oota-llvm.git
10 years ago[OCaml] Refactor Llvm_target tests
Peter Zotov [Sun, 3 Nov 2013 08:27:13 +0000 (08:27 +0000)]
[OCaml] Refactor Llvm_target tests

Llvm_target tests did not check for return values. This actually
caused them to miss a bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193949 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[OCaml] Llvm_target: fix typo (Int_val instead of Int64_val)
Peter Zotov [Sun, 3 Nov 2013 08:27:08 +0000 (08:27 +0000)]
[OCaml] Llvm_target: fix typo (Int_val instead of Int64_val)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193948 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Sparc] Expand FP_TO_UINT, UINT_TO_FP for fp128.
Venkatraman Govindaraju [Sun, 3 Nov 2013 08:00:19 +0000 (08:00 +0000)]
[Sparc] Expand FP_TO_UINT, UINT_TO_FP for fp128.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193947 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[OCaml] Llvm_scalar_opts: add missing transforms
Peter Zotov [Sun, 3 Nov 2013 07:54:17 +0000 (07:54 +0000)]
[OCaml] Llvm_scalar_opts: add missing transforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193946 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[OCaml] Llvm_ipo: add missing transforms
Peter Zotov [Sun, 3 Nov 2013 07:54:08 +0000 (07:54 +0000)]
[OCaml] Llvm_ipo: add missing transforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193945 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConvert calls to __sinpi and __cospi into __sincospi_stret
Bob Wilson [Sun, 3 Nov 2013 06:48:38 +0000 (06:48 +0000)]
Convert calls to __sinpi and __cospi into __sincospi_stret

This adds an SimplifyLibCalls case which converts the special __sinpi and
__cospi (float & double variants) into a __sincospi_stret where appropriate to
remove duplicated work.

Patch by Tim Northover

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193943 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoEnable optimization of sin / cos pair into call to __sincos_stret for iOS7+.
Bob Wilson [Sun, 3 Nov 2013 06:14:38 +0000 (06:14 +0000)]
Enable optimization of sin / cos pair into call to __sincos_stret for iOS7+.

rdar://12856873
Patch by Evan Cheng, with a fix for rdar://13209539 by Tilmann Scheller

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193942 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[SparcV9] Add ctpop instruction for i64. Also, expand ctlz, cttz and bswap.
Venkatraman Govindaraju [Sun, 3 Nov 2013 05:59:07 +0000 (05:59 +0000)]
[SparcV9] Add ctpop instruction for i64.  Also, expand ctlz, cttz and bswap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193941 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConfigure with --enable-terminfo=no for Apple's llvmCore builds.
Bob Wilson [Sun, 3 Nov 2013 05:49:36 +0000 (05:49 +0000)]
Configure with --enable-terminfo=no for Apple's llvmCore builds.

Patch by Pete Cooper.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193940 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a comment to note that LLVMDisablePrettyStackTrace() is likely not a good long...
Filip Pizlo [Sun, 3 Nov 2013 04:38:31 +0000 (04:38 +0000)]
Add a comment to note that LLVMDisablePrettyStackTrace() is likely not a good long-term solution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193939 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoWhen LLVM is embedded in a larger application, it's not OK for LLVM to intercept...
Filip Pizlo [Sun, 3 Nov 2013 00:29:47 +0000 (00:29 +0000)]
When LLVM is embedded in a larger application, it's not OK for LLVM to intercept crashes.  LLVM already has
the ability to disable this functionality.  This patch exposes it via the C API.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193937 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agomove getSymbolNMTypeChar to the one program that needs it: nm.
Rafael Espindola [Sat, 2 Nov 2013 21:16:09 +0000 (21:16 +0000)]
move getSymbolNMTypeChar to the one program that needs it: nm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193933 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConvert another use of getSymbolNMTypeChar.
Rafael Espindola [Sat, 2 Nov 2013 20:10:07 +0000 (20:10 +0000)]
Convert another use of getSymbolNMTypeChar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193932 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAvoid some getSymbolNMTypeChar uses in COFFObjectFile.cpp itself.
Rafael Espindola [Sat, 2 Nov 2013 18:07:48 +0000 (18:07 +0000)]
Avoid some getSymbolNMTypeChar uses in COFFObjectFile.cpp itself.

This is a fixed version of 193928 which keeps these uses in sync.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193931 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Don't use getSymbolNMTypeChar for implementing COFFObjectFile::getSymbolFileO...
Rafael Espindola [Sat, 2 Nov 2013 17:12:49 +0000 (17:12 +0000)]
Revert "Don't use getSymbolNMTypeChar for implementing COFFObjectFile::getSymbolFileOffset."

Investigating a bot failure.

This reverts commit r193928.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193929 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDon't use getSymbolNMTypeChar for implementing COFFObjectFile::getSymbolFileOffset.
Rafael Espindola [Sat, 2 Nov 2013 16:55:21 +0000 (16:55 +0000)]
Don't use getSymbolNMTypeChar for implementing COFFObjectFile::getSymbolFileOffset.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193928 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSLPVectorizer: Remove duplicated function.
Benjamin Kramer [Sat, 2 Nov 2013 14:46:27 +0000 (14:46 +0000)]
SLPVectorizer: Remove duplicated function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193927 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorize: Remove quadratic behavior the local CSE.
Benjamin Kramer [Sat, 2 Nov 2013 13:39:00 +0000 (13:39 +0000)]
LoopVectorize: Remove quadratic behavior the local CSE.

Doing this with a hash map doesn't change behavior and avoids calling
isIdenticalTo O(n^2) times. This should probably eventually move into a utility
class shared with EarlyCSE and the limited CSE in the SLPVectorizer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193926 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoA better fix that also works on ppc: add a target tripple.
Rafael Espindola [Sat, 2 Nov 2013 06:00:09 +0000 (06:00 +0000)]
A better fix that also works on ppc: add a target tripple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193915 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix this test to pass on darwin now that llvm-nm is working.
Rafael Espindola [Sat, 2 Nov 2013 05:29:22 +0000 (05:29 +0000)]
Fix this test to pass on darwin now that llvm-nm is working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193914 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix llvm-nm to mach OS X's nm on some tests.
Rafael Espindola [Sat, 2 Nov 2013 05:03:24 +0000 (05:03 +0000)]
Fix llvm-nm to mach OS X's nm on some tests.

There is still a long way to go for llvm-nm, but at least we now match
nm's letter output in the cases we test for.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193912 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix PR17764
Michael Liao [Sat, 2 Nov 2013 00:10:02 +0000 (00:10 +0000)]
Fix PR17764

- When selecting BLEND from vselect, the operands need swapping as due to the
  difference between vselect and SSE/AVX's BLEND insn

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193900 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdded command-line option to output llvm-cov to file.
Yuchen Wu [Sat, 2 Nov 2013 00:09:17 +0000 (00:09 +0000)]
Added command-line option to output llvm-cov to file.

Added -o option to llvm-cov. If no output file is specified, it defaults
to STDOUT.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193899 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorizer: Move cse code into its own function
Arnold Schwaighofer [Fri, 1 Nov 2013 23:28:54 +0000 (23:28 +0000)]
LoopVectorizer: Move cse code into its own function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193895 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoComment some and reformat for clarity beginFunction.
Eric Christopher [Fri, 1 Nov 2013 23:14:17 +0000 (23:14 +0000)]
Comment some and reformat for clarity beginFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193894 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: regenerate test case from Clang to adjust for fixes/improvements
David Blaikie [Fri, 1 Nov 2013 22:29:28 +0000 (22:29 +0000)]
DebugInfo: regenerate test case from Clang to adjust for fixes/improvements

I hit some problems with future work due to the member subprogram of
'a_b's type having a subprogram (an implicit default ctor, !52 in the
pre-commit source) with no name. Clang now generates a name for such a
function but in this case doesn't even emit debug info for it as it is
unused (Clang never emits the body of the ctor, instead just emitting
memset if needed).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193892 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorizer: Perform redundancy elimination on induction variables
Arnold Schwaighofer [Fri, 1 Nov 2013 22:18:19 +0000 (22:18 +0000)]
LoopVectorizer: Perform redundancy elimination on induction variables

When the loop vectorizer was part of the SCC inliner pass manager gvn would
run after the loop vectorizer followed by instcombine. This way redundancy
(multiple uses) were removed and instcombine could perform scalarization on the
induction variables. Having moved the loop vectorizer to later we no longer run
any form of redundancy elimination before we perform instcombine. This caused
vectorized induction variables to survive that did not before.

On a recent iMac this helps linpack back from 6000Mflops to 7000Mflops.

This should also help lpbench and paq8p.

I ran a Release (without Asserts) build over the test-suite and did not see any
negative impact on compile time.

radar://15339680

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193891 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Improve readability of test case added in r193878
David Blaikie [Fri, 1 Nov 2013 20:59:53 +0000 (20:59 +0000)]
DebugInfo: Improve readability of test case added in r193878

The point is to ensure that the attribute in question
(DW_AT_data_member_location) is associated with the prior tag, so ensure
that we don't see another tag starting between the intended tag and the
desired attribute.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193884 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: add a test case for data member locations (coverage for r193835)
David Blaikie [Fri, 1 Nov 2013 18:25:55 +0000 (18:25 +0000)]
DebugInfo: add a test case for data member locations (coverage for r193835)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193878 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a test case broken by r193872
David Blaikie [Fri, 1 Nov 2013 18:18:16 +0000 (18:18 +0000)]
Fix a test case broken by r193872

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193876 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd comments.
Manman Ren [Fri, 1 Nov 2013 18:06:25 +0000 (18:06 +0000)]
Add comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193874 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Make pubnames header printing similar to unit header printing
David Blaikie [Fri, 1 Nov 2013 17:53:30 +0000 (17:53 +0000)]
DebugInfo: Make pubnames header printing similar to unit header printing

In a failed attempt to allow the gnu-public-names.ll test case to not
hardcode the size of the unit that the pubnames section referred to I've
at least managed to have unit headers and pubnames headers print out in
a similar style.

This failed to achieve the desired goal because the header in a unit
specifies the length of the unit without the length element of the
header whereas the length in the pubnames includes this element, so the
numbers are off by 4 bytes. I don't know of any arithmetic powers in
FileCheck so the test case can't simply say "CU_LENGTH + 4".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193872 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Stackmap] Remove erroneous assert.
Juergen Ributzka [Fri, 1 Nov 2013 17:53:27 +0000 (17:53 +0000)]
[Stackmap] Remove erroneous assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193871 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse isa<> instead of dyn_cast<> with unused value
Matt Arsenault [Fri, 1 Nov 2013 17:39:26 +0000 (17:39 +0000)]
Use isa<> instead of dyn_cast<> with unused value

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193869 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Simplify a few of the instruction patterns. No functional change intended.
Chad Rosier [Fri, 1 Nov 2013 17:13:44 +0000 (17:13 +0000)]
[AArch64] Simplify a few of the instruction patterns. No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193867 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Fix assembly string formatting and other coding standard violations.
Chad Rosier [Fri, 1 Nov 2013 17:13:42 +0000 (17:13 +0000)]
[AArch64] Fix assembly string formatting and other coding standard violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193866 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove linkonce_odr_auto_hide.
Rafael Espindola [Fri, 1 Nov 2013 17:09:14 +0000 (17:09 +0000)]
Remove linkonce_odr_auto_hide.

linkonce_odr_auto_hide was in incomplete attempt to implement a way
for the linker to hide symbols that are known to be available in every
TU and whose addresses are not relevant for a particular DSO.

It was redundant in that it all its uses are equivalent to
linkonce_odr+unnamed_addr. Unlike those, it has never been connected
to clang or llvm's optimizers, so it was effectively dead.

Given that nothing produces it, this patch just nukes it
(other than the llvm-c enum value).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193865 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCommenting out this assert because it is causing the build bots to fail. This effect...
Aaron Ballman [Fri, 1 Nov 2013 15:12:23 +0000 (15:12 +0000)]
Commenting out this assert because it is causing the build bots to fail.  This effectively reverts r193861, but needs to be fixed as part of r193769.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193862 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFixing an order of evaluation error in an assert.
Aaron Ballman [Fri, 1 Nov 2013 14:53:14 +0000 (14:53 +0000)]
Fixing an order of evaluation error in an assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193861 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorize: Look for consecutive acces in GEPs with trailing zero indices
Benjamin Kramer [Fri, 1 Nov 2013 14:09:50 +0000 (14:09 +0000)]
LoopVectorize: Look for consecutive acces in GEPs with trailing zero indices

If we have a pointer to a single-element struct we can still build wide loads
and stores to it (if there is no padding).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193860 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] Add Virtualization subtarget feature and more build attributes in this area
Bradley Smith [Fri, 1 Nov 2013 13:27:35 +0000 (13:27 +0000)]
[ARM] Add Virtualization subtarget feature and more build attributes in this area

Add a Virtualization ARM subtarget feature along with adding proper build
attribute emission for Tag_Virtualization_use (encodes Virtualization and
TrustZone) and Tag_MPextension_use.

Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to
something that is more maintainable. This changes the focus of this
testcase away from testing CPU defaults (which is tested elsewhere), onto
specifically testing that attributes are encoded correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ARM] Fix Tag_ABI_HardFP_use build attribute
Bradley Smith [Fri, 1 Nov 2013 11:21:16 +0000 (11:21 +0000)]
[ARM] Fix Tag_ABI_HardFP_use build attribute

Fix Tag_ABI_HardFP_use build attribute to handle single precision FP,
replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add
some tests for Tag_ABI_VFP_args.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoConsider (x == -1) unlikely in BranchProbabilityInfo
Hal Finkel [Fri, 1 Nov 2013 10:58:22 +0000 (10:58 +0000)]
Consider (x == -1) unlikely in BranchProbabilityInfo

This adds another heuristic to BPI, similar to the existing heuristic that
considers (x == 0) unlikely to be true. As suggested in the PACT'98 paper by
Deitrich, Cheng, and Hwu, -1 is often used to indicate an invalid index, and
equality comparisons with -1 are also unlikely to succeed. Local
experimentation supports this hypothesis: This yields a 1-2% speedup in the
test-suite sqlite benchmark on the PPC A2 core, with no significant
regressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193855 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorizer: If dependency checks fail try runtime checks
Arnold Schwaighofer [Fri, 1 Nov 2013 03:05:07 +0000 (03:05 +0000)]
LoopVectorizer: If dependency checks fail try runtime checks

When a dependence check fails we can still try to vectorize loops with runtime
array bounds checks.

This helps linpack to vectorize a loop in dgefa. And we are back to 2x of the
scalar performance on a corei7-avx.

radar://15339680

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193853 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorizer: Clear all member data structures in RuntimeCheck.reset()
Arnold Schwaighofer [Fri, 1 Nov 2013 03:05:04 +0000 (03:05 +0000)]
LoopVectorizer: Clear all member data structures in RuntimeCheck.reset()

Clear all data structures when resetting the RuntimeCheck data structure.

No test case. This was exposed by an upcomming change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193852 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOCaml bindings: fix typo
Sylvestre Ledru [Fri, 1 Nov 2013 02:29:13 +0000 (02:29 +0000)]
OCaml bindings: fix typo

Patch by Peter Zotov

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193851 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOCaml bindings: remove unused DONT_BUILD_RELINKED from Makefiles
Sylvestre Ledru [Fri, 1 Nov 2013 01:14:24 +0000 (01:14 +0000)]
OCaml bindings: remove unused DONT_BUILD_RELINKED from Makefiles

Patch by Peter Zotov

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193847 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse \01 to disable the mangler. Should fix the 32 bit windows bots.
Rafael Espindola [Fri, 1 Nov 2013 01:14:20 +0000 (01:14 +0000)]
Use \01 to disable the mangler. Should fix the 32 bit windows bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193846 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOCaml bindings: fix typo
Sylvestre Ledru [Fri, 1 Nov 2013 01:10:58 +0000 (01:10 +0000)]
OCaml bindings: fix typo

Patch by Peter Zotov

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193845 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOCaml bindings: fix typo in documentation
Sylvestre Ledru [Fri, 1 Nov 2013 00:30:02 +0000 (00:30 +0000)]
OCaml bindings: fix typo in documentation

Patch by Peter Zotov

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193838 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoOCaml bindings: formatting
Sylvestre Ledru [Fri, 1 Nov 2013 00:26:01 +0000 (00:26 +0000)]
OCaml bindings: formatting
This commit only changes comments and documentation in OCaml bindings. The official name of the language is OCaml, and the usage is now consistent.

Patch by Peter Zotov

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193836 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Emit member variable locations as data instead of expressions in blocks
David Blaikie [Fri, 1 Nov 2013 00:25:45 +0000 (00:25 +0000)]
DebugInfo: Emit member variable locations as data instead of expressions in blocks

Drive by space optimization. Also makes the DIEs more regular which
might speed up DWARF parsing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193835 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd to the disassembler C API output reference types for
Kevin Enderby [Fri, 1 Nov 2013 00:00:07 +0000 (00:00 +0000)]
Add to the disassembler C API output reference types for
Objective-C data structures.

This is allows tools such as darwin's otool(1) that uses the
LLVM disassembler take a pointer value being loaded by
an instruction and add a comment to what it is being referenced
to make following disassembly of Objective-C programs
more readable.

For example disassembling the Mac OS X TextEdit app one
will see comments like the following:

movq    0x20684(%rip), %rsi ## Objc selector ref: standardUserDefaults
movq    0x21985(%rip), %rdi ## Objc class ref: _OBJC_CLASS_$_NSUserDefaults
movq    0x1d156(%rip), %r14 ## Objc message: +[NSUserDefaults standardUserDefaults]
leaq    0x23615(%rip), %rdx ## Objc cfstring ref: @"SelectLinePanel"
callq   0x10001386c ## Objc message: -[[%rdi super] initWithWindowNibName:]

These diffs also include putting quotes around C strings
in literal pools and uses "symbol address" in the comment
when adding a symbol name to the comment to tell these
types of references apart:

leaq 0x4f(%rip), %rax ## literal pool for: "Hello world"
movq    0x1c3ea(%rip), %rax ## literal pool symbol address: ___stack_chk_guard

Of course the easy changes are in the LLVM disassembler and
the hard work is up to the implementer of the SymbolLookUp()
call back.

rdar://10602439

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193833 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix unused variable warnings.
Dan Gohman [Thu, 31 Oct 2013 22:58:11 +0000 (22:58 +0000)]
Fix unused variable warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193823 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThese test cases for experimental features are a bit too darwin-specific still. Use...
Andrew Trick [Thu, 31 Oct 2013 22:46:51 +0000 (22:46 +0000)]
These test cases for experimental features are a bit too darwin-specific still. Use a triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193820 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUnused variable
Andrew Trick [Thu, 31 Oct 2013 22:42:20 +0000 (22:42 +0000)]
Unused variable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193819 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Add support for NEON scalar fixed-point convert to floating-point instructions.
Chad Rosier [Thu, 31 Oct 2013 22:36:59 +0000 (22:36 +0000)]
[AArch64] Add support for NEON scalar fixed-point convert to floating-point instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193816 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoModify CMakeLists.txt to work around pr17763 and bring some bots back.
Rafael Espindola [Thu, 31 Oct 2013 22:13:41 +0000 (22:13 +0000)]
Modify CMakeLists.txt to work around pr17763 and bring some bots back.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193813 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd new calling convention for WebKit Java Script.
Andrew Trick [Thu, 31 Oct 2013 22:12:01 +0000 (22:12 +0000)]
Add new calling convention for WebKit Java Script.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193812 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd support for stack map generation in the X86 backend.
Andrew Trick [Thu, 31 Oct 2013 22:11:56 +0000 (22:11 +0000)]
Add support for stack map generation in the X86 backend.

Originally implemented by Lang Hames.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193811 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRelax check line to match what llvm-nm prints for COFF.
Rafael Espindola [Thu, 31 Oct 2013 22:07:46 +0000 (22:07 +0000)]
Relax check line to match what llvm-nm prints for COFF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193810 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDo not convert "call asm" to "invoke asm" in Inliner.
Manman Ren [Thu, 31 Oct 2013 21:56:03 +0000 (21:56 +0000)]
Do not convert "call asm" to "invoke asm" in Inliner.

Given that backend does not handle "invoke asm" correctly ("invoke asm" will be
handled by SelectionDAGBuilder::visitInlineAsm, which does not have the right
setup for LPadToCallSiteMap) and we already made the assumption that inline asm
does not throw in InstCombiner::visitCallSite, we are going to make the same
assumption in Inliner to make sure we don't convert "call asm" to "invoke asm".

If it becomes necessary to add support for "invoke asm" later on, we will need
to modify the backend as well as remove the assumptions that inline asm does
not throw.

Fix rdar://15317907

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193808 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoXFAIL on ppc64 too.
Rafael Espindola [Thu, 31 Oct 2013 21:27:02 +0000 (21:27 +0000)]
XFAIL on ppc64 too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193804 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoXFAIL this for now.
Rafael Espindola [Thu, 31 Oct 2013 21:22:43 +0000 (21:22 +0000)]
XFAIL this for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193802 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse LTO_SYMBOL_SCOPE_DEFAULT_CAN_BE_HIDDEN instead of the "dso list".
Rafael Espindola [Thu, 31 Oct 2013 20:51:58 +0000 (20:51 +0000)]
Use LTO_SYMBOL_SCOPE_DEFAULT_CAN_BE_HIDDEN instead of the "dso list".

There are two ways one could implement hiding of linkonce_odr symbols in LTO:
* LLVM tells the linker which symbols can be hidden if not used from native
  files.
* The linker tells LLVM which symbols are not used from other object files,
  but will be put in the dso symbol table if present.

GOLD's API is the second option. It was implemented almost 1:1 in llvm by
passing the list down to internalize.

LLVM already had partial support for the first option. It is also very similar
to how ld64 handles hiding these symbols when *not* doing LTO.

This patch then
* removes the APIs for the DSO list.
* marks LTO_SYMBOL_SCOPE_DEFAULT_CAN_BE_HIDDEN all linkonce_odr unnamed_addr
  global values and other linkonce_odr whose address is not used.
* makes the gold plugin responsible for handling the API mismatch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193800 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Add diagnostic tests for NEON scalar shift immediate instructions (see...
Chad Rosier [Thu, 31 Oct 2013 20:11:32 +0000 (20:11 +0000)]
[AArch64] Add diagnostic tests for NEON scalar shift immediate instructions (see: r193790).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193798 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse StringRef::startswith_lower. No functionality change.
Rui Ueyama [Thu, 31 Oct 2013 19:59:55 +0000 (19:59 +0000)]
Use StringRef::startswith_lower. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193796 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ConstantRange] improve my previous patch per Nick suggestion
Nuno Lopes [Thu, 31 Oct 2013 19:53:53 +0000 (19:53 +0000)]
[ConstantRange] improve my previous patch per Nick suggestion

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193795 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Add support for NEON scalar shift immediate instructions.
Chad Rosier [Thu, 31 Oct 2013 19:28:44 +0000 (19:28 +0000)]
[AArch64] Add support for NEON scalar shift immediate instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193790 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoSparcV9 doesnt have rem instruction either.
Roman Divacky [Thu, 31 Oct 2013 19:22:33 +0000 (19:22 +0000)]
SparcV9 doesnt have rem instruction either.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193789 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse a larger invalid attribute bitcode number
Reid Kleckner [Thu, 31 Oct 2013 19:12:36 +0000 (19:12 +0000)]
Use a larger invalid attribute bitcode number

That way the test won't start faililng when someone adds a new attribute
and wants to use the next logical enum (38) for bitcode.  The new
bitcode file tries to use the number 48 as an attribute instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193787 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDWARFDebugArangeSet: remove dead code
Alexey Samsonov [Thu, 31 Oct 2013 18:54:20 +0000 (18:54 +0000)]
DWARFDebugArangeSet: remove dead code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193785 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd FileCheck tests for @LINE
Matt Arsenault [Thu, 31 Oct 2013 18:18:09 +0000 (18:18 +0000)]
Add FileCheck tests for @LINE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193782 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips] XFAIL several MCJIT remote tests
Petar Jovanovic [Thu, 31 Oct 2013 18:10:25 +0000 (18:10 +0000)]
[mips] XFAIL several MCJIT remote tests

Two of the tests are new test cases (cross-module-a.ll, multi-module-a.ll)
not yet supported on MIPS, while XFAIL for the other two tests was
accidentally removed in r193570 and this change reverts those lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193781 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDWARFUnit: kill dead code and make a couple of functions private. No functionality...
Alexey Samsonov [Thu, 31 Oct 2013 18:05:02 +0000 (18:05 +0000)]
DWARFUnit: kill dead code and make a couple of functions private. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193780 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info: remove duplication of DIEs when a DIE can be shared across CUs.
Manman Ren [Thu, 31 Oct 2013 17:54:35 +0000 (17:54 +0000)]
Debug Info: remove duplication of DIEs when a DIE can be shared across CUs.

We add a map in DwarfDebug to map MDNodes that are shareable across CUs to the
corresponding DIEs: MDTypeNodeToDieMap. These DIEs can be shared across CUs,
that is why we keep the maps in DwarfDebug instead of CompileUnit.

We make the assumption that if a DIE is not added to an owner yet, we assume
it belongs to the current CU. Since DIEs for the type system are added to
their owners immediately after creation, and other DIEs belong to the current
CU, the assumption should be true.

A testing case is added to show that we only create a single DIE for a type
MDNode and we use ref_addr to refer to the type DIE.

We also add a testing case to show ref_addr relocations for non-darwin
platforms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193779 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMerge and filecheckize.
Roman Divacky [Thu, 31 Oct 2013 17:50:45 +0000 (17:50 +0000)]
Merge and filecheckize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193778 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoCleanup: update comments.
Manman Ren [Thu, 31 Oct 2013 17:25:22 +0000 (17:25 +0000)]
Cleanup: update comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193773 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDWARFAbbreviationDeclaration: remove dead code, refactor parsing code and make it...
Alexey Samsonov [Thu, 31 Oct 2013 17:20:14 +0000 (17:20 +0000)]
DWARFAbbreviationDeclaration: remove dead code, refactor parsing code and make it more robust. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193770 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLower stackmap intrinsics directly to their target opcode in the DAG builder.
Andrew Trick [Thu, 31 Oct 2013 17:18:24 +0000 (17:18 +0000)]
Lower stackmap intrinsics directly to their target opcode in the DAG builder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193769 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd Verifier test case for variable argument intrinsics.
Andrew Trick [Thu, 31 Oct 2013 17:18:17 +0000 (17:18 +0000)]
Add Verifier test case for variable argument intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193768 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd experimental stackmap intrinsics to definition file and documenation.
Andrew Trick [Thu, 31 Oct 2013 17:18:14 +0000 (17:18 +0000)]
Add experimental stackmap intrinsics to definition file and documenation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193767 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoEnable variable arguments support for intrinsics.
Andrew Trick [Thu, 31 Oct 2013 17:18:11 +0000 (17:18 +0000)]
Enable variable arguments support for intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193766 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agowhitespace
Andrew Trick [Thu, 31 Oct 2013 17:18:07 +0000 (17:18 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193765 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove another unused flag.
Rafael Espindola [Thu, 31 Oct 2013 15:58:33 +0000 (15:58 +0000)]
Remove another unused flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193756 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unused flag.
Rafael Espindola [Thu, 31 Oct 2013 15:49:39 +0000 (15:49 +0000)]
Remove unused flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193752 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRules adjustments in order to build on DragonFly BSD.
Rafael Espindola [Thu, 31 Oct 2013 14:35:00 +0000 (14:35 +0000)]
Rules adjustments in order to build on DragonFly BSD.

Patch by Robin Hahling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193750 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove the --shrink-wrap option.
Rafael Espindola [Thu, 31 Oct 2013 14:07:59 +0000 (14:07 +0000)]
Remove the --shrink-wrap option.

It had no tests, was unused and was "experimental at best".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193749 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd AVX512 unmasked integer broadcast intrinsics and support.
Cameron McInally [Thu, 31 Oct 2013 13:56:31 +0000 (13:56 +0000)]
Add AVX512 unmasked integer broadcast intrinsics and support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193748 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAVX-512: Implemented CMOV for 512-bit vectors
Elena Demikhovsky [Thu, 31 Oct 2013 13:15:32 +0000 (13:15 +0000)]
AVX-512: Implemented CMOV for 512-bit vectors

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193747 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[SystemZ] Automatically detect zEC12 and z196 hosts
Richard Sandiford [Thu, 31 Oct 2013 12:14:17 +0000 (12:14 +0000)]
[SystemZ] Automatically detect zEC12 and z196 hosts

As on other hosts, the CPU identification instruction is priveleged,
so we need to look through /proc/cpuinfo.  I copied the PowerPC way of
handling "generic".

Several tests were implicitly assuming z10 and so failed on z196.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193742 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64] Make the use of FP instructions optional, but enabled by default.
Amara Emerson [Thu, 31 Oct 2013 09:32:11 +0000 (09:32 +0000)]
[AArch64] Make the use of FP instructions optional, but enabled by default.

This adds a new subtarget feature called FPARMv8 (implied by NEON), and
predicates the support of the FP instructions and registers on this feature.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193739 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm/test/Bitcode/invalid.ll: Tweak expresion to mach "llvm-dis.EXE:"
NAKAMURA Takumi [Thu, 31 Oct 2013 06:21:00 +0000 (06:21 +0000)]
llvm/test/Bitcode/invalid.ll: Tweak expresion to mach "llvm-dis.EXE:"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193738 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a use after free on invalid input.
Rafael Espindola [Thu, 31 Oct 2013 04:20:23 +0000 (04:20 +0000)]
Fix a use after free on invalid input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193737 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix most memory leaks in tablegen.
Rafael Espindola [Thu, 31 Oct 2013 04:07:41 +0000 (04:07 +0000)]
Fix most memory leaks in tablegen.

Found by the valgrind bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193736 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoMerge CallGraph and BasicCallGraph.
Rafael Espindola [Thu, 31 Oct 2013 03:03:55 +0000 (03:03 +0000)]
Merge CallGraph and BasicCallGraph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193734 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUpdated llvm-cov's OVERVIEW description
Yuchen Wu [Thu, 31 Oct 2013 02:01:24 +0000 (02:01 +0000)]
Updated llvm-cov's OVERVIEW description

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193732 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLegalize: Improve legalization of long vector extends.
Jim Grosbach [Thu, 31 Oct 2013 00:20:48 +0000 (00:20 +0000)]
Legalize: Improve legalization of long vector extends.

When an extend more than doubles the size of the elements (e.g., a zext
from v16i8 to v16i32), the normal legalization method of splitting the
vectors will run into problems as by the time the destination vector is
legal, the source vector is illegal. The end result is the operation
often becoming scalarized, with the typical horrible performance. For
example, on x86_64, the simple input of:
define void @bar(<16 x i8> %a, <16 x i32>* %p) nounwind {
  %tmp = zext <16 x i8> %a to <16 x i32>
  store <16 x i32> %tmp, <16 x i32>*%p
  ret void
}

Generates:
  .section  __TEXT,__text,regular,pure_instructions
  .section  __TEXT,__const
  .align  5
LCPI0_0:
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .section  __TEXT,__text,regular,pure_instructions
  .globl  _bar
  .align  4, 0x90
_bar:
  vpunpckhbw  %xmm0, %xmm0, %xmm1
  vpunpckhwd  %xmm0, %xmm1, %xmm2
  vpmovzxwd %xmm1, %xmm1
  vinsertf128 $1, %xmm2, %ymm1, %ymm1
  vmovaps LCPI0_0(%rip), %ymm2
  vandps  %ymm2, %ymm1, %ymm1
  vpmovzxbw %xmm0, %xmm3
  vpunpckhwd  %xmm0, %xmm3, %xmm3
  vpmovzxbd %xmm0, %xmm0
  vinsertf128 $1, %xmm3, %ymm0, %ymm0
  vandps  %ymm2, %ymm0, %ymm0
  vmovaps %ymm0, (%rdi)
  vmovaps %ymm1, 32(%rdi)
  vzeroupper
  ret

So instead we can check if there are legal types that enable us to split
more cleverly when the input vector is already legal such that we don't
turn it into an illegal type. If the extend is such that it's more than
doubling the size of the input we check if
  - the number of vector elements is even,
  - the source type is legal,
  - the type of a split source is illegal,
  - the type of an extended (by doubling element size) source is legal, and
  - the type of that extended source when split is legal.
If the conditions are met, instead of just splitting both the
destination and the source types, we create an extend that only goes up
one "step" (doubling the element width), and the continue legalizing the
rest of the operation normally. The result is that this operates as a
new, more effecient, termination condition for the loop of "split the
operation until the destination type is legal."

With this change, the above example now compiles to:
_bar:
  vpxor %xmm1, %xmm1, %xmm1
  vpunpcklbw  %xmm1, %xmm0, %xmm2
  vpunpckhwd  %xmm1, %xmm2, %xmm3
  vpunpcklwd  %xmm1, %xmm2, %xmm2
  vinsertf128 $1, %xmm3, %ymm2, %ymm2
  vpunpckhbw  %xmm1, %xmm0, %xmm0
  vpunpckhwd  %xmm1, %xmm0, %xmm3
  vpunpcklwd  %xmm1, %xmm0, %xmm0
  vinsertf128 $1, %xmm3, %ymm0, %ymm0
  vmovaps %ymm0, 32(%rdi)
  vmovaps %ymm2, (%rdi)
  vzeroupper
  ret

This generalizes a custom lowering that was added a while back to the
ARM backend. That lowering is no longer necessary, and is removed. The
testcases for it, however, provide excellent ARM tests for this change
and so remain.

rdar://14735100

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193727 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix a few typos
Matt Arsenault [Wed, 30 Oct 2013 23:43:29 +0000 (23:43 +0000)]
Fix a few typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193723 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix CodeGen for unaligned loads with address spaces
Matt Arsenault [Wed, 30 Oct 2013 23:30:05 +0000 (23:30 +0000)]
Fix CodeGen for unaligned loads with address spaces

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193721 91177308-0d34-0410-b5e6-96231b3b80d8