oota-llvm.git
10 years agoAdd back LLVM_FINAL macro. lld still needs it.
Craig Topper [Sun, 2 Mar 2014 08:34:05 +0000 (08:34 +0000)]
Add back LLVM_FINAL macro. lld still needs it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202619 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSwitch all uses of LLVM_FINAL to just use 'final', and remove the macro.
Craig Topper [Sun, 2 Mar 2014 08:08:51 +0000 (08:08 +0000)]
Switch all uses of LLVM_FINAL to just use 'final', and remove the macro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202618 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoremove an old entry whose link is broken anyway
Chris Lattner [Sun, 2 Mar 2014 06:37:03 +0000 (06:37 +0000)]
remove an old entry whose link is broken anyway

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202617 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for parsing branches and conditional move instructions with ...
Venkatraman Govindaraju [Sun, 2 Mar 2014 06:28:15 +0000 (06:28 +0000)]
[Sparc] Add support for parsing branches and conditional move instructions with %fcc1-%fcc3 conditional registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202616 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Make floating point branch instruction formats to accept %fcc0-%fcc1 conditio...
Venkatraman Govindaraju [Sun, 2 Mar 2014 04:43:45 +0000 (04:43 +0000)]
[Sparc] Make floating point branch instruction formats to accept %fcc0-%fcc1 conditional registers as input.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202614 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Switch all uses of the llvm_move macro to use std::move
Chandler Carruth [Sun, 2 Mar 2014 04:08:41 +0000 (04:08 +0000)]
[C++11] Switch all uses of the llvm_move macro to use std::move
directly, and remove the macro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202612 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for parsing fcmp with %fcc registers.
Venkatraman Govindaraju [Sun, 2 Mar 2014 03:39:39 +0000 (03:39 +0000)]
[Sparc] Add support for parsing fcmp with %fcc registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202610 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Add support for OwningPtr<T> to be converted to and from
Chandler Carruth [Sun, 2 Mar 2014 03:38:32 +0000 (03:38 +0000)]
[C++11] Add support for OwningPtr<T> to be converted to and from
std::unique_ptr<T>.

Patch by Ahmed Charles!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202609 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Add unit tests for OwningPtr<T> in preparation for changes to make
Chandler Carruth [Sun, 2 Mar 2014 03:26:39 +0000 (03:26 +0000)]
[C++11] Add unit tests for OwningPtr<T> in preparation for changes to make
it interoperate (minimally) with std::unique_ptr<T>. This is part of my
plan to migrate LLVM to use std::unique_ptr with a minimal impact on
out-of-tree code.

Patch by Ahmed Charles with some minor cleanups (and bool casts) by me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202608 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Expand and eliminate the LLVM_ENUM_INT_TYPE() macro
Alp Toker [Sun, 2 Mar 2014 03:20:38 +0000 (03:20 +0000)]
[C++11] Expand and eliminate the LLVM_ENUM_INT_TYPE() macro

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202607 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add register class for floating point conditional flags (%fcc0 - %fcc3).
Venkatraman Govindaraju [Sun, 2 Mar 2014 02:12:33 +0000 (02:12 +0000)]
[Sparc] Add register class for floating point conditional flags (%fcc0 - %fcc3).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202604 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Fix some Sphinx warnings.
Sean Silva [Sun, 2 Mar 2014 00:21:42 +0000 (00:21 +0000)]
[docs] Fix some Sphinx warnings.

The docs now build cleanly. Yay!

The following warnings were fixed:

/home/sean/pg/llvm/llvm/docs/HowToReleaseLLVM.rst:364: WARNING: Enumerated list ends without a blank line; unexpected unindent.
/home/sean/pg/llvm/llvm/docs/InAlloca.rst:: WARNING: document isn't included in any toctree

/home/sean/pg/llvm/llvm/docs/CodingStandards.rst:85: WARNING: Title underline too short.

Supported C++11 Language and Library Features
-------------------------------------------
/home/sean/pg/llvm/llvm/docs/CodingStandards.rst:85: WARNING: Title underline too short.

Supported C++11 Language and Library Features
-------------------------------------------
/home/sean/pg/llvm/llvm/docs/GettingStarted.rst:185: WARNING: Explicit markup ends without a blank line; unexpected unindent.
/home/sean/pg/llvm/llvm/docs/GettingStarted.rst:565: WARNING: Explicit markup ends without a blank line; unexpected unindent.
/home/sean/pg/llvm/llvm/docs/GettingStarted.rst:567: WARNING: Block quote ends without a blank line; unexpected unindent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202603 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[SparcV9] Add support for parsing branch instructions with prediction.
Venkatraman Govindaraju [Sat, 1 Mar 2014 22:03:07 +0000 (22:03 +0000)]
[SparcV9] Add support for parsing branch instructions with prediction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202602 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Add failing control flow tests.
Matt Arsenault [Sat, 1 Mar 2014 21:45:41 +0000 (21:45 +0000)]
R600: Add failing control flow tests.

Simple cases hit a variety of problems at -O0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202601 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove extra truncs/exts around i32 bit operations on PPC64
Hal Finkel [Sat, 1 Mar 2014 21:36:57 +0000 (21:36 +0000)]
Remove extra truncs/exts around i32 bit operations on PPC64

This generalizes the code to eliminate extra truncs/exts around i1 bit
operations to also do the same on PPC64 for i32 bit operations. This eliminates
a fairly prevalent code wart:

int foo(int a) {
  return a == 5 ? 7 : 8;
}

On PPC64, because of the extension implied by the ABI, this would generate:

cmplwi 0, 3, 5
li 12, 8
li 4, 7
isel 3, 4, 12, 2
rldicl 3, 3, 0, 32
blr

where the 'rldicl 3, 3, 0, 32', the extension, is completely unnecessary. At
least for the single-BB case (which is all that the DAG combine mechanism can
handle), this unnecessary extension is no longer generated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202600 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for parsing annulled branch instructions.
Venkatraman Govindaraju [Sat, 1 Mar 2014 20:08:48 +0000 (20:08 +0000)]
[Sparc] Add support for parsing annulled branch instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202599 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for parsing sparcv9 instructions addc/subc/addccc/subccc.
Venkatraman Govindaraju [Sat, 1 Mar 2014 18:54:52 +0000 (18:54 +0000)]
[Sparc] Add support for parsing sparcv9 instructions addc/subc/addccc/subccc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202598 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add missing ALU instruction patterns.
Venkatraman Govindaraju [Sat, 1 Mar 2014 17:51:00 +0000 (17:51 +0000)]
[Sparc] Add missing ALU instruction patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202597 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake helper function static.
Benjamin Kramer [Sat, 1 Mar 2014 17:24:40 +0000 (17:24 +0000)]
Make helper function static.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202596 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd missing FileCheck in test command line.
Sasa Stankovic [Sat, 1 Mar 2014 16:14:29 +0000 (16:14 +0000)]
Add missing FileCheck in test command line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202594 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoNow that we have C++11, turn simple functors into lambdas and remove a ton of boilerp...
Benjamin Kramer [Sat, 1 Mar 2014 11:47:00 +0000 (11:47 +0000)]
Now that we have C++11, turn simple functors into lambdas and remove a ton of boilerplate.

No intended functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202588 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Remove LLVM_HAS_CXX11_STDLIB now that it is just on.
Chandler Carruth [Sat, 1 Mar 2014 10:57:19 +0000 (10:57 +0000)]
[C++11] Remove LLVM_HAS_CXX11_STDLIB now that it is just on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202587 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Remove LLVM_HAS_CXX11_TYPETRAITS now that it is a constant due
Chandler Carruth [Sat, 1 Mar 2014 10:48:14 +0000 (10:48 +0000)]
[C++11] Remove LLVM_HAS_CXX11_TYPETRAITS now that it is a constant due
to the build being C++11.

There is clearly still plenty of simplification than can be done here by
using standard type traits instead of rolling our own in many places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202586 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Now that we're not using it, remove the
Chandler Carruth [Sat, 1 Mar 2014 10:31:48 +0000 (10:31 +0000)]
[C++11] Now that we're not using it, remove the
LLVM_HAS_RVALUE_REFERENCES macro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202585 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Remove uses of LLVM_HAS_RVALUE_REFERENCES from the unittests.
Chandler Carruth [Sat, 1 Mar 2014 09:36:06 +0000 (09:36 +0000)]
[C++11] Remove uses of LLVM_HAS_RVALUE_REFERENCES from the unittests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202583 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Remove the use of LLVM_HAS_RVALUE_REFERENCES from the rest of
Chandler Carruth [Sat, 1 Mar 2014 09:32:03 +0000 (09:32 +0000)]
[C++11] Remove the use of LLVM_HAS_RVALUE_REFERENCES from the rest of
the core LLVM libraries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202582 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support to decode unimp instruction.
Venkatraman Govindaraju [Sat, 1 Mar 2014 09:28:18 +0000 (09:28 +0000)]
[Sparc] Add support to decode unimp instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202581 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Remove the R-value reference #if usage from the ADT and Support
Chandler Carruth [Sat, 1 Mar 2014 09:27:28 +0000 (09:27 +0000)]
[C++11] Remove the R-value reference #if usage from the ADT and Support
libraries. It is now always 1 in LLVM builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202580 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Force the other C++11 standard library detection macros to be
Chandler Carruth [Sat, 1 Mar 2014 09:12:30 +0000 (09:12 +0000)]
[C++11] Force the other C++11 standard library detection macros to be
on unconditionally. Continuing to break down the C++98 support,
hopefully without breaking anything.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202579 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support to decode negative simm13 operands in the sparc disassembler.
Venkatraman Govindaraju [Sat, 1 Mar 2014 09:11:57 +0000 (09:11 +0000)]
[Sparc] Add support to decode negative simm13 operands in the sparc disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202578 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for decoding call instructions in the sparc disassembler.
Venkatraman Govindaraju [Sat, 1 Mar 2014 08:30:58 +0000 (08:30 +0000)]
[Sparc] Add support for decoding call instructions in the sparc disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202577 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRewrite a terrible comment about the machine model.
Andrew Trick [Sat, 1 Mar 2014 07:57:02 +0000 (07:57 +0000)]
Rewrite a terrible comment about the machine model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202576 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support to disassemble sparc memory instructions.
Venkatraman Govindaraju [Sat, 1 Mar 2014 07:46:33 +0000 (07:46 +0000)]
[Sparc] Add support to disassemble sparc memory instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202575 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Turn off compiler-based detection of R-value references, relying
Chandler Carruth [Sat, 1 Mar 2014 06:31:00 +0000 (06:31 +0000)]
[C++11] Turn off compiler-based detection of R-value references, relying
on the fact that we now build in C++11 mode with modern compilers. This
should flush out any issues. If the build bots are happy with this, I'll
GC all the code for coping without R-value references.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202574 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd support for parsing sun-style section flags in ELFAsmParser.
Venkatraman Govindaraju [Sat, 1 Mar 2014 06:21:00 +0000 (06:21 +0000)]
Add support for parsing sun-style section flags in ELFAsmParser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202573 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Implement writeNopData. Emit actual NOP instruction instead of just filling...
Venkatraman Govindaraju [Sat, 1 Mar 2014 05:45:09 +0000 (05:45 +0000)]
[Sparc] Implement writeNopData. Emit actual NOP instruction instead of just filling with zeroes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202572 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Teach SparcAsmParser to emit correct relocations for PIC code.
Venkatraman Govindaraju [Sat, 1 Mar 2014 05:07:21 +0000 (05:07 +0000)]
[Sparc] Teach SparcAsmParser to emit correct relocations for PIC code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202571 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix RWMutex to be thread-safe when pthread_rwlock is not available
Mark Seaborn [Sat, 1 Mar 2014 04:30:32 +0000 (04:30 +0000)]
Fix RWMutex to be thread-safe when pthread_rwlock is not available

lib/Support/RWMutex.cpp contains an implementation of RWMutex that
uses pthread_rwlock, but when pthread_rwlock is not available (such as
under NaCl, when using newlib), it silently falls back to using the
no-op definition in lib/Support/Unix/RWMutex.inc, which is not
thread-safe.

Fix this case to be thread-safe by using a normal mutex.

Differential Revision: http://llvm-reviews.chandlerc.com/D2892

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202570 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Replace autoconf --enable-cxx11 with --enable-cxx1y. The
Chandler Carruth [Sat, 1 Mar 2014 03:33:08 +0000 (03:33 +0000)]
[C++11] Replace autoconf --enable-cxx11 with --enable-cxx1y. The
baseline is now C++11, and we unconditionally add -std=c++11 to the
flags.

This has the dim potential to break some non-GNU-compatible compiler (in
terms of -std flags) using the makefiles, but those makefiles are
littered with GNU-style compile flags so it would be very surprising to
me for it to actually happen in practice. As always, do let me know if
there is a toolchain you're using where this doesn't work, and I'll be
watching the bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202569 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Switch the CMake option from LLVM_ENABLE_CXX11 (default on) to
Chandler Carruth [Sat, 1 Mar 2014 03:16:07 +0000 (03:16 +0000)]
[C++11] Switch the CMake option from LLVM_ENABLE_CXX11 (default on) to
LLVM_ENABLE_CXX1Y (default *off*). =D C++98 is dead. Long live C++11.

I don't exactly recommend using C++1y just yet though...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202567 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Update the docs to remove my hedging about C++98 vs. C++11. =]
Chandler Carruth [Sat, 1 Mar 2014 02:48:03 +0000 (02:48 +0000)]
[docs] Update the docs to remove my hedging about C++98 vs. C++11. =]
The switch has been thrown. While I'm still watching for any failures or
problems with this, the documentation can go ahead and move forward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202566 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] 80 column rule. No functionality change.
Venkatraman Govindaraju [Sat, 1 Mar 2014 02:28:34 +0000 (02:28 +0000)]
[Sparc] 80 column rule. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202565 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Add support for parsing directives in SparcAsmParser.
Venkatraman Govindaraju [Sat, 1 Mar 2014 02:18:04 +0000 (02:18 +0000)]
[Sparc] Add support for parsing directives in SparcAsmParser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202564 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Emit 'restore' instead of 'restore %g0, %g0, %g0'. This improves the readabil...
Venkatraman Govindaraju [Sat, 1 Mar 2014 01:04:26 +0000 (01:04 +0000)]
[Sparc] Emit 'restore' instead of 'restore %g0, %g0, %g0'. This improves the readability of the generated code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202563 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSpillPlacement: fix a bug in iterate.
Manman Ren [Fri, 28 Feb 2014 23:05:31 +0000 (23:05 +0000)]
SpillPlacement: fix a bug in iterate.

Inside iterate, we scan backwards then scan forwards in a loop. When iteration
is not zero, the last node was just updated so we can skip it. But when
iteration is zero, we can't skip the last node.

For the testing case, fixing this will save a spill and move register copies
from hot path to cold path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202557 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake] Remove dead C backend option
Reid Kleckner [Fri, 28 Feb 2014 22:51:27 +0000 (22:51 +0000)]
[CMake] Remove dead C backend option

Patch by Jevin Sweval!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202556 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReflow isProfitableToMakeFastCC
Reid Kleckner [Fri, 28 Feb 2014 22:50:08 +0000 (22:50 +0000)]
Reflow isProfitableToMakeFastCC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202555 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoJumped the gun with r202551 and broke some bots that weren't yet C++11ified.
Lang Hames [Fri, 28 Feb 2014 22:44:44 +0000 (22:44 +0000)]
Jumped the gun with r202551 and broke some bots that weren't yet C++11ified.

Reverting until the C++11 switch is complete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202554 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoNew PBQP solver, and updates to the PBQP graph.
Lang Hames [Fri, 28 Feb 2014 22:25:24 +0000 (22:25 +0000)]
New PBQP solver, and updates to the PBQP graph.

The previous PBQP solver was very robust but consumed a lot of memory,
performed a lot of redundant computation, and contained some unnecessarily tight
coupling that prevented experimentation with novel solution techniques. This new
solver is an attempt to address these shortcomings.

Important/interesting changes:

1) The domain-independent PBQP solver class, HeuristicSolverImpl, is gone.
It is replaced by a register allocation specific solver, PBQP::RegAlloc::Solver
(see RegAllocSolver.h).

The optimal reduction rules and the backpropagation algorithm have been extracted
into stand-alone functions (see ReductionRules.h), which can be used to build
domain specific PBQP solvers. This provides many more opportunities for
domain-specific knowledge to inform the PBQP solvers' decisions. In theory this
should allow us to generate better solutions. In practice, we can at least test
out ideas now.

As a side benefit, I believe the new solver is more readable than the old one.

2) The solver type is now a template parameter of the PBQP graph.

This allows the graph to notify the solver of any modifications made (e.g. by
domain independent rules) without the overhead of a virtual call. It also allows
the solver to supply policy information to the graph (see below).

3) Significantly reduced memory overhead.

Memory management policy is now an explicit property of the PBQP graph (via
the CostAllocator typedef on the graph's solver template argument). Because PBQP
graphs for register allocation tend to contain many redundant instances of
single values (E.g. the value representing an interference constraint between
GPRs), the new RASolver class uses a uniquing scheme. This massively reduces
memory consumption for large register allocation problems. For example, looking
at the largest interference graph in each of the SPEC2006 benchmarks (the
largest graph will always set the memory consumption high-water mark for PBQP),
the average memory reduction for the PBQP costs was 400x. That's times, not
percent. The highest was 1400x. Yikes. So - this is fixed.

"PBQP: No longer feasting upon every last byte of your RAM".

Minor details:

- Fully C++11'd. Never copy-construct another vector/matrix!

- Cute tricks with cost metadata: Metadata that is derived solely from cost
matrices/vectors is attached directly to the cost instances themselves. That way
if you unique the costs you never have to recompute the metadata. 400x less
memory means 400x less cost metadata (re)computation.

Special thanks to Arnaud de Grandmaison, who has been the source of much
encouragement, and of many very useful test cases.

This new solver forms the basis for future work, of which there's plenty to do.
I will be adding TODO notes shortly.

- Lang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202551 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Clarify that there isn't much to be done other than watch build
Chandler Carruth [Fri, 28 Feb 2014 21:59:51 +0000 (21:59 +0000)]
[docs] Clarify that there isn't much to be done other than watch build
bots when using the standard library facilities. The missing pieces here
aren't always in useful discreet chunks.

Fortunately, the missing pieces are few and far between, and we can
emulate most of them in our headers as needed.

Based on feedback from Lang and Dave.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202548 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Switch autoconf and make to use C++11 by default. Now both build
Chandler Carruth [Fri, 28 Feb 2014 21:47:41 +0000 (21:47 +0000)]
[C++11] Switch autoconf and make to use C++11 by default. Now both build
systems have the default as C++11, but retain the ability to build with
C++98.

Again, please restrain your enthusiasm a bit in case this needs to be
reverted. =]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202546 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix >> to be > > for non-c++11.
Eric Christopher [Fri, 28 Feb 2014 21:37:28 +0000 (21:37 +0000)]
Fix >> to be > > for non-c++11.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202545 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Verify all instructions in the AsmPrinter on debug builds
Tom Stellard [Fri, 28 Feb 2014 21:36:41 +0000 (21:36 +0000)]
R600: Verify all instructions in the AsmPrinter on debug builds

Make a call to R600's implementation of verifyInstruction() to
check that instructions are only using legal operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202544 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Expand all v16[if]32 operations
Tom Stellard [Fri, 28 Feb 2014 21:36:37 +0000 (21:36 +0000)]
R600/SI: Expand all v16[if]32 operations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202543 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Switch CMake to use C++11 by default! Next up, autoconf/make!
Chandler Carruth [Fri, 28 Feb 2014 21:30:03 +0000 (21:30 +0000)]
[C++11] Switch CMake to use C++11 by default! Next up, autoconf/make!

Now, please don't get too excited. I've just toggled the default to suss
out the last remaining bot problems. This does *not* mean we can all go
write lots of C++11 code yet. I at least want to let the dust settle
from the bots first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202542 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago80-col.
Eric Christopher [Fri, 28 Feb 2014 21:27:59 +0000 (21:27 +0000)]
80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202541 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a crasher where when we're attempting to replace a type
Eric Christopher [Fri, 28 Feb 2014 21:27:57 +0000 (21:27 +0000)]
Fix a crasher where when we're attempting to replace a type
during the finalization for CGDebugInfo in clang we would RAUW
a type and it would result in a corrupted MDNode for an
imported declaration.

Testcase pending as reducing has been difficult.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202540 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd more whitespace to fix more bullets.
Richard Smith [Fri, 28 Feb 2014 21:14:25 +0000 (21:14 +0000)]
Add more whitespace to fix more bullets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202538 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd whitespace to try to fix bulleted list.
Richard Smith [Fri, 28 Feb 2014 21:11:28 +0000 (21:11 +0000)]
Add whitespace to try to fix bulleted list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202537 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix some links to C++11 feature papers in the Coding Standards
Ben Langmuir [Fri, 28 Feb 2014 19:37:20 +0000 (19:37 +0000)]
Fix some links to C++11 feature papers in the Coding Standards

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202532 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoadd missing 3.4 release
Gabor Greif [Fri, 28 Feb 2014 19:20:48 +0000 (19:20 +0000)]
add missing 3.4 release

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202531 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCommandLine: Exit successfully for -version and -help
Justin Bogner [Fri, 28 Feb 2014 19:08:01 +0000 (19:08 +0000)]
CommandLine: Exit successfully for -version and -help

Tools that use the CommandLine library currently exit with an error
when invoked with -version or -help. This is unusual and non-standard,
so we'll fix them to exit successfully instead.

I don't expect that anyone relies on the current behaviour, so this
should be a fairly safe change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202530 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTest commit
Adam Nemet [Fri, 28 Feb 2014 18:44:39 +0000 (18:44 +0000)]
Test commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202528 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFixed operand of SC microMIPS instruction.
Zoran Jovanovic [Fri, 28 Feb 2014 18:22:56 +0000 (18:22 +0000)]
Fixed operand of SC microMIPS instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202526 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDrop libtool from llvm.
Rafael Espindola [Fri, 28 Feb 2014 18:17:54 +0000 (18:17 +0000)]
Drop libtool from llvm.

We were only using it so find the shared library extension and nm. There are
simpler ways to do those things :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202524 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFixed encoding of SYSCALL microMIPS instruction.
Zoran Jovanovic [Fri, 28 Feb 2014 18:17:08 +0000 (18:17 +0000)]
Fixed encoding of SYSCALL microMIPS instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202523 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert revision 202518 because of wrong commit message.
Zoran Jovanovic [Fri, 28 Feb 2014 18:14:16 +0000 (18:14 +0000)]
Revert revision 202518  because of wrong commit message.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202521 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix operand of SC instruction.
Zoran Jovanovic [Fri, 28 Feb 2014 18:02:17 +0000 (18:02 +0000)]
Fix operand of SC instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202518 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoWith rpaths being set correctly, SHLIBPATH_VAR is not needed anymore.
Rafael Espindola [Fri, 28 Feb 2014 16:16:51 +0000 (16:16 +0000)]
With rpaths being set correctly, SHLIBPATH_VAR is not needed anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202510 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCorrectly set rpath for unittests.
Rafael Espindola [Fri, 28 Feb 2014 16:11:03 +0000 (16:11 +0000)]
Correctly set rpath for unittests.

This lets us run the unittest from the command line without setting
LD_LIBRARY_PATH.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202509 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCentralize the handling of install_name and rpath.
Rafael Espindola [Fri, 28 Feb 2014 13:48:03 +0000 (13:48 +0000)]
Centralize the handling of install_name and rpath.

This centralizes the Makefile handling of -install_name and -rpath. It also
moves the cmake build to using @rpath. The reason being that libclang needs it,
and it works for everything else.

A followup patch will move clang to using this and then there will be a single
point to edit to support other systems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202499 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Add a section to the coding standards about languages and such.
Chandler Carruth [Fri, 28 Feb 2014 13:35:54 +0000 (13:35 +0000)]
[docs] Add a section to the coding standards about languages and such.

A lot of this is writing down common knowledge and things often
communicated on mailing lists and in discussions. It could live in the
Programmer's Manual alternatively, but that felt slightly less
well-fitting.

It also includes (and was motivated by) the section on the relevant
language standards for LLVM and the specific features that will be
enabled with the switch to C++11.

With this, all of the documentation for the C++11 switch is, I think, in
place. I plan to flip the switch RSN. =]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202497 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoX86Operand is extracted into individual header.
Evgeniy Stepanov [Fri, 28 Feb 2014 12:28:07 +0000 (12:28 +0000)]
X86Operand is extracted into individual header.

X86Operand is extracted into individual header, because it allows to create an
arbitrary memory operand and append it to MCInst. It'll be reused in X86 inline
assembly instrumentation.

Patch by Yuri Gorshenin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202496 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] A slight tweak to the intro for the golden rule in the coding
Chandler Carruth [Fri, 28 Feb 2014 12:24:18 +0000 (12:24 +0000)]
[docs] A slight tweak to the intro for the golden rule in the coding
standards.

It claims the document intentionally doesn't give fixed standards for
brace placement or spacing, and then the document goes on to do
precisely that in several places. Instead, try to highlight that even
these rules are simply *guidance* which may be trumped by some other
circumstance or the local conventions of code.

I'm not trying to change the thrust of this part of the document, and if
folks think this does so, I'm happy to re-wordsmith it. I just don't
want it to be so self-contradicting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202495 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Tweak the example to match what is apparantly the desired form
Chandler Carruth [Fri, 28 Feb 2014 12:14:56 +0000 (12:14 +0000)]
[docs] Tweak the example to match what is apparantly the desired form
for the style templates we're using.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202494 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Switch to external hyperlink references. Much more readable and
Chandler Carruth [Fri, 28 Feb 2014 12:09:14 +0000 (12:09 +0000)]
[docs] Switch to external hyperlink references. Much more readable and
hopefully easier to get the formatting right for ReST.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202493 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Fix my links to use the correct ReST syntax.
Chandler Carruth [Fri, 28 Feb 2014 11:12:14 +0000 (11:12 +0000)]
[docs] Fix my links to use the correct ReST syntax.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202490 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Fix 80-column wrap that I messed up.
Chandler Carruth [Fri, 28 Feb 2014 11:11:41 +0000 (11:11 +0000)]
[docs] Fix 80-column wrap that I messed up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202489 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Tweak discussion of BSDs based on feedback from Roman Divacky.
Chandler Carruth [Fri, 28 Feb 2014 11:09:33 +0000 (11:09 +0000)]
[docs] Tweak discussion of BSDs based on feedback from Roman Divacky.
FreeBSD 10.0 and newer have a modern Clang toolchain that should work
well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202488 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[docs] Add a big section with details about how to go about acquiring
Chandler Carruth [Fri, 28 Feb 2014 10:56:57 +0000 (10:56 +0000)]
[docs] Add a big section with details about how to go about acquiring
a more modern host C++ toolchain for Linux distros where folks sometimes
don't have a good option to get one as part of their system.

This is a first cut, so feedback, testing, and suggestions are very,
very welcom. This is one of the last real documentation changes that was
specifically requested prior to switching LLVM and Clang to build in
C++11 mode by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202486 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReorder Mips/MCTargetDesc/CMakeLists.txt.
NAKAMURA Takumi [Fri, 28 Feb 2014 10:18:21 +0000 (10:18 +0000)]
Reorder Mips/MCTargetDesc/CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202483 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips] Add MipsNaClELFStreamer.cpp to CMakeLists.txt.
Sasa Stankovic [Fri, 28 Feb 2014 10:14:12 +0000 (10:14 +0000)]
[mips] Add MipsNaClELFStreamer.cpp to CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202482 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips] Implement NaCl sandboxing of indirect jumps:
Sasa Stankovic [Fri, 28 Feb 2014 10:00:38 +0000 (10:00 +0000)]
[mips] Implement NaCl sandboxing of indirect jumps:

  * Align targets of indirect jumps to instruction bundle boundaries (in MI layer).
  * Add masking instructions before indirect jumps (in MC layer).

Differential Revision: http://llvm-reviews.chandlerc.com/D2847

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202479 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd 'remark' diagnostic type in LLVM
Tobias Grosser [Fri, 28 Feb 2014 09:08:45 +0000 (09:08 +0000)]
Add 'remark' diagnostic type in LLVM

A 'remark' is information that is not an error or a warning, but rather some
additional information provided to the user. In contrast to a 'note' a 'remark'
is an independent diagnostic, whereas a 'note' always depends on another
diagnostic.

A typical use case for remark nodes is information provided to the user, e.g.
information provided by the vectorizer about loops that have been vectorized.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202474 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTurn static inline functions to inline, following Rafael's suggestion
Alexey Samsonov [Fri, 28 Feb 2014 08:23:09 +0000 (08:23 +0000)]
Turn static inline functions to inline, following Rafael's suggestion

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202473 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSwap PPC isel operands to allow for 0-folding
Hal Finkel [Fri, 28 Feb 2014 06:11:16 +0000 (06:11 +0000)]
Swap PPC isel operands to allow for 0-folding

The PPC isel instruction can fold 0 into the first operand (thus eliminating
the need to materialize a zero-containing register when the 'true' result of
the isel is 0). When the isel is fed by a bit register operation that we can
invert, do so as part of the bit-register-operation peephole routine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202469 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-objdump: Fix crash bug with printing unwind info on stripped file.
Rui Ueyama [Fri, 28 Feb 2014 05:21:29 +0000 (05:21 +0000)]
llvm-objdump: Fix crash bug with printing unwind info on stripped file.

The current COFF unwind printer tries to print SEH handler function names,
assuming that it can always find function names in string table. It crashes
if file being read has no symbol table (i.e. executable).

With this patch, llvm-objdump prints SEH handler's RVA if there's no symbol
table entry for that RVA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202466 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoStyle fix.
Rui Ueyama [Fri, 28 Feb 2014 05:21:26 +0000 (05:21 +0000)]
Style fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202465 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoNow that it is possible, use the mangler in IRObjectFile.
Rafael Espindola [Fri, 28 Feb 2014 02:17:23 +0000 (02:17 +0000)]
Now that it is possible, use the mangler in IRObjectFile.

A really simple patch marks the end of a lot of yak shaving :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202463 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRename coff_pdata_x64 -> coff_runtime_function_x64.
Rui Ueyama [Fri, 28 Feb 2014 01:18:58 +0000 (01:18 +0000)]
Rename coff_pdata_x64 -> coff_runtime_function_x64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202460 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTrying to unbreak the darwin11 builder
Hal Finkel [Fri, 28 Feb 2014 01:17:25 +0000 (01:17 +0000)]
Trying to unbreak the darwin11 builder

The CR bit tracking code broke PPC/Darwin; trying to get it working again...

(the darwin11 builder, which defaults to the darwin ABI when running PPC tests,
asserted when running test/CodeGen/PowerPC/inverted-bool-compares.ll)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202459 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTry to unbreak the C++11 build
Hal Finkel [Fri, 28 Feb 2014 00:45:27 +0000 (00:45 +0000)]
Try to unbreak the C++11 build

Cannot use negative numbers in case statements without running afoul of -Wc++11-narrowing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202455 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake] llvm_add_library(SHARED|STATIC): Fix broken OUTPUT_NAME for *_static.
NAKAMURA Takumi [Fri, 28 Feb 2014 00:28:13 +0000 (00:28 +0000)]
[CMake] llvm_add_library(SHARED|STATIC): Fix broken OUTPUT_NAME for *_static.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202454 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd CR-bit tracking to the PowerPC backend for i1 values
Hal Finkel [Fri, 28 Feb 2014 00:27:01 +0000 (00:27 +0000)]
Add CR-bit tracking to the PowerPC backend for i1 values

This change enables tracking i1 values in the PowerPC backend using the
condition register bits. These bits can be treated on PowerPC as separate
registers; individual bit operations (and, or, xor, etc.) are supported.
Tracking booleans in CR bits has several advantages:

 - Reduction in register pressure (because we no longer need GPRs to store
   boolean values).

 - Logical operations on booleans can be handled more efficiently; we used to
   have to move all results from comparisons into GPRs, perform promoted
   logical operations in GPRs, and then move the result back into condition
   register bits to be used by conditional branches. This can be very
   inefficient, because the throughput of these CR <-> GPR moves have high
   latency and low throughput (especially when other associated instructions
   are accounted for).

 - On the POWER7 and similar cores, we can increase total throughput by using
   the CR bits. CR bit operations have a dedicated functional unit.

Most of this is more-or-less mechanical: Adjustments were needed in the
calling-convention code, support was added for spilling/restoring individual
condition-register bits, and conditional branch instruction definitions taking
specific CR bits were added (plus patterns and code for generating bit-level
operations).

This is enabled by default when running at -O2 and higher. For -O0 and -O1,
where the ability to debug is more important, this feature is disabled by
default. Individual CR bits do not have assigned DWARF register numbers,
and storing values in CR bits makes them invisible to the debugger.

It is critical, however, that we don't move i1 values that have been promoted
to larger values (such as those passed as function arguments) into bit
registers only to quickly turn around and move the values back into GPRs (such
as happens when values are returned by functions). A pair of target-specific
DAG combines are added to remove the trunc/extends in:
  trunc(binary-ops(binary-ops(zext(x), zext(y)), ...)
and:
  zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...)
In short, we only want to use CR bits where some of the i1 values come from
comparisons or are used by conditional branches or selects. To put it another
way, if we can do the entire i1 computation in GPRs, then we probably should
(on the POWER7, the GPR-operation throughput is higher, and for all cores, the
CR <-> GPR moves are expensive).

POWER7 test-suite performance results (from 10 runs in each configuration):

SingleSource/Benchmarks/Misc/mandel-2: 35% speedup
MultiSource/Benchmarks/Prolangs-C++/city/city: 21% speedup
MultiSource/Benchmarks/MiBench/automotive-susan: 23% speedup
SingleSource/Benchmarks/CoyoteBench/huffbench: 13% speedup
SingleSource/Benchmarks/Misc-C++/Large/sphereflake: 13% speedup
SingleSource/Benchmarks/Misc-C++/mandel-text: 10% speedup

SingleSource/Benchmarks/Misc-C++-EH/spirit: 10% slowdown
MultiSource/Applications/lemon/lemon: 8% slowdown

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202451 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd an OutPatFrag TableGen class
Hal Finkel [Fri, 28 Feb 2014 00:26:56 +0000 (00:26 +0000)]
Add an OutPatFrag TableGen class

Unfortunately, it is currently impossible to use a PatFrag as part of an output
pattern (the part of the pattern that has instructions in it) in TableGen.
Looking at the current implementation, this was clearly intended to work (there
is already code in place to expand patterns in the output DAG), but is
currently broken by the baked-in type-checking assumption and the order in which
the pattern fragments are processed (output pattern fragments need to be
processed after the instruction definitions are processed).

Fixing this is fairly simple, but requires some way of differentiating output
patterns from the existing input patterns. The simplest way to handle this
seems to be to create a subclass of PatFrag, and so that's what I've done here.

As a simple example, this allows us to write:

def crnot : OutPatFrag<(ops node:$in),
                       (CRNOR $in, $in)>;

def       : Pat<(not i1:$in),
                (crnot $in)>;

which captures the core use case: handling of repeated subexpressions inside
of complicated output patterns.

This will be used by an upcoming commit to the PowerPC backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202450 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix visitTRUNCATE for legal i1 values
Hal Finkel [Fri, 28 Feb 2014 00:26:45 +0000 (00:26 +0000)]
Fix visitTRUNCATE for legal i1 values

This extract-and-trunc vector optimization cannot work for i1 values as
currently implemented, and so I'm disabling this for now for i1 values. In the
future, this can be fixed properly.

Soon I'll commit support for i1 CR bit tracking in the PowerPC backend, and
this will be covered by one of the existing regression tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202449 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unnecessary temporary variable.
Rui Ueyama [Fri, 28 Feb 2014 00:06:20 +0000 (00:06 +0000)]
Remove unnecessary temporary variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202445 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoObject/COFF: Add a struct for the function table in .pdata.
Rui Ueyama [Thu, 27 Feb 2014 22:20:07 +0000 (22:20 +0000)]
Object/COFF: Add a struct for the function table in .pdata.

This is the data structure listed on Microsoft PE/COFF Spec Revision 8.3, p. 80.
The name of the struct is not mentioned in the Microsoft PE/COFF spec, so I made
it up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202438 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoProvide a target override for the latest regalloc heuristic.
Andrew Trick [Thu, 27 Feb 2014 21:37:33 +0000 (21:37 +0000)]
Provide a target override for the latest regalloc heuristic.

This is a temporary workaround for native arm linux builds:
PR18996: Changing regalloc order breaks "lencod" on native arm linux builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202433 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDrive-by comment fix. This regalloc comment was not accurate.
Andrew Trick [Thu, 27 Feb 2014 21:37:30 +0000 (21:37 +0000)]
Drive-by comment fix. This regalloc comment was not accurate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202432 91177308-0d34-0410-b5e6-96231b3b80d8