oota-llvm.git
10 years ago[CMake] llvm_add_library(MODULE) may use CMAKE_MODULE_LINKER_FLAGS instead of target...
NAKAMURA Takumi [Thu, 13 Feb 2014 11:19:11 +0000 (11:19 +0000)]
[CMake] llvm_add_library(MODULE) may use CMAKE_MODULE_LINKER_FLAGS instead of target property LINK_FLAGS.

I mis-dropped Darwin's link flags (in clang side) since r201073.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201317 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake] LLVM_PLUGIN_EXT: Use CMAKE_SHARED_LIBRARY_SUFFIX rather than CMAKE_SHARED_MOD...
NAKAMURA Takumi [Thu, 13 Feb 2014 11:19:00 +0000 (11:19 +0000)]
[CMake] LLVM_PLUGIN_EXT: Use CMAKE_SHARED_LIBRARY_SUFFIX rather than CMAKE_SHARED_MODULE_SUFFIX in llvm tree.

FIXME: llvm/test may be aware of LLVM_PLUGIN_EXT, like as clang/test does.
FIXME: CMAKE_*_SUFFIX may be set in HandleLLVMOptions if those variables could be writable, rather than to set one as target properties.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201316 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm/test/CodeGen/AArch64/cpus.ll: Tweak to use -mtriple=aarch64-unknown-unknown...
NAKAMURA Takumi [Thu, 13 Feb 2014 11:06:23 +0000 (11:06 +0000)]
llvm/test/CodeGen/AArch64/cpus.ll: Tweak to use -mtriple=aarch64-unknown-unknown, or this would crash for targeting pecoff like *-mingw32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201315 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: remove floating-point patterns for @llvm.arm.neon.vabs
Tim Northover [Thu, 13 Feb 2014 10:44:30 +0000 (10:44 +0000)]
ARM: remove floating-point patterns for @llvm.arm.neon.vabs

The front-end is now generating the generic @llvm.fabs for this
operation now, so the extra patterns are no longer needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201314 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd Cortex-A53 and Cortex-A57 cores to the AArch64 backend
Oliver Stannard [Thu, 13 Feb 2014 09:46:11 +0000 (09:46 +0000)]
Add Cortex-A53 and Cortex-A57 cores to the AArch64 backend

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201305 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unused method declaration.
Craig Topper [Thu, 13 Feb 2014 07:12:40 +0000 (07:12 +0000)]
Remove unused method declaration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201301 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove filtering concept from X86 disassembler table generation. It's no longer neces...
Craig Topper [Thu, 13 Feb 2014 07:07:16 +0000 (07:07 +0000)]
Remove filtering concept from X86 disassembler table generation. It's no longer necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201299 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64]Fix the problems that can't select mul/add/sub of v1i8/v1i16/v1i32 types.
Hao Liu [Thu, 13 Feb 2014 05:42:33 +0000 (05:42 +0000)]
[AArch64]Fix the problems that can't select mul/add/sub of v1i8/v1i16/v1i32 types.
As this problems are similar to shl/sra/srl, also add patterns for shift nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201298 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[RegAlloc] Fix the assertion in the last chance recoloring to match the
Quentin Colombet [Thu, 13 Feb 2014 05:17:37 +0000 (05:17 +0000)]
[RegAlloc] Fix the assertion in the last chance recoloring to match the
condition at the call site.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201296 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCopy dll storage in copyAttributes.
Rafael Espindola [Thu, 13 Feb 2014 05:11:35 +0000 (05:11 +0000)]
Copy dll storage in copyAttributes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201295 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[DAG] Fix the recognition of opaque constants in the SelectionDAGBuilder.
Juergen Ributzka [Thu, 13 Feb 2014 04:19:26 +0000 (04:19 +0000)]
[DAG] Fix the recognition of opaque constants in the SelectionDAGBuilder.

This fix checks the original LLVM IR node to identify opaque constants by
looking for the bitcast-constant pattern. Originally we looked at the generated
SDNode, but this might lead to incorrect results. The SDNode could have been
generated by an constant expression that was folded to a constant.

This fixes <rdar://problem/16050719>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201291 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse simpler version of sys::fs::exists when possible.
Rafael Espindola [Thu, 13 Feb 2014 04:00:35 +0000 (04:00 +0000)]
Use simpler version of sys::fs::exists when possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201289 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64]Add support for spilling FPR8/FPR16.
Hao Liu [Thu, 13 Feb 2014 02:36:58 +0000 (02:36 +0000)]
[AArch64]Add support for spilling FPR8/FPR16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201287 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoGlobalOpt: Aliases don't have sections, don't copy them when replacing
Reid Kleckner [Thu, 13 Feb 2014 02:18:36 +0000 (02:18 +0000)]
GlobalOpt: Aliases don't have sections, don't copy them when replacing

As defined in LangRef, aliases do not have sections.  However, LLVM's
GlobalAlias class inherits from GlobalValue, which means we can read and
set its section.  We should probably ban that as a separate change,
since it doesn't make much sense for an alias to have a section that
differs from its aliasee.

Fixes PR18757, where the section was being lost on the global in code
from Clang like:

extern "C" {
__attribute__((used, section("CUSTOM"))) static int in_custom_section;
}

Reviewers: rafael.espindola

Differential Revision: http://llvm-reviews.chandlerc.com/D2758

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201286 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CMake] llvm_add_library: Add handling of the parameter ADDITIONAL_HEADERS to pass...
NAKAMURA Takumi [Thu, 13 Feb 2014 01:00:52 +0000 (01:00 +0000)]
[CMake] llvm_add_library: Add handling of the parameter ADDITIONAL_HEADERS to pass through to process_source.

I was insightless then about unknown optional parameters.
(Consider that LINK_LIBS foo bar ADDITIONAL_HEADERS qux quux)

Suggested by Michael Kruse. Thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201283 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix misleading comment.
Lang Hames [Thu, 13 Feb 2014 00:31:14 +0000 (00:31 +0000)]
Fix misleading comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201279 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThe new MCJIT C-API unit test is generating objects without constant data
Lang Hames [Thu, 13 Feb 2014 00:16:36 +0000 (00:16 +0000)]
The new MCJIT C-API unit test is generating objects without constant data
sections, at least on MachO. Relax expectations to keep the bots green while I
investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201277 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove a very old instcombine where we would turn sequences of selects into
Owen Anderson [Wed, 12 Feb 2014 23:54:07 +0000 (23:54 +0000)]
Remove a very old instcombine where we would turn sequences of selects into
logical operations on the i1's driving them.  This is a bad idea for every
target I can think of (confirmed with micro tests on all of: x86-64, ARM,
AArch64, Mips, and PowerPC) because it forces the i1 to be materialized into
a general purpose register, whereas consuming it directly into a select generally
allows it to exist only transiently in a predicate or flags register.

Chandler ran a set of performance tests with this change, and reported no
measurable change on x86-64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201275 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Vectorizer] Add a new 'OperandValueKind' in TargetTransformInfo called
Andrea Di Biagio [Wed, 12 Feb 2014 23:43:47 +0000 (23:43 +0000)]
[Vectorizer] Add a new 'OperandValueKind' in TargetTransformInfo called
'OK_NonUniformConstValue' to identify operands which are constants but
not constant splats.

The cost model now allows returning 'OK_NonUniformConstValue'
for non splat operands that are instances of ConstantVector or
ConstantDataVector.

With this change, targets are now able to compute different costs
for instructions with non-uniform constant operands.
For example, On X86 the cost of a vector shift may vary depending on whether
the second operand is a uniform or non-uniform constant.

This patch applies the following changes:
 - The cost model computation now takes into account non-uniform constants;
 - The cost of vector shift instructions has been improved in
   X86TargetTransformInfo analysis pass;
 - BBVectorize, SLPVectorizer and LoopVectorize now know how to distinguish
   between non-uniform and uniform constant operands.

Added a new test to verify that the output of opt
'-cost-model -analyze' is valid in the following configurations: SSE2,
SSE4.1, AVX, AVX2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201272 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[X86] Teach the backend how to lower vector shift left into multiply rather than...
Andrea Di Biagio [Wed, 12 Feb 2014 23:42:28 +0000 (23:42 +0000)]
[X86] Teach the backend how to lower vector shift left into multiply rather than scalarizing it.

Instead of expanding a packed shift into a sequence of scalar shifts,
the backend now tries (when possible) to convert the vector shift into a
vector multiply.

Before this change, a shift of a MVT::v8i16 vector by a
build_vector of constants was always scalarized into a long sequence of "vector
extracts + scalar shifts + vector insert".
With this change, if there is SSE2 support, we emit a single vector multiply.

This change also affects SSE4.1, AVX, AVX2 shifts:
 - A shift of a MVT::v4i32 vector by a build_vector of non uniform constants
is now lowered when possible into a single SSE4.1 vector multiply.
 - Packed v16i16 shift left by constant build_vector are now expanded when
possible into a single AVX2 vpmullw.
This change also improves the lowering of AVX512f vector shifts.

Added test CodeGen/X86/vec_shift6.ll with some code examples that are affected
by this change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201271 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoVS Integration: install toolsets for x64 too (PR18738)
Hans Wennborg [Wed, 12 Feb 2014 23:26:29 +0000 (23:26 +0000)]
VS Integration: install toolsets for x64 too (PR18738)

This does exactly the same thing as for Win32, except it passes -m64 to
the compiler and the files go in a different directory.

Differential Revision: http://llvm-reviews.chandlerc.com/D2749

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201269 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Demonstrate that we're not currently uniquing address table entries in...
David Blaikie [Wed, 12 Feb 2014 23:03:54 +0000 (23:03 +0000)]
DebugInfo: Demonstrate that we're not currently uniquing address table entries in fission

Since I just discovered this while poking at other things, here's the
test case so I have it to come back to later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201267 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Merge fission and non-fission (and 32 and 64 bit) tests for TLS support.
David Blaikie [Wed, 12 Feb 2014 23:03:51 +0000 (23:03 +0000)]
DebugInfo: Merge fission and non-fission (and 32 and 64 bit) tests for TLS support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201266 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReformat a few lines with clang-format.
Eric Christopher [Wed, 12 Feb 2014 22:47:09 +0000 (22:47 +0000)]
Reformat a few lines with clang-format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201265 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago80-col.
Eric Christopher [Wed, 12 Feb 2014 22:38:04 +0000 (22:38 +0000)]
80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201264 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmaps] Improve the stackmap lowering code in the SelectionDAGBuilder.
Juergen Ributzka [Wed, 12 Feb 2014 22:17:13 +0000 (22:17 +0000)]
[Stackmaps] Improve the stackmap lowering code in the SelectionDAGBuilder.

We are now no longer relying on the target-specific call lowering implementation
to lower a stackmap intrinsic call. Instead we perform the call lowering in a
target-independent way directly in the stackmap lowering code. This simplifies
the code and removes the need to fixup the code after the target-specific call
lowering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201263 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmaps] Fix the ID type to be i64 also for stackmaps (as we claim in the documena...
Juergen Ributzka [Wed, 12 Feb 2014 22:17:10 +0000 (22:17 +0000)]
[Stackmaps] Fix the ID type to be i64 also for stackmaps (as we claim in the documenation)

The ID type for the stackmap and patchpoint intrinsics are in both cases i64.
This fixes an zero extend in the SelectionDAGBuilder that still used i32. This
also updates the target independent instructions STACKMAP and PATCHPOINT to use
the correct type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201262 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDon't try to build ExceptionDemo on ARM.
Rafael Espindola [Wed, 12 Feb 2014 21:47:15 +0000 (21:47 +0000)]
Don't try to build ExceptionDemo on ARM.

The demo uses _Unwind_GetIP which is not available on that platform.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201261 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoExtend RTDyld API to enable optionally precomputing the total amount of memory
Lang Hames [Wed, 12 Feb 2014 21:30:07 +0000 (21:30 +0000)]
Extend RTDyld API to enable optionally precomputing the total amount of memory
required for all sections in a module. This can be useful when targets or
code-models place strict requirements on how sections must be laid out
in memory.

If RTDyldMemoryManger::needsToReserveAllocationSpace() is overridden to return
true then the JIT will call the following method on the memory manager, which
can be used to preallocate the necessary memory.

void RTDyldMemoryManager::reserveAllocationSpace(uintptr_t CodeSize,
                                                 uintptr_t DataSizeRO,
                                                 uintptr_t DataSizeRW)

Patch by Vaidas Gasiunas. Thanks very much Viadas!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201259 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRename Windows.h to WindowsSupport.h to avoid ambiguity
Reid Kleckner [Wed, 12 Feb 2014 21:26:20 +0000 (21:26 +0000)]
Rename Windows.h to WindowsSupport.h to avoid ambiguity

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201258 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago_CS_DARWIN_USER macros available on darwin>=9. Thanks, Dave Odell!
David Fang [Wed, 12 Feb 2014 21:02:12 +0000 (21:02 +0000)]
_CS_DARWIN_USER macros available on darwin>=9.  Thanks, Dave Odell!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201255 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug info: Bugfix for r201190: DW_OP_piece takes bytes, not bits.
Adrian Prantl [Wed, 12 Feb 2014 19:34:44 +0000 (19:34 +0000)]
Debug info: Bugfix for r201190: DW_OP_piece takes bytes, not bits.
rdar://problem/16015314

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201253 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoPass edges weights to MachineBasicBlock::addSuccessor in TailDuplicatePass to
Akira Hatanaka [Wed, 12 Feb 2014 18:09:18 +0000 (18:09 +0000)]
Pass edges weights to MachineBasicBlock::addSuccessor in TailDuplicatePass to
preserve branch probability information.

<rdar://problem/15893208>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201245 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoPC-rel implemented in AArch64, test now pass
Renato Golin [Wed, 12 Feb 2014 17:17:41 +0000 (17:17 +0000)]
PC-rel implemented in AArch64, test now pass

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201243 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove more self-contained functionality away from tools/opt/opt.cpp
Eli Bendersky [Wed, 12 Feb 2014 16:48:02 +0000 (16:48 +0000)]
Move more self-contained functionality away from tools/opt/opt.cpp

BreakpointPrinter moves to its own module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201242 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r201237+r201238: Demote EmitRawText call in AsmPrinter::EmitInlineAsm() and...
Daniel Sanders [Wed, 12 Feb 2014 15:39:20 +0000 (15:39 +0000)]
Revert r201237+r201238: Demote EmitRawText call in AsmPrinter::EmitInlineAsm() and remove hasRawTextSupport() call

It introduced multiple test failures in the buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201241 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDemote EmitRawText call in AsmPrinter::EmitInlineAsm() and remove hasRawTextSupport...
Daniel Sanders [Wed, 12 Feb 2014 14:44:54 +0000 (14:44 +0000)]
Demote EmitRawText call in AsmPrinter::EmitInlineAsm() and remove hasRawTextSupport() call

Summary:
AsmPrinter::EmitInlineAsm() will no longer use the EmitRawText() call for targets with mature MC support. Such targets will always parse the inline assembly (even when emitting assembly). Targets without mature MC support continue to use EmitRawText() for assembly output.

The hasRawTextSupport() check in AsmPrinter::EmitInlineAsm() has been replaced with MCAsmInfo::UseIntegratedAs which when true, causes the integrated assembler to parse inline assembly (even when emitting assembly output). UseIntegratedAs is set to true for targets that consider any failure to parse valid assembly to be a bug. Target specific subclasses generally enable the integrated assembler in their constructor. The default value can be overridden with -no-integrated-as.

All tests that rely on inline assembly supporting invalid assembly (for example, those that use mnemonics such as 'foo' or 'hello world') have been updated to disable the integrated assembler.

Reviewers: rafael

Reviewed By: rafael

CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D2686

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201237 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoWindows/Path.inc: Move <shlobj.h> after "Windows.h" for some API available.
NAKAMURA Takumi [Wed, 12 Feb 2014 11:50:22 +0000 (11:50 +0000)]
Windows/Path.inc: Move <shlobj.h> after "Windows.h" for some API available.

  I found that swapping the order of some header files helped fix a
  build issue that we're seeing on mingw32. Without the swap, windows.h
  was being included before _WIN32_WINNT was being defined and the
  CreateHardLinkW function was #ifdef'd out.

  It looks like the header is mainly used to get the SHGetFolderPathW
  function, so I don't think that there'll be much fallout from the
  switch.

Suggested by Alex Crichton. Thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201230 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTeach tablegen() macro to check needed variables
NAKAMURA Takumi [Wed, 12 Feb 2014 11:50:11 +0000 (11:50 +0000)]
Teach tablegen() macro to check needed variables

This macro depends on several variables to be set in the calling
context.  Check them and report an error if they are not set.
Without this, custom commands may be silently specified that
will fail at build time.

Patch by Brad King.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201229 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Always implement both versions of isTruncateFree and add a sanity check.
Benjamin Kramer [Wed, 12 Feb 2014 10:17:54 +0000 (10:17 +0000)]
R600: Always implement both versions of isTruncateFree and add a sanity check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201222 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove special case filtering for instructions with lock prefix as they are all marke...
Craig Topper [Wed, 12 Feb 2014 08:09:20 +0000 (08:09 +0000)]
Remove special case filtering for instructions with lock prefix as they are all marked with isCodeGenOnly already.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201216 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMark XACQUIRE_PREFIX/XRELEASE_PREFIX as isAsmParserOnly so they'll disappear from...
Craig Topper [Wed, 12 Feb 2014 08:02:29 +0000 (08:02 +0000)]
Mark XACQUIRE_PREFIX/XRELEASE_PREFIX as isAsmParserOnly so they'll disappear from the disassembler table build without custom filtering code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201215 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PR18809] Mark DebugInfo/empty.ll as XFAIL:cygming.
NAKAMURA Takumi [Wed, 12 Feb 2014 07:15:05 +0000 (07:15 +0000)]
[PR18809] Mark DebugInfo/empty.ll as XFAIL:cygming.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201211 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTest case I forgot to 'add' for r201126.
Craig Topper [Wed, 12 Feb 2014 03:58:47 +0000 (03:58 +0000)]
Test case I forgot to 'add' for r201126.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201207 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unneeded include
Matt Arsenault [Wed, 12 Feb 2014 01:00:24 +0000 (01:00 +0000)]
Remove unneeded include

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201201 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfUnit: Include type unit's file strings in the defining compile unit's file_names...
David Blaikie [Wed, 12 Feb 2014 00:40:47 +0000 (00:40 +0000)]
DwarfUnit: Include type unit's file strings in the defining compile unit's file_names table

There's still one piece missing here, which is adding the
DW_AT_stmt_list to the type unit that refer's to the compile unit's line
table. Working on that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201198 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix some formatting in my last commit (r201196)
David Blaikie [Wed, 12 Feb 2014 00:32:05 +0000 (00:32 +0000)]
Fix some formatting in my last commit (r201196)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201197 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfUnit: Provide a reference to a defining DwarfCompileUnit from DwarfTypeUnit.
David Blaikie [Wed, 12 Feb 2014 00:31:30 +0000 (00:31 +0000)]
DwarfUnit: Provide a reference to a defining DwarfCompileUnit from DwarfTypeUnit.

Type units need to insert their file strings into the compile unit's
line/file table. This is preliminary work to that end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201196 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfUnit: Refactor DW_AT_file creation into a common function.
David Blaikie [Wed, 12 Feb 2014 00:11:25 +0000 (00:11 +0000)]
DwarfUnit: Refactor DW_AT_file creation into a common function.

This is preliminary work to fix type unit file strings so they appear in
their originating CU's line table - but it's also just good/simple
cleanup, so I'm committing it ahead of time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201195 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfUnit: Replace unnecessary conditionals with asserts.
David Blaikie [Tue, 11 Feb 2014 23:57:03 +0000 (23:57 +0000)]
DwarfUnit: Replace unnecessary conditionals with asserts.

We used to be pretty vague about what debug entities were what, with
many conditionals to silently drop/skip/accept things. These don't seem
to be relevant anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201194 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTweak ARM fastcc by adopting these two AAPCS rules:
Evan Cheng [Tue, 11 Feb 2014 23:49:31 +0000 (23:49 +0000)]
Tweak ARM fastcc by adopting these two AAPCS rules:
* CPRCs may be allocated to co-processor registers or the stack – they may never be allocated to core registers
* When a CPRC is allocated to the stack, all other VFP registers should be marked as unavailable

The difference is only noticeable in rare cases where there are a large number of floating point arguments (e.g.
7 doubles + additional float, double arguments). Although it's probably still better to avoid vmov as it can cause
stalls in some older ARM cores. The other, more subtle benefit, is to minimize difference between the various
calling conventions.

rdar://16039676

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201193 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReapply r201180 with an additional error path.
Adrian Prantl [Tue, 11 Feb 2014 22:22:15 +0000 (22:22 +0000)]
Reapply r201180 with an additional error path.
Debug info: Emit values in subregisters that do not have a separate
DWARF register number by emitting a super-register + DW_OP_bit_piece.
This is necessary because on x86_64, there are no DWARF register numbers
for i386-style subregisters.
Fixes a bunch of FIXMEs.

rdar://problem/16015314

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201190 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Debug info: Emit values in subregisters that do not have a separate"
Adrian Prantl [Tue, 11 Feb 2014 22:03:30 +0000 (22:03 +0000)]
Revert "Debug info: Emit values in subregisters that do not have a separate"

This reverts commit r201179 for buildbot breakage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201188 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Don't include the name of the CU file in the line table file list when...
David Blaikie [Tue, 11 Feb 2014 21:49:46 +0000 (21:49 +0000)]
DebugInfo: Don't include the name of the CU file in the line table file list when it's unneeded

This comes up in empty files or files containing #file directives that
never reference the actual source file name. Came up in a small test of
line tables I was playing with.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201187 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Remove dependence on file numbering in the line table.
David Blaikie [Tue, 11 Feb 2014 21:46:46 +0000 (21:46 +0000)]
DebugInfo: Remove dependence on file numbering in the line table.

These tests were unnecessarily sensitive to the presence and ordering of
elements in the line table file_names list which will break on a future
change I'm working on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201185 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agowhitespace
Adrian Prantl [Tue, 11 Feb 2014 21:23:02 +0000 (21:23 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201181 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug info: Emit values in subregisters that do not have a separate
Adrian Prantl [Tue, 11 Feb 2014 21:22:59 +0000 (21:22 +0000)]
Debug info: Emit values in subregisters that do not have a separate
DWARF register number by emitting a super-register + DW_OP_bit_piece.
This is necessary because on x86_64, there are no DWARF register numbers
for i386-style subregisters.
Fixes a bunch of FIXMEs.

rdar://problem/16015314

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201180 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agomake llvm-dwarfdump a little more resilient when parsing .debug_loc
Adrian Prantl [Tue, 11 Feb 2014 21:22:53 +0000 (21:22 +0000)]
make llvm-dwarfdump a little more resilient when parsing .debug_loc
sections. The call to data.getUnsigned(&Offset, AddressSize) only
increments Offset if the read succeeds, which will result in an infinite
loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201179 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix assertion on infinite loops.
Matt Arsenault [Tue, 11 Feb 2014 21:12:38 +0000 (21:12 +0000)]
R600/SI: Fix assertion on infinite loops.

This isn't the most useful case to fix in the real world,
but bugpoint runs into this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201177 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoInstCombine: Teach icmp merging about the equivalence of bit tests and UGE/ULT with...
Benjamin Kramer [Tue, 11 Feb 2014 21:09:03 +0000 (21:09 +0000)]
InstCombine: Teach icmp merging about the equivalence of bit tests and UGE/ULT with a power of 2.

This happens in bitfield code. While there reorganize the existing code
a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201176 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTidy up a bit. Formatting only.
Jim Grosbach [Tue, 11 Feb 2014 20:48:41 +0000 (20:48 +0000)]
Tidy up a bit. Formatting only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201174 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: Thumb2 LDR(literal) can target SP.
Jim Grosbach [Tue, 11 Feb 2014 20:48:39 +0000 (20:48 +0000)]
ARM: Thumb2 LDR(literal) can target SP.

Fix a slightly overzealous destination register restriction for the
'without .w' alias. Add some explicit testcases.

rdar://16033140

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201173 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSCEV: Cast switched values to make -Wswitch more useful.
Benjamin Kramer [Tue, 11 Feb 2014 19:02:55 +0000 (19:02 +0000)]
SCEV: Cast switched values to make -Wswitch more useful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201170 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImprove the declaration when LDPO_PIE is not available.
Sylvestre Ledru [Tue, 11 Feb 2014 17:30:18 +0000 (17:30 +0000)]
Improve the declaration when LDPO_PIE is not available.
Thanks to İsmail Dönmez for the better declaration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201163 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoIf LDPO_PIE is not defined (before binutils 2.23 version),
Sylvestre Ledru [Tue, 11 Feb 2014 17:11:32 +0000 (17:11 +0000)]
If LDPO_PIE is not defined (before binutils 2.23 version),
use the hardcoded declaration 3
See r201110 for the initial change

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201161 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a note about using "Differential Revision:" in commit messages
Mark Seaborn [Tue, 11 Feb 2014 16:58:03 +0000 (16:58 +0000)]
Add a note about using "Differential Revision:" in commit messages

I noticed this convention from the commit logs.  It seems like it
would be useful to document it, to encourage other committers to link
back to code reviews in their commits.

Differential Revision: http://llvm-reviews.chandlerc.com/D2678

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201160 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoScalarEvolution: Analyze trip count of loops with a switch guarding the exit.
Benjamin Kramer [Tue, 11 Feb 2014 15:44:32 +0000 (15:44 +0000)]
ScalarEvolution: Analyze trip count of loops with a switch guarding the exit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201159 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTeach the DAGCombiner how to fold concat_vector nodes when the input is two
Robert Lougher [Tue, 11 Feb 2014 15:42:46 +0000 (15:42 +0000)]
Teach the DAGCombiner how to fold concat_vector nodes when the input is two
BUILD_VECTOR nodes, e.g.:

(concat_vectors (BUILD_VECTOR a1, a2, a3, a4), (BUILD_VECTOR b1, b2, b3, b4))
->
(BUILD_VECTOR a1, a2, a3, a4, b1, b2, b3, b4)

This fixes an issue with AVX, where a sequence was not recognized as a 256-bit
vbroadcast due to the concat_vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201158 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Add missing PCRel relocations for AArch64 in RuntimeDyldELF
Bradley Smith [Tue, 11 Feb 2014 12:59:09 +0000 (12:59 +0000)]
[AArch64] Add missing PCRel relocations for AArch64 in RuntimeDyldELF

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201149 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[LPM] Switch LICM to actively use LCSSA in addition to preserving it.
Chandler Carruth [Tue, 11 Feb 2014 12:52:27 +0000 (12:52 +0000)]
[LPM] Switch LICM to actively use LCSSA in addition to preserving it.

Fixes PR18753 and PR18782.

This is necessary for LICM to preserve LCSSA correctly and efficiently.
There is still some active discussion about whether we should be using
LCSSA, but we can't just immediately stop using it and we *need* LICM to
preserve it while we are using it. We can restore the old SSAUpdater
driven code if and when there is a serious effort to remove the reliance
on LCSSA from all of the loop passes.

However, this also serves as a great example of why LCSSA is very nice
to have. This change significantly simplifies the process of sinking
instructions for LICM, and makes it quite a bit less expensive.

It wouldn't even be as complex as it is except that I had to start the
process of removing the big recursive LCSSA formation hammer in order to
switch even this much of the re-forming code to asserting that LCSSA was
preserved. I'll fully remove that next just to tidy things up until the
LCSSA debate settles one way or the other.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201148 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoXCore target: fix const section handling
Robert Lytton [Tue, 11 Feb 2014 10:36:26 +0000 (10:36 +0000)]
XCore target: fix const section handling

Xcore target ABI requires const data that is externally visible
to be handled differently if it has C-language linkage rather than
C++ language linkage.

Clang now emits ".cp.rodata" section information.

All other externally visible constant data will be placed in the DP section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201144 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago XCore target: Lower ATOMIC_LOAD & ATOMIC_STORE
Robert Lytton [Tue, 11 Feb 2014 10:36:18 +0000 (10:36 +0000)]
 XCore target: Lower ATOMIC_LOAD & ATOMIC_STORE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201143 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAVX: fixed a bug in LowerVECTOR_SHUFFLE
Elena Demikhovsky [Tue, 11 Feb 2014 10:21:53 +0000 (10:21 +0000)]
AVX: fixed a bug in LowerVECTOR_SHUFFLE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201140 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove TimeValue::toPosixTime() -- it is buggy, semantics are unclear, and its
Dmitri Gribenko [Tue, 11 Feb 2014 09:11:18 +0000 (09:11 +0000)]
Remove TimeValue::toPosixTime() -- it is buggy, semantics are unclear, and its
only current user should be using toEpochTime() instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201136 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAVX-512: Optimized BUILD_VECTOR pattern;
Elena Demikhovsky [Tue, 11 Feb 2014 07:25:59 +0000 (07:25 +0000)]
AVX-512: Optimized BUILD_VECTOR pattern;
fixed encoding of VEXTRACTPS instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201134 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CodeGenPrepare] Test case for the promotions that bypass the
Quentin Colombet [Tue, 11 Feb 2014 06:55:43 +0000 (06:55 +0000)]
[CodeGenPrepare] Test case for the promotions that bypass the
profitability check due to some other checks in the addressing
mode matcher. I.e., test case for commit r201121.

<rdar://problem/16020230>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201132 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoIn RuntimeDyldImpl::emitSection, make Allocate (section size to be allocated) a
Lang Hames [Tue, 11 Feb 2014 05:28:24 +0000 (05:28 +0000)]
In RuntimeDyldImpl::emitSection, make Allocate (section size to be allocated) a
uintptr_t. An unsigned could overflow for large sections.

No test case - anything big enough to overflow an unsigned is going to take an
appreciable time to zero when the test passes.

The choice of uintptr_t was made to match the RTDyldMemoryManager APIs, but
these should probably be hardcoded to uint64_ts: It is legitimate to JIT for
64-bit targets from a 32-bit host/compiler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201127 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoChanged attributes of all gather intrinsics from IntrReadMem to IntrReadArgMem as...
Craig Topper [Tue, 11 Feb 2014 04:05:33 +0000 (04:05 +0000)]
Changed attributes of all gather intrinsics from IntrReadMem to IntrReadArgMem as they access only memory based on argument. Patch by Robert Khasanov.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201126 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUsing the helper API for random number generation.
Aaron Ballman [Tue, 11 Feb 2014 03:40:14 +0000 (03:40 +0000)]
Using the helper API for random number generation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201125 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoHopefully fixing the MinGW 32 build, which was broken by r200767. Not using rand_s...
Aaron Ballman [Tue, 11 Feb 2014 02:47:33 +0000 (02:47 +0000)]
Hopefully fixing the MinGW 32 build, which was broken by r200767. Not using rand_s() since MinGW does not have an implementation for it, but instead using the underlying CryptGenRandom APIs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201124 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CodeGenPrepare] Undo changes that happened for the profitability check.
Quentin Colombet [Tue, 11 Feb 2014 01:59:02 +0000 (01:59 +0000)]
[CodeGenPrepare] Undo changes that happened for the profitability check.

The addressing mode matcher checks at some point the profitability of folding an
instruction into the addressing mode. When the instruction to be folded has
several uses, it checks that the instruction can be folded in each use.
To do so, it creates a new matcher for each use and check if the instruction is
in the list of the matched instructions of this new matcher.

The new matchers may promote some instructions and this has to be undone to keep
the state of the original matcher consistent.

A test case will follow.

<rdar://problem/16020230>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201121 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Use existing symbol rather than creating it again.
David Blaikie [Tue, 11 Feb 2014 01:23:52 +0000 (01:23 +0000)]
DebugInfo: Use existing symbol rather than creating it again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201119 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove the *PassPrinter into their own module.
Eli Bendersky [Mon, 10 Feb 2014 23:34:23 +0000 (23:34 +0000)]
Move the *PassPrinter into their own module.

These are self-contained in functionality so it makes sense to separate them,
as opt.cpp has grown quite big already.

Following Eric's suggestions, if this code is ever deemed useful outside of
tools/opt, it will make sense to move it to one of the LLVM libraries like IR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201116 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmaps] Cleanup code. No functional change intended.
Juergen Ributzka [Mon, 10 Feb 2014 23:30:26 +0000 (23:30 +0000)]
[Stackmaps] Cleanup code. No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201115 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLTO API: add lto_module_create_from_memory_with_path.
Manman Ren [Mon, 10 Feb 2014 23:26:14 +0000 (23:26 +0000)]
LTO API: add lto_module_create_from_memory_with_path.

This function adds an extra path argument to lto_module_create_from_memory.
The path argument will be passed to makeBuffer to make sure the MemoryBuffer
has a name and the created module has a module identifier.

This is mainly for emitting warning messages from the linker. When we emit
warning message on a module, we can use the module identifier.

rdar://15985737

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201114 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMark the methods in the Mangler const.
Rafael Espindola [Mon, 10 Feb 2014 21:25:13 +0000 (21:25 +0000)]
Mark the methods in the Mangler const.

A const ObjectFile needs to be able to provide its name. For an IRObjectFile,
that means being able to call the mangler. Since each IRObjectFile can have
a different mangling, it is natural for them to contain a Mangler which is
therefore also const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201113 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agocmake: Do not export targets when installing toolchain only
Reid Kleckner [Mon, 10 Feb 2014 20:44:57 +0000 (20:44 +0000)]
cmake: Do not export targets when installing toolchain only

Patch by Brad King!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201111 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix an old FIXME. LDPO_PIE is available since 2.23, realeased 2012-10-22.
Rafael Espindola [Mon, 10 Feb 2014 20:38:38 +0000 (20:38 +0000)]
Fix an old FIXME. LDPO_PIE is available since 2.23, realeased 2012-10-22.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201110 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoChange the begin and end methods in ObjectFile to match the style guide.
Rafael Espindola [Mon, 10 Feb 2014 20:24:04 +0000 (20:24 +0000)]
Change the begin and end methods in ObjectFile to match the style guide.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201108 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Implement isTruncateFree
Matt Arsenault [Mon, 10 Feb 2014 19:57:42 +0000 (19:57 +0000)]
R600: Implement isTruncateFree

Truncation is just accessing a subregister for any multiple of
the register size, so it's free.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201107 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[LPM] A terribly simple fix to a terribly complex bug: PR18773.
Chandler Carruth [Mon, 10 Feb 2014 19:39:35 +0000 (19:39 +0000)]
[LPM] A terribly simple fix to a terribly complex bug: PR18773.

The crux of the issue is that LCSSA doesn't preserve stateful alias
analyses. Before r200067, LICM didn't cause LCSSA to run in the LTO pass
manager, where LICM runs essentially without any of the other loop
passes. As a consequence the globalmodref-aa pass run before that loop
pass manager was able to survive the loop pass manager and be used by
DSE to eliminate stores in the function called from the loop body in
Adobe-C++/loop_unroll (and similar patterns in other benchmarks).

When LICM was taught to preserve LCSSA it had to require it as well.
This caused it to be run in the loop pass manager and because it did not
preserve AA, the stateful AA was lost. Most of LLVM's AA isn't stateful
and so this didn't manifest in most cases. Also, in most cases LCSSA was
already running, and so there was no interesting change.

The real kicker is that LCSSA by its definition (injecting PHI nodes
only) trivially preserves AA! All we need to do is mark it, and then
everything goes back to working as intended. It probably was blocking
some other weird cases of stateful AA but the only one I have is
a 1000-line IR test case from loop_unroll, so I don't really have a good
test case here.

Hopefully this fixes the regressions on performance that have been seen
since that revision.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201104 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCopy the ThreadLocalMode in GlobalVariable::copyAttributesFrom
Hans Wennborg [Mon, 10 Feb 2014 17:13:56 +0000 (17:13 +0000)]
Copy the ThreadLocalMode in GlobalVariable::copyAttributesFrom

This fixes the oversight from r159077.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201098 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Initialize M0 and emit S_WQM_B64 whenever DS instructions are used
Tom Stellard [Mon, 10 Feb 2014 16:58:30 +0000 (16:58 +0000)]
R600/SI: Initialize M0 and emit S_WQM_B64 whenever DS instructions are used

DS instructions that access local memory can only uses addresses that
are less than or equal to the value of M0.  When M0 is uninitialized,
then we experience undefined behavior.

This patch also changes the behavior to emit S_WQM_B64 on pixel shaders
no matter what kind of DS instruction is used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201097 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Only use S_WQM_B64 in pixel shaders
Tom Stellard [Mon, 10 Feb 2014 16:58:27 +0000 (16:58 +0000)]
R600/SI: Only use S_WQM_B64 in pixel shaders

This doesn't change any functionality, since we only have two shader
types (compute and pixel) that use local memory.  We're just changing
the logic to match the documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201096 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove some prototype code accidentally committed in r201043
David Blaikie [Mon, 10 Feb 2014 16:49:07 +0000 (16:49 +0000)]
Remove some prototype code accidentally committed in r201043

Thanks to Chandler for the catch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201095 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: use natural LLVM IR for vshll instructions
Tim Northover [Mon, 10 Feb 2014 16:20:29 +0000 (16:20 +0000)]
ARM: use natural LLVM IR for vshll instructions

Similarly to the vshrn instructions, these are simple zext/sext + trunc
operations. Using normal LLVM IR should allow for better code, and more sharing
with the AArch64 backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201093 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Handle aliases of conditional branches without b.pred form.
Chad Rosier [Mon, 10 Feb 2014 15:43:11 +0000 (15:43 +0000)]
[AArch64] Handle aliases of conditional branches without b.pred form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201091 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: r12 is callee-saved for interrupt handlers
Oliver Stannard [Mon, 10 Feb 2014 14:24:23 +0000 (14:24 +0000)]
ARM: r12 is callee-saved for interrupt handlers

For A- and R-class processors, r12 is not normally callee-saved, but is for
interrupt handlers. See AAPCS, 5.3.1.1, "Use of IP by the linker".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201089 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake succ_iterator a real random access iterator and clean up a couple of users.
Benjamin Kramer [Mon, 10 Feb 2014 14:17:42 +0000 (14:17 +0000)]
Make succ_iterator a real random access iterator and clean up a couple of users.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201088 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoGlobalsModRef: Unify and clean up duplicated pointer analysis code.
Benjamin Kramer [Mon, 10 Feb 2014 14:17:30 +0000 (14:17 +0000)]
GlobalsModRef: Unify and clean up duplicated pointer analysis code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201087 91177308-0d34-0410-b5e6-96231b3b80d8