platform/upstream/llvm.git
9 years agoAdd AtosSymbolizer and DlAddrSymbolizer as fallbacks for OS X
Kuba Brecka [Sun, 22 Mar 2015 10:02:50 +0000 (10:02 +0000)]
Add AtosSymbolizer and DlAddrSymbolizer as fallbacks for OS X

This patch changes the symbolizer chain on OS X (which currently only uses 1
symbolizer at most) to use this behavior:

* By default, use LLVMSymbolizer -> DlAddrSymbolizer.
* If the llvm-symbolizer binary is not found, use AtosSymbolizer
    -> DlAddrSymbolizer.
* If the user specifies ASAN_SYMBOLIZER_PATH=.../atos, then use AtosSymbolizer
    -> DlAddrSymbolizer.
* If neither llvm-symbolizer or atos is found, or external symbolication is
    disabled with ASAN_SYMBOLIZER_PATH="", use DlAddrSymbolizer.

Reviewed at http://reviews.llvm.org/D8285

llvm-svn: 232908

9 years agoReverting 232853 and 232870 because they depend on 232793,
Vince Harron [Sun, 22 Mar 2015 08:47:07 +0000 (08:47 +0000)]
Reverting 232853 and 232870 because they depend on 232793,
which was reverted because it was causing LLDB test failures

llvm-svn: 232907

9 years agoMS ABI: Implement driver-level support for thread-safe statics
David Majnemer [Sun, 22 Mar 2015 08:39:22 +0000 (08:39 +0000)]
MS ABI: Implement driver-level support for thread-safe statics

Decide whether or not to use thread-safe statics depending on whether or
not we have an explicit request from the driver.  If we don't have an
explicit request, infer which behavior to use depending on the
compatibility version we are targeting.

N.B. CodeGen support is still ongoing.

llvm-svn: 232906

9 years agoReverting r232793 until some new LLDB failures are discussed
Vince Harron [Sun, 22 Mar 2015 05:59:59 +0000 (05:59 +0000)]
Reverting r232793 until some new LLDB failures are discussed

llvm-svn: 232905

9 years agoReverted r232883 due to failing tests.
Vince Harron [Sun, 22 Mar 2015 05:47:00 +0000 (05:47 +0000)]
Reverted r232883 due to failing tests.

llvm-svn: 232904

9 years ago[SimplifyLibCalls] Fix negative shifts being produced by the memchr -> bitfield trans...
Benjamin Kramer [Sat, 21 Mar 2015 22:04:26 +0000 (22:04 +0000)]
[SimplifyLibCalls] Fix negative shifts being produced by the memchr -> bitfield transform.

llvm-svn: 232903

9 years ago[SimplifyLibCalls] Turn memchr(const, C, const) into a bitfield check.
Benjamin Kramer [Sat, 21 Mar 2015 21:09:33 +0000 (21:09 +0000)]
[SimplifyLibCalls] Turn memchr(const, C, const) into a bitfield check.

strchr("123!", C) != nullptr is a common pattern to check if C is one
of 1, 2, 3 or !. If the largest element of the string is smaller than
the target's register size we can easily create a bitfield and just
do a simple test for set membership.

int foo(char C) { return strchr("123!", C) != nullptr; } now becomes

cmpl $64, %edi ## range check
sbbb %al, %al
movabsq $0xE000200000001, %rcx
btq %rdi, %rcx ## bit test
sbbb %cl, %cl
andb %al, %cl ## and the two conditions
andb $1, %cl
movzbl %cl, %eax ## returning an int
ret

(imho the backend should expand this into a series of branches, but
that's a different story)

The code is currently limited to bit fields that fit in a register, so
usually 64 or 32 bits. Sadly, this misses anything using alpha chars
or {}. This could be fixed by just emitting a i128 bit field, but that
can generate really ugly code so we have to find a better way. To some
degree this is also recreating switch lowering logic, but we can't
simply emit a switch instruction and thus change the CFG within
instcombine.

llvm-svn: 232902

9 years agoR600: Cleanup test with multiple check prefixes
Matt Arsenault [Sat, 21 Mar 2015 19:15:46 +0000 (19:15 +0000)]
R600: Cleanup test with multiple check prefixes

llvm-svn: 232901

9 years ago-Wdynamic-class-memaccess: Also warn about array types.
Nico Weber [Sat, 21 Mar 2015 17:56:44 +0000 (17:56 +0000)]
-Wdynamic-class-memaccess: Also warn about array types.

It looks like not warning on this was an oversight in the original
implementation of this warning.

llvm-svn: 232900

9 years agoDedent code for -Wdynamic-class-memaccess warning. No behavior change.
Nico Weber [Sat, 21 Mar 2015 17:37:46 +0000 (17:37 +0000)]
Dedent code for -Wdynamic-class-memaccess warning. No behavior change.

The diff looks intimidating, but this just moves the -Wdynamic-class-memaccess
code out a scope, protected by a

    if (PointeeTy == QualType())
      continue;

check so that it still only runs when it should.

llvm-svn: 232899

9 years agoStringRef: Just forward StringRef::find to libc's memchr.
Benjamin Kramer [Sat, 21 Mar 2015 16:42:35 +0000 (16:42 +0000)]
StringRef: Just forward StringRef::find to libc's memchr.

Modern libc's have an SSE version of memchr which is a lot faster than our
hand-rolled version. In the past I was reluctant to use it because Darwin's
memchr used a naive ridiculously slow implementation, but that has been fixed
some versions ago.

Should have zero functional impact.

llvm-svn: 232898

9 years agoRevert accidental commit.
Benjamin Kramer [Sat, 21 Mar 2015 15:37:32 +0000 (15:37 +0000)]
Revert accidental commit.

While this is a fun change, I didn't really test it :)

llvm-svn: 232897

9 years agoSimplifyLibCalls: Add basic optimization of memchr calls.
Benjamin Kramer [Sat, 21 Mar 2015 15:36:21 +0000 (15:36 +0000)]
SimplifyLibCalls: Add basic optimization of memchr calls.

This is just memchr(x, y, 0) -> nullptr and constant folding.

llvm-svn: 232896

9 years agoValueTracking: Forward getConstantStringInfo's TrimAtNul param into recursive invocation
Benjamin Kramer [Sat, 21 Mar 2015 15:36:06 +0000 (15:36 +0000)]
ValueTracking: Forward getConstantStringInfo's TrimAtNul param into recursive invocation

Currently this is only used to tweak the backend's memcpy inlining
heuristics, testing that isn't very helpful. A real test case will
follow in the next commit, where this behavior would cause a real
miscompilation.

llvm-svn: 232895

9 years agoTidied up vec_zero_cse.ll test. NFCI.
Simon Pilgrim [Sat, 21 Mar 2015 14:05:12 +0000 (14:05 +0000)]
Tidied up vec_zero_cse.ll test. NFCI.

Added target triple and refactored the CHECKs to be per function.

llvm-svn: 232894

9 years agoSmall test fix for r232890
Alexander Musman [Sat, 21 Mar 2015 11:49:27 +0000 (11:49 +0000)]
Small test fix for r232890

llvm-svn: 232893

9 years agoCompare pointers directly instead of ::strcmp in SBXxx::EventIsXxxEvent()
Ilia K [Sat, 21 Mar 2015 11:11:07 +0000 (11:11 +0000)]
Compare pointers directly instead of ::strcmp in SBXxx::EventIsXxxEvent()

llvm-svn: 232892

9 years agoFix handling of CommandInterpreter's events in lldb-mi
Ilia K [Sat, 21 Mar 2015 10:53:37 +0000 (10:53 +0000)]
Fix handling of CommandInterpreter's events in lldb-mi

Summary:
Previously lldb-mi contains a stub for that but it didn't work and all CommanInterpreter's events were ignored.
This commit adds a handling of CommandInterpreter's events in lldb-mi.

Steps:
# Fix CMICmnLLDBDebugger::InitSBListener
# Add SBCommandInterpreter::EventIsCommandInterpreterEvent
# Exit on lldb::SBCommandInterpreter::eBroadcastBitQuitCommandReceived

All tests pass on OS X.

In further we can remove "quit" hack in lldb-mi.

Test Plan:
# Create start_script file:
```
target create ~/p/hello
b main
r
quit
```

# Run lldb-mi --interpreter
# Execute start_script file by following command:
```
-interpreter-exec console "command source start_script"
```

Log:
```
$ bin/lldb-mi --interpreter
(gdb)
-interpreter-exec console "command source start_script"
Executing commands in '/Users/IliaK/p/llvm/build_ninja/start_script'.
(lldb) target create ~/p/hello
Current executable set to '~/p/hello' (x86_64).
(lldb) b main
Breakpoint 1: where = hello`main + 29 at hello.cpp:12, address = 0x0000000100000e2d
(lldb) r
Process 1582 launched: '/Users/IliaK/p/hello' (x86_64)
(lldb) quit
^done
(gdb)
=thread-created,id="1",group-id="i1"
=thread-selected,id="1"
(gdb)
=shlibs-added,shlib-info=[num="1",name="hello",dyld-addr="-",reason="dyld",path="/Users/IliaK/p/hello",loaded_addr="-",dsym-objpath="/Users/IliaK/p/hello.dSYM/Contents/Resources/DWARF/hello"]
...
=shlibs-added,shlib-info=[num="132",name="libDiagnosticMessagesClient.dylib",dyld-addr="0x7fff91705000",reason="dyld",path="/usr/lib/libDiagnosticMessagesClient.dylib",loaded_addr="0x7fff91705000"]
(gdb)
*stopped,reason="breakpoint-hit",disp="del",bkptno="1",frame={addr="0x100000e2d",func="main",args=[{name="argc",value="1"},{name="argv",value="0x00007fff5fbffc88"}],file="hello.cpp",fullname="/Users/IliaK/p/hello.cpp",line="12"},thread-id="1",stopped-threads="all"
(gdb)<press Enter>

MI: Program exited OK
```

Reviewers: abidh, clayborg

Reviewed By: abidh

Subscribers: jingham, lldb-commits, clayborg, abidh

Differential Revision: http://reviews.llvm.org/D8382

llvm-svn: 232891

9 years ago[OPENMP] CodeGen of the 'linear' clause for the 'omp simd' directive.
Alexander Musman [Sat, 21 Mar 2015 10:12:56 +0000 (10:12 +0000)]
[OPENMP] CodeGen of the 'linear' clause for the 'omp simd' directive.
The linear variable is privatized (similar to 'private') and its
value on current iteration is calculated, similar to the loop
counter variables.
Differential revision: http://reviews.llvm.org/D8375

llvm-svn: 232890

9 years agoMemoryDependenceAnalysis: Don't miscompile atomics
David Majnemer [Sat, 21 Mar 2015 06:19:17 +0000 (06:19 +0000)]
MemoryDependenceAnalysis: Don't miscompile atomics

r216771 introduced a change to MemoryDependenceAnalysis that allowed it
to reason about acquire/release operations.  However, this change does
not ensure that the acquire/release operations pair.  Unfortunately,
this leads to miscompiles as we won't see an acquire load as properly
memory effecting.  This largely reverts r216771.

This fixes PR22708.

llvm-svn: 232889

9 years agoAdd CodeGen support for adding cpu attributes on functions based on
Eric Christopher [Sat, 21 Mar 2015 06:15:15 +0000 (06:15 +0000)]
Add CodeGen support for adding cpu attributes on functions based on
the target-cpu, if different from the triple's cpu, and
target-features as they're written that are passed down from the
driver.

Together with LLVM r232885 this should allow the LTO'ing of binaries
that contain modules compiled with different code generation options
on a subset of architectures with full backend support (x86, powerpc,
aarch64).

llvm-svn: 232888

9 years agoRemove unneeded redeclaration of reference_wrapper.
Eric Fiselier [Sat, 21 Mar 2015 06:05:45 +0000 (06:05 +0000)]
Remove unneeded redeclaration of reference_wrapper.

llvm-svn: 232887

9 years agoAArch64: simplify test case
Tim Northover [Sat, 21 Mar 2015 04:37:08 +0000 (04:37 +0000)]
AArch64: simplify test case

llvm-svn: 232886

9 years agoRemove the target independent TargetMachine::getSubtarget and
Eric Christopher [Sat, 21 Mar 2015 04:22:23 +0000 (04:22 +0000)]
Remove the target independent TargetMachine::getSubtarget and
TargetMachine::getSubtargetImpl routines.

This keeps the target independent code free of bare subtarget
calls while the remainder of the backends are migrated, or not
if they don't wish to support per-function subtargets as would
be needed for function multiversioning or LTO of disparate
cpu subarchitecture types, e.g.

clang -msse4.2 -c foo.c -emit-llvm -o foo.bc
clang -c bar.c -emit-llvm -o bar.bc
llvm-link foo.bc bar.bc -o baz.bc
llc baz.bc

and get appropriate code for what the command lines requested.

llvm-svn: 232885

9 years agoRemove the bare getSubtargetImpl call from the AArch64 port. As part
Eric Christopher [Sat, 21 Mar 2015 04:04:50 +0000 (04:04 +0000)]
Remove the bare getSubtargetImpl call from the AArch64 port. As part
of this add a test that shows we can generate code for functions
that specifically enable a subtarget feature.

llvm-svn: 232884

9 years agoFix ComputeSupportExeDirectory for Linux (merge with Android).
Chaoren Lin [Sat, 21 Mar 2015 03:59:29 +0000 (03:59 +0000)]
Fix ComputeSupportExeDirectory for Linux (merge with Android).

Summary:
ComputeSupportExeDirectory relied on ComputeSharedLibraryDirectory which was
not always reliable. Using procfs seems to be the best way to deal with it on
Linux (since it's already done on Android, might as well merge it).

Reviewers: ovyalov

Reviewed By: ovyalov

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D8511

llvm-svn: 232883

9 years agoRemove the bare getSubtargetImpl call from the PPC port. As part
Eric Christopher [Sat, 21 Mar 2015 03:36:02 +0000 (03:36 +0000)]
Remove the bare getSubtargetImpl call from the PPC port. As part
of this add a test that shows we can generate code with
for functions that differ by subtarget feature.

llvm-svn: 232882

9 years agoForward the Function based getSubtarget call to the appropriate Impl
Eric Christopher [Sat, 21 Mar 2015 03:32:45 +0000 (03:32 +0000)]
Forward the Function based getSubtarget call to the appropriate Impl
call.

llvm-svn: 232881

9 years agoGrab a subtarget off of an AMDGPUTargetMachine rather than a
Eric Christopher [Sat, 21 Mar 2015 03:17:25 +0000 (03:17 +0000)]
Grab a subtarget off of an AMDGPUTargetMachine rather than a
bare target machine in preparation for the TargetMachine bare
getSubtarget/getSubtargetImpl calls going away.

llvm-svn: 232880

9 years agoCache the Function dependent subtarget on the MachineFunction.
Eric Christopher [Sat, 21 Mar 2015 03:13:10 +0000 (03:13 +0000)]
Cache the Function dependent subtarget on the MachineFunction.

As preparation for removing the getSubtargetImpl() call from
TargetMachine go ahead and flip the switch on caching the function
dependent subtarget and remove the bare getSubtargetImpl call
from the X86 port. As part of this add a few tests that show we
can generate code and assemble on X86 based on features/cpu on
the Function.

llvm-svn: 232879

9 years agoGrab the cached subtarget off of the MachineFunction.
Eric Christopher [Sat, 21 Mar 2015 03:13:07 +0000 (03:13 +0000)]
Grab the cached subtarget off of the MachineFunction.

llvm-svn: 232878

9 years agoGrab a subtarget off of a MipsTargetMachine rather than a
Eric Christopher [Sat, 21 Mar 2015 03:13:05 +0000 (03:13 +0000)]
Grab a subtarget off of a MipsTargetMachine rather than a
bare target machine in preparation for the TargetMachine bare
getSubtarget/getSubtargetImpl calls going away.

llvm-svn: 232877

9 years agoSimplify the query for a subtarget in the NVPTX pass manager.
Eric Christopher [Sat, 21 Mar 2015 03:13:03 +0000 (03:13 +0000)]
Simplify the query for a subtarget in the NVPTX pass manager.

llvm-svn: 232876

9 years agoChange getISAEncoding to use the target triple to determine
Eric Christopher [Sat, 21 Mar 2015 03:13:01 +0000 (03:13 +0000)]
Change getISAEncoding to use the target triple to determine
thumb-ness similar to the rest of the Module level asm printing
infrastructure as debug info finalization happens after the function
may be missing.

llvm-svn: 232875

9 years agoMake the Hexagon ISelDAGToDAG pass set the subtarget dynamically
Eric Christopher [Sat, 21 Mar 2015 03:12:59 +0000 (03:12 +0000)]
Make the Hexagon ISelDAGToDAG pass set the subtarget dynamically
on each runOnMachineFunction invocation.

llvm-svn: 232874

9 years ago[sanitizer] experimental tracing for cmp instructions
Kostya Serebryany [Sat, 21 Mar 2015 01:29:36 +0000 (01:29 +0000)]
[sanitizer] experimental tracing for cmp instructions

llvm-svn: 232873

9 years ago[CodeGen][IfCvt] Don't re-ifcvt blocks with unanalyzable terminators.
Ahmed Bougacha [Sat, 21 Mar 2015 01:23:15 +0000 (01:23 +0000)]
[CodeGen][IfCvt] Don't re-ifcvt blocks with unanalyzable terminators.

If we couldn't analyze its terminator (i.e., it's an indirectbr, or some
other weirdness), we can't safely re-if-convert a predicated block,
because we can't tell whether the predicated terminator can
fallthrough (it does).

Currently, we would completely ignore the fallthrough successor. In
the added testcase, this means we used to generate:

    ...
  @ %entry:
    cmp   r5, #21
    ittt  ne
  @ %cc1f:
    cmpne r7, #42
  @ %cc2t:
    strne.w       r5, [r8]
    movne pc, r10
  @ %cc1t:
    ...

Whereas the successor of %cc1f was originally %bb1.
With the fix, we get the correct:

    ...
  @ %entry:
    cmp   r5, #21
    itt   eq
  @ %cc1t:
    streq.w       r5, [r11]
    moveq pc, r0
  @ %cc1f:
    cmp   r7, #42
    itt   ne
  @ %cc2t:
    strne.w       r5, [r8]
    movne pc, r10
  @ %bb1:
    ...

rdar://20192768
Differential Revision: http://reviews.llvm.org/D8509

llvm-svn: 232872

9 years ago[AArch64] Prefer UZP for concat_vector of illegal truncs.
Ahmed Bougacha [Sat, 21 Mar 2015 01:08:39 +0000 (01:08 +0000)]
[AArch64] Prefer UZP for concat_vector of illegal truncs.

Follow-up to r232459: prefer a UZP shuffle to the intermediate truncs.

llvm-svn: 232871

9 years ago[modules] When either redecl chain merging or an update record causes us to
Richard Smith [Sat, 21 Mar 2015 00:58:54 +0000 (00:58 +0000)]
[modules] When either redecl chain merging or an update record causes us to
give an exception specification to a declaration that didn't have an exception
specification in any of our imported modules, emit an update record ourselves.
Without this, code importing the current module would not see an exception
specification that we could see and might have relied on.

llvm-svn: 232870

9 years agoOnly add -fno-rtti if KernelOrKext or in C++ mode.
Filipe Cabecinhas [Fri, 20 Mar 2015 23:51:15 +0000 (23:51 +0000)]
Only add -fno-rtti if KernelOrKext or in C++ mode.

llvm-svn: 232869

9 years ago[ELF] GLOBAL_OFFSET_TABLE is a hidden symbol
Shankar Easwaran [Fri, 20 Mar 2015 23:47:08 +0000 (23:47 +0000)]
[ELF] GLOBAL_OFFSET_TABLE is a hidden symbol

llvm-svn: 232868

9 years ago[ELF] Dont add local symbols for dynamic lookup.
Shankar Easwaran [Fri, 20 Mar 2015 23:47:05 +0000 (23:47 +0000)]
[ELF] Dont add local symbols for dynamic lookup.

Unable to add a unit test for this, as there is only one local undefined
symbol in regular shared libraries without a name.

llvm-svn: 232867

9 years ago[ELF] OrderPass : Order atoms only by ordinals.
Shankar Easwaran [Fri, 20 Mar 2015 23:47:03 +0000 (23:47 +0000)]
[ELF] OrderPass : Order atoms only by ordinals.

Move the init array/fini array sorting to the Output ELF writer.

AFAIK, this is only needed by the X86_64/ARM ABI.

This shaves time taken to self host lld by 0.2 seconds.

Before patch
----------------
4791.062059 task-clock                #    1.634 CPUs utilized            ( +-  0.28% )
     61,107 context-switches          #    0.013 M/sec                    ( +-  0.56% )
2.932902671 seconds time elapsed                                          ( +-  0.84% )

After patch
-------------
4608.417248 task-clock                #    1.669 CPUs utilized            ( +-  0.30% )
     61,616 context-switches          #    0.013 M/sec                    ( +-  0.63% )
2.761012703 seconds time elapsed                                          ( +-  0.63% )

llvm-svn: 232866

9 years ago[ELF][X86_64] Use anonymous namespace only for class declarations
Shankar Easwaran [Fri, 20 Mar 2015 23:47:01 +0000 (23:47 +0000)]
[ELF][X86_64] Use anonymous namespace only for class declarations

llvm-svn: 232865

9 years ago[ELF] Use unordered_map for AdditionalSegments
Shankar Easwaran [Fri, 20 Mar 2015 23:47:00 +0000 (23:47 +0000)]
[ELF] Use unordered_map for AdditionalSegments

Fix a leftover class during implementation.

llvm-svn: 232864

9 years ago[ELF] Use MapVector.
Shankar Easwaran [Fri, 20 Mar 2015 23:46:58 +0000 (23:46 +0000)]
[ELF] Use MapVector.

Order is still deterministic and we dont need a sorted order.

llvm-svn: 232863

9 years ago[ELF][X86_64] Remove relative references to include files
Shankar Easwaran [Fri, 20 Mar 2015 23:46:56 +0000 (23:46 +0000)]
[ELF][X86_64] Remove relative references to include files

llvm-svn: 232862

9 years ago[ELF][X86_64] Sort includes
Shankar Easwaran [Fri, 20 Mar 2015 23:46:54 +0000 (23:46 +0000)]
[ELF][X86_64] Sort includes

llvm-svn: 232861

9 years agoDon't claim exception related arguments when looking at RTTIMode
Filipe Cabecinhas [Fri, 20 Mar 2015 23:33:23 +0000 (23:33 +0000)]
Don't claim exception related arguments when looking at RTTIMode

Summary:
We were claiming the -f*exceptions arguments when looking for the
RTTIMode. This makes us not warn about unused arguments if compiling a C
file with -fcxx-exceptions.

This patch fixes it by not claiming the exception-related arguments at
that point.

Reviewers: rsmith, samsonov

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D8507

llvm-svn: 232860

9 years agoMake getLastArgNoClaim work for up to 4 arguments.
Filipe Cabecinhas [Fri, 20 Mar 2015 23:32:58 +0000 (23:32 +0000)]
Make getLastArgNoClaim work for up to 4 arguments.

Summary:
This is needed for http://reviews.llvm.org/D8507
I have no idea what stand-alone tests could be done, if needed.

Reviewers: Bigcheese, craig.topper, samsonov

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8508

llvm-svn: 232859

9 years agoImprove automatic detection of filetype for sym_check
Eric Fiselier [Fri, 20 Mar 2015 23:07:38 +0000 (23:07 +0000)]
Improve automatic detection of filetype for sym_check

llvm-svn: 232858

9 years agofixed vperm2* intrinsics to check for shuffles
Sanjay Patel [Fri, 20 Mar 2015 22:37:20 +0000 (22:37 +0000)]
fixed vperm2* intrinsics to check for shuffles

This corresponds to llvm r232852:
http://reviews.llvm.org/rL232852

llvm-svn: 232857

9 years agoFix use of incorrect package name in sym_check
Eric Fiselier [Fri, 20 Mar 2015 22:13:37 +0000 (22:13 +0000)]
Fix use of incorrect package name in sym_check

llvm-svn: 232856

9 years agoAdd symbol checking script to libc++ to help manage exported symbols.
Eric Fiselier [Fri, 20 Mar 2015 22:09:29 +0000 (22:09 +0000)]
Add symbol checking script to libc++ to help manage exported symbols.

Summary:
Add symbol checking scripts for extracting a list of symbols from shared libraries and for comparing symbol lists for differences.

Reviewers: mclow.lists, danalbert, EricWF

Reviewed By: EricWF

Subscribers: majnemer, emaste, cfe-commits

Differential Revision: http://reviews.llvm.org/D4946

llvm-svn: 232855

9 years agoTell lit.cfg about more Windows triples.
Yunzhong Gao [Fri, 20 Mar 2015 22:08:40 +0000 (22:08 +0000)]
Tell lit.cfg about more Windows triples.
For example, the host triple on my 64-bit PC is x86_64-pc-windows-msvc.

llvm-svn: 232854

9 years ago[modules] Remove temporary IdentifierInfo lookup results when we're done with them.
Richard Smith [Fri, 20 Mar 2015 21:56:50 +0000 (21:56 +0000)]
[modules] Remove temporary IdentifierInfo lookup results when we're done with them.

llvm-svn: 232853

9 years ago[X86, AVX] instcombine common cases of vperm2* intrinsics into shuffles
Sanjay Patel [Fri, 20 Mar 2015 21:47:56 +0000 (21:47 +0000)]
[X86, AVX] instcombine common cases of vperm2* intrinsics into shuffles

vperm2* intrinsics are just shuffles.
In a few special cases, they're not even shuffles.

Optimizing intrinsics in InstCombine is better than
handling this in the front-end for at least two reasons:

1. Optimizing custom-written SSE intrinsic code at -O0 makes vector coders
   really angry (and so I have regrets about some patches from last week).

2. Doing mask conversion logic in header files is hard to write and
   subsequently read.

There are a couple of TODOs in this patch to complete this optimization.

Differential Revision: http://reviews.llvm.org/D8486

llvm-svn: 232852

9 years agoFixing a bug with WinEH PHI handling
Andrew Kaylor [Fri, 20 Mar 2015 21:42:54 +0000 (21:42 +0000)]
Fixing a bug with WinEH PHI handling

llvm-svn: 232851

9 years ago[X86] Prefer blendps over insertps codegen for one special case
Sanjay Patel [Fri, 20 Mar 2015 21:19:52 +0000 (21:19 +0000)]
[X86] Prefer blendps over insertps codegen for one special case

With this patch, for this one exact case, we'll generate:

  blendps %xmm0, %xmm1, $1

instead of:

  insertps %xmm0, %xmm1, $0

If there's a memory operand available for load folding and we're
optimizing for size, we'll still generate the insertps.

The detailed performance data motivation for this may be found in D7866;
in summary, blendps has 2-3x throughput vs. insertps on widely used chips.

Differential Revision: http://reviews.llvm.org/D8332

llvm-svn: 232850

9 years ago[analyzer] RetainCountChecker: Don't assume +0 for ivars backing readonly properties.
Jordan Rose [Fri, 20 Mar 2015 21:12:27 +0000 (21:12 +0000)]
[analyzer] RetainCountChecker: Don't assume +0 for ivars backing readonly properties.

Similarly, don't assume +0 if the property's setter is manually implemented.
In both cases, if the property's ownership is explicitly written, then we /do/
assume the ivar has the same ownership.

rdar://problem/20218183

llvm-svn: 232849

9 years agoX86: Make helper functions static. NFC.
Benjamin Kramer [Fri, 20 Mar 2015 21:07:30 +0000 (21:07 +0000)]
X86: Make helper functions static. NFC.

llvm-svn: 232848

9 years agoRemove dead calls and function arguments dealing with TRI in StackMaps.
Eric Christopher [Fri, 20 Mar 2015 21:05:18 +0000 (21:05 +0000)]
Remove dead calls and function arguments dealing with TRI in StackMaps.

llvm-svn: 232847

9 years ago[asan] update the sized_delete_test following the change of clang flags in r232788.
Kostya Serebryany [Fri, 20 Mar 2015 20:45:42 +0000 (20:45 +0000)]
[asan] update the sized_delete_test following the change of clang flags in r232788.

llvm-svn: 232845

9 years agoDebugInfo: Require valid DIDescriptors
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 20:17:07 +0000 (20:17 +0000)]
DebugInfo: Require valid DIDescriptors

As part of PR22777, switch from `dyn_cast_or_null<>` to `cast<>` in most
`DIDescriptor` accessors.  These classes are lightweight wrappers around
pointers, so the users should check for valid pointers before using
them.

This survives a Darwin clang -g bootstrap (after fixing testcases), but
it's possible the bots will complain about other configurations.  I'll
fix any fallout as quickly as I can!  Once this bakes for a bit I'll
remove the macros.

Note that `DebugLoc` implicitly gets stricter with this change as well,
since it forward to `DILocation`.  Any code that's using `DebugLoc`
accessors should check `DebugLoc::isUnknown()` first.  (BTW, I'm also
partway through a cleanup of the `DebugLoc` API to make it more obvious
what it is (a glorified pointer wrapper) and remove cruft from before
the Metadata/Value split.  I'll commit soon.)

llvm-svn: 232844

9 years agoUpdate for llvm api change.
Rafael Espindola [Fri, 20 Mar 2015 20:00:30 +0000 (20:00 +0000)]
Update for llvm api change.

llvm-svn: 232843

9 years agoDon't declare all text sections at the start of the .s
Rafael Espindola [Fri, 20 Mar 2015 20:00:01 +0000 (20:00 +0000)]
Don't declare all text sections at the start of the .s

The code this patch removes was there to make sure the text sections went
before the dwarf sections. That is necessary because MachO uses offsets
relative to the start of the file, so adding a section can change relaxations.

The dwarf sections were being printed at the start just to produce symbols
pointing at the start of those sections.

The underlying issue was fixed in r231898. The dwarf sections are now printed
when they are about to be used, which is after we printed the text sections.

To make sure we don't regress, the patch makes the MachO streamer assert
if CodeGen puts anything unexpected after the DWARF sections.

llvm-svn: 232842

9 years agoAvoid output on successful execution of the SWIG scripts
Ed Maste [Fri, 20 Mar 2015 19:59:35 +0000 (19:59 +0000)]
Avoid output on successful execution of the SWIG scripts

Standard practice for tools like this upon success is no output and a
zero return value.

llvm-svn: 232841

9 years agoFixup whitespace in finishSwigWrapperClasses.py
Ed Maste [Fri, 20 Mar 2015 19:53:49 +0000 (19:53 +0000)]
Fixup whitespace in finishSwigWrapperClasses.py

- replace hard tabs with 4-space indents
- delete EOL whitespace

llvm-svn: 232840

9 years agoBugpoint: Fix invalid 'inlinedAt:' references in testcase
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 19:51:34 +0000 (19:51 +0000)]
Bugpoint: Fix invalid 'inlinedAt:' references in testcase

These are causing crashes in `DebugInfoFinder` after a WIP patch to
increase strictness of `DIDescriptor` accessors.

llvm-svn: 232839

9 years agoAsmPrinter: Check subprogram before using it
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 19:50:00 +0000 (19:50 +0000)]
AsmPrinter: Check subprogram before using it

Check return of `getDISubprogram()` before using it.  A WIP patch makes
`DIDescriptor` accessors more strict (and would crash on this).

llvm-svn: 232838

9 years agoReorganize the x86 ELF relocation selection logic.
Rafael Espindola [Fri, 20 Mar 2015 19:48:54 +0000 (19:48 +0000)]
Reorganize the x86 ELF relocation selection logic.

The main differences are:

* Split in 32 and 64 bit functions.
* First switch on the Modifier so that we have only one non fully covered
  switch.
* Map the fixup kind first to a x86_64 (or i386) specific enum, to make
  it easy to handle cases like X86::reloc_riprel_4byte_movq_load.
* Switch on IsPCRel last, which reduces code duplication.

Fixes pr22308.

llvm-svn: 232837

9 years agoDwarfDebug: Check for null DebugLocs
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 19:37:03 +0000 (19:37 +0000)]
DwarfDebug: Check for null DebugLocs

`DL` might be null, so check for that before using accessors.  A WIP
patch to make `DIDescriptors` more strict fails otherwise.

As a bonus, I think the logic is easier to follow now (despite the extra
nesting depth).

llvm-svn: 232836

9 years agoVerifier: Check that !dbg attachments have the right type
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 19:26:58 +0000 (19:26 +0000)]
Verifier: Check that !dbg attachments have the right type

A WIP patch makes `DIDescriptor` accessors more strict, which in turn
causes the `DebugInfoFinder` to crash on wrongly typed `!dbg`
attachments.  Catch that error up front in
`Verifier::visitInstruction()`.

Also remove a test that we "handle" invalid `!dbg` attachments, added
back in r99938.  We don't want to handle those anymore.

Note: I'm *not* recursing and verifying the debug info graph reachable
from this node; that work is already done by `verifyDebugInfo()`.

llvm-svn: 232834

9 years agoDebugInfoFinder: Check for null imported entities
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 19:13:53 +0000 (19:13 +0000)]
DebugInfoFinder: Check for null imported entities

Don't use the accessors in `DIImportedEntity` on a null pointer.  (A WIP
patch to make `DIDescriptor` accessors more strict crashes here
otherwise.)

llvm-svn: 232833

9 years agoSanitizerCoverage: Check for null DebugLocs
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 18:48:45 +0000 (18:48 +0000)]
SanitizerCoverage: Check for null DebugLocs

After a WIP patch to make `DIDescriptor` accessors more strict, this
started asserting.

llvm-svn: 232832

9 years agoSelectionDAGBuilder: Rangeify a loop. NFC.
Hans Wennborg [Fri, 20 Mar 2015 18:48:40 +0000 (18:48 +0000)]
SelectionDAGBuilder: Rangeify a loop. NFC.

llvm-svn: 232831

9 years agoSelectionDAGBuilder::handleJTSwitchCase, simplify loop; NFC
Hans Wennborg [Fri, 20 Mar 2015 18:48:31 +0000 (18:48 +0000)]
SelectionDAGBuilder::handleJTSwitchCase, simplify loop; NFC

llvm-svn: 232830

9 years ago[UBSan] Don't allow to use UBSan with anything except for ASan.
Alexey Samsonov [Fri, 20 Mar 2015 18:45:06 +0000 (18:45 +0000)]
[UBSan] Don't allow to use UBSan with anything except for ASan.

We are not able to make a reliable solution for using UBSan together
with other sanitizers with runtime support (and sanitizer_common).
Instead, we want to follow the path used for LSan: have a "standalone"
UBSan tool, and plug-in UBSan that would be explicitly embedded into
specific sanitizers (in short term, it will be only ASan).

llvm-svn: 232829

9 years agoRewrite test/Feature/md_on_instruction.ll
Duncan P. N. Exon Smith [Fri, 20 Mar 2015 18:34:53 +0000 (18:34 +0000)]
Rewrite test/Feature/md_on_instruction.ll

This test is supposed to be testing whether metadata attachments to
instructions work, but it was using invalid debug info to do so.  (This
was causing assertion failures in the `DebugInfoFinder` with a WIP patch
to be more strict about `DIDescriptor` accessors.)

Rather than fix the debug info -- which is better tested elsewhere --
just test the IR feature directly.

llvm-svn: 232828

9 years agoCorrectly estimate SROA savings for store operands in inline cost analysis.
Wei Mi [Fri, 20 Mar 2015 18:33:12 +0000 (18:33 +0000)]
Correctly estimate SROA savings for store operands in inline cost analysis.

When estimating SROA savings, we want to see if an address is derived
off an alloca in the caller. For store instructions, operand 1 is the
address operand, but the current code uses operand 0.  Use
getPointerOperand for loads and stores to fix this.

Patch by Easwaran Raman.
http://reviews.llvm.org/D8425

llvm-svn: 232827

9 years agoSmall optimization to avoid getting pass info when we will not run loop
Daniel Berlin [Fri, 20 Mar 2015 18:05:49 +0000 (18:05 +0000)]
Small optimization to avoid getting pass info when we will not run loop

llvm-svn: 232826

9 years ago[ARM] Fix handling of thumb1 out-of-range frame offsets
John Brawn [Fri, 20 Mar 2015 17:20:07 +0000 (17:20 +0000)]
[ARM] Fix handling of thumb1 out-of-range frame offsets

LocalStackSlotPass assumes that isFrameOffsetLegal doesn't change its
answer when the base register changes. Unfortunately this isn't true
in thumb1, where SP-based loads allow a larger offset than
non-SP-based loads, and this causes the base register reuse code to
generate instructions that are unencodable, causing an assertion
failure.

Solve this by adding a BaseReg parameter to isFrameOffsetLegal, which
ARMBaseRegisterInfo can then make use of to give the correct answer.

Differential Revision: http://reviews.llvm.org/D8419

llvm-svn: 232825

9 years agoMS ABI: Accept calls to an unprototyped declaration of _setjmp
David Majnemer [Fri, 20 Mar 2015 17:03:35 +0000 (17:03 +0000)]
MS ABI: Accept calls to an unprototyped declaration of _setjmp

This fixes PR22961.

llvm-svn: 232824

9 years agoTry to fix configure/make build.
Zachary Turner [Fri, 20 Mar 2015 16:23:40 +0000 (16:23 +0000)]
Try to fix configure/make build.

llvm-svn: 232823

9 years agoStripped trailing whitespace. NFC.
Simon Pilgrim [Fri, 20 Mar 2015 16:08:17 +0000 (16:08 +0000)]
Stripped trailing whitespace. NFC.

llvm-svn: 232822

9 years agoRewrite StackMap location handling to pre-compute the dwarf register
Eric Christopher [Fri, 20 Mar 2015 16:03:42 +0000 (16:03 +0000)]
Rewrite StackMap location handling to pre-compute the dwarf register
numbers before emission.

This removes a dependency on being able to access TRI at the module
level and is similar to the DwarfExpression handling. I've modified
the debug support into print/dump routines that'll do the same dumping
but is now callable anywhere and if TRI isn't available will go ahead
and just print out raw register numbers.

llvm-svn: 232821

9 years agoAt the beginning of doFinalization set the MachineFunction to
Eric Christopher [Fri, 20 Mar 2015 16:03:39 +0000 (16:03 +0000)]
At the beginning of doFinalization set the MachineFunction to
nullptr so that users get an earlier dereferencing error and
so that we can use it to conditionalize access to MachineFunction
specific data.

llvm-svn: 232820

9 years agoTypo.
Chad Rosier [Fri, 20 Mar 2015 15:45:14 +0000 (15:45 +0000)]
Typo.

llvm-svn: 232819

9 years agoR600/SI: Refactor VOP2 instruction defs
Tom Stellard [Fri, 20 Mar 2015 15:14:23 +0000 (15:14 +0000)]
R600/SI: Refactor VOP2 instruction defs

llvm-svn: 232817

9 years agoR600/SI: Refactor VOP1 instruction defs
Tom Stellard [Fri, 20 Mar 2015 15:14:21 +0000 (15:14 +0000)]
R600/SI: Refactor VOP1 instruction defs

llvm-svn: 232816

9 years agoAdd missing cases to NativeProcessLinux LogThreadStopInfo
Pavel Labath [Fri, 20 Mar 2015 14:45:13 +0000 (14:45 +0000)]
Add missing cases to NativeProcessLinux LogThreadStopInfo

Test Plan: No tests, this is just a debug logging function.

Reviewers: tberghammer

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D8453

llvm-svn: 232815

9 years agoReduce indentation after return. NFC.
Rafael Espindola [Fri, 20 Mar 2015 14:33:25 +0000 (14:33 +0000)]
Reduce indentation after return. NFC.

llvm-svn: 232814

9 years agoUse early returns. NFC.
Rafael Espindola [Fri, 20 Mar 2015 14:23:46 +0000 (14:23 +0000)]
Use early returns. NFC.

llvm-svn: 232813

9 years ago[ASan] fix litling warnings for coverage-order-pcs.cc
Alexander Potapenko [Fri, 20 Mar 2015 14:11:05 +0000 (14:11 +0000)]
[ASan] fix litling warnings for coverage-order-pcs.cc

llvm-svn: 232812

9 years agoFold a llvm_unreachable into an assert. NFC.
Rafael Espindola [Fri, 20 Mar 2015 13:50:15 +0000 (13:50 +0000)]
Fold a llvm_unreachable into an assert. NFC.

llvm-svn: 232811

9 years agoclang-format a function. NFC.
Rafael Espindola [Fri, 20 Mar 2015 13:47:40 +0000 (13:47 +0000)]
clang-format a function. NFC.

llvm-svn: 232810

9 years ago[ASan] Move a couple of Posix-specific tests to Posix/
Alexander Potapenko [Fri, 20 Mar 2015 13:42:11 +0000 (13:42 +0000)]
[ASan] Move a couple of Posix-specific tests to Posix/
Add a comment about potential breakage of coverage-maybe-open-file.cc on Win

llvm-svn: 232809

9 years ago[ASan] Move the coverage tests that work on Darwin to common testcase dir.
Alexander Potapenko [Fri, 20 Mar 2015 13:31:03 +0000 (13:31 +0000)]
[ASan] Move the coverage tests that work on Darwin to common testcase dir.

llvm-svn: 232808

9 years ago[Mips] Create special PLT entry in case of MIPS R6 ABI
Simon Atanasyan [Fri, 20 Mar 2015 11:28:22 +0000 (11:28 +0000)]
[Mips] Create special PLT entry in case of MIPS R6 ABI

llvm-svn: 232806

9 years agoFix ninja check-lldb
Pavel Labath [Fri, 20 Mar 2015 11:24:36 +0000 (11:24 +0000)]
Fix ninja check-lldb

Ninja apparently has issues with commands writing nul characters '\0' to stdout. When it
encounters a nul character, the rest of the output is skipped (I will file a bug with ninja about
that). This breaks the linux buildbot since it parses the ninja check-lldb output to get a list
of failures.

Recently, we have started outputting nul characters in lldb-server tests. This is probably
a bug in itself, but I am not addressing that now. This is just a workaround commit, which
prevents the nul characters from appearing in the output, and gets the buildbot operational again.

llvm-svn: 232805