oota-llvm.git
10 years agoFix missing C++ mode comment
Matt Arsenault [Sun, 10 Nov 2013 01:03:59 +0000 (01:03 +0000)]
Fix missing C++ mode comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194339 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse variable for register name in test
Matt Arsenault [Sun, 10 Nov 2013 00:57:17 +0000 (00:57 +0000)]
Use variable for register name in test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194338 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMostly finish up constant islands port for Mips for load constants.
Reed Kotler [Sun, 10 Nov 2013 00:09:26 +0000 (00:09 +0000)]
Mostly finish up constant islands port for Mips for load constants.
Still need to finish the branch part. Still lots more review of the code,
clean up and testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194337 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove some unnecessary temporary strings.
Benjamin Kramer [Sat, 9 Nov 2013 22:48:13 +0000 (22:48 +0000)]
Remove some unnecessary temporary strings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194335 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[arm] Refine ARMBuildAttrs.h.
Logan Chien [Sat, 9 Nov 2013 14:16:52 +0000 (14:16 +0000)]
[arm] Refine ARMBuildAttrs.h.

This commit cleans up some comments in ARMBuildAttrs.h.
Besides, this commit fixes an error related to AllowWMMXv1
and AllowWMMXv2 (although they are not used currently.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194327 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Start sketching out the new module and function pass manager.
Chandler Carruth [Sat, 9 Nov 2013 13:09:08 +0000 (13:09 +0000)]
[PM] Start sketching out the new module and function pass manager.

This is still just a skeleton. I'm trying to pull together the
experimentation I've done into committable chunks, and this is the first
coherent one. Others will follow in hopefully short order that move this
more toward a useful initial implementation. I still expect the design
to continue evolving in small ways as I work through the different
requirements and features needed here though.

Keep in mind, all of this is off by default.

Currently, this mostly exercises the use of a polymorphic smart pointer
and templates to hide the polymorphism for the pass manager from the
pass implementation. The next step will be more significant, adding the
first framework of analysis support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194325 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove the old pass manager infrastructure into a legacy namespace and
Chandler Carruth [Sat, 9 Nov 2013 12:26:54 +0000 (12:26 +0000)]
Move the old pass manager infrastructure into a legacy namespace and
give the files a legacy prefix in the right directory. Use forwarding
headers in the old locations to paper over the name change for most
clients during the transitional period.

No functionality changed here! This is just clearing some space to
reduce renaming churn later on with a new system.

Even when the new stuff starts to go in, it is going to be hidden behind
a flag and off-by-default as it is still WIP and under development.

This patch is specifically designed so that very little out-of-tree code
has to change. I'm going to work as hard as I can to keep that the case.
Only direct forward declarations of the PassManager class are impacted
by this change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194324 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThis exposes the new calling conventions (WebKit_JS and AnyReg) via the C API by...
Filip Pizlo [Sat, 9 Nov 2013 06:00:03 +0000 (06:00 +0000)]
This exposes the new calling conventions (WebKit_JS and AnyReg) via the C API by adding them to the enumeration in Core.h.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194323 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSwitch to allow implicit construction. In many cases, we're wrapping
Chandler Carruth [Sat, 9 Nov 2013 05:55:03 +0000 (05:55 +0000)]
Switch to allow implicit construction. In many cases, we're wrapping
a derived type and this makes it *much* easier to write this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194321 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTest the polymorphic behavior of this utility.
Chandler Carruth [Sat, 9 Nov 2013 04:58:13 +0000 (04:58 +0000)]
Test the polymorphic behavior of this utility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194320 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse something really explicit to test "move semantics" on builds without
Chandler Carruth [Sat, 9 Nov 2013 04:49:27 +0000 (04:49 +0000)]
Use something really explicit to test "move semantics" on builds without
r-value references. I still want to test that when we have them,
llvm_move is actually a move.

Have I mentioned that I really want to move to C++11? ;]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194318 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd the critically missing 'clone' method. =]
Chandler Carruth [Sat, 9 Nov 2013 04:32:34 +0000 (04:32 +0000)]
Add the critically missing 'clone' method. =]

Clang managed to never instantiate the copy constructor. Added tests to
ensure this path is tested.

We could still use tests for the polymorphic nature. Those coming up
next.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194317 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove the test type out of the function and into the anonymous namespace
Chandler Carruth [Sat, 9 Nov 2013 04:09:50 +0000 (04:09 +0000)]
Move the test type out of the function and into the anonymous namespace
to fix C++98 builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194316 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a polymorphic_ptr<T> smart pointer data type. It's a somewhat silly
Chandler Carruth [Sat, 9 Nov 2013 04:06:02 +0000 (04:06 +0000)]
Add a polymorphic_ptr<T> smart pointer data type. It's a somewhat silly
unique ownership smart pointer which is *deep* copyable by assuming it
can call a T::clone() method to allocate a copy of the owned data.

This is mostly useful with containers or other collections of uniquely
owned data in C++98 where they *might* copy. With C++11 we can likely
remove this in favor of move-only types and containers wrapped around
those types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194315 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoinclude/llvm/CodeGen/PBQP: Update @param(s) in comments. [-Wdocumentation]
NAKAMURA Takumi [Sat, 9 Nov 2013 03:54:05 +0000 (03:54 +0000)]
include/llvm/CodeGen/PBQP: Update @param(s) in comments. [-Wdocumentation]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194314 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix whitespace.
NAKAMURA Takumi [Sat, 9 Nov 2013 03:53:55 +0000 (03:53 +0000)]
Fix whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194313 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRe-apply r194300 with fixes for warnings.
Lang Hames [Sat, 9 Nov 2013 03:08:56 +0000 (03:08 +0000)]
Re-apply r194300 with fixes for warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194311 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips] Make sure there is a chain edge dependency between loads that read
Akira Hatanaka [Sat, 9 Nov 2013 02:38:51 +0000 (02:38 +0000)]
[mips] Make sure there is a chain edge dependency between loads that read
formal arguments on the stack and stores created afterwards. We need this to
ensure tail call optimized function calls do not write over the argument area
of the stack before it is read out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194309 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r194300 which broke the build.
Nick Lewycky [Sat, 9 Nov 2013 02:01:25 +0000 (02:01 +0000)]
Revert r194300 which broke the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194308 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmap] Materialize the jump address within the patchpoint noop slide.
Juergen Ributzka [Sat, 9 Nov 2013 01:51:33 +0000 (01:51 +0000)]
[Stackmap] Materialize the jump address within the patchpoint noop slide.

This patch moves the jump address materialization inside the noop slide. This
enables patching of the materialization itself or its complete removal. This
patch also adds the ability to define scratch registers that can be used safely
by the code called from the patchpoint intrinsic. At least one scratch register
is required, because that one is used for the materialization of the jump
address. This patch depends on D2009.

Differential Revision: http://llvm-reviews.chandlerc.com/D2074

Reviewed by Andy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194306 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Move copying of global initializers below the cloning of functions."
Adrian Prantl [Sat, 9 Nov 2013 00:43:18 +0000 (00:43 +0000)]
Revert "Move copying of global initializers below the cloning of functions."

This would cause internal symbols that are only referenced by global initializers to be removed.

This reverts commit 194219.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194304 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Run clang-format on file."
Adrian Prantl [Sat, 9 Nov 2013 00:43:12 +0000 (00:43 +0000)]
Revert "Run clang-format on file."

This reverts commit 194219.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194303 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRewrite the PBQP graph data structure.
Lang Hames [Sat, 9 Nov 2013 00:14:07 +0000 (00:14 +0000)]
Rewrite the PBQP graph data structure.

The new graph structure replaces the node and edge linked lists with vectors.
Free lists (well, free vectors) are used for fast insertion/deletion.

The ultimate aim is to make PBQP graphs cheap to clone. The motivation is that
the PBQP solver destructively consumes input graphs while computing a solution,
forcing the graph to be fully reconstructed for each round of PBQP. This
imposes a high cost on large functions, which often require several rounds of
solving/spilling to find a final register allocation. If we can cheaply clone
the PBQP graph and incrementally update it between rounds then hopefully we can
reduce this cost. Further, once we begin pooling matrix/vector values (future
work), we can cache some PBQP solver metadata and share it between cloned
graphs, allowing the PBQP solver to re-use some of the computation done in
earlier rounds.

For now this is just a data structure update. The allocator and solver still
use the graph the same way as before, fully reconstructing it between each
round. I expect no material change from this update, although it may change
the iteration order of the nodes, causing ties in the solver to break in
different directions, and this could perturb the generated allocations
(hopefully in a completely benign way).

Thanks very much to Arnaud Allard de Grandmaison for encouraging me to get back
to work on this, and for a lot of discussion and many useful PBQP test cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194300 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmap] Add AnyReg calling convention support for patchpoint intrinsic.
Juergen Ributzka [Fri, 8 Nov 2013 23:28:16 +0000 (23:28 +0000)]
[Stackmap] Add AnyReg calling convention support for patchpoint intrinsic.

The idea of the AnyReg Calling Convention is to provide the call arguments in
registers, but not to force them to be placed in a paticular order into a
specified set of registers. Instead it is up tp the register allocator to assign
any register as it sees fit. The same applies to the return value (if
applicable).

Differential Revision: http://llvm-reviews.chandlerc.com/D2009

Reviewed by Andy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194293 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoincrease the accuracy of register pressure computation in the presence of dead defini...
Pedro Artigas [Fri, 8 Nov 2013 22:46:28 +0000 (22:46 +0000)]
increase the accuracy of register pressure computation in the presence of dead definitions by using live intervals, if available, to identify dead definitions and proceed accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194286 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoX86: Assembly files with .cfi_cfa_def shouldn't hit llvm_unreachable()
Jim Grosbach [Fri, 8 Nov 2013 22:33:06 +0000 (22:33 +0000)]
X86: Assembly files with .cfi_cfa_def shouldn't hit llvm_unreachable()

On darwin, when trying to create compact unwind info, a .cfi_cfa_def
directive would case an llvm_unreachable() to be hit. Back off when we
see this directive and generate the regular DWARF style eh_frame.

rdar://15406518

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194285 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix some minor issues with r194282 to get the tree healthy again.
Lang Hames [Fri, 8 Nov 2013 22:30:52 +0000 (22:30 +0000)]
Fix some minor issues with r194282 to get the tree healthy again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194284 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a method to get the object-file appropriate stack map section.
Lang Hames [Fri, 8 Nov 2013 22:14:49 +0000 (22:14 +0000)]
Add a method to get the object-file appropriate stack map section.

Thanks to Eric Christopher for the tips on the appropriate way to do this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194282 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove dead code from LoopUnswitch
Hal Finkel [Fri, 8 Nov 2013 19:58:21 +0000 (19:58 +0000)]
Remove dead code from LoopUnswitch

LoopUnswitch's code simplification routine has logic to convert conditional
branches into unconditional branches, after unswitching makes the condition
constant, and then remove any blocks that renders dead. Unfortunately, this
code is dead, currently broken, and furthermore, has never been alive (at least
as far back at 2006).

No functionality change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194277 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "CalculateSpillWeights does not need to be a pass"
Arnaud A. de Grandmaison [Fri, 8 Nov 2013 18:19:19 +0000 (18:19 +0000)]
Revert "CalculateSpillWeights does not need to be a pass"

Temporarily revert my previous commit until I understand why it breaks 3 target tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194272 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[VirtRegMap] Fix for PR17825. Do not ignore noreturn definitions when setting
Quentin Colombet [Fri, 8 Nov 2013 18:14:17 +0000 (18:14 +0000)]
[VirtRegMap] Fix for PR17825. Do not ignore noreturn definitions when setting
isPhysRegUsed if the unwind information is required.
Indeed, the runtime may need a correct stack to be able to unwind the call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194271 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMake PrintAsmOperand call to the superclass to handle 'n' and 'c' operand modifiers.
Richard Barton [Fri, 8 Nov 2013 18:09:57 +0000 (18:09 +0000)]
Make PrintAsmOperand call to the superclass to handle 'n' and 'c' operand modifiers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194270 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCalculateSpillWeights does not need to be a pass
Arnaud A. de Grandmaison [Fri, 8 Nov 2013 17:56:29 +0000 (17:56 +0000)]
CalculateSpillWeights does not need to be a pass

Based on discussions with Lang Hames and Jakob Stoklund Olesen at the hacker's lab, and in the light of upcoming work on the PBQP register allocator, it was though that CalcSpillWeights does not need to be a pass. This change will enable to customize / tune the spill weight computation depending on the allocator.

Update the documentation style while there.

No functionnal change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194269 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd ImmutableSet profiling info for 'bool'.
Jordan Rose [Fri, 8 Nov 2013 17:23:49 +0000 (17:23 +0000)]
Add ImmutableSet profiling info for 'bool'.

Useful for tri-state maps: true, false, and "no data yet".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194266 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: fold prologue/epilogue sp updates into push/pop for code size
Tim Northover [Fri, 8 Nov 2013 17:18:07 +0000 (17:18 +0000)]
ARM: fold prologue/epilogue sp updates into push/pop for code size

ARM prologues usually look like:
    push {r7, lr}
    sub sp, sp, #4

If code size is extremely important, this can be optimised to the single
instruction:
    push {r6, r7, lr}

where we don't actually care about the contents of r6, but pushing it subtracts
4 from sp as a side effect.

This should implement such a conversion, predicated on the "minsize" function
attribute (-Oz) since I've yet to find any code it actually makes faster.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194264 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8...
Artyom Skrobov [Fri, 8 Nov 2013 16:25:50 +0000 (16:25 +0000)]
[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8 (Thumb encodings)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194263 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8...
Artyom Skrobov [Fri, 8 Nov 2013 16:17:14 +0000 (16:17 +0000)]
[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8 (ARM encodings)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194262 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8...
Artyom Skrobov [Fri, 8 Nov 2013 16:16:30 +0000 (16:16 +0000)]
[ARM] Handling for coprocessor instructions that are undefined starting from ARMv8 (ARM encodings)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194261 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoExport MCDisassembler's SubtargetInfo, to allow architecture-aware disassembly
Artyom Skrobov [Fri, 8 Nov 2013 16:07:43 +0000 (16:07 +0000)]
Export MCDisassembler's SubtargetInfo, to allow architecture-aware disassembly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194260 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCalculateSpillWeights cleanup: remove unneeded includes
Arnaud A. de Grandmaison [Fri, 8 Nov 2013 15:13:05 +0000 (15:13 +0000)]
CalculateSpillWeights cleanup: remove unneeded includes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194259 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTest for microMIPS trap instructions.
Zoran Jovanovic [Fri, 8 Nov 2013 14:55:31 +0000 (14:55 +0000)]
Test for microMIPS trap instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194258 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-ar: Let opening a directory failed in llvm-ar.
NAKAMURA Takumi [Fri, 8 Nov 2013 12:35:56 +0000 (12:35 +0000)]
llvm-ar: Let opening a directory failed in llvm-ar.

Linux cannot open directories with open(2), although cygwin and *bsd can.

Motivation: The test, Object/directory.ll, had been failing with --target=cygwin on Linux. XFAIL was improper for host issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194257 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips][msa] Update encoding of LDI instruction.
Matheus Almeida [Fri, 8 Nov 2013 10:43:11 +0000 (10:43 +0000)]
[mips][msa] Update encoding of LDI instruction.

The encoding was updated in MSA r1.07.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194255 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] In ARMAsmParser, MatchCoprocessorOperandName() permitted p10 and p11 as operand...
Artyom Skrobov [Fri, 8 Nov 2013 09:16:31 +0000 (09:16 +0000)]
[ARM] In ARMAsmParser, MatchCoprocessorOperandName() permitted p10 and p11 as operands for coprocessor instructions, resulting in encodings that clash with FP/NEON instruction encodings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194253 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove ^M from the file.
Bill Wendling [Fri, 8 Nov 2013 08:13:15 +0000 (08:13 +0000)]
Remove ^M from the file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194251 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoIR: Properly canonicalize PointerType in ConstantExpr GEPs
David Majnemer [Thu, 7 Nov 2013 22:29:42 +0000 (22:29 +0000)]
IR: Properly canonicalize PointerType in ConstantExpr GEPs

No additional test was needed, Other/constant-fold-gep.ll detects this
just fine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194221 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoIR: Do not canonicalize constant GEPs into an out-of-bounds array access
David Majnemer [Thu, 7 Nov 2013 22:15:53 +0000 (22:15 +0000)]
IR: Do not canonicalize constant GEPs into an out-of-bounds array access

Summary:
Consider a GEP of:
i8* getelementptr ({ [2 x i8], i32, i8, [3 x i8] }* @main.c, i32 0, i32 0, i64 0)

If we proceeded to GEP the aforementioned object by 8, would form a GEP of:
i8* getelementptr ({ [2 x i8], i32, i8, [3 x i8] }* @main.c, i32 0, i32 0, i64 8)

Note that we would go through the first array member, causing an
out-of-bounds accesses.  This is problematic because we might get fooled
if we are trying to evaluate loads using this GEP, for example, based
off of an object with a constant initializer where the array is zero.

This fixes PR17732.

Reviewers: nicholas, chandlerc, void

Reviewed By: void

CC: llvm-commits, echristo, void, aemerson
Differential Revision: http://llvm-reviews.chandlerc.com/D2093

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194220 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRun clang-format on file.
Bill Wendling [Thu, 7 Nov 2013 20:18:21 +0000 (20:18 +0000)]
Run clang-format on file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194219 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMove copying of global initializers below the cloning of functions.
Bill Wendling [Thu, 7 Nov 2013 20:14:51 +0000 (20:14 +0000)]
Move copying of global initializers below the cloning of functions.

The BlockAddress doesn't have access to the correct basic blocks until the
functions have been cloned. This causes the BlockAddress to point to the old
values. Just wait until the functions have been cloned before copying the
initializers.
PR13163

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194218 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSupport for microMIPS trap instructions 1.
Zoran Jovanovic [Thu, 7 Nov 2013 14:35:24 +0000 (14:35 +0000)]
Support for microMIPS trap instructions 1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194205 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-c/Support.h: Add a newline at eof.
NAKAMURA Takumi [Thu, 7 Nov 2013 13:54:24 +0000 (13:54 +0000)]
llvm-c/Support.h: Add a newline at eof.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194203 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDisable some code that is causing some warnings. It's in the process
Reed Kotler [Thu, 7 Nov 2013 11:56:33 +0000 (11:56 +0000)]
Disable some code that is causing some warnings. It's in the process
of being converted and this path is not relevant to anything at this time
so I have just disabled it for a few days while I'm at the LLVM conference
and don't have time to complete it or properly fix it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194201 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd the fact that we anticipate switching to use (some subset of) C++11
Chandler Carruth [Thu, 7 Nov 2013 00:23:08 +0000 (00:23 +0000)]
Add the fact that we anticipate switching to use (some subset of) C++11
after the 3.4 release to the release notes. See the *lengthy* llvmdev
and cfe-dev threads on this subject. There will be more emails,
discussion and announcements, but I want to make noise in as many places
as I can to get everyone's concerns voiced and understood.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194183 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Fix LowerUDIVREM
Vincent Lejeune [Wed, 6 Nov 2013 17:36:04 +0000 (17:36 +0000)]
R600: Fix LowerUDIVREM

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194153 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Remove NEON from "generic" CPU target.
Amara Emerson [Wed, 6 Nov 2013 16:19:08 +0000 (16:19 +0000)]
[AArch64] Remove NEON from "generic" CPU target.

We can change this back when NEON support is complete and ready to become
enabled by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194152 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd DT_VERSYM dynamic table entry tag definition.
Simon Atanasyan [Wed, 6 Nov 2013 12:23:52 +0000 (12:23 +0000)]
Add DT_VERSYM dynamic table entry tag definition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194149 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[SystemZ] Handle vectors in getSetCCResultType
Richard Sandiford [Wed, 6 Nov 2013 12:16:02 +0000 (12:16 +0000)]
[SystemZ] Handle vectors in getSetCCResultType

I don't have a standalone testcase for this, but it should allow r193676
to be reapplied.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194148 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd test case for PR12377, it was fixed by r194116.
Benjamin Kramer [Wed, 6 Nov 2013 11:55:41 +0000 (11:55 +0000)]
Add test case for PR12377, it was fixed by r194116.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194147 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Add parameter names in Target.h for C99 compliance
Peter Zotov [Wed, 6 Nov 2013 11:52:40 +0000 (11:52 +0000)]
[llvm-c] Add parameter names in Target.h for C99 compliance

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194146 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement gpword directive for mips, test case added. Stype changes using clang-forma...
Vladimir Medic [Wed, 6 Nov 2013 11:27:05 +0000 (11:27 +0000)]
Implement gpword directive for mips, test case added. Stype changes using clang-format are also included.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194145 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd newline at EOF in DynamicLibrary.cpp
Peter Zotov [Wed, 6 Nov 2013 11:12:46 +0000 (11:12 +0000)]
Add newline at EOF in DynamicLibrary.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194144 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Improve TargetMachine bindings
Peter Zotov [Wed, 6 Nov 2013 10:25:18 +0000 (10:25 +0000)]
[llvm-c] Improve TargetMachine bindings

Original patch by Chris Wailes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194143 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Specify explicit namespace in LLVMLoadLibraryPermanently
Peter Zotov [Wed, 6 Nov 2013 09:45:58 +0000 (09:45 +0000)]
[llvm-c] Specify explicit namespace in LLVMLoadLibraryPermanently

Presence of using namespace llvm depended on several #ifdef's, and
this broke the build on mswin32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194142 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Correctly check for existence of native AsmParser, AsmPrinter, Disassembler
Peter Zotov [Wed, 6 Nov 2013 09:45:53 +0000 (09:45 +0000)]
[llvm-c] Correctly check for existence of native AsmParser, AsmPrinter, Disassembler

Also, properly name the functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194141 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Add functions for initializing native AsmPrinter, AsmParser & Disassembler
Peter Zotov [Wed, 6 Nov 2013 09:21:35 +0000 (09:21 +0000)]
[llvm-c] Add functions for initializing native AsmPrinter, AsmParser & Disassembler

Original patch by Chris Wailes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194140 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Expose LLVMLoadLibraryPermanently
Peter Zotov [Wed, 6 Nov 2013 09:21:31 +0000 (09:21 +0000)]
[llvm-c] Expose LLVMLoadLibraryPermanently

Original patch by Chris Wailes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194139 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[OCaml] Impement Llvm_irreader, bindings to LLVM assembly parser
Peter Zotov [Wed, 6 Nov 2013 09:21:25 +0000 (09:21 +0000)]
[OCaml] Impement Llvm_irreader, bindings to LLVM assembly parser

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194138 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Expose IRReader interface
Peter Zotov [Wed, 6 Nov 2013 09:21:15 +0000 (09:21 +0000)]
[llvm-c] Expose IRReader interface

Original patch by Chris Wailes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194137 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[OCaml] Implement Llvm.string_of_llvalue
Peter Zotov [Wed, 6 Nov 2013 09:21:08 +0000 (09:21 +0000)]
[OCaml] Implement Llvm.string_of_llvalue

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194136 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[llvm-c] Implement LLVMPrintValueToString
Peter Zotov [Wed, 6 Nov 2013 09:21:01 +0000 (09:21 +0000)]
[llvm-c] Implement LLVMPrintValueToString

Original patch by Chris Wailes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194135 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix definition for Mips16 pc relative load word instructions.
Reed Kotler [Wed, 6 Nov 2013 04:29:52 +0000 (04:29 +0000)]
Fix definition for Mips16 pc relative load word instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194126 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement AArch64 Neon instruction set Perm.
Jiangning Liu [Wed, 6 Nov 2013 03:35:27 +0000 (03:35 +0000)]
Implement AArch64 Neon instruction set Perm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194123 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement AArch64 Neon instruction set Bitwise Extract.
Jiangning Liu [Wed, 6 Nov 2013 02:25:49 +0000 (02:25 +0000)]
Implement AArch64 Neon instruction set Bitwise Extract.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194118 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRewrite SCEV's backedge taken count computation.
Andrew Trick [Wed, 6 Nov 2013 02:08:26 +0000 (02:08 +0000)]
Rewrite SCEV's backedge taken count computation.

Patch by Michele Scandale!

Rewrite of the functions used to compute the backedge taken count of a
loop on LT and GT comparisons.

I decided to split the handling of LT and GT cases becasue the trick
"a > b == -a < -b" in some cases prevents the trip count computation
due to the multiplication by -1 on the two operands of the
comparison. This issue comes from the conservative computation of
value range of SCEVs: taking the negative SCEV of an expression that
have a small positive range (e.g. [0,31]), we would have a SCEV with a
fullset as value range.

Indeed, in the new rewritten function I tried to better handle the
maximum backedge taken count computation when MAX/MIN expression are
used to handle the cases where no entry guard is found.

Some test have been modified in order to check the new value correctly
(I manually check them and reasoning on possible overflow the new
values seem correct).

I finally added a new test case related to the multiplication by -1
issue on GT comparisons.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194116 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove another unused, and IMHO, not very desirable feature of ErrorOr.
Rafael Espindola [Tue, 5 Nov 2013 23:41:57 +0000 (23:41 +0000)]
Remove another unused, and IMHO, not very desirable feature of ErrorOr.

One of the uses of the IsValid flag is to support default constructing
a ErrorOr that is not a Error or a Value. There is not much value in
doing that IMHO. If ErrorOr was to have a default constructor, it
should be implemented by default constructing the value, but even that
looks unnecessary.

The other use is to avoid calling destructors on moved objects. This
looks wrong. If the data being moved has non trivial treatment of
moves (an std::vector for example), it is its destructor that should
handle it, not ~ErrorOr.

With this change ErrorOr becomes a fairly simple wrapper and should
always be better than using an error_code + value in an API.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194109 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoGet rid of current calculation function and adjustment scheme
Reed Kotler [Tue, 5 Nov 2013 23:36:58 +0000 (23:36 +0000)]
Get rid of current calculation function and adjustment scheme
from MipsConstantIslands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194108 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSlightly change the way stackmap and patchpoint intrinsics are lowered.
Andrew Trick [Tue, 5 Nov 2013 22:44:04 +0000 (22:44 +0000)]
Slightly change the way stackmap and patchpoint intrinsics are lowered.

MorphNodeTo is not safe to call during DAG building. It eagerly
deletes dependent DAG nodes which invalidates the NodeMap. We could
expose a safe interface for morphing nodes, but I don't think it's
worth it. Just create a new MachineNode and replaceAllUsesWith.

My understaning of the SD design has been that we want to support
early target opcode selection. That isn't very well supported, but
generally works. It seems reasonable to rely on this feature even if
it isn't widely used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194102 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoGet rid of all references to soimm in MipsConstantIslands pass because
Reed Kotler [Tue, 5 Nov 2013 22:34:29 +0000 (22:34 +0000)]
Get rid of all references to soimm in MipsConstantIslands pass because
we don't have such an operand.
Suprisingly enough, this is never actually accounted for in the
ARM version when determining offset ranges. In both places there is the
comment:
-    // FIXME: Make use full range of soimm values.
(soimm = shift operand immediate).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194101 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCleanup getUserOffset. Issues related to inline assembler length and
Reed Kotler [Tue, 5 Nov 2013 21:39:57 +0000 (21:39 +0000)]
Cleanup getUserOffset. Issues related to inline assembler length and
alignment will be handled differently than in ARM constant islands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194096 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoARM: permit bare dmb/dsb/isb aliases on Cortex-M0
Tim Northover [Tue, 5 Nov 2013 21:36:02 +0000 (21:36 +0000)]
ARM: permit bare dmb/dsb/isb aliases on Cortex-M0

Cortex-M0 supports these 32-bit instructions despite being Thumb1 only
(mostly). We knew about that but not that the aliases without the default "sy"
operand were also permitted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194094 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConvert comments to documentation comments (// -> ///)
Dmitri Gribenko [Tue, 5 Nov 2013 21:28:42 +0000 (21:28 +0000)]
Convert comments to documentation comments (// -> ///)

Patch by MathOnNapkins

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194093 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse error_code in GVMaterializer.
Rafael Espindola [Tue, 5 Nov 2013 19:36:34 +0000 (19:36 +0000)]
Use error_code in GVMaterializer.

They just propagate out the bitcode reader error, so we don't need a new enum.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194091 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoImplement AArch64 Neon Crypto instruction classes AES, SHA, and 3 SHA.
Jiangning Liu [Tue, 5 Nov 2013 17:42:05 +0000 (17:42 +0000)]
Implement AArch64 Neon Crypto instruction classes AES, SHA, and 3 SHA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194085 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoConvert FindFunctionInStream to return an error_code.
Rafael Espindola [Tue, 5 Nov 2013 17:16:08 +0000 (17:16 +0000)]
Convert FindFunctionInStream to return an error_code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194084 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[objc-arc] Convert the one directional retain/release relation assert to a conditiona...
Michael Gottesman [Tue, 5 Nov 2013 16:02:40 +0000 (16:02 +0000)]
[objc-arc] Convert the one directional retain/release relation assert to a conditional check + fail.

Due to the previously added overflow checks, we can have a retain/release
relation that is one directional. This occurs specifically when we run into an
additive overflow causing us to drop state in only one direction. If that
occurs, we should bail and not optimize that retain/release instead of
asserting.

Apologies for the size of the testcase. It is necessary to cause the additive
cfg overflow to trigger.

rdar://15377890

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194083 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoProvide a test input for opt
Alp Toker [Tue, 5 Nov 2013 13:57:34 +0000 (13:57 +0000)]
Provide a test input for opt

This was only working previously due to a quirk in the way lit
concatenates script commands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194078 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSilence GCC warning about dropping off a fully covered switch.
Benjamin Kramer [Tue, 5 Nov 2013 13:45:09 +0000 (13:45 +0000)]
Silence GCC warning about dropping off a fully covered switch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194077 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] (PR16190) Add ValueKinds for ConstantDataSequential and subclasses
Peter Zotov [Tue, 5 Nov 2013 12:55:43 +0000 (12:55 +0000)]
[OCaml] (PR16190) Add ValueKinds for ConstantDataSequential and subclasses

Original patch by David Monniaux

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194075 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[llvm-c] (PR16190) Add LLVMIsA* functions for ConstantDataSequential and subclasses
Peter Zotov [Tue, 5 Nov 2013 12:55:37 +0000 (12:55 +0000)]
[llvm-c] (PR16190) Add LLVMIsA* functions for ConstantDataSequential and subclasses

Original patch by David Monniaux

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194074 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] (PR10016) Add a few missing line in OCamlLangImpl2.rst
Peter Zotov [Tue, 5 Nov 2013 12:14:04 +0000 (12:14 +0000)]
[OCaml] (PR10016) Add a few missing line in OCamlLangImpl2.rst

Original patch by Damien Schoof

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194067 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove the word "thumb" from comments. Remove also an incorrect
Reed Kotler [Tue, 5 Nov 2013 12:04:37 +0000 (12:04 +0000)]
Remove the word "thumb" from comments. Remove also an incorrect
command regarding the porting from the ARM version (was an old comment).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194066 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] (PR16318) Add missing argument to Llvm.const_intcast
Peter Zotov [Tue, 5 Nov 2013 11:56:20 +0000 (11:56 +0000)]
[OCaml] (PR16318) Add missing argument to Llvm.const_intcast

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194065 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] (PR11717) Make declare_qualified_global respect address argument
Peter Zotov [Tue, 5 Nov 2013 11:56:13 +0000 (11:56 +0000)]
[OCaml] (PR11717) Make declare_qualified_global respect address argument

Original patch by Jonathan Ragan-Kelley

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194064 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoX86 Disassembler: remove unused bool typedef-name
David Majnemer [Tue, 5 Nov 2013 10:34:42 +0000 (10:34 +0000)]
X86 Disassembler: remove unused bool typedef-name

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194062 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSuppress OS crash dialog in llvm-rtdyld
Alp Toker [Tue, 5 Nov 2013 09:33:43 +0000 (09:33 +0000)]
Suppress OS crash dialog in llvm-rtdyld

All other tools have this -- it's needed to avoid hanging lit on Windows in
case of a crash.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194060 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] Properly tag the custom operations of Llvm.llbuilder
Peter Zotov [Tue, 5 Nov 2013 09:13:46 +0000 (09:13 +0000)]
[OCaml] Properly tag the custom operations of Llvm.llbuilder

All other custom operations tags have LLVM prefix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194058 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[OCaml] Llvm_linker: do not use external in module interface
Peter Zotov [Tue, 5 Nov 2013 09:13:39 +0000 (09:13 +0000)]
[OCaml] Llvm_linker: do not use external in module interface

Workaround for an OCaml bug:
http://caml.inria.fr/mantis/view.php?id=4166

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194057 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix r194019 as requested by Eric Christopher.
Reed Kotler [Tue, 5 Nov 2013 08:14:14 +0000 (08:14 +0000)]
Fix r194019 as requested by Eric Christopher.
Submit the basic port of the rest of ARM constant islands code to Mips.
Two test cases are added which reflect the next level of functionality:
constants getting moved to water areas that are out of range from the
initial placement at the end of the function and basic blocks being split to
create water when none exists that can be used. There is a bunch of this
code that is not complete and has been marked with IN_PROGRESS. I will
finish cleaning this all up during the next week or two and submit the
rest of the test cases. I have elminated some code for dealing with
inline assembly because to me it unecessarily complicates things and
some of the newer features of llvm like function attributies and builtin
assembler give me better tools to solve the alignment issues created
there. Also, for Mips16 I even have the option of not doing constant
islands in the present of inline assembler if I chose. When everything
has been completed I will summarize the port and notify people that
are knowledgable regarding the ARM Constant Islands code so they can
review it in it's entirety if they wish.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194053 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix symbol defines in config.h.cmake
Alp Toker [Tue, 5 Nov 2013 07:27:18 +0000 (07:27 +0000)]
Fix symbol defines in config.h.cmake

These were incorrectly pointing to HAVE_LOG despite being checked for
correctly in config-ix.cmake.

Patch by James Lyon!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194051 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLift alignment restrictions on load folding for a significant portion of AVX instruct...
Craig Topper [Tue, 5 Nov 2013 06:31:43 +0000 (06:31 +0000)]
Lift alignment restrictions on load folding for a significant portion of AVX instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194048 91177308-0d34-0410-b5e6-96231b3b80d8