on the eh.exception intrinsic
being in the landing pad: the loop optimizers can move it out.

llvm-svn: 37463
 2007-06-06 10:05:18 +00:00
Evan Cheng
4c42a30bad If a unconditional branch is added to branch to the false path during ifcvt, the predicated block cannot be iteratively ifcvted. 
llvm-svn: 37456
 2007-06-06 02:08:52 +00:00
Evan Cheng
cc7186df2b Minor statistics counting bug. 
llvm-svn: 37451
 2007-06-06 01:12:44 +00:00
Evan Cheng
2eefb6f4c3 Fix a couple of typos and be smarter about order of blocks when ifcvt a diamond. 
llvm-svn: 37449
 2007-06-06 00:57:55 +00:00
Evan Cheng
3db28ad9f2 Fix diamond shape ifcvt bugs. 
llvm-svn: 37444
 2007-06-05 23:46:14 +00:00
Evan Cheng
acf719bf83 ReplaceUsesOfBlockWith() can modify the predecessors list. 
llvm-svn: 37441
 2007-06-05 22:03:53 +00:00
Evan Cheng
f2a2a5a025 Do not ifcvt if either true / false path is a backedge. Not profitable in almost all cases. 
llvm-svn: 37440
 2007-06-05 20:38:42 +00:00
Evan Cheng
d495b2f9b9 I had a senior moment. 
llvm-svn: 37433
 2007-06-05 07:05:25 +00:00
Evan Cheng
a7f7c92dc9 If the predicated block requires an early exit, end the block there and add a unconditional branch to false block. AnalyzeBranch() does not understand early exits. 
llvm-svn: 37430
 2007-06-05 01:31:40 +00:00
Evan Cheng
7b113514d4 Fix some subtle bugs: bug during succeessor copying; incorrectly updating states of ifcvted blocks. 
llvm-svn: 37429
 2007-06-05 00:07:37 +00:00
Dale Johannesen
a68662e4a4 Tail merging wasn't working for predecessors of landing pads. PR 1496. 
llvm-svn: 37427
 2007-06-04 23:52:54 +00:00
Evan Cheng
8728a6c3fd Global ctors / dtors alignment shouldn't be hard-coded at 4. e.g. It could be 8 for 64-bit targets. 
llvm-svn: 37421
 2007-06-04 20:39:18 +00:00
Evan Cheng
79f53ff66d Forgot to check for if iterator reached the end. 
llvm-svn: 37420
 2007-06-04 20:33:36 +00:00
Dan Gohman
e6c827b537 Pass the DAG to SDNode::dump to let it do more detailed dumps in some cases. 
llvm-svn: 37413
 2007-06-04 16:17:33 +00:00
Dan Gohman
b69321ca1d Resolve implicit alignment before computing the FoldingSet information so 
that the CSE map always contains explicit alignment information. This allows
more loads to be CSE'd when there is a mix of explicit-alignment loads and
implicit-alignment loads.

Also, in SelectionDAG::FindModifiedNodeSlot, add the operands to the
FoldingSetNodeID before the load/store information instead of after, so
that it matches what is done elsewhere.

llvm-svn: 37411
 2007-06-04 15:49:41 +00:00
Evan Cheng
c86e3086da Let IfConverter loose. Allow more aggressive subsumptions; reorder basic blocks to expose more ifcvt opportunities; code clean up and fixes. 
llvm-svn: 37409
 2007-06-04 06:47:22 +00:00
Evan Cheng
0dbe4d04db Move ReplaceUsesOfBlockWith() out of BranchFolding into a MachineBasicBlock general facility. 
llvm-svn: 37408
 2007-06-04 06:44:01 +00:00
Duncan Sands
f37bc8207e The semantics of invoke require that we always jump to the unwind block 
(landing pad) when an exception unwinds through the call.  This doesn't
quite match the way the dwarf unwinder works: by default it only jumps to
the landing pad if the catch or filter specification matches, and otherwise
it keeps on unwinding.  There are two ways of specifying to the unwinder
that it should "always" (more on why there are quotes here later) jump to
the landing pad: follow the specification by a 0 typeid, or follow it by
the typeid for the NULL typeinfo.  GCC does the first, and this patch makes
LLVM do the same as gcc.  However there is a problem: the unwinder performs
optimizations based on C++ semantics (it only expects destructors to be
run if the 0 typeid fires - known as "cleanups"), meaning it assumes that no
exceptions will be raised and that the raised exception will be reraised
at the end of the cleanup code.  So if someone writes their own LLVM code
using the exception intrinsics they will get a nasty surprise if they don't
follow these rules.  The other possibility of using the typeid corresponding
to NULL (catch-all) causes the unwinder to make no assumptions, so this is
probably what we should use in the long-run.  However since we are still
having trouble getting exception handling working properly, for the moment
it seems best to closely imitate GCC.

llvm-svn: 37399
 2007-06-02 17:16:06 +00:00
Duncan Sands
37925d3f39 Integrate exception filter support and exception catch support. This 
simplifies the code in DwarfWriter, allows for multiple filters and
makes it trivial to specify filters accompanied by cleanups or catch-all
specifications (see next patch).  What a deal!  Patch blessed by Anton.

llvm-svn: 37398
 2007-06-02 16:53:42 +00:00
Zhou Sheng
60714d7922 Make LowerCTPOP() support arbitrary bitwidth integer type. 
llvm-svn: 37397
 2007-06-02 04:10:33 +00:00
Dale Johannesen
63d49a09aa Fix CorrectExtraCFGEdges to allow for multiple LandingPad targets. 
llvm-svn: 37394
 2007-06-02 00:08:15 +00:00
Dale Johannesen
071dbb1f87 Fancier algorithm in tail-merge comment implemented, so remove comment. 
llvm-svn: 37393
 2007-06-01 23:04:28 +00:00
Dale Johannesen
fd7f1ad2f7 Implement smarter algorithm for choosing which blocks to tail-merge. 
See test/CodeGen/X86/test-pic-jtbl.ll for a case where it works well;
shaves another 10K off our favorite benchmark.  I was hesitant about
this because of compile speed, but seems to do OK on a bootstrap.

llvm-svn: 37392
 2007-06-01 23:02:45 +00:00
Evan Cheng
d7793e4561 Correctly mark early-exit on the false path. 
llvm-svn: 37387
 2007-06-01 20:29:21 +00:00
Duncan Sands
b1cc3f9881 Since TypeInfos are passed as i8 pointers, a NULL TypeInfo should be passed 
as a null i8 pointer not as a 0 i32.

llvm-svn: 37383
 2007-06-01 08:18:30 +00:00
Evan Cheng
ce6bae7667 Ifcvt triangle: don't ifcvt 'true' BB if it has other predecessors; don't merge 'false' BB if it has other predecessors. 
llvm-svn: 37382
 2007-06-01 07:41:07 +00:00
Evan Cheng
8e188afada Remove a bogus check. Even terminators in a ifcvt need to be predicated. Unconditional branches can usually be converted to conditional ones. 
llvm-svn: 37380
 2007-06-01 00:55:26 +00:00
Evan Cheng
7c081d0d02 Allow multiple ifcvt candidates to share children blocks; add some debugging code. 
llvm-svn: 37379
 2007-06-01 00:12:12 +00:00
Dale Johannesen
4ca6f07cf0 Arrange for only 1 of multiple branches to landing pad to be kept. 
Do not remove empty landing pads (EH table needs to be updated)

llvm-svn: 37375
 2007-05-31 21:54:00 +00:00
Evan Cheng
7a6edf8eab Fix a typo. 
llvm-svn: 37374
 2007-05-31 20:53:33 +00:00
Chris Lattner
72e07238ba Fix the asmprinter so that a globalvalue can specify an explicit alignment 
smaller than the preferred alignment, but so that the target can actually
specify a minimum alignment if needed.  This fixes some objc protocol
failures Devang tracked down.

llvm-svn: 37373
 2007-05-31 18:57:45 +00:00
Lauro Ramos Venancio
b5186a0078 Fix PR1424. 
When a function has FP, the register scavenging spill slot offset already
was calculated.

llvm-svn: 37371
 2007-05-31 18:27:58 +00:00
Evan Cheng
c9c6bbd273 Change traversal order to bottom up in preparation for more aggressive if-conversion. 
llvm-svn: 37365
 2007-05-30 19:49:19 +00:00
Chris Lattner
14bcacb40b Fix CodeGen/PowerPC/2007-05-30-dagcombine-miscomp.ll, and PR1473. 
llvm-svn: 37362
 2007-05-30 16:30:06 +00:00
Dale Johannesen
b306dfb861 Changed per review comment. 
llvm-svn: 37355
 2007-05-30 00:32:01 +00:00
Dale Johannesen
1ba04f7d87 Make stable_sort in tail merging actually be stable (it never was, but didn't 
matter until my last change).  Reenable tail merging by default.

llvm-svn: 37354
 2007-05-29 23:47:50 +00:00
Evan Cheng
a1300649ba Don't merge in tail block of a diamond if it has more than one predecessors after if-conversion. 
llvm-svn: 37353
 2007-05-29 23:37:20 +00:00
Evan Cheng
b1ba206831 If there is an empty block between a source and its successor block, it still requires a unconditional branch. 
llvm-svn: 37344
 2007-05-29 22:31:16 +00:00
Evan Cheng
d8b25a2091 Add missing const qualifiers. 
llvm-svn: 37341
 2007-05-29 18:35:22 +00:00
Zhou Sheng
3e018c3e3e Correct the logic in LowerPartSet which cleared the bits from 0 to low-1. 
llvm-svn: 37331
 2007-05-26 03:43:13 +00:00
Chris Lattner
5161203d52 tighten up recursion depth again 
llvm-svn: 37330
 2007-05-25 02:19:06 +00:00