Commit Graph

8585 Commits

Author SHA1 Message Date
Jim Grosbach
b123a9cbc0 Convert to FileCheck
llvm-svn: 89007
2009-11-17 00:20:26 +00:00
Jim Grosbach
2f09113304 Convert to FileCheck
llvm-svn: 89002
2009-11-17 00:03:38 +00:00
Jim Grosbach
299e4e76c4 Cleanup. Missed removing these when converting. Oops.
llvm-svn: 89001
2009-11-17 00:00:33 +00:00
Dan Gohman
c2979de134 Fix this test - there don't appear to be any actual Reload Reuses
in this testcase.

llvm-svn: 88998
2009-11-16 23:49:55 +00:00
Dan Gohman
c35e84e1f5 Revert r87049, which was the workaround for the regression triggered
by the recent FixedStackPseudoSourceValue-related changes, now that
the specific bug that affected it is fixed, in r88954.

llvm-svn: 88997
2009-11-16 23:43:42 +00:00
Jeffrey Yasskin
9f05d972b7 Revert the test from r88984. It relies on being able to mmap 16GB of
address space (though it only uses a small fraction of that), and the
buildbots disallow that.

Also add a comment to the Makefile's ulimit line warning future
developers that changing it won't work.

llvm-svn: 88994
2009-11-16 23:32:30 +00:00
Jim Grosbach
95cf7fad36 Convert to FileCheck
llvm-svn: 88991
2009-11-16 23:19:29 +00:00
Jeffrey Yasskin
0f846dbb3e Make X86-64 in the Large model always emit 64-bit calls.
The large code model is documented at
http://www.x86-64.org/documentation/abi.pdf and says that calls should
assume their target doesn't live within the 32-bit pc-relative offset
that fits in the call instruction.

To do this, we turn off the global-address->target-global-address
conversion in X86TargetLowering::LowerCall(). The first attempt at
this broke the lazy JIT because it can separate the movabs(imm->reg)
from the actual call instruction. The lazy JIT receives the address of
the movabs as a relocation and needs to record the return address from
the call; and then when that call happens, it needs to patch the
movabs with the newly-compiled target. We could thread the call
instruction into the relocation and record the movabs<->call mapping
explicitly, but that seems to require at least as much new
complication in the code generator as this change.

To fix this, we make lazy functions _always_ go through a call
stub. You'd think we'd only have to force lazy calls through a stub on
difficult platforms, but that turns out to break indirect calls
through a function pointer. The right fix for that is to distinguish
between calls and address-of operations on uncompiled functions, but
that's complex enough to leave for someone else to do.

Another attempt at this defined a new CALL64i pseudo-instruction,
which expanded to a 2-instruction sequence in the assembly output and
was special-cased in the X86CodeEmitter's emitInstruction()
function. That broke indirect calls in the same way as above.

This patch also removes a hack forcing Darwin to the small code model.
Without far-call-stubs, the small code model requires things of the
JITMemoryManager that the DefaultJITMemoryManager can't provide.

Thanks to echristo for lots of testing!

llvm-svn: 88984
2009-11-16 22:41:33 +00:00
Evan Cheng
78be20d62e - Check memoperand alignment instead of checking stack alignment. Most load / store folding instructions are not referencing spill stack slots.
- Mark MOVUPSrm re-materializable.

llvm-svn: 88974
2009-11-16 21:56:03 +00:00
Jim Grosbach
f4abb1280a Convert to FileCheck
llvm-svn: 88947
2009-11-16 20:04:15 +00:00
Lang Hames
6a5810c037 Added a testcase for PR5495.
llvm-svn: 88946
2009-11-16 20:03:13 +00:00
Jim Grosbach
deee4fbd5d Convert to FileCheck
llvm-svn: 88942
2009-11-16 19:46:46 +00:00
Jim Grosbach
0ba7bb08d7 tbb opt off by default
llvm-svn: 88921
2009-11-16 17:24:45 +00:00
David Greene
6469fa6824 Support spill comments.
Have the asm printer emit a comment if an instruction is a spill or
reload and have the spiller mark copies it introdues so the asm printer
can also annotate those.

llvm-svn: 88911
2009-11-16 15:12:23 +00:00
Evan Cheng
ea46259f53 Check if subreg index is zero.
llvm-svn: 88899
2009-11-16 06:31:49 +00:00
Evan Cheng
2fa416debd For some targets, a copy can use a register multiple times, e.g. ppc.
llvm-svn: 88895
2009-11-16 05:52:06 +00:00
Evan Cheng
5c06a152a8 xfail for now. It has been failing.
llvm-svn: 88892
2009-11-16 05:44:04 +00:00
Bruno Cardoso Lopes
21ca44ba49 - Fix a small bug while handling target constant pools (one param was missing).
- Add a smarter constant pool loading, instead of:

lui $2, %hi($CPI1_0)
addiu $2, $2, %lo($CPI1_0)
lwc1 $f0, 0($2)

Generate:

lui $2, %hi($CPI1_0)
lwc1 $f0, %lo($CPI1_0)($2)

llvm-svn: 88886
2009-11-16 04:33:42 +00:00
Jim Grosbach
1aa571da3c Detect need for autoalignment of the stack earlier to catch spills more
conservatively. eliminateFrameIndex() machinery adjust to handle addr mode
6 (vld1/vst1) used for spills. Fix tests to expect aligned Q-reg spilling

llvm-svn: 88874
2009-11-15 21:45:34 +00:00
Nick Lewycky
f05946faff Revert r88830 and r88831 which appear to have caused a selfhost buildbot some
grief. I suspect this patch merely exposed a bug else.

llvm-svn: 88841
2009-11-15 07:47:32 +00:00
Nick Lewycky
62f00c17eb Correct typo.
llvm-svn: 88831
2009-11-15 06:16:57 +00:00
Nick Lewycky
14a2122db3 Teach instcombine to look for booleans in wider integers when it encounters a
zext(icmp). It may be able to optimize that away. This fixes one of the cases
in PR5438.

llvm-svn: 88830
2009-11-15 05:55:17 +00:00
Jim Grosbach
6028068e88 remove xfail
llvm-svn: 88817
2009-11-14 21:57:35 +00:00
Richard Osborne
8748f55236 Add XCore support for arbitrary-sized aggregate returns.
llvm-svn: 88802
2009-11-14 19:33:35 +00:00
Nick Lewycky
fba1efd298 Teach BasicAA that a constant expression can't alias memory provably not
allocated until runtime (such as an alloca). Patch by Hans Wennborg!

llvm-svn: 88760
2009-11-14 06:15:14 +00:00
Evan Cheng
b8c04e1226 Added getSubRegIndex(A,B) that returns subreg index of A to B. Use it to replace broken code in VirtRegRewriter.
llvm-svn: 88753
2009-11-14 03:42:17 +00:00
Evan Cheng
9b46e74f42 - Change TargetInstrInfo::reMaterialize to pass in TargetRegisterInfo.
- If destination is a physical register and it has a subreg index, use the
  sub-register instead.
This fixes PR5423.

llvm-svn: 88745
2009-11-14 02:55:43 +00:00
Evan Cheng
3781b2e7b3 Add radar number.
llvm-svn: 88739
2009-11-14 02:11:32 +00:00
Evan Cheng
c56b0a0f14 Fix PR5412: Fix an inverted check and another missing sub-register check.
llvm-svn: 88738
2009-11-14 02:09:09 +00:00
Dan Gohman
b36274632d Enable the tail call optimization when the caller returns undef.
llvm-svn: 88737
2009-11-14 02:06:30 +00:00
Evan Cheng
e43198c166 When expanding t2STRDi8 r, r to two stores, add kill markers correctly.
llvm-svn: 88734
2009-11-14 01:50:00 +00:00
Evan Cheng
e2907b91de Fix PR5411. Bug in UpdateKills. A reg def partially define its super-registers.
llvm-svn: 88719
2009-11-13 23:16:41 +00:00
David Greene
97a1f550d1 Move DebugInfo checks into EmitComments and remove them from
target-specific AsmPrinters.  Not all comments need DebugInfo.

Re-enable the line numbers comment test.

llvm-svn: 88697
2009-11-13 21:34:57 +00:00
Dan Gohman
972293611d When optimizing for size, don't tail-merge unless it's likely to be a
code-size win, and not when it's only likely to be code-size neutral,
such as when only a single instruction would be eliminated and a new
branch would be required.

This fixes rdar://7392894.

llvm-svn: 88692
2009-11-13 21:02:15 +00:00
Evan Cheng
f629fdcab2 Fix PR5410: LiveVariables lost subreg def:
D0<def,dead> = ...
...
             = S0<use, kill>
S0<def>      = ...
...
D0<def>      = 

The first D0 def is correctly marked dead, however, livevariables should have
added an implicit def of S0 or we end up with a use without a def.

llvm-svn: 88690
2009-11-13 20:36:40 +00:00
Dan Gohman
01b65e1e48 Don't let a noalias difference disrupt the tailcall optimization.
llvm-svn: 88672
2009-11-13 18:49:38 +00:00
Dale Johannesen
f57a58c4fe Adjust isConstantSplat to allow for big-endian targets.
PPC is such a target; make it work.

llvm-svn: 87060
2009-11-13 01:45:18 +00:00
Daniel Dunbar
cdeab5257c Update test.
llvm-svn: 87049
2009-11-13 01:01:58 +00:00
Jim Grosbach
8ffdb5d109 Clean up testcase a bit. Simplify case blocks and adjust switch instruction to not take an undefined value as input.
llvm-svn: 86997
2009-11-12 17:19:09 +00:00
Benjamin Kramer
86592507dc Fix typo in run line.
llvm-svn: 86984
2009-11-12 12:35:27 +00:00
Gabor Greif
15afbc8fa9 typo
llvm-svn: 86980
2009-11-12 09:44:17 +00:00
Chris Lattner
00a9240c9c implement a nice little efficiency hack in the inliner. Since we're now
running IPSCCP early, and we run functionattrs interlaced with the inliner,
we often (particularly for small or noop functions) completely propagate
all of the information about a call to its call site in IPSSCP (making a call
dead) and functionattrs is smart enough to realize that the function is
readonly (because it is interlaced with inliner).

To improve compile time and make the inliner threshold more accurate, realize
that we don't have to inline dead readonly function calls.  Instead, just 
delete the call.  This happens all the time for C++ codes, here are some
counters from opt/llvm-ld counting the number of times calls were deleted vs
inlined on various apps:

Tramp3d opt:
  5033 inline                - Number of call sites deleted, not inlined
 24596 inline                - Number of functions inlined
llvm-ld:
  667 inline           - Number of functions deleted because all callers found
  699 inline           - Number of functions inlined

483.xalancbmk opt:
  8096 inline                - Number of call sites deleted, not inlined
 62528 inline                - Number of functions inlined
llvm-ld:
   217 inline           - Number of allocas merged together
  2158 inline           - Number of functions inlined

471.omnetpp:
  331 inline                - Number of call sites deleted, not inlined
 8981 inline                - Number of functions inlined
llvm-ld:
  171 inline           - Number of functions deleted because all callers found
  629 inline           - Number of functions inlined


Deleting a call is much faster than inlining it, and is insensitive to the
size of the callee. :)

llvm-svn: 86975
2009-11-12 07:56:08 +00:00
Evan Cheng
deacae0dd9 RegScavenger::enterBasicBlock should always reset register state.
llvm-svn: 86972
2009-11-12 07:49:10 +00:00
Evan Cheng
b0a193db31 - Teach LSR to avoid changing cmp iv stride if it will create an immediate that
cannot be folded into target cmp instruction.
- Avoid a phase ordering issue where early cmp optimization would prevent the
  later count-to-zero optimization.
- Add missing checks which could cause LSR to reuse stride that does not have
  users.
- Fix a bug in count-to-zero optimization code which failed to find the pre-inc
  iv's phi node.
- Remove, tighten, loosen some incorrect checks disable valid transformations.
- Quite a bit of code clean up.

llvm-svn: 86969
2009-11-12 07:35:05 +00:00
Chris Lattner
01fddcec53 use getPredicateOnEdge to fold comparisons through PHI nodes,
which implements GCC PR18046.  This also gets us 360 more
jump threads on 176.gcc.

llvm-svn: 86953
2009-11-12 05:24:05 +00:00
Chris Lattner
b976027c91 should not commit when distracted.
llvm-svn: 86929
2009-11-12 02:04:17 +00:00
Chris Lattner
32bb09e867 We now thread some impossible condition information with LVI.
llvm-svn: 86927
2009-11-12 01:55:20 +00:00
Chris Lattner
68f3b53ddc with the new code we can thread non-instruction values. This
allows us to handle the test10 testcase.

llvm-svn: 86924
2009-11-12 01:41:34 +00:00
Chris Lattner
46056d81aa move some stuff into DEBUG's and turn on lazy-value-info for
the basic.ll testcase.

llvm-svn: 86918
2009-11-12 01:22:16 +00:00
Dan Gohman
f8ec4856e4 Tail merge at any size when there are two potentials blocks and one
can be made to fall through into the other.

llvm-svn: 86909
2009-11-12 00:39:10 +00:00