oota-llvm.git
10 years ago[Modules] Fix potential ODR violations by sinking the DEBUG_TYPE
Chandler Carruth [Tue, 22 Apr 2014 02:48:03 +0000 (02:48 +0000)]
[Modules] Fix potential ODR violations by sinking the DEBUG_TYPE
definition below all the header #include lines, lib/Analysis/...
edition.

This one has a bit extra as there were *other* #define's before #include
lines in addition to DEBUG_TYPE. I've sunk all of them as a block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206843 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Fix potential ODR violations by sinking the DEBUG_TYPE
Chandler Carruth [Tue, 22 Apr 2014 02:41:26 +0000 (02:41 +0000)]
[Modules] Fix potential ODR violations by sinking the DEBUG_TYPE
definition below all of the header #include lines, lib/Target/...
edition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206842 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[cleanup] Fix two headers where we included a standard library header
Chandler Carruth [Tue, 22 Apr 2014 02:28:45 +0000 (02:28 +0000)]
[cleanup] Fix two headers where we included a standard library header
after including the generated code from tablegen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206841 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[cleanup] Fix another place where we were including the tablegen'ed code
Chandler Carruth [Tue, 22 Apr 2014 02:25:17 +0000 (02:25 +0000)]
[cleanup] Fix another place where we were including the tablegen'ed code
of a '.inc' file before including actual headers. In this case we had
both duplicated a header's include and were including a standard header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206840 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThis reverts r206828 until David has time to figure out that is going on.
Quentin Colombet [Tue, 22 Apr 2014 02:17:11 +0000 (02:17 +0000)]
This reverts r206828 until David has time to figure out that is going on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206839 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[cleanup] Lift using directives, DEBUG_TYPE definitions, and even some
Chandler Carruth [Tue, 22 Apr 2014 02:03:14 +0000 (02:03 +0000)]
[cleanup] Lift using directives, DEBUG_TYPE definitions, and even some
system headers above the includes of generated '.inc' files that
actually contain code. In a few targets this was already done pretty
consistently, but it wasn't done *really* consistently anywhere. It is
strictly cleaner IMO and necessary in a bunch of places where the
DEBUG_TYPE is referenced from the generated code. Consistency with the
necessary places trumps. Hopefully the build bots are OK with the
movement of intrin.h...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206838 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Remove potential ODR violations by sinking the DEBUG_TYPE
Chandler Carruth [Tue, 22 Apr 2014 02:02:50 +0000 (02:02 +0000)]
[Modules] Remove potential ODR violations by sinking the DEBUG_TYPE
define below all header includes in the lib/CodeGen/... tree. While the
current modules implementation doesn't check for this kind of ODR
violation yet, it is likely to grow support for it in the future. It
also removes one layer of macro pollution across all the included
headers.

Other sub-trees will follow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206837 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[CodeGenPrepare] Use APInt to check the value of the immediate in a and
Quentin Colombet [Tue, 22 Apr 2014 01:20:34 +0000 (01:20 +0000)]
[CodeGenPrepare] Use APInt to check the value of the immediate in a and
while checking candidate for bit field extract.
Otherwise the value may not fit in uint64_t and this will trigger an
assertion.

This fixes PR19503.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206834 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Followup to r206822 to add a DEBUG_TYPE which is used on ARM
Chandler Carruth [Mon, 21 Apr 2014 23:58:10 +0000 (23:58 +0000)]
[Modules] Followup to r206822 to add a DEBUG_TYPE which is used on ARM
and PPC, but not x86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206830 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to handle ownership of Value*s in Cloning unit tests.
David Blaikie [Mon, 21 Apr 2014 23:47:47 +0000 (23:47 +0000)]
Use unique_ptr to handle ownership of Value*s in Cloning unit tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206828 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix wrong iterator type
Rui Ueyama [Mon, 21 Apr 2014 23:00:42 +0000 (23:00 +0000)]
Fix wrong iterator type

ELFEntityIterator does not implement RandomAccessIterator. It does
not even implement BidirectionalIterator.

This patch fixes LLD build issue when compiled with MSVC2013 with
debug: MSVC's find_if checks if the start iterator is before the end
iterator in the sense of operator< if it declares implementing
RandomAccessIterator. If a class does not have operator<, it fails
to compile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206825 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSimplify DFAPacketizerEmitter State copy/move semantics to use compiler defaults.
David Blaikie [Mon, 21 Apr 2014 22:58:00 +0000 (22:58 +0000)]
Simplify DFAPacketizerEmitter State copy/move semantics to use compiler defaults.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206824 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Make Support/Debug.h modular. This requires it to not change
Chandler Carruth [Mon, 21 Apr 2014 22:55:11 +0000 (22:55 +0000)]
[Modules] Make Support/Debug.h modular. This requires it to not change
behavior based on other files defining DEBUG_TYPE, which means it cannot
define DEBUG_TYPE at all. This is actually better IMO as it forces folks
to define relevant DEBUG_TYPEs for their files. However, it requires all
files that currently use DEBUG(...) to define a DEBUG_TYPE if they don't
already. I've updated all such files in LLVM and will do the same for
other upstream projects.

This still leaves one important change in how LLVM uses the DEBUG_TYPE
macro going forward: we need to only define the macro *after* header
files have been #include-ed. Previously, this wasn't possible because
Debug.h required the macro to be pre-defined. This commit removes that.
By defining DEBUG_TYPE after the includes two things are fixed:

- Header files that need to provide a DEBUG_TYPE for some inline code
  can do so by defining the macro before their inline code and undef-ing
  it afterward so the macro does not escape.

- We no longer have rampant ODR violations due to including headers with
  different DEBUG_TYPE definitions. This may be mostly an academic
  violation today, but with modules these types of violations are easy
  to check for and potentially very relevant.

Where necessary to suppor headers with DEBUG_TYPE, I have moved the
definitions below the includes in this commit. I plan to move the rest
of the DEBUG_TYPE macros in LLVM in subsequent commits; this one is big
enough.

The comments in Debug.h, which were hilariously out of date already,
have been updated to reflect the recommended practice going forward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206822 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix builds that use an stl missing std::set::emplace
David Blaikie [Mon, 21 Apr 2014 22:46:09 +0000 (22:46 +0000)]
Fix builds that use an stl missing std::set::emplace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206821 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoStore State objects by value in TableGen's DFAPacketizerEmitter
David Blaikie [Mon, 21 Apr 2014 22:35:11 +0000 (22:35 +0000)]
Store State objects by value in TableGen's DFAPacketizerEmitter

Removes some extra manual dynamic memory allocation/management. It does
get a bit quirky having to make State's members mutable and
pointers/references to const rather than non-const, but that's a
necessary workaround to dealing with the std::set elements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206807 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSet default value of HasExtractBitsInsn to false
Yi Jiang [Mon, 21 Apr 2014 22:22:44 +0000 (22:22 +0000)]
Set default value of HasExtractBitsInsn to false

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206803 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: Refactor away a few redundant helpers.
Jim Grosbach [Mon, 21 Apr 2014 22:13:57 +0000 (22:13 +0000)]
ARM64: Refactor away a few redundant helpers.

The comment claimed that the register class information wasn't available
in the assembly parser, but that's not really true. It's just annoying to
get to. Replace the helper functions with references to the auto-generated
information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206802 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSimplify a vpermil* with constant mask.
Rafael Espindola [Mon, 21 Apr 2014 22:06:04 +0000 (22:06 +0000)]
Simplify a vpermil* with constant mask.

With a constant mask a vpermil* is just a shufflevector. This patch implements
that simplification. This allows us to produce denser code. It should also
allow more folding down the line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206801 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse Regex objects by value (rather than 'new'ed) in CodeGenSchedule.
David Blaikie [Mon, 21 Apr 2014 21:49:08 +0000 (21:49 +0000)]
Use Regex objects by value (rather than 'new'ed) in CodeGenSchedule.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206800 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: Improve diagnostics for malformed reg+reg addressing mode.
Jim Grosbach [Mon, 21 Apr 2014 21:45:57 +0000 (21:45 +0000)]
ARM64: Improve diagnostics for malformed reg+reg addressing mode.

Make sure only general purpose registers are valid for offset regs and
that 32-bit regs are only valid for sxtw and uxtw extends.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206799 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove helper functions earlier in the file.
Jim Grosbach [Mon, 21 Apr 2014 21:45:53 +0000 (21:45 +0000)]
Move helper functions earlier in the file.

No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206798 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: Extended addressing mode source reg is 64-bit.
Jim Grosbach [Mon, 21 Apr 2014 21:45:44 +0000 (21:45 +0000)]
ARM64: Extended addressing mode source reg is 64-bit.

The canonical form for the extended addressing mode (e.g.,
"[x1, w2, uxtw #3]" is for the MCInst to have the second register be the
full 64-bit GPR64 register class. The instruction printer cleans up
the output for display to show the 32-bit register instead, per the
specification.

This simplifies 205893 now that the aliasing is handled in the printer
in 206495 so that the codegen path and the disassembler path give the
same MCInst form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206797 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage ownership of GCOVFunctions, Blocks, and Edges.
David Blaikie [Mon, 21 Apr 2014 21:40:16 +0000 (21:40 +0000)]
Use unique_ptr to manage ownership of GCOVFunctions, Blocks, and Edges.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206796 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove seemingly-unneeded artificial dependency
Hal Finkel [Mon, 21 Apr 2014 21:30:25 +0000 (21:30 +0000)]
Remove seemingly-unneeded artificial dependency

The rationale for this artificial dependency seems to have been lost to the
ravages of time, it is covered by no regression tests, and has no impact on
test-suite performance numbers on either x86 or PPC.

For the test suite, on both x86 and PPC, I ran the test suite 10 times (both as
a baseline and with this change), and found no statistically-significant
changes.  For PPC, I used a P7 box. For x86, I used an Intel Xeon E5430. Both
with -O3 -mcpu=native.

This was discussed on-list back in January, but I've not had a chance to run
the performance tests until today.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206795 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSimplify destruction of Modules in LLVContextImpl.
David Blaikie [Mon, 21 Apr 2014 21:27:19 +0000 (21:27 +0000)]
Simplify destruction of Modules in LLVContextImpl.

This avoids copying the container by simply deleting until empty.

While I'd rather move to a stricter ownership semantic (unique_ptr),
SmallPtrSet can't cope with unique_ptr and the ownership semantics here
are a bit incestuous (Module sort of owns itself, but sort of doesn't
(if the LLVMContext is destroyed before the Module, then it deregisters
itself from the context... )).

Ideally Modules would be given to the context, or possibly an
emplace-like function to construct them there. Modules then shouldn't be
destroyed by LLVM API clients, but by interacting with the owner
(LLVMContext) directly (but even then, passing a Module* to LLVMContext
doesn't provide an easy way to destroy the Module, since the set would
be over unique_ptrs and you'd need a heterogenous lookup function which
SmallPtrSet doesn't have either).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206794 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoHandle _GLOBAL_OFFSET_TABLE_ in 64 bit mode.
Rafael Espindola [Mon, 21 Apr 2014 21:15:45 +0000 (21:15 +0000)]
Handle _GLOBAL_OFFSET_TABLE_ in 64 bit mode.

With this MC is able to handle _GLOBAL_OFFSET_TABLE_ in 64 bit mode, which is
needed for medium and large code models.

This fixes pr19470.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206793 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoclang-format this function.
Rafael Espindola [Mon, 21 Apr 2014 21:00:58 +0000 (21:00 +0000)]
clang-format this function.

No functionality change, it will just make the next patch easier to read.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206792 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to handle GlobalOpt's Evaluator members
David Blaikie [Mon, 21 Apr 2014 20:49:36 +0000 (20:49 +0000)]
Use unique_ptr to handle GlobalOpt's Evaluator members

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206790 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix PR7272 in -tailcallelim instead of the inliner
Reid Kleckner [Mon, 21 Apr 2014 20:48:47 +0000 (20:48 +0000)]
Fix PR7272 in -tailcallelim instead of the inliner

The -tailcallelim pass should be checking if byval or inalloca args can
be captured before marking calls as tail calls.  This was the real root
cause of PR7272.

With a better fix in place, revert the inliner change from r105255.  The
test case it introduced still passes and has been moved to
test/Transforms/Inline/byval-tail-call.ll.

Reviewers: chandlerc

Differential Revision: http://reviews.llvm.org/D3403

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206789 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSimplify expression that was explicitly naming an operator overload in a call.
David Blaikie [Mon, 21 Apr 2014 20:43:51 +0000 (20:43 +0000)]
Simplify expression that was explicitly naming an operator overload in a call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206788 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to handle ownership of GCOVFunctions in GCOVProfiler.
David Blaikie [Mon, 21 Apr 2014 20:41:55 +0000 (20:41 +0000)]
Use unique_ptr to handle ownership of GCOVFunctions in GCOVProfiler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206786 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to handle ownership of UserValues in LiveDebugVariablesImpl
David Blaikie [Mon, 21 Apr 2014 20:37:07 +0000 (20:37 +0000)]
Use unique_ptr to handle ownership of UserValues in LiveDebugVariablesImpl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206785 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to manage objects owned by the ScheduleDAGMI.
David Blaikie [Mon, 21 Apr 2014 20:32:32 +0000 (20:32 +0000)]
Use unique_ptr to manage objects owned by the ScheduleDAGMI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206784 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse value semantics to manage DbgVariables rather than dynamic allocation/pointers.
David Blaikie [Mon, 21 Apr 2014 20:13:09 +0000 (20:13 +0000)]
Use value semantics to manage DbgVariables rather than dynamic allocation/pointers.

Requires switching some vectors to lists to maintain pointer validity.
These could be changed to forward_lists (singly linked) with a bit more
work - I've left comments to that effect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206780 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRename X86insrtps to the proper instruction name.
Filipe Cabecinhas [Mon, 21 Apr 2014 20:07:29 +0000 (20:07 +0000)]
Rename X86insrtps to the proper instruction name.

Summary:
The INSERTPS pattern fragment was called insrtps (mising 'e'), which
would make it harder to grep for the patterns related to this instruction.
Renaming it to use the proper instruction name.

Reviewers: nadav

CC: llvm-commits
Differential Revision: http://reviews.llvm.org/D3443

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206779 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Consolidate the DEBUG_TYPE defines in NVPTX to the top of the
Chandler Carruth [Mon, 21 Apr 2014 19:53:55 +0000 (19:53 +0000)]
[Modules] Consolidate the DEBUG_TYPE defines in NVPTX to the top of the
cpp file rather than in the header and then again in the cpp file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206778 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Sink all the DEBUG_TYPE defines for InstCombine out of the
Chandler Carruth [Mon, 21 Apr 2014 19:51:41 +0000 (19:51 +0000)]
[Modules] Sink all the DEBUG_TYPE defines for InstCombine out of the
header files and into the cpp files.

These files will require more touches as the header files actually use
DEBUG(). Eventually, I'll have to introduce a matched #define and #undef
of DEBUG_TYPE for the header files, but that comes as step N of many to
clean all of this up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206777 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Modules] Sink the DEBUG_TYPE macro out of LegalizeTypes.h and into the
Chandler Carruth [Mon, 21 Apr 2014 19:43:07 +0000 (19:43 +0000)]
[Modules] Sink the DEBUG_TYPE macro out of LegalizeTypes.h and into the
various .cpp files. This macro is inherently non-modular, and it wasn't
even needed in this header file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206775 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM64: Combine shifts and uses from different basic block to bit-extract instruction
Yi Jiang [Mon, 21 Apr 2014 19:34:27 +0000 (19:34 +0000)]
ARM64: Combine shifts and uses from different basic block to bit-extract instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206774 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "[rtdyld,c++11] Range'ify symbol table walking."
Jim Grosbach [Mon, 21 Apr 2014 19:23:59 +0000 (19:23 +0000)]
Revert "[rtdyld,c++11] Range'ify symbol table walking."

Tentative revert for
http://lab.llvm.org:8011/builders/llvm-mips-linux/builds/8305.

This reverts commit c2a58efff07294fca724f89500538f2ddbcd12ff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206773 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix unnecessary line break
Matt Arsenault [Mon, 21 Apr 2014 18:39:13 +0000 (18:39 +0000)]
Fix unnecessary line break

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206772 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoblockfreq: Some cleanup of UnsignedFloat
Duncan P. N. Exon Smith [Mon, 21 Apr 2014 18:31:58 +0000 (18:31 +0000)]
blockfreq: Some cleanup of UnsignedFloat

Change `PositiveFloat` to `UnsignedFloat`, and fix some of the comments
to indicate that it's disappearing eventually.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206771 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[rtdyld,c++11] Range'ify symbol table walking.
Jim Grosbach [Mon, 21 Apr 2014 18:10:31 +0000 (18:10 +0000)]
[rtdyld,c++11] Range'ify symbol table walking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206769 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTidy up. Remove extraneous typedef.
Jim Grosbach [Mon, 21 Apr 2014 18:10:29 +0000 (18:10 +0000)]
Tidy up. Remove extraneous typedef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206768 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoObject: iterator_range accessors for ObjectImage symbols and sections.
Jim Grosbach [Mon, 21 Apr 2014 18:10:26 +0000 (18:10 +0000)]
Object: iterator_range accessors for ObjectImage symbols and sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206767 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReapply "blockfreq: Rewrite BlockFrequencyInfoImpl"
Duncan P. N. Exon Smith [Mon, 21 Apr 2014 17:57:07 +0000 (17:57 +0000)]
Reapply "blockfreq: Rewrite BlockFrequencyInfoImpl"

This reverts commit r206707, reapplying r206704.  The preceding commit
to CalcSpillWeights should have sorted out the failing buildbots.

<rdar://problem/14292693>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206766 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCalcSpillWeights: Hack to prevent x87 nonsense
Duncan P. N. Exon Smith [Mon, 21 Apr 2014 17:57:01 +0000 (17:57 +0000)]
CalcSpillWeights: Hack to prevent x87 nonsense

This gross hack forces `hweight` into memory, preventing hidden
precision from making `1 > 1` occasionally equal `true`.

<rdar://problem/14292693>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206765 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix the test: DCE optimized away everything.
Eli Bendersky [Mon, 21 Apr 2014 17:23:12 +0000 (17:23 +0000)]
Fix the test: DCE optimized away everything.

Use volatile store to protect the generated PTX from DCE.

Patch by Jingyue Wu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206763 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[msan] Enable out-of-line instrumentation for large functions by default.
Evgeniy Stepanov [Mon, 21 Apr 2014 15:04:05 +0000 (15:04 +0000)]
[msan] Enable out-of-line instrumentation for large functions by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206759 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAppease autoconf build since X86Disassembler.c has been disappeared in r206717.
NAKAMURA Takumi [Mon, 21 Apr 2014 14:59:11 +0000 (14:59 +0000)]
Appease autoconf build since X86Disassembler.c has been disappeared in r206717.

It can be reverted a few days later, after X86Disassembler.d is updated not to contain "X86Disassembler.c".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206758 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[asan] add a run-time flag detect_container_overflow=true/false
Kostya Serebryany [Mon, 21 Apr 2014 14:35:00 +0000 (14:35 +0000)]
[asan] add a run-time flag detect_container_overflow=true/false

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206756 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConvert getFileOffset to getOffset and move it to its only user.
Rafael Espindola [Mon, 21 Apr 2014 13:45:32 +0000 (13:45 +0000)]
Convert getFileOffset to getOffset and move it to its only user.

We normally don't drop functions from the C API's, but in this case I think we
can:

* The old implementation of getFileOffset was fairly broken
* The introduction of LLVMGetSymbolFileOffset was itself a C api breaking
  change as it removed LLVMGetSymbolOffset.
* It is an incredibly specialized use case. The only reason MCJIT needs it is
  because of its odd position of being a dynamic linker of .o files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206750 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReapply r206732. This time without optimization of branches.
Michael Zolotukhin [Mon, 21 Apr 2014 12:01:33 +0000 (12:01 +0000)]
Reapply r206732. This time without optimization of branches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206749 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[asan] add llvm-ish test for memset/etc instrumentation
Kostya Serebryany [Mon, 21 Apr 2014 11:57:43 +0000 (11:57 +0000)]
[asan] add llvm-ish test for memset/etc instrumentation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206747 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[asan] instead of inserting inline instrumentation around memset/memcpy/memmove,...
Kostya Serebryany [Mon, 21 Apr 2014 11:50:42 +0000 (11:50 +0000)]
[asan] instead of inserting inline instrumentation around memset/memcpy/memmove, replace the intrinsic with __asan_memset/etc. This makes the memset/etc handling more complete and consistent with what we do in msan. It may slowdown some cases (when the intrinsic was actually inlined) and speedup other cases (when it was not inlined)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206746 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Add a new-PM-style CGSCC pass manager using the newly added
Chandler Carruth [Mon, 21 Apr 2014 11:12:00 +0000 (11:12 +0000)]
[PM] Add a new-PM-style CGSCC pass manager using the newly added
LazyCallGraph analysis framework. Wire it up all the way through the opt
driver and add some very basic testing that we can build pass pipelines
including these components. Still a lot more to do in terms of testing
that all of this works, but the basic pieces are here.

There is a *lot* of boiler plate here. It's something I'm going to
actively look at reducing, but I don't have any immediate ideas that
don't end up making the code terribly complex in order to fold away the
boilerplate. Until I figure out something to minimize the boilerplate,
almost all of this is based on the code for the existing pass managers,
copied and heavily adjusted to suit the needs of the CGSCC pass
management layer.

The actual CG management still has a bunch of FIXMEs in it. Notably, we
don't do *any* updating of the CG as it is potentially invalidated.
I wanted to get this in place to motivate the new analysis, and add
update APIs to the analysis and the pass management layers in concert to
make sure that the *right* APIs are present.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206745 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Fix a bug where we didn't properly clear the list map when the list
Chandler Carruth [Mon, 21 Apr 2014 11:11:54 +0000 (11:11 +0000)]
[PM] Fix a bug where we didn't properly clear the list map when the list
became empty. This would manifest later as an assert failure due to
a non-empty list map but an empty result map. This doesn't easily
manifest with just the module pass manager and the function pass
manager, but the next commit will add the CGSCC pass manager that hits
this assert immediately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206744 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm/test/CodeGen/X86/bmi.ll: Relax expressions for targeting win32.
NAKAMURA Takumi [Mon, 21 Apr 2014 11:01:46 +0000 (11:01 +0000)]
llvm/test/CodeGen/X86/bmi.ll: Relax expressions for targeting win32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206743 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[asan] temporary disable generating __asan_loadN/__asan_storeN
Kostya Serebryany [Mon, 21 Apr 2014 10:28:13 +0000 (10:28 +0000)]
[asan] temporary disable generating __asan_loadN/__asan_storeN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206741 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Replace OwningPtr with std::unique_ptr in places where it doesn't break the...
Benjamin Kramer [Mon, 21 Apr 2014 09:34:48 +0000 (09:34 +0000)]
[C++11] Replace OwningPtr with std::unique_ptr in places where it doesn't break the API.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206740 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Wire the analysis passes (such as they are) into the registry, and
Chandler Carruth [Mon, 21 Apr 2014 08:20:10 +0000 (08:20 +0000)]
[PM] Wire the analysis passes (such as they are) into the registry, and
teach the opt driver to use it rather than a manual list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206739 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[X86] ISEL (and X, <constant mask>) to BZHI when BMI2 is available.
Lang Hames [Mon, 21 Apr 2014 08:18:53 +0000 (08:18 +0000)]
[X86] ISEL (and X, <constant mask>) to BZHI when BMI2 is available.

Generating BZHI in the variable mask case, i.e. (and X, (sub (shl 1, N), 1)),
was already supported, but we were missing the constant-mask case. This patch
fixes that.

<rdar://problem/15480077>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206738 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Add a nice low-tech registry of passes as a boring macro expansion
Chandler Carruth [Mon, 21 Apr 2014 08:08:50 +0000 (08:08 +0000)]
[PM] Add a nice low-tech registry of passes as a boring macro expansion
file. This will make it easy to scale up the number of passes supported.
Currently, it just supports the function and module transformation
passes that were already supported in the opt tool explicitly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206737 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r206732 which is causing llc to crash on most of the build bots.
Chandler Carruth [Mon, 21 Apr 2014 07:11:15 +0000 (07:11 +0000)]
Revert r206732 which is causing llc to crash on most of the build bots.

Original commit message:
  Implement builtins for safe division: safe.sdiv.iN, safe.udiv.iN,
  safe.srem.iN, safe.urem.iN (iN = i8, i61, i32, or i64).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206735 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[asan] insert __asan_loadN/__asan_storeN as out-lined asan checks, llvm part
Kostya Serebryany [Mon, 21 Apr 2014 07:10:43 +0000 (07:10 +0000)]
[asan] insert __asan_loadN/__asan_storeN as out-lined asan checks, llvm part

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206734 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement builtins for safe division: safe.sdiv.iN, safe.udiv.iN, safe.srem.iN,
Michael Zolotukhin [Mon, 21 Apr 2014 05:33:09 +0000 (05:33 +0000)]
Implement builtins for safe division: safe.sdiv.iN, safe.udiv.iN, safe.srem.iN,
safe.urem.iN (iN = i8, i16, i32, or i64).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206732 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[LCG] Add some basic debug output to the LCG pass.
Chandler Carruth [Mon, 21 Apr 2014 05:04:24 +0000 (05:04 +0000)]
[LCG] Add some basic debug output to the LCG pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206730 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoProtect the ArgList dtor
David Blaikie [Sun, 20 Apr 2014 23:59:00 +0000 (23:59 +0000)]
Protect the ArgList dtor

It could even be made non-virtual if it weren't for bad compiler
warnings.

This demonstrates that ArgList objects aren't destroyed polymorphically
and possibly that they aren't even used polymorphically. If that's the
case, it might be possible to refactor the two ArgList types more
separately and simplify the Arg ownership model. *continues
experimenting*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206727 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd missing #include found by modules build.
Richard Smith [Sun, 20 Apr 2014 23:39:19 +0000 (23:39 +0000)]
Add missing #include found by modules build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206726 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove comment that hasn't been true for 5 years
David Blaikie [Sun, 20 Apr 2014 22:40:43 +0000 (22:40 +0000)]
Remove comment that hasn't been true for 5 years

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206725 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse unique_ptr to handle ownership of synthesized args in DerivedArgList
David Blaikie [Sun, 20 Apr 2014 22:37:46 +0000 (22:37 +0000)]
Use unique_ptr to handle ownership of synthesized args in DerivedArgList

This might be able to be simplified further by using Arg as a value type
in a linked list (to maintain pointer validity), but here's something
simple to start with.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206724 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoC++ has a bool type! (And C's had one too, for 15 years...)
Richard Smith [Sun, 20 Apr 2014 22:15:37 +0000 (22:15 +0000)]
C++ has a bool type! (And C's had one too, for 15 years...)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206723 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMore C++ification.
Richard Smith [Sun, 20 Apr 2014 22:10:16 +0000 (22:10 +0000)]
More C++ification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206722 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove some more C junk from these files.
Richard Smith [Sun, 20 Apr 2014 21:56:02 +0000 (21:56 +0000)]
Remove some more C junk from these files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206721 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDon't provide two different definitions of ModRMDecision, OpcodeDecision, and Context...
Richard Smith [Sun, 20 Apr 2014 21:52:16 +0000 (21:52 +0000)]
Don't provide two different definitions of ModRMDecision, OpcodeDecision, and ContextDecision in different source files (depending on #define magic).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206720 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDon't define llvm::X86Disassembler::InstructionSpecifier in different ways in
Richard Smith [Sun, 20 Apr 2014 21:35:26 +0000 (21:35 +0000)]
Don't define llvm::X86Disassembler::InstructionSpecifier in different ways in
different source files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206719 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMaybe if I touch this file the buildbots will actually rerun configure like they...
Richard Smith [Sun, 20 Apr 2014 21:28:33 +0000 (21:28 +0000)]
Maybe if I touch this file the buildbots will actually rerun configure like they need to...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206718 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoWhat year is it! This file has no reason to be written in C, and has doubly no
Richard Smith [Sun, 20 Apr 2014 21:07:34 +0000 (21:07 +0000)]
What year is it! This file has no reason to be written in C, and has doubly no
reason to expose a global symbol 'decodeInstruction' nor to pollute the global
scope with a bunch of external linkage entities (some of which conflict with
others elsewhere in LLVM).

This is just the initial transition to C++; more cleanups to follow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206717 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Mips] Add more special values for the st_other field in the symbol
Simon Atanasyan [Sun, 20 Apr 2014 21:05:36 +0000 (21:05 +0000)]
[Mips] Add more special values for the st_other field in the symbol
table entry for MIPS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206716 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[C++11] Range-based loop simplification.
Simon Atanasyan [Sun, 20 Apr 2014 21:05:30 +0000 (21:05 +0000)]
[C++11] Range-based loop simplification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206715 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix redefinition of default argument, found by modules build. It's not
Richard Smith [Sun, 20 Apr 2014 20:26:39 +0000 (20:26 +0000)]
Fix redefinition of default argument, found by modules build.  It's not
entirely clear whether this should be valid with modules enabled, but the fixed
code is cleaner regardless.

Also fix a TU-local type that accidentally had external linkage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206714 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove some empty statements
Alp Toker [Sat, 19 Apr 2014 23:56:35 +0000 (23:56 +0000)]
Remove some empty statements

Cleanup only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206710 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoProfileData: Remove an extra semicolon
Justin Bogner [Sat, 19 Apr 2014 23:42:50 +0000 (23:42 +0000)]
ProfileData: Remove an extra semicolon

Spotted by Nick Lewycky in review, thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206708 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "blockfreq: Rewrite BlockFrequencyInfoImpl"
Duncan P. N. Exon Smith [Sat, 19 Apr 2014 22:46:00 +0000 (22:46 +0000)]
Revert "blockfreq: Rewrite BlockFrequencyInfoImpl"

This reverts commit r206704, as expected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206707 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "blockfreq: Temporarily turn on -debug-only=block-freq"
Duncan P. N. Exon Smith [Sat, 19 Apr 2014 22:45:44 +0000 (22:45 +0000)]
Revert "blockfreq: Temporarily turn on -debug-only=block-freq"

This reverts commit r206705, as planned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206706 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoblockfreq: Temporarily turn on -debug-only=block-freq
Duncan P. N. Exon Smith [Sat, 19 Apr 2014 22:40:56 +0000 (22:40 +0000)]
blockfreq: Temporarily turn on -debug-only=block-freq

These tests fail after my BlockFrequencyInfo rewrite on two buildbots
[1][2].  I can't reproduce it locally, so I'm temporarily turning on
-debug-only=block-freq so I can find the problem.

[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1860
[2]: http://llvm-amd64.freebsd.your.org/b/builders/clang-i386-freebsd/builds/18477

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206705 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReapply "blockfreq: Rewrite BlockFrequencyInfoImpl"
Duncan P. N. Exon Smith [Sat, 19 Apr 2014 22:34:26 +0000 (22:34 +0000)]
Reapply "blockfreq: Rewrite BlockFrequencyInfoImpl"

This reverts commit r206677, reapplying my BlockFrequencyInfo rewrite.

I've done a careful audit, added some asserts, and fixed a couple of
bugs (unfortunately, they were in unlikely code paths).  There's a small
chance that this will appease the failing bots [1][2].  (If so, great!)

If not, I have a follow-up commit ready that will temporarily add
-debug-only=block-freq to the two failing tests, allowing me to compare
the code path between what the failing bots and what my machines (and
the rest of the bots) are doing.  Once I've triggered those builds, I'll
revert both commits so the bots go green again.

[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1816
[2]: http://llvm-amd64.freebsd.your.org/b/builders/clang-i386-freebsd/builds/18445

<rdar://problem/14292693>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206704 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoPatch by Vadim Chugunov
Yaron Keren [Sat, 19 Apr 2014 13:47:43 +0000 (13:47 +0000)]
Patch by Vadim Chugunov

Win64 stack unwinder gets confused when execution flow "falls through" after
a call to 'noreturn' function. This fixes the "missing epilogue" problem by
emitting a trap instruction for IR 'unreachable' on x86_x64-pc-windows.

A secondary use for it would be for anyone wanting to make double-sure that
'noreturn' functions, indeed, do not return.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206684 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoPatch by Ray Donnelly to print register names instead of numbers.
Yaron Keren [Sat, 19 Apr 2014 05:40:09 +0000 (05:40 +0000)]
Patch by Ray Donnelly to print register names instead of numbers.

http://reviews.llvm.org/D3422

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206683 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd parens to appease GCC warning.
David Blaikie [Sat, 19 Apr 2014 00:50:15 +0000 (00:50 +0000)]
Add parens to appease GCC warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206678 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "blockfreq: Rewrite BlockFrequencyInfoImpl" (#2)
Duncan P. N. Exon Smith [Sat, 19 Apr 2014 00:42:46 +0000 (00:42 +0000)]
Revert "blockfreq: Rewrite BlockFrequencyInfoImpl" (#2)

This reverts commit r206666, as planned.

Still stumped on why the bots are failing.  Sanitizer bots haven't
turned anything up.  If anyone can help me debug either of the failures
(referenced in r206666) I'll owe them a beer.  (In the meantime, I'll be
auditing my patch for undefined behaviour.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206677 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOnDiskHashTable: Audit types and use offset_type consistently
Justin Bogner [Sat, 19 Apr 2014 00:33:15 +0000 (00:33 +0000)]
OnDiskHashTable: Audit types and use offset_type consistently

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206675 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoProfileData: Avoid UB when reading
Justin Bogner [Sat, 19 Apr 2014 00:33:12 +0000 (00:33 +0000)]
ProfileData: Avoid UB when reading

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206674 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOnDiskHashTable: Fix a think-o with offset_type
Justin Bogner [Fri, 18 Apr 2014 23:50:07 +0000 (23:50 +0000)]
OnDiskHashTable: Fix a think-o with offset_type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206672 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCheck whether functions have any lines associated before emitting coverage info for...
Nick Lewycky [Fri, 18 Apr 2014 23:32:28 +0000 (23:32 +0000)]
Check whether functions have any lines associated before emitting coverage info for them. This isn't just a size/time saving, gcov may crash on these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206671 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-profdata: Avoid writing to /dev/null in tests
Justin Bogner [Fri, 18 Apr 2014 23:25:35 +0000 (23:25 +0000)]
llvm-profdata: Avoid writing to /dev/null in tests

We fseek on our output file in llvm-profdata, which errors on some
systems. Avoid getting into the situation by writing to /dev/null

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206670 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoChange the ARM assembler to require a :lower16: or :upper16 on non-constant
Kevin Enderby [Fri, 18 Apr 2014 23:06:39 +0000 (23:06 +0000)]
Change the ARM assembler to require a :lower16: or :upper16 on non-constant
expressions for mov instructions instead of silently truncating by default.

For the ARM assembler, we want to avoid misleadingly allowing something
like "mov r0, <symbol>" especially when we turn it into a movw and the
expression <symbol> does not have a :lower16: or :upper16" as part of the
expression.  We don't want the behavior of silently truncating, which can be
unexpected and lead to bugs that are difficult to find since this is an easy
mistake to make.

This does change the previous behavior of llvm but actually matches an
older gnu assembler that would not allow this but print less useful errors
of like “invalid constant (0x927c0) after fixup” and “unsupported relocation on
symbol foo”.  The error for llvm is "immediate expression for mov requires
:lower16: or :upper16" with correct location information on the operand
as shown in the added test cases.

rdar://12342160

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206669 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agotest: Add extra run lines to investigate an error on the bots
Justin Bogner [Fri, 18 Apr 2014 23:05:31 +0000 (23:05 +0000)]
test: Add extra run lines to investigate an error on the bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206668 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReapply "blockfreq: Rewrite BlockFrequencyInfoImpl" (#2)
Duncan P. N. Exon Smith [Fri, 18 Apr 2014 22:30:03 +0000 (22:30 +0000)]
Reapply "blockfreq: Rewrite BlockFrequencyInfoImpl" (#2)

This reverts commit r206628, reapplying r206622 (and r206626).

Two tests are failing only on buildbots [1][2]: i.e., I can't reproduce
on Darwin, and Chandler can't reproduce on Linux.  Asan and valgrind
don't tell us anything, but we're hoping the msan bot will catch it.

So, I'm applying this again to get more feedback from the bots.  I'll
leave it in long enough to trigger builds in at least the sanitizer
buildbots (it was failing for reasons unrelated to my commit last time
it was in), and hopefully a few others.... and then I expect to revert a
third time.

[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1816
[2]: http://llvm-amd64.freebsd.your.org/b/builders/clang-i386-freebsd/builds/18445

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206666 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-symbolizer] Print file/line for a PC even if there is no DIE describing it.
Alexey Samsonov [Fri, 18 Apr 2014 22:22:44 +0000 (22:22 +0000)]
[llvm-symbolizer] Print file/line for a PC even if there is no DIE describing it.

This is important for symbolizing executables with debug info in
unavailable .dwo files. Even if all DIE entries are missing, we can
still symbolize an address: function name can be fetched from symbol table,
and file/line info can be fetched from line table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206665 91177308-0d34-0410-b5e6-96231b3b80d8