RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-23 04:52:54 +02:00
Code Issues Projects Releases Wiki Activity
59,242 Commits 4 Branches 5 Tags 704 MiB
c83be77854
Commit Graph

9343 Commits

Author SHA1 Message Date
Devang Patel
dd5c1fcba9 Fix comment. 
llvm-svn: 98830
 2010-03-18 16:41:16 +00:00
Jakob Stoklund Olesen
d35fc676e4 Revert "Change coalescer complexity from N^2 to N logN by changing one letter." 
This reverts commit 98776. It broke the llvm-gcc boot strap.

llvm-svn: 98784
 2010-03-18 00:23:47 +00:00
Devang Patel
9ad039e65f Debug info intrinsic does not intefer during tail call optimization. 
llvm-svn: 98778
 2010-03-17 23:52:37 +00:00
Jakob Stoklund Olesen
e05919807e Change coalescer complexity from N^2 to N logN by changing one letter. 
Remove ugly hack that aborted the coalescer before using N^2 time.

This affects functions with very complicated live intervals for physical
registers, i.e. functions with thousands of function calls.

llvm-svn: 98776
 2010-03-17 23:21:58 +00:00
Devang Patel
4df8850a32 Fix EmitSectionOffset incorrect argument. DwarfDebug is emitting debug info so isEH is always false. This was hiding until now from compilers because of default arguments. This was hiding from dwarf debug info users because for most of the platform isAbsoluteEHSectionOffsets() is same as isAbsoluteDebugSectionOffsets(). But Chris found it while updating dwarf printer to use MC*. 
llvm-svn: 98743
 2010-03-17 17:29:55 +00:00
Chris Lattner
962d3d700b fix GetOrCreateTemporarySymbol to require a name, clients 
should use CreateTempSymbol() if they don't care about the
name.

llvm-svn: 98712
 2010-03-17 05:41:18 +00:00
Bob Wilson
7a9bf0aa55 Remove a check that can no longer be true, after r84803. 
llvm-svn: 98694
 2010-03-16 23:40:32 +00:00
Evan Cheng
fb978ca2c5 Fix liveintervals handling of dbg_value instructions. 
llvm-svn: 98686
 2010-03-16 21:51:27 +00:00
Chris Lattner
cf7f134913 reapply r98656 unmodified, which exposed the asmprinter not 
handling constant unions.

llvm-svn: 98680
 2010-03-16 21:25:55 +00:00
Devang Patel
16c77b14f2 Ignore debug value instructions while analyzing BB for tail duplication. 
llvm-svn: 98675
 2010-03-16 21:02:07 +00:00
Daniel Dunbar
faade5305c Revert r98656, its breaking all over the place. 
llvm-svn: 98662
 2010-03-16 19:35:34 +00:00
Chris Lattner
7a96045d0a improve support for uniontype and ConstantUnion, patch by Tim Northover! 
llvm-svn: 98656
 2010-03-16 19:15:03 +00:00
Bill Wendling
44b1fe3ee1 Use getFirstTerminator(). 
llvm-svn: 98604
 2010-03-16 02:01:51 +00:00
Bill Wendling
5f6361464d When checking if something's killed, don't rely simply on whether it's marked as 
"used outside of the block". If the block ends in a return, then it won't be
used outside of it.

llvm-svn: 98599
 2010-03-16 01:05:35 +00:00
Chris Lattner
adff4d133f Fix the third (and last known) case of code update problems due 
to LLVM IR changes with addr label weirdness.  In the testcase, we
generate references to the two bb's when codegen'ing the first
function:

_test1:                                 ## @test1
	leaq	Ltmp0(%rip), %rax
..
	leaq	Ltmp1(%rip), %rax

Then continue to codegen the second function where the blocks
get merged.  We're now smart enough to emit both labels, producing
this code:

_test_fun:                              ## @test_fun
## BB#0:                                ## %entry
Ltmp1:                                  ## Block address taken
Ltmp0:
## BB#1:                                ## %ret
	movl	$-1, %eax
	ret

Rejoice.

llvm-svn: 98595
 2010-03-16 00:29:39 +00:00
Bill Wendling
5493cf6715 Now that the default for Darwin platforms is to place the LSDA into the TEXT 
section, remove the target-specific code that performs this.

llvm-svn: 98580
 2010-03-15 21:09:38 +00:00
Chris Lattner
45a0ae21b8 Implement support for the case when a reference to a addr-of-bb 
label is generated, but then the block is deleted.  Since the
value is undefined, we just emit the label right after the entry 
label of the function.  It might matter that the label is in the
same section as the function was afterall.

llvm-svn: 98579
 2010-03-15 20:39:00 +00:00
Chris Lattner
0ae3e35f22 use Mang->getSymbol() more. 
llvm-svn: 98577
 2010-03-15 20:37:38 +00:00
Devang Patel
db8c479e0d Create SDDbgValue for dbg_value intrinsics and remember its connections with DAG nodes. 
This is a work in progress. Patch by Dale Johannesen!

llvm-svn: 98568
 2010-03-15 19:15:44 +00:00
Chris Lattner
802ebf9561 Fix the case when a reference to an address taken BB is emitted in one 
function, then the BB is RAUW'd before the definition is emitted.  There
are still two cases not being handled, but this should improve us back to
the situation before I touched anything.

llvm-svn: 98566
 2010-03-15 19:09:43 +00:00
Chris Lattner
9f4f6954f7 revert r98550, it isn't necessary or sufficient. 
llvm-svn: 98558
 2010-03-15 18:42:01 +00:00
Devang Patel
3244526a3b Emit dwarf variable info communicated by code generator through DBG_VALUE machine instructions. 
This is a work in progress.

llvm-svn: 98556
 2010-03-15 18:33:46 +00:00
Chris Lattner
a8214d0775 don't eliminate address-taken blocks here. 
llvm-svn: 98550
 2010-03-15 16:37:42 +00:00
Chris Lattner
a8e4282df3 SIGN_EXTEND from the same type as the dest is valid. 
llvm-svn: 98548
 2010-03-15 16:15:56 +00:00
Chris Lattner
89c2d22d3d sink the call to VT.getSizeInBits() down into its uses, 
not all unary nodes necessarily have a simple result type.

llvm-svn: 98547
 2010-03-15 16:05:15 +00:00
Chris Lattner
b24982ba65 fix MCSectionELF to not leak memory, just like I did for MCSymbol. 
MCSectionMachO is already fine (yay for fixed size arrays?),
MCSectionCOFF still leaks.

llvm-svn: 98537
 2010-03-15 06:23:52 +00:00
Chris Lattner
37c5893297 remove dead method. 
llvm-svn: 98526
 2010-03-15 00:00:42 +00:00
Benjamin Kramer
be3d9aadbd Remove pointless forward declaration, MSVC got confused by this. 
llvm-svn: 98519
 2010-03-14 22:00:28 +00:00
Duncan Sands
217cec1786 Turn calls to copysignl into an FCOPYSIGN node. Handle FCOPYSIGN nodes 
with ppc_f128 type by having the type legalizer turn these back into a
call to copysignl.

llvm-svn: 98514
 2010-03-14 21:08:40 +00:00
Evan Cheng
f91618ae9f Rename SDDbgValue.h to SDNodeDbgValue.h for consistency. 
llvm-svn: 98513
 2010-03-14 19:56:39 +00:00
Chris Lattner
70eca8f78e fix ShrinkDemandedOps to not leave dead nodes around, 
fixing PR6607

llvm-svn: 98512
 2010-03-14 19:46:02 +00:00
Chris Lattner
5393443ed6 rewrite ShrinkDemandedOps to be faster and indent less, 
no functionality change.

llvm-svn: 98511
 2010-03-14 19:43:04 +00:00
Chris Lattner
f656ba2910 make -view-isel-dags print after the 'ShrinkDemandedOps' pass. 
llvm-svn: 98509
 2010-03-14 19:27:55 +00:00
Anton Korobeynikov
2cb4451ae6 Make default expansion for FP16 <-> FP32 nodes into libcalls 
llvm-svn: 98501
 2010-03-14 18:42:24 +00:00
Anton Korobeynikov
95f830f289 Add DAG nodes to represent FP16 <-> FP32 intrinsics 
llvm-svn: 98500
 2010-03-14 18:42:15 +00:00
Chris Lattner
2bdb0765f8 fix AsmPrinter::GetBlockAddressSymbol to always return a unique 
label instead of trying to form one based on the BB name (which
causes collisions if the name is empty).  This fixes PR6608

llvm-svn: 98495
 2010-03-14 17:53:23 +00:00
Chris Lattner
9331acc6d7 get MMI out of the label uniquing business, just go to MCContext 
to get unique assembler temporary labels.

llvm-svn: 98489
 2010-03-14 08:36:50 +00:00
Chris Lattner
5af59dfd1b don't use getDWLabel("label" anymore, always go through MMI. 
llvm-svn: 98485
 2010-03-14 08:18:13 +00:00
Chris Lattner
49130c09d0 remove now-dead code, all labels use MCSymbols. 
llvm-svn: 98484
 2010-03-14 08:17:53 +00:00
Chris Lattner
5ee7841af2 change SrcLineInfo to contain a label instead of a label ID. 
llvm-svn: 98483
 2010-03-14 08:15:55 +00:00
Chris Lattner
23ec7363b8 Now that DBG_LABEL is updated, we can finally make MachineMove 
contain an MCSymbol instead of a label index.

llvm-svn: 98482
 2010-03-14 08:12:40 +00:00
Chris Lattner
ba3b320aa3 change the DBG_LABEL MachineInstr to always be created 
with an MCSymbol instead of an immediate.

llvm-svn: 98481
 2010-03-14 07:56:48 +00:00
Chris Lattner
45aceed227 unbreak the build, grr symlinks. 
llvm-svn: 98477
 2010-03-14 07:36:49 +00:00
Chris Lattner
35d8b4e3e5 Change this code to allocate temporary labels from mccontext, 
not from MMI.

llvm-svn: 98475
 2010-03-14 07:29:45 +00:00
Chris Lattner
853b4f38df switch GC_LABEL to use an MCSymbol operand instead of a label ID operand. 
llvm-svn: 98474
 2010-03-14 07:27:07 +00:00
Chris Lattner
ee71bdd77f Fix some EH failures on NNT I introduced in r98461 
llvm-svn: 98471
 2010-03-14 07:02:50 +00:00
Chris Lattner
961e202df4 fix a bug I introduced in r98459, causing some NNT failures. 
llvm-svn: 98470
 2010-03-14 06:50:56 +00:00
Chris Lattner
5fef80c5aa change the LabelSDNode to be EHLabelSDNode and make it hold 
an MCSymbol.  Make the EH_LABEL MachineInstr hold its label
with an MCSymbol instead of ID.  Fix a bug in MMI.cpp which
would return labels named "Label4" instead of "label4".

llvm-svn: 98463
 2010-03-14 02:33:54 +00:00
Chris Lattner
de49dbc188 eliminate InvalidateLabel and LabelIDList from MMI and replace 
them with a counter.

llvm-svn: 98462
 2010-03-14 02:24:55 +00:00
Chris Lattner
f856c407af use Label->isDefined() instead of isLabelDeleted() now that we 
consistently use MCSymbol and only call this predicate after
they should have been emitted.

llvm-svn: 98461
 2010-03-14 02:20:58 +00:00
Chris Lattner
149cf816bb change EH related stuff (other than EH_LABEL) to use MCSymbol 
instead of label ID's.  This cleans up and regularizes a bunch 
of code and makes way for future progress.

Unfortunately, this pointed out to me that JITDwarfEmitter.cpp
is largely copy and paste from DwarfException/MachineModuleInfo
and other places.  This is very sad and disturbing. :(

One major change here is that TidyLandingPads moved from being
called in DwarfException::BeginFunction to being called in
DwarfException::EndFunction.  There should not be any 
functionality change from doing this, but I'm not an EH expert.

llvm-svn: 98459
 2010-03-14 01:41:15 +00:00
Chris Lattner
ceaa2343e7 eliminate the now-unneeded context argument of MBB::getSymbol() 
llvm-svn: 98451
 2010-03-13 21:04:28 +00:00
Chris Lattner
e7538fa303 rearrange MCContext ownership. Before LLVMTargetMachine created it 
and passing off ownership to AsmPrinter.  Now MachineModuleInfo
creates it and owns it by value.  This allows us to use MCSymbols
more consistently throughout the rest of the code generator, and
simplifies a bit of code.  This also allows MachineFunction to 
keep an MCContext reference handy, and cleans up the TargetRegistry
interfaces for AsmPrinters.

llvm-svn: 98450
 2010-03-13 20:55:24 +00:00
Chris Lattner
533e5c8ffd add support for MCSymbols as operands to MachineInstrs. 
llvm-svn: 98433
 2010-03-13 08:14:18 +00:00
Chris Lattner
f26e0336c2 simplify EmitFrameMoves to take BaseLabel in as a symbol 
instead of as a stem+idx pair, simplify the "is a new 
location" check to use symbol comparison.

llvm-svn: 98432
 2010-03-13 08:05:25 +00:00
Chris Lattner
ade5faaac4 factor some labels, simplify some code. 
llvm-svn: 98429
 2010-03-13 07:40:56 +00:00
Chris Lattner
836e8b5d1b various cleanups. 
llvm-svn: 98426
 2010-03-13 07:26:18 +00:00
Chris Lattner
bfdbd2f3a2 reimplement the string pool used for inlined function 
entries to not thrash std::strings and MCSymbols.

llvm-svn: 98415
 2010-03-13 02:17:42 +00:00
Chris Lattner
14ab9784f4 remove gone method, grr symlinks. 
llvm-svn: 98392
 2010-03-12 21:30:49 +00:00
Chris Lattner
7a2e6110bc inline the now-trivial implementation of GetGlobalValueSymbol into 
some of its callers.

llvm-svn: 98388
 2010-03-12 21:09:07 +00:00
Chris Lattner
c101ad818c give Mangler access to TargetData. 
llvm-svn: 98378
 2010-03-12 20:47:28 +00:00
Chris Lattner
01e70df9b2 finally give Mangler a getSymbol method, which returns an MCSymbol 
for a global instead of messing around with string buffers.

llvm-svn: 98366
 2010-03-12 18:55:20 +00:00
Chris Lattner
04cd48f865 remove dead code. 
llvm-svn: 98365
 2010-03-12 18:49:32 +00:00
Chris Lattner
956582f876 make the mangler take an MCContext instead of an MAI. 
No functionality change.

llvm-svn: 98363
 2010-03-12 18:44:54 +00:00
Chris Lattner
f4dce6a6d8 remove MAI argument from createAsmStreamer since it 
can get it from the context now.

llvm-svn: 98361
 2010-03-12 18:28:53 +00:00
Chris Lattner
4cc9a9004f fix a bug emitting .secrel32 that I introduced, PR6587, patch 
by A.Mazur!

llvm-svn: 98360
 2010-03-12 18:10:35 +00:00
Jeffrey Yasskin
824bd1b30d Free DbgScopes in DwarfDebug::endFunction(). Also increased the const-ness of 
several fields to make it easier to figure out where bugs might be creeping in.

llvm-svn: 98358
 2010-03-12 17:45:06 +00:00
Duncan Sands
086a80eee3 Revert turning copysignl into a COPYSIGN node for the moment: 
ppc calls copysignl with a 128 bit ppc long double, resulting
in a node that the type legalizer doesn't know how to expand.

llvm-svn: 98357
 2010-03-12 17:41:34 +00:00
Duncan Sands
aeed41b97a Now that it's supported, turn copysignl into a COPYSIGN node. 
llvm-svn: 98348
 2010-03-12 12:13:59 +00:00
Duncan Sands
5dcc3b328d Fix PR6522: implement copysign expansion for x86 long double 
(it seems that FreeBSD doesn't have copysignl).  Done by
removing a bunch of assumptions from the code.  This may also
help with sparc 128 bit floats.

llvm-svn: 98346
 2010-03-12 11:45:06 +00:00
Chris Lattner
80ab250a1c fix PR6577, a bug in sdbuilder lowering select instructions 
whose true value was not Val#0.

llvm-svn: 98336
 2010-03-12 07:15:36 +00:00
Devang Patel
93aa350213 There is no need to create specification DIE for definitions at DIFile level. 
llvm-svn: 98302
 2010-03-11 23:44:52 +00:00
Jakob Stoklund Olesen
1c8c4e2bb2 Extract methods from LocalRewriter::RewriteMBB bringing it down to 666 lines. 
llvm-svn: 98295
 2010-03-11 23:04:34 +00:00
Chris Lattner
debc026df2 change MCContext to always have an MCAsmInfo. 
llvm-svn: 98293
 2010-03-11 22:53:35 +00:00
Chris Lattner
5aefaeb42e fix a fixme in TargetLoweringObjectFile::getExprForDwarfReference 
where we used ot create an MCSymbol for ".".  Now emit an assembler
temporary label and reference it instead of "." textually.

rdar://7739457

llvm-svn: 98292
 2010-03-11 21:55:20 +00:00
Dan Gohman
6b1b9e37d7 Remove getWidenVectorType, which is no longer used. 
llvm-svn: 98289
 2010-03-11 21:39:57 +00:00
Chris Lattner
1ce9dfd498 rename getSymbolForDwarf* to getExprForDwarf* since it returns 
an MCExpr and not an MCSymbol.  Change it to take an MCStreamer,
which is currently unused.

No functionality change.

llvm-svn: 98278
 2010-03-11 19:41:58 +00:00
Evan Cheng
e1c0438e7b In case of tail call size of Ins and InVals may not match. 
llvm-svn: 98277
 2010-03-11 19:38:18 +00:00
Jeffrey Yasskin
239beb0e85 Avoid leaking CompileUnits in DwarfDebug.cpp. 
llvm-svn: 98268
 2010-03-11 18:29:55 +00:00
Richard Osborne
9c71a0a5ba Add a new jump table encoding to indicate jump tables entries 
are inside the function by the target at the point of use.

llvm-svn: 98255
 2010-03-11 14:58:16 +00:00
Evan Cheng
4ef6d8fa15 The check for coalescing a virtual register to a physical register, e.g. 
cl = EXTRACT_SUBREG reg1024, 1, is overly conservative. It should check
for overlaps of vr's live interval with the super registers of the
physical register (ECX in this case) and let JoinIntervals() handle checking
the coalescing feasibility against the physical register (cl in this case).

llvm-svn: 98251
 2010-03-11 08:20:21 +00:00
Daniel Dunbar
b24134670c Remove dead include. 
llvm-svn: 98225
 2010-03-11 02:28:48 +00:00
Dale Johannesen
15ff882aa1 Fix debug_value handling. 
llvm-svn: 98224
 2010-03-11 02:10:24 +00:00
Daniel Dunbar
687d99cfa8 MC: Provide MCAssembler with a TargetAsmBackend. 
llvm-svn: 98222
 2010-03-11 01:34:27 +00:00
Jakob Stoklund Olesen
87f8d0c4f2 VirtRegRewriter spring cleaning. No functional change. 
Move methods out of line and M-x whitespace-cleanup.
Promote common method arguments to member variables.

llvm-svn: 98207
 2010-03-11 00:11:33 +00:00
Chris Lattner
7cd70b8066 fix PR6533 by updating the br(xor) code to remember the case 
when it looked past a trunc.

llvm-svn: 98203
 2010-03-10 23:46:44 +00:00
Dale Johannesen
29afbd39e4 Cosmetic: lengthen names and improve comments. 
llvm-svn: 98202
 2010-03-10 23:37:24 +00:00
Bill Wendling
df325dee9d Add a bit along with the MCSymbols stored in the MachineModuleInfo maps that 
indicates that an MCSymbol is external or not. (It's true if it's external.)
This will be used to specify the correct information to add to non-lazy
pointers. That will be explained further when this bit is used.

llvm-svn: 98199
 2010-03-10 22:34:10 +00:00
Dale Johannesen
987770c05d Progress towards shepherding debug info through SelectionDAG. 
No functional effect yet.  This is still evolving and should
not be viewed as final.

llvm-svn: 98195
 2010-03-10 22:13:47 +00:00
Dan Gohman
4c22c7a665 Fix another bitwidth calculation to handle vector types; based on a 
patch by Micah Villmow for PR6572.

llvm-svn: 98188
 2010-03-10 21:04:53 +00:00
Dale Johannesen
5edf11aad3 Fix another place where DEBUG_VALUE affected codegen. 
llvm-svn: 98181
 2010-03-10 19:57:56 +00:00
Dale Johannesen
b68b96c0f6 Fix a bug in DEBUG_VALUE handling Devang ran into. 
I'll get this loop right yet.

llvm-svn: 98155
 2010-03-10 15:06:26 +00:00
Chris Lattner
e34dde1e83 move three lowering hooks from MAI to TLOF and make one of them 
semantic instead of syntactic.  This completes MCization of
darwin/x86[-64]!

llvm-svn: 98145
 2010-03-10 07:20:42 +00:00
Dale Johannesen
c9611b6d0a This survived a bootstrap, so let's try 98104 again. 
llvm-svn: 98137
 2010-03-10 05:45:47 +00:00
Evan Cheng
668ceddeec Enable machine cse pass. 
llvm-svn: 98132
 2010-03-10 03:07:41 +00:00
Chris Lattner
270a61db96 mcize the rest of EH emission, only one more directive missing 
for darwin/x86 to be completely mcized.

llvm-svn: 98130
 2010-03-10 02:48:06 +00:00
Chris Lattner
60b9c3a333 add missing filename! 
llvm-svn: 98125
 2010-03-10 02:29:31 +00:00
Chris Lattner
ed0b8d36e4 set the temporary bit on MCSymbols correctly. 
llvm-svn: 98124
 2010-03-10 02:25:11 +00:00
Evan Cheng
f75338cfcb Add a couple more heuristics to neuter machine cse some more. 
1. Be careful with cse "cheap" expressions. e.g. constant materialization. Only cse them when the common expression is local or in a direct predecessor. We don't want cse of cheap instruction causing other expressions to be spilled.
2. Watch out for the case where the expression doesn't itself uses a virtual register. e.g. lea of frame object. If the common expression itself is used by copies (common for passing addresses to function calls), don't perform the cse. Since these expressions do not use a register, it creates a live range but doesn't close any, we want to be very careful with increasing register pressure.

Note these are heuristics so machine cse doesn't make register allocator unhappy. Once we have proper live range splitting and re-materialization support in place, these should be evaluated again.

Now machine cse is almost always a win on llvm nightly tests on x86 and x86_64.

llvm-svn: 98121
 2010-03-10 02:12:03 +00:00
Chris Lattner
e284f80948 inline away a form of IsPCRelative, eliminating the 
dead IsPCRel argument.

llvm-svn: 98117
 2010-03-10 01:17:49 +00:00
Chris Lattner
cff79125fb add some fixme's for MCizing. EH still has a few things that 
need to be MCized, but the last debug info thing are LEB and
cygwin specific (which the MC api doesn't support yet) and
one specific form of EmitReference which I'll tackle next.

llvm-svn: 98116
 2010-03-10 01:04:13 +00:00
Jim Grosbach
aa8617389c Clear up the last (famous last words) frame index value reuse issues for Thumb1. 
llvm-svn: 98109
 2010-03-10 00:13:42 +00:00
Dale Johannesen
3800f76c1a Speculatively revert 98104; could be what's causing crashes 
llvm-svn: 98108
 2010-03-10 00:11:34 +00:00
Chris Lattner
915a362856 mcize uses of PrintRelDirective and eliminate it. 
llvm-svn: 98107
 2010-03-10 00:09:21 +00:00
Chris Lattner
84561d5f46 inline the bool form of PrintRelDirective away, leaving just the unsigned form. 
llvm-svn: 98106
 2010-03-09 23:54:52 +00:00
Chris Lattner
d5d96f76d0 eliminate EOL, adding all comments with the OutStreamer.AddComment 
method.  With this, comments should end up on the same lines as the .byte
directives (for example) and we now get no output with:

$ llc CodeGen/X86/2009-02-12-DebugInfoVLA.ll -o - -filetype=null -asm-verbose

woot.

llvm-svn: 98105
 2010-03-09 23:52:58 +00:00
Dale Johannesen
02f3bfbecc Ever more complicated DEBUG_VALUE fixes for branch folding. 
llvm-svn: 98104
 2010-03-09 23:52:37 +00:00
Chris Lattner
58829fcfe9 eliminate a bunch of \n's that are being printed to O. Next up is to kill 
off "EOL".

llvm-svn: 98102
 2010-03-09 23:38:23 +00:00
Chris Lattner
2ede66313d convert the non-"ispcrel" case of EmitReference to MC, 
significant debug info testcases are now all going through
MCStreamer, though they print a lot of extraneous newlines to "O".

llvm-svn: 98101
 2010-03-09 23:19:15 +00:00
Jim Grosbach
f328c5334f Change the Value argument to eliminateFrameIndex to a type-tagged value. This 
is preparatory to having PEI's scavenged frame index value reuse logic
properly distinguish types of frame values (e.g., whether the value is
stack-pointer relative or frame-pointer relative).

No functionality change.

llvm-svn: 98086
 2010-03-09 21:45:49 +00:00
Evan Cheng
625a7248d6 Allow more cross-rc coalescing. 
llvm-svn: 98048
 2010-03-09 06:38:17 +00:00
Chris Lattner
d51f1bdc58 reapply r98035: 
Now that setStartLabel takes an MCSymbol, we can de-ID'ize
beginScope and RecordSourceLine.

llvm-svn: 98047
 2010-03-09 04:54:43 +00:00
Chris Lattner
43b812b6a8 add some extra checks. I'm not sure why, but this does unbreak a 
failure remaining on mainline.

llvm-svn: 98046
 2010-03-09 04:48:35 +00:00
Evan Cheng
1b43094229 Revert accidental commit. 
llvm-svn: 98045
 2010-03-09 04:04:38 +00:00
Jakob Stoklund Olesen
cc6b4fdf11 Don't do illegal cross-class coalescing. 
llvm-svn: 98044
 2010-03-09 03:56:06 +00:00
Evan Cheng
8277a90f75 - Make the machine cse dumb coalescer (as opposed to the more awesome simple 
coalescer) handle sub-register classes.
- Add heuristics to avoid non-profitable cse. Given the current lack of live
  range splitting, avoid cse when an expression has PHI use and the would be
  new use is in a BB where the expression wasn't already being used.

llvm-svn: 98043
 2010-03-09 03:21:12 +00:00
Eric Christopher
8875eba06a Speculatively revert r98035. It appears to have caused a set of buildbot 
failures.

llvm-svn: 98039
 2010-03-09 02:36:31 +00:00
Dan Gohman
0d69c61fec Attempt to make this debug output meaningful, both in the case of 
multibyte opcodes and in the case of multiple scopes.

llvm-svn: 98036
 2010-03-09 02:15:05 +00:00
Chris Lattner
c4a74bf48b Now that setStartLabel takes an MCSymbol, we can de-ID'ize 
beginScope and RecordSourceLine.

llvm-svn: 98035
 2010-03-09 02:08:02 +00:00
Chris Lattner
f7ad2622cc change DbgScope to keep track of the start/end label as MCSymbol* 
now that the dependence on ID is removed from MMI.

llvm-svn: 98034
 2010-03-09 01:58:53 +00:00
Chris Lattner
a72a32a06b remove a useless optimization: now that label replacement never 
happens, the start/end of a scope can never be the same.

llvm-svn: 98032
 2010-03-09 01:52:43 +00:00
Chris Lattner
d4ce7a73e8 strength reduce MMI::MappedLabel to MMI::isLabelDeleted, 
and add a FIXME about how we are eventually going to zap this
lookup table once mc world domination is complete.

llvm-svn: 98031
 2010-03-09 01:51:43 +00:00
Chris Lattner
4494a28b3c mcstreamerize AsmPrinter::printLabel. 
llvm-svn: 98025
 2010-03-09 01:02:30 +00:00
Jakob Stoklund Olesen
3d0157ae7e Disable physical register coalescing when the number of live ranges for the 
physreg becomes ridiculously high.

std::upper_bound may be log(N), but for sufficiently large live intervals, it
becomes log(N)*cachemiss = a long long time.

This patch improves coalescer time by 4500x for a function with 20000
function calls. The generated code is different, but not significantly worse -
the allocator hints are almost as good as physreg coalescing anyway.

llvm-svn: 98023
 2010-03-09 00:59:48 +00:00
Chris Lattner
8147c144d9 add a EmitSymbolValue convenience method to MCStreamer. 
llvm-svn: 98017
 2010-03-09 00:39:24 +00:00
Chris Lattner
db93b91db1 make InlineInfoLabels hold MCSymbol*'s, avoiding 
recomputation of the labels.

llvm-svn: 98016
 2010-03-09 00:31:02 +00:00
Chris Lattner
a583d01e79 mc'ize the last use of PrintLabelName and eliminate PrintLabelName. 
llvm-svn: 98015
 2010-03-09 00:26:09 +00:00
Chris Lattner
644850459d eliminate an argument from PrintRelDirective, sinking 
the one special case into EmitSectionOffset.  MCize
the non-special case in EmitSectionOffset.

llvm-svn: 98014
 2010-03-09 00:17:58 +00:00
Dan Gohman
cc7ed51fa3 Print the correct index in the "match failed at index" message. 
llvm-svn: 98013
 2010-03-09 00:07:36 +00:00
Chris Lattner
06f5a077de remove the suffix form of PrintLabelName, which was only 
used for 'flavor'.

llvm-svn: 98012
 2010-03-09 00:00:57 +00:00
Chris Lattner
183e87502e now that the debug and eh emitters use a common .set counter, 
we can eliminate "flavor".

llvm-svn: 98011
 2010-03-09 00:00:15 +00:00
Chris Lattner
99ca33d324 move .set generation out of DwarfPrinter into AsmPrinter and 
MCize it.

llvm-svn: 98010
 2010-03-08 23:58:37 +00:00
Evan Cheng
8afdea401a Don't waste time trying to CSE labels, phis, inline asm. Definitely avoid cse implicit-def for obvious performance reason. 
llvm-svn: 98009
 2010-03-08 23:49:12 +00:00
Evan Cheng
ca86fb517c Restrict machine cse to really trivial coalescing. Leave the heavy lifting to a real coalescer. 
llvm-svn: 98007
 2010-03-08 23:28:08 +00:00
Chris Lattner
10d571f349 simplify EmitSectionOffset to always use .set if it is 
available, the only thing this affects is that we produce
.set in one case we didn't before, which shouldn't harm
anything.  Make EmitSectionOffset call EmitDifference
instead of duplicating it.

llvm-svn: 98005
 2010-03-08 23:23:25 +00:00
Chris Lattner
9dfcbeaca1 Remove a version of EmitDifference. 
llvm-svn: 98002
 2010-03-08 23:02:59 +00:00
Chris Lattner
81bf3f933d eliminate a form of PrintLabelName. 
llvm-svn: 97999
 2010-03-08 22:52:49 +00:00
Chris Lattner
bbd8af41d3 remove another form of EmitReference. 
llvm-svn: 97998
 2010-03-08 22:50:36 +00:00
Chris Lattner
4ccfa3d315 eliminate the non-MCSymbol versions of EmitReference. 
llvm-svn: 97997
 2010-03-08 22:47:57 +00:00
Chris Lattner
2bbaab13d8 mc'ize EmitLabel. 
llvm-svn: 97996
 2010-03-08 22:44:40 +00:00
Chris Lattner
bfd38653e6 merge DIEObjectLabel and DIEDwarfLabel into DIELabel. 
Yes, DIE you fiendish labels, die all of you.

llvm-svn: 97995
 2010-03-08 22:31:46 +00:00
Chris Lattner
cbce2df237 elimiante the DWLabel class, using MCSymbol instead. Start 
switching some stuff over to passing around MCSymbol* instead
of stem+ID.

llvm-svn: 97993
 2010-03-08 22:23:36 +00:00
Devang Patel
ea239f6e0a Derive DIType from DIScope. This simplifies getContext() where for members the context is a type. This also eliminates need of CompileUnitMaps maintained by dwarf writer. 
llvm-svn: 97990
 2010-03-08 22:02:50 +00:00
Devang Patel
176dc29a62 Avoid using DIDescriptor.isNull(). 
This is a first step towards eliminating checks in Descriptor constructors.

llvm-svn: 97975
 2010-03-08 20:52:55 +00:00
Devang Patel
a716e313d6 Revert r97947. 
llvm-svn: 97963
 2010-03-08 19:20:38 +00:00
Devang Patel
1527b2657b Avoid using DIDescriptor.isNull(). 
This is a first step towards eliminating unncessary constructor checks in light weight DIDescriptor wrappers.

llvm-svn: 97947
 2010-03-08 18:25:48 +00:00
Dale Johannesen
b87c6c82e6 Add Order to SDDbgValue 
llvm-svn: 97939
 2010-03-08 05:39:50 +00:00
Dale Johannesen
d610f0a82a Fix dbg value handling in tail merging. 
llvm-svn: 97938
 2010-03-08 05:38:13 +00:00
Douglas Gregor
c52fc8f922 Revert r97917, which was causing Clang Debug self-host failures. 
llvm-svn: 97932
 2010-03-08 02:58:37 +00:00
Jeffrey Yasskin
2c23c1d1a1 Roll r97906 forward again, without double-deleting CompileUnit::IndexTyDie. 
llvm-svn: 97917
 2010-03-07 17:10:13 +00:00
Chris Lattner
f3ee582f23 Use Other as a sentinel instead of iAny. 
llvm-svn: 97914
 2010-03-07 07:45:08 +00:00
Jeffrey Yasskin
c68dc05867 _2_ gcc crashes, ah, ah, ah... 
(Rolling back r97906.)

llvm-svn: 97909
 2010-03-07 07:16:49 +00:00
Jeffrey Yasskin
c0123f0033 Avoid leaking CompileUnits and DbgScopes from DwarfDebug. Leaks found by Valgrind! 
llvm-svn: 97906
 2010-03-07 06:55:35 +00:00
Chris Lattner
6dbe3336f9 tidy up 
llvm-svn: 97895
 2010-03-07 04:28:09 +00:00
Evan Cheng
d8e0138666 Don't update physical register def. 
llvm-svn: 97861
 2010-03-06 01:14:19 +00:00
Dale Johannesen
f0c8e76a85 Add some new bits of debug info handling. No 
functional change yet.

llvm-svn: 97855
 2010-03-06 00:03:23 +00:00
Dan Gohman
00a652eea0 Reapply r97778 and r97779, enabled only for unsigned i64 to f64 
conversions.

llvm-svn: 97854
 2010-03-06 00:00:55 +00:00
Charles Davis
faa2f44081 Don't emit global symbols into the (__TEXT,__ustring) section on Darwin. This 
is a workaround for <rdar://problem/7672401/> (which I filed).

This let's us build Wine on Darwin, and it gets the Qt build there a little bit
further (so Doug says).

llvm-svn: 97845
 2010-03-05 22:28:45 +00:00
Jakob Stoklund Olesen
4e033d2070 Better handling of dead super registers in LiveVariables. We used to do this: 
CALL ... %RAX<imp-def>
   ... [not using %RAX]
   %EAX = ..., %RAX<imp-use, kill>
   RET %EAX<imp-use,kill>

Now we do this:

   CALL ... %RAX<imp-def, dead>
   ... [not using %RAX]
   %EAX = ...
   RET %EAX<imp-use,kill>

By not artificially keeping %RAX alive, we lower register pressure a bit.

The correct number of instructions for 2008-08-05-SpillerBug.ll is obviously
55, anybody can see that. Sheesh.

llvm-svn: 97838
 2010-03-05 21:49:17 +00:00
Jakob Stoklund Olesen
1fce28720a We don't really care about correct register liveness information after the 
post-ra scheduler has run. Disable the verifier checks that late in the game.

llvm-svn: 97837
 2010-03-05 21:49:13 +00:00
Jakob Stoklund Olesen
67476519d7 Avoid creating bad PHI instructions when BR is being const-folded. 
llvm-svn: 97836
 2010-03-05 21:49:10 +00:00
Chris Lattner
80aaccb987 Fix PR6497, a bug where we'd fold a load into an addc 
node which has a flag.  That flag in turn was used by an
already-selected adde which turned into an ADC32ri8 which
used a selected load which was chained to the load we
folded.  This flag use caused us to form a cycle.  Fix
this by not ignoring chains in IsLegalToFold even in
cases where the isel thinks it can.

llvm-svn: 97791
 2010-03-05 06:19:13 +00:00
Chris Lattner
d7353d219f inline a small function with one call site. 
llvm-svn: 97789
 2010-03-05 05:49:45 +00:00
Dan Gohman
ebdb1743d3 Revert r97778 and r97779. They're somehow breaking llvm-gcc builds. 
llvm-svn: 97781
 2010-03-05 02:40:23 +00:00
Dan Gohman
0a01ba144d Fix these constants to be more portable. 
llvm-svn: 97779
 2010-03-05 02:13:10 +00:00
Dan Gohman
e5b9ea020f Rewrite i64-to-f64 conversion using an algorithm which handles 
rounding correctly. This implementation is a generalization of
the x86_64 code in compiler-rt.

This fixes rdar://7683708.

llvm-svn: 97778
 2010-03-05 02:00:46 +00:00
Dale Johannesen
b1fd546dd0 Fix some more places where dbg_value affected codegen. 
llvm-svn: 97765
 2010-03-05 00:02:59 +00:00
Jeffrey Yasskin
e4e796c00b Fix memcheck-found leaks: one false positive from using new[], and one true 
positive where pointers would be leaked on llvm_shutdown.

llvm-svn: 97759
 2010-03-04 22:15:01 +00:00
Jim Grosbach
330c1b777e For SJLJ exception handling, make sure that all calls that are not marked 
as nounwind are marked with a -1 call-site value. This is necessary to, for
example, correctly process exceptions thrown from within an "unexpected"
execption handler (see SingleSource/Regression/C++/EH/expection_spec_test.cpp).

llvm-svn: 97757
 2010-03-04 22:07:46 +00:00
Evan Cheng
9b35d553df Run machine licm before machine cse to avoid messing up licm opportunities. 
llvm-svn: 97752
 2010-03-04 21:28:09 +00:00
Evan Cheng
a9828532ff Avoid cse load instructions unless they are known to be invariant loads. 
llvm-svn: 97747
 2010-03-04 21:18:08 +00:00
Chris Lattner
860cbbb031 add a statistic for # times fastisel fails. 
llvm-svn: 97738
 2010-03-04 19:46:56 +00:00
Dan Gohman
632e2a2b8c Fix a typo Duncan noticed. 
llvm-svn: 97735
 2010-03-04 19:11:28 +00:00
Evan Cheng
df8f4b6142 Rename -machine-cse to -enable-machine-cse. 
llvm-svn: 97713
 2010-03-04 02:08:04 +00:00
Evan Cheng
9120412740 Look ahead a bit to determine if a physical register def that is not marked dead is really alive. This is necessary to catch a lot of common cse opportunities for targets like x86. 
llvm-svn: 97706
 2010-03-04 01:33:55 +00:00
Chris Lattner
2bbca2de9e change the new isel matcher to emit ComplexPattern matches 
as the very last thing before node emission.  This should
dramatically reduce the number of times we do 'MatchAddress'
on X86, speeding up compile time.  This also improves comments
in the tables and shrinks the table a bit, now down to 
80506 bytes for x86.

llvm-svn: 97703
 2010-03-04 01:23:08 +00:00
Dan Gohman
9f6d374ab7 Fix more code to work properly with vector operands. Based on 
a patch my Micah Villmow for PR6465.

llvm-svn: 97692
 2010-03-04 00:23:16 +00:00
Chris Lattner
19007009c8 inline CannotYetSelectIntrinsic into CannotYetSelect and simplify. 
llvm-svn: 97690
 2010-03-04 00:21:16 +00:00
Evan Cheng
9729f2a2b4 Fix a logic error. An instruction that has a live physical register def cannot be CSE'ed, but it *can* be used to replace a common subexpression. 
llvm-svn: 97688
 2010-03-03 23:59:08 +00:00
Evan Cheng
7310728cf9 Remove PHINodeTraits and use MachineInstrExpressionTrait instead. 
llvm-svn: 97687
 2010-03-03 23:55:49 +00:00
Evan Cheng
8c608dffd5 Move MachineInstrExpressionTrait::getHashValue() out of line so it can skip over only virtual register defs. This matches what isEqual() is doing. 
llvm-svn: 97680
 2010-03-03 23:37:30 +00:00
Evan Cheng
36a7a94029 Re-apply r97667 but with a little bit of thought put into the patch. This implements a special DenseMapInfo trait for DenseMap<MachineInstr*> that compare the value of the MachineInstr rather than the pointer value. Since the hashing and equality test functions ignore defs it's useful for doing CSE kind optimization. 
llvm-svn: 97678
 2010-03-03 23:27:36 +00:00
Dan Gohman
fe6b41169a Revert 97667. It broke a bunch of tests. 
llvm-svn: 97673
 2010-03-03 22:40:03 +00:00
Evan Cheng
57064125c4 Fix funky indentation and add comments. 
llvm-svn: 97670
 2010-03-03 21:54:14 +00:00
Evan Cheng
338c7aed4c Move DenseMapInfo for MachineInstr* to MachineInstr.h 
llvm-svn: 97667
 2010-03-03 21:47:16 +00:00
Dan Gohman
cdc603ecae Fix a bug in SelectionDAG's ReplaceAllUsesWith in the case where 
CSE and recursive RAUW calls delete a node from the use list,
invalidating the use list iterator. There's currently no known
way to reproduce this in an unmodified LLVM, however there's no
fundamental reason why a SelectionDAG couldn't be formed which
would trigger this case.

llvm-svn: 97665
 2010-03-03 21:33:37 +00:00
Evan Cheng
918ac5d21f Machine CSE work in progress. It's doing some CSE now. But implicit def of physical registers are getting in the way. 
llvm-svn: 97664
 2010-03-03 21:20:05 +00:00
Evan Cheng
810a26689e Add MachineRegisterInfo::hasOneUse and hasOneNonDBGUse. 
llvm-svn: 97663
 2010-03-03 21:18:38 +00:00
Bill Wendling
f0bd51c4e3 Revert... 
--- Reverse-merging r97592 into '.':
U    lib/CodeGen/TargetLoweringObjectFileImpl.cpp

llvm-svn: 97657
 2010-03-03 19:31:05 +00:00
Chris Lattner
9e7f00c3aa add some of the more obscure predicate types to the 
Scope accelerator.

llvm-svn: 97652
 2010-03-03 07:46:25 +00:00
Chris Lattner
9889ed8c45 speed up scope node processing: if the first element of a scope 
entry we're about to process is obviously going to fail, don't
bother pushing a scope only to have it immediately be popped.
This avoids a lot of scope stack traffic in common cases.

Unfortunately, this requires duplicating some of the predicate
dispatch.  To avoid duplicating the actual logic I pulled each
predicate out to its own static function which gets used in
both places.

llvm-svn: 97651
 2010-03-03 07:31:15 +00:00
Chris Lattner
92a814205f introduce a new SwitchTypeMatcher node (which is analogous to 
SwitchOpcodeMatcher) and have DAGISelMatcherOpt form it.  This
speeds up selection, particularly for X86 which has lots of 
variants of instructions with only type differences.

llvm-svn: 97645
 2010-03-03 06:28:15 +00:00
Evan Cheng
13f096bf0a Work in progress. Finding some cse now. 
llvm-svn: 97635
 2010-03-03 02:48:20 +00:00
Bill Wendling
65baaf9499 Use APInt instead of zext value. 
llvm-svn: 97631
 2010-03-03 01:58:01 +00:00
Evan Cheng
70b7ecdef9 - Change MachineInstr::isIdenticalTo to take a new option that determines whether it should skip checking defs or at least virtual register defs. This subsumes part of the TargetInstrInfo::isIdentical functionality. 
- Eliminate TargetInstrInfo::isIdentical and replace it with produceSameValue. In the default case, produceSameValue just checks whether two machine instructions are identical (except for virtual register defs). But targets may override it to check for unusual cases (e.g. ARM pic loads from constant pools).

llvm-svn: 97628
 2010-03-03 01:44:33 +00:00
Evan Cheng
ae376081d2 Add an option to enable machine cse (it's not doing anything yet. 
llvm-svn: 97627
 2010-03-03 01:38:35 +00:00
Bill Wendling
d1f658563d This test case: 
long test(long x) { return (x & 123124) | 3; }

Currently compiles to:

_test:
        orl     $3, %edi
        movq    %rdi, %rax
        andq    $123127, %rax
        ret

This is because instruction and DAG combiners canonicalize

  (or (and x, C), D) -> (and (or, D), (C | D))

However, this is only profitable if (C & D) != 0. It gets in the way of the
3-addressification because the input bits are known to be zero.

llvm-svn: 97616
 2010-03-03 00:35:56 +00:00
Chris Lattner
9c9c1158cb Fix some issues in WalkChainUsers dealing with 
CopyToReg/CopyFromReg/INLINEASM.  These are annoying because
they have the same opcode before an after isel.  Fix this by
setting their NodeID to -1 to indicate that they are selected,
just like what automatically happens when selecting things that
end up being machine nodes.

With that done, give IsLegalToFold a new flag that causes it to
ignore chains.  This lets the HandleMergeInputChains routine be
the one place that validates chains after a match is successful,
enabling the new hotness in chain processing.  This smarter
chain processing eliminates the need for "PreprocessRMW" in the
X86 and MSP430 backends and enables MSP to start matching it's
multiple mem operand instructions more aggressively.

I currently #if out the dead code in the X86 backend and MSP 
backend, I'll remove it for real in a follow-on patch.

The testcase changes are:
  test/CodeGen/X86/sse3.ll: we generate better code
  test/CodeGen/X86/store_op_load_fold2.ll: PreprocessRMW was 
      miscompiling this before, we now generate correct code
      Convert it to filecheck while I'm at it.
  test/CodeGen/MSP430/Inst16mm.ll: Add a testcase for mem/mem
      folding to make anton happy. :)

llvm-svn: 97596
 2010-03-02 22:20:06 +00:00
Bill Wendling
1ad5825d64 Okay. One last attempt: 
Place the LSDA into the TEXT section on Mach-O. This saves space.

llvm-svn: 97592
 2010-03-02 21:50:35 +00:00
Chris Lattner
0c14477270 run HandleMergeInputChains even if we only have one input chain. 
llvm-svn: 97581
 2010-03-02 19:34:59 +00:00