Enrico Granata [Mon, 15 Apr 2013 19:57:32 +0000 (19:57 +0000)]
Enabling test case to write the average+stddev pair to the results
The sketch test case writes avg+stddev for all its metrics:
<key>fetch-frames</key>
<dict>
<key>description</key>
<string>time to dump backtrace for every frame in every thread</string>
<key>stddev</key>
<real>0.
006270938361432314</real>
<key>value</key>
<real>0.
011568079851851851</real>
</dict>
llvm-svn: 179550
Rafael Espindola [Mon, 15 Apr 2013 19:28:45 +0000 (19:28 +0000)]
Fix endianness on some MSVC versions.
Looks like it was evaluating undef == undef to true.
llvm-svn: 179549
Enrico Granata [Mon, 15 Apr 2013 19:07:38 +0000 (19:07 +0000)]
- Adding a relaunch feature to the performance tester: you can use the relaunch if you want to measure multiple runs of your app keeping the same metrics alive. New arguments must be supplied - and the step counter will not be reset (this makes it easy to avoid endless loops)
- Having the Sketch test case relaunch itself
llvm-svn: 179548
Argyrios Kyrtzidis [Mon, 15 Apr 2013 18:47:22 +0000 (18:47 +0000)]
Revert "Speed-up ObjCMethodDecl::getOverriddenMethods()."
This reverts commit r179436.
Due to caching, it was possible that we could miss overridden methods that
were introduced by categories later on.
Along with reverting the commit I also included a test case that would have caught this.
llvm-svn: 179547
Tom Stellard [Mon, 15 Apr 2013 17:51:35 +0000 (17:51 +0000)]
R600/SI: Emit config values in register value pairs.
Instead of emitting config values in a predefined order, the code
emitter will now emit a 32-bit register index followed by the 32-bit
config value.
llvm-svn: 179546
Tom Stellard [Mon, 15 Apr 2013 17:51:30 +0000 (17:51 +0000)]
R600/SI: Emit configuration value in the .AMDGPU.config ELF section
llvm-svn: 179545
Tom Stellard [Mon, 15 Apr 2013 17:51:21 +0000 (17:51 +0000)]
R600: Emit ELF formatted code rather than raw ISA.
llvm-svn: 179544
Sean Callanan [Mon, 15 Apr 2013 17:44:44 +0000 (17:44 +0000)]
Removed a duplicate copy of the contents of
Materializer.h that somehow crept in, maybe
during a patch operation.
llvm-svn: 179543
Jim Grosbach [Mon, 15 Apr 2013 17:40:48 +0000 (17:40 +0000)]
Fix a typo in comment.
llvm-svn: 179542
Jim Grosbach [Mon, 15 Apr 2013 17:40:45 +0000 (17:40 +0000)]
Simplify the MCInst operator iterator declaration.
llvm-svn: 179541
John Criswell [Mon, 15 Apr 2013 17:38:06 +0000 (17:38 +0000)]
Grammar and punctuation fixes.
No content changes.
llvm-svn: 179540
Sean Callanan [Mon, 15 Apr 2013 17:12:47 +0000 (17:12 +0000)]
Audited the existing Materializer code to ensure
that it works in the absence of a process. Codepaths
in the Materializer now use the best execution context
scope available to them.
llvm-svn: 179539
Rafael Espindola [Mon, 15 Apr 2013 17:06:15 +0000 (17:06 +0000)]
Remove XFAIL now that the test is standalone.
llvm-svn: 179538
Argyrios Kyrtzidis [Mon, 15 Apr 2013 16:52:57 +0000 (16:52 +0000)]
[PCH/test] Make test/PCH/cxx-typeid.cpp self-contained by including the relevant standard library declarations
instead of depending on a system header inclusion.
llvm-svn: 179537
Rafael Espindola [Mon, 15 Apr 2013 16:46:43 +0000 (16:46 +0000)]
Try to fix the mingw builds.
llvm-svn: 179536
Arnold Schwaighofer [Mon, 15 Apr 2013 16:11:25 +0000 (16:11 +0000)]
Fix bit size of v64i8 and v32i16 vector types.
Patch by Cameron McInally <cameron.mcinally@nyu.edu>.
llvm-svn: 179535
Rafael Espindola [Mon, 15 Apr 2013 16:08:02 +0000 (16:08 +0000)]
Remove getters now that we can specialize structs on the host endianness.
llvm-svn: 179534
Tim Northover [Mon, 15 Apr 2013 15:49:13 +0000 (15:49 +0000)]
Avoid outputting temporary test file into source tree.
llvm-svn: 179532
Alexander Kornienko [Mon, 15 Apr 2013 15:47:34 +0000 (15:47 +0000)]
Fix unused variable warning with assertions disabled.
llvm-svn: 179531
Rafael Espindola [Mon, 15 Apr 2013 15:13:10 +0000 (15:13 +0000)]
Remove unused function.
llvm-svn: 179530
Rafael Espindola [Mon, 15 Apr 2013 15:10:35 +0000 (15:10 +0000)]
Use llvm::sys::IsBigEndianHost.
llvm-svn: 179529
Edwin Vane [Mon, 15 Apr 2013 14:50:35 +0000 (14:50 +0000)]
Adding support for -include/-exclude to cpp11-migrate
This commit adds initial support for the -include/-exclude options which are
both currently marked as hidden. This support is the first step toward
supporting transformations in headers included from source files.
Added unittests to test include/exclude support.
Author: Jack Yang <jack.yang@intel.com>
llvm-svn: 179528
Rafael Espindola [Mon, 15 Apr 2013 14:44:24 +0000 (14:44 +0000)]
Make the host endianness check an integer constant expression.
I will remove the isBigEndianHost function once I update clang.
The ifdef logic is designed to
* not use configure/cmake to avoid breaking -arch i686 -arch ppc.
* default to little endian
* be as small as possible
It looks like sys/endian.h is the preferred header on most modern BSD systems,
but it is better to change this in a followup patch as machine/endian.h is
available on FreeBSD, OpenBSD, NetBSD and OS X.
llvm-svn: 179527
Alexander Kornienko [Mon, 15 Apr 2013 14:28:00 +0000 (14:28 +0000)]
Unified token breaking logic for strings and block comments.
Summary:
Both strings and block comments are broken into lines in
breakProtrudingToken. Logic specific for strings or block comments is abstracted
in implementations of the BreakToken interface. Among other goodness, this
change fixes placement of backslashes after a block comment inside a
preprocessor directive (see removed FIXMEs in unit tests).
The code is far from being polished, and some parts of it will be changed for
line comments support.
Reviewers: klimek
Reviewed By: klimek
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D665
llvm-svn: 179526
Timur Iskhodzhanov [Mon, 15 Apr 2013 14:16:31 +0000 (14:16 +0000)]
Add a missing space
llvm-svn: 179525
Evgeniy Stepanov [Mon, 15 Apr 2013 13:35:05 +0000 (13:35 +0000)]
[msan] Fix sigaction test.
Restore SIGPROF handler to the original state after the test.
llvm-svn: 179524
Rafael Espindola [Mon, 15 Apr 2013 12:49:13 +0000 (12:49 +0000)]
Remove hasExternalLinkageUncached.
It was being used correctly, but it is a very dangerous API to have around.
Instead, move the logic from the filtering to when we are deciding if we should
link two decls.
llvm-svn: 179523
Evgeniy Stepanov [Mon, 15 Apr 2013 12:41:52 +0000 (12:41 +0000)]
[msan] Really disable replacement new and delete.
llvm-svn: 179522
Rafael Espindola [Mon, 15 Apr 2013 12:38:20 +0000 (12:38 +0000)]
Fix the storage class of method instantiations.
We keep the "as written" storage class, but that is a fuzzy concept for
instantiations. With this patch instantiations of methods of class templates
now get a storage class that is based on the semantics of isStatic(). With this
can simplify isStatic() itself.
llvm-svn: 179521
Andy Gibbs [Mon, 15 Apr 2013 12:06:32 +0000 (12:06 +0000)]
Replace uses of the deprecated std::auto_ptr with OwningPtr.
This is a rework of the broken parts in r179373 which were subsequently reverted in r179374 due to incompatibility with C++98 compilers. This version should be ok under C++98.
llvm-svn: 179520
Tim Northover [Mon, 15 Apr 2013 11:55:27 +0000 (11:55 +0000)]
Remove reference to MSVC only building X86 backend.
This is no longer true.
llvm-svn: 179519
Tim Northover [Mon, 15 Apr 2013 11:53:05 +0000 (11:53 +0000)]
Enable all targets by default on Visual Studio.
llvm-svn: 179518
Richard Smith [Mon, 15 Apr 2013 08:33:22 +0000 (08:33 +0000)]
Local thread_local variables are implicitly 'static'. (This doesn't apply to _Thread_local nor __thread.)
llvm-svn: 179517
Richard Smith [Mon, 15 Apr 2013 08:07:34 +0000 (08:07 +0000)]
Properly check for a constant initializer for a thread-local variable.
llvm-svn: 179516
Richard Smith [Mon, 15 Apr 2013 08:02:05 +0000 (08:02 +0000)]
Add triple to another test.
llvm-svn: 179515
Richard Smith [Mon, 15 Apr 2013 08:00:15 +0000 (08:00 +0000)]
Add triples to these tests since they're now using TLS, which isn't available on all targets.
llvm-svn: 179514
Eric Christopher [Mon, 15 Apr 2013 07:31:37 +0000 (07:31 +0000)]
Revert "Recommit r179497 after fixing uninitialized variable." until
I can fix the testcases here:
http://lab.llvm.org:8011/builders/clang-native-arm-cortex-a9/builds/6952
This reverts commit r179512 due to testcases specifying triples
that they didn't actually mean and causing failures on other platforms.
llvm-svn: 179513
Eric Christopher [Mon, 15 Apr 2013 07:07:21 +0000 (07:07 +0000)]
Recommit r179497 after fixing uninitialized variable.
llvm-svn: 179512
Nadav Rotem [Mon, 15 Apr 2013 05:56:55 +0000 (05:56 +0000)]
Document our desire to enable the loop vectorizer on -Os in future releases.
llvm-svn: 179511
Nadav Rotem [Mon, 15 Apr 2013 05:53:23 +0000 (05:53 +0000)]
Docs: merge the description of the BB and SLP vectorizers and document the -fslp-vectorize-aggressive flag.
llvm-svn: 179510
Nadav Rotem [Mon, 15 Apr 2013 05:39:58 +0000 (05:39 +0000)]
Add an option -vectorize-slp-aggressive for running the BB vectorizer. Make -fslp-vectorize run the slp-vectorizer.
llvm-svn: 179508
Nadav Rotem [Mon, 15 Apr 2013 05:38:41 +0000 (05:38 +0000)]
Add a new flag -vectorize-slp-aggressive to enable Hals BB vectorizer.
llvm-svn: 179507
Nadav Rotem [Mon, 15 Apr 2013 04:57:18 +0000 (04:57 +0000)]
Rename the slp-vectorizer clang/llvm flags. No functionality change.
llvm-svn: 179506
Nadav Rotem [Mon, 15 Apr 2013 04:54:42 +0000 (04:54 +0000)]
Rename the slp-vectorizer clang/llvm flags. No functionality change.
llvm-svn: 179505
Nadav Rotem [Mon, 15 Apr 2013 04:25:27 +0000 (04:25 +0000)]
SLPVectorizer: Add support for vectorizing trees that start at compare instructions.
llvm-svn: 179504
Jia Liu [Mon, 15 Apr 2013 03:26:13 +0000 (03:26 +0000)]
fix include path in doc Extending LLVM
llvm-svn: 179503
Hal Finkel [Mon, 15 Apr 2013 02:37:46 +0000 (02:37 +0000)]
Mark all PPC comparison instructions as not having side effects
Now that the CR spilling issues have been resolved, we can remove the
unmodeled-side-effect attributes from the comparison instructions (and also
mark them as isCompare). By allowing these, by default, to have unmodeled side
effects, we were hiding problems with CR spilling; but everything seems much
happier now.
llvm-svn: 179502
Dmitri Gribenko [Mon, 15 Apr 2013 02:31:50 +0000 (02:31 +0000)]
Comment command table: use inheritance instead of duplicating code
llvm-svn: 179501
Hal Finkel [Mon, 15 Apr 2013 02:07:05 +0000 (02:07 +0000)]
Fix PPC64 CR spill location for callee-saved registers
This fixes an ABI bug for non-Darwin PPC64. For the callee-saved condition
registers, the spill location is specified relative to the stack pointer (SP +
8). However, this is not relative to the SP after the new stack frame is
established, but instead relative to the caller's stack pointer (it is stored
into the linkage area of the parent's stack frame).
So, like with the link register, we don't directly spill the CRs with other
callee-saved registers, but just mark them to be spilled during prologue
generation.
In practice, this reverts r179457 for PPC64 (but leaves it in place for PPC32).
llvm-svn: 179500
Eric Christopher [Sun, 14 Apr 2013 23:35:36 +0000 (23:35 +0000)]
Revert "Remove some unused triple and data layout."
This reverts commit r179497 and the accompanying commit as it broke random platforms that aren't osx.
llvm-svn: 179499
Eric Christopher [Sun, 14 Apr 2013 23:32:44 +0000 (23:32 +0000)]
Remove some unused triple and data layout.
llvm-svn: 179498
Eric Christopher [Sun, 14 Apr 2013 23:32:40 +0000 (23:32 +0000)]
If we've specified a triple on the command line then go ahead
and use that as the default triple for the module and target
data layout.
llvm-svn: 179497
Richard Smith [Sun, 14 Apr 2013 23:01:42 +0000 (23:01 +0000)]
CodeGen support for function-local static thread_local variables with
non-constant constructors or non-trivial destructors. Plus bugfixes for
thread_local references bound to temporaries (the temporaries themselves are
lifetime-extended to become thread_local), and the corresponding case for
std::initializer_list.
llvm-svn: 179496
Rafael Espindola [Sun, 14 Apr 2013 22:08:07 +0000 (22:08 +0000)]
Removed #if 0 code that doesn't compiled if uncommented.
llvm-svn: 179495
Nico Rieck [Sun, 14 Apr 2013 21:18:36 +0000 (21:18 +0000)]
Use object file specific section type for initial text section
llvm-svn: 179494
David Majnemer [Sun, 14 Apr 2013 21:15:43 +0000 (21:15 +0000)]
Reorders two transforms that collide with each other
One performs: (X == 13 | X == 14) -> X-13 <u 2
The other: (A == C1 || A == C2) -> (A & ~(C1 ^ C2)) == C1
The problem is that there are certain values of C1 and C2 that
trigger both transforms but the first one blocks out the second,
this generates suboptimal code.
Reordering the transforms should be better in every case and
allows us to do interesting stuff like turn:
%shr = lshr i32 %X, 4
%and = and i32 %shr, 15
%add = add i32 %and, -14
%tobool = icmp ne i32 %add, 0
into:
%and = and i32 %X, 240
%tobool = icmp ne i32 %and, 224
llvm-svn: 179493
Nadav Rotem [Sun, 14 Apr 2013 20:13:05 +0000 (20:13 +0000)]
Make the command line triple match the module triple.
llvm-svn: 179492
Richard Smith [Sun, 14 Apr 2013 20:11:31 +0000 (20:11 +0000)]
Diagnose if a __thread or _Thread_local variable has a non-constant initializer
or non-trivial destructor.
llvm-svn: 179491
Anna Zaks [Sun, 14 Apr 2013 18:36:51 +0000 (18:36 +0000)]
[analyzer] Add a link to the Building a Checker in 24 Hours talk to the developer manual
llvm-svn: 179490
Simon Atanasyan [Sun, 14 Apr 2013 14:07:51 +0000 (14:07 +0000)]
[Mips] Support -mmicromips / -mno-micromips command line options.
llvm-svn: 179489
Simon Atanasyan [Sun, 14 Apr 2013 14:07:41 +0000 (14:07 +0000)]
[Mips] Fix indentation.
llvm-svn: 179488
Simon Atanasyan [Sun, 14 Apr 2013 14:07:36 +0000 (14:07 +0000)]
[Mips] Follow-up to r179481. Reduce code duplication. Use
AddTargetFeature() routine to handle -msingle-float / -mdouble-float
options.
llvm-svn: 179487
Simon Atanasyan [Sun, 14 Apr 2013 14:07:30 +0000 (14:07 +0000)]
[Mips] Follow-up to r179481. Consider "single-float" as a separate
independent of float ABI feature in the MipsTargetInfoBase class.
llvm-svn: 179486
Tobias Grosser [Sun, 14 Apr 2013 13:15:59 +0000 (13:15 +0000)]
SCEVValidator: Correctly store 'k * p' as a parameter
We do not only need to understand that 'k * p' is a parameter expression, but
also need to store this expression in the set of parameters. Before this patch
we wrongly stored the two individual parameters %k and %p.
Reported by: Sebastian Pop <spop@codeaurora.org>
llvm-svn: 179485
Rafael Espindola [Sun, 14 Apr 2013 10:14:21 +0000 (10:14 +0000)]
Add driver support for fedora 18 on ARM.
llvm-svn: 179484
Benjamin Kramer [Sun, 14 Apr 2013 09:33:08 +0000 (09:33 +0000)]
Miscellaneous cleanups for VecUtils.h
llvm-svn: 179483
John McCall [Sun, 14 Apr 2013 08:50:55 +0000 (08:50 +0000)]
Handle incompatible redeclarations of library builtins better.
Invalid redeclarations of valid explicit declarations shouldn't
take the same path as redeclarations of implicit declarations,
and invalid local extern declarations shouldn't foul things up
for everybody else.
llvm-svn: 179482
Simon Atanasyan [Sun, 14 Apr 2013 08:37:15 +0000 (08:37 +0000)]
[Mips] Remove "single" from the list of valid MIPS float ABI names. Add
two new options –msingle-float and –mdouble-float. These options can be
used simultaneously with float ABI selection options (-mfloat-abi,
-mhard-float, -msoft-float). They mark whether a floating-point
coprocessor supports double-precision operations.
llvm-svn: 179481
Nadav Rotem [Sun, 14 Apr 2013 07:42:25 +0000 (07:42 +0000)]
Document the SLP infrastructure.
llvm-svn: 179480
Nadav Rotem [Sun, 14 Apr 2013 07:22:22 +0000 (07:22 +0000)]
SLP: Document the scalarization cost method.
llvm-svn: 179479
Nadav Rotem [Sun, 14 Apr 2013 05:55:18 +0000 (05:55 +0000)]
Document the decision to assume that the cost of floats is twice as much as integers.
llvm-svn: 179478
Jakob Stoklund Olesen [Sun, 14 Apr 2013 05:48:50 +0000 (05:48 +0000)]
Use i32 for all SPARC shift amounts, even in 64-bit mode.
Test case by llvm-stress.
llvm-svn: 179477
Nadav Rotem [Sun, 14 Apr 2013 05:47:04 +0000 (05:47 +0000)]
Remove unused function attributes.
llvm-svn: 179476
Nadav Rotem [Sun, 14 Apr 2013 05:15:53 +0000 (05:15 +0000)]
SLPVectorizer: Add support for trees that don't start at binary operators, and add the cost of extracting values from the roots of the tree.
llvm-svn: 179475
Jakob Stoklund Olesen [Sun, 14 Apr 2013 05:10:36 +0000 (05:10 +0000)]
Add support for the abs64 SPARC v9 code model.
For when 16 TB just isn't enough.
llvm-svn: 179474
Jakob Stoklund Olesen [Sun, 14 Apr 2013 04:57:51 +0000 (04:57 +0000)]
Add support for the SPARC v9 abs44 code model.
This is the default model for non-PIC 64-bit code. It supports
text+data+bss linked anywhere in the low 16 TB of the address space.
llvm-svn: 179473
Jakob Stoklund Olesen [Sun, 14 Apr 2013 04:35:19 +0000 (04:35 +0000)]
Use target flags for printing SPARC asm operands.
64-bit code models need multiple relocations that can't be inferred from
the opcode like they can in 32-bit code.
llvm-svn: 179472
Jakob Stoklund Olesen [Sun, 14 Apr 2013 04:35:16 +0000 (04:35 +0000)]
Also put target flags on SPARC constant pool references.
Constant pool entries are accessed exactly the same way as global
variables.
llvm-svn: 179471
Nadav Rotem [Sun, 14 Apr 2013 03:22:20 +0000 (03:22 +0000)]
SLPVectorizer: add initial support for reduction variable vectorization.
llvm-svn: 179470
Jakob Stoklund Olesen [Sun, 14 Apr 2013 01:53:23 +0000 (01:53 +0000)]
Fix patterns for 64-bit pointers.
This fixes the pic32 code model for SPARC v9.
llvm-svn: 179469
Jakob Stoklund Olesen [Sun, 14 Apr 2013 01:33:32 +0000 (01:33 +0000)]
Add target flags to SPARC address operands.
SDNodes and MachineOperands get target flags representing the %hi() and
%lo() assembly annotations that eventually become relocations.
Also define flags to be used by the 64-bit code models.
llvm-svn: 179468
Howard Hinnant [Sun, 14 Apr 2013 00:01:13 +0000 (00:01 +0000)]
Accidentally disallowed explicit tuple conversions when all elements of the tuple can be explicitly converted.
llvm-svn: 179467
Greg Clayton [Sat, 13 Apr 2013 23:17:23 +0000 (23:17 +0000)]
Fixed issues with the way ELF symbols are parsed:
- Do not add symbols with no names
- Make sure that symbols from ELF symbol tables know that the byte size is correct. Previously the symbols would calculate their sizes by looking for the next symbol and take symbols that had zero size and make them have invalid sizes.
- Added the ability to dump raw ELF symbols by adding a Dump method to ELFSymbol
Also removed some unused code from lldb_private::Symtab.
llvm-svn: 179466
Hal Finkel [Sat, 13 Apr 2013 23:06:15 +0000 (23:06 +0000)]
Mark all PPC CR registers to be spilled as live-in and tag MFCR appropriately
Leaving MFCR has having unmodeled side effects is not enough to prevent
unwanted instruction reordering post-RA. We could probably apply a stronger
barrier attribute, but there is a better way: Add all (not just the first) CR
to be spilled as live-in to the entry block, and add all CRs to the MFCR
instruction as implicitly killed.
Unfortunately, I don't have a small test case.
llvm-svn: 179465
Rafael Espindola [Sat, 13 Apr 2013 22:26:02 +0000 (22:26 +0000)]
Simplify test so that it is more portable.
I have checked that the test still fails when the "|| !P.isRegularFile()" from
the original patch is removed.
llvm-svn: 179464
Jakob Stoklund Olesen [Sat, 13 Apr 2013 19:02:23 +0000 (19:02 +0000)]
Define SPARC code models.
Currently, only abs32 and pic32 are implemented. Add a test case for
abs32 with 64-bit code. 64-bit PIC code is currently broken.
llvm-svn: 179463
Jakob Stoklund Olesen [Sat, 13 Apr 2013 19:02:16 +0000 (19:02 +0000)]
Use the correct types when matching ADDRri patterns from frame indexes.
It doesn't seem like anybody is checking types this late in isel, so no
test case.
llvm-svn: 179462
Howard Hinnant [Sat, 13 Apr 2013 18:19:25 +0000 (18:19 +0000)]
Set failbit when strtold sets errno to ERANGE when parsing floating point values.
llvm-svn: 179461
Benjamin Kramer [Sat, 13 Apr 2013 16:11:14 +0000 (16:11 +0000)]
GlobalDCE: Fix an oversight in my last commit that could lead to crashes.
There is a Constant with non-constant operands: blockaddress.
llvm-svn: 179460
Sylvestre Ledru [Sat, 13 Apr 2013 13:20:12 +0000 (13:20 +0000)]
Remove the useless SRCROOT declaration from the call of build-swig-wrapper-classes.sh & finish-swig-wrapper-classes.sh
Two reasons for that:
* the declaration is not used. the LLDB_SOURCE_DIR is provided as the first argument in the script ($1) (called SRC_ROOT in the source code)
* add_custom_command is quoting the first argument of the command. Usually, it is the script itself (and then the full path to the script) but, here, it is the declaration of a variable.
It was failing with:
cd "/llvm-toolchain-3.3~svn179457/build-llvm/tools/lldb/scripts" && "SRCROOT=/llvm-toolchain-3.3~svn179457/tools/lldb" /llvm-toolchain-3.3~svn179457/tools/lldb/scripts/build-swig-wrapper-classes.sh /llvm-toolchain-3.3~svn179457/tools/lldb /llvm-toolchain-3.3~svn179457/build-llvm/tools/lldb/scripts /llvm-toolchain-3.3~svn179457/build-llvm/tools/lldb/scripts /llvm-toolchain-3.3~svn179457/build-llvm -m
/bin/sh: 1: SRCROOT=/llvm-toolchain-3.3~svn179457/tools/lldb: not found
llvm-svn: 179459
Benjamin Kramer [Sat, 13 Apr 2013 12:53:18 +0000 (12:53 +0000)]
Fix a scalability issue with complex ConstantExprs.
This is basically the same fix in three different places. We use a set to avoid
walking the whole tree of a big ConstantExprs multiple times.
For example: (select cmp, (add big_expr 1), (add big_expr 2))
We don't want to visit big_expr twice here, it may consist of thousands of
nodes.
The testcase exercises this by creating an insanely large ConstantExprs out of
a loop. It's questionable if the optimizer should ever create those, but this
can be triggered with real C code. Fixes PR15714.
llvm-svn: 179458
Hal Finkel [Sat, 13 Apr 2013 08:09:20 +0000 (08:09 +0000)]
Spill and restore PPC CR registers using the FP when we have one
For functions that need to spill CRs, and have dynamic stack allocations, the
value of the SP during the restore is not what it was during the save, and so
we need to use the FP in these cases (as for all of the other spills and
restores, but the CR restore has a special code path because its reserved slot,
like the link register, is specified directly relative to the adjusted SP).
llvm-svn: 179457
Andrew Trick [Sat, 13 Apr 2013 07:37:27 +0000 (07:37 +0000)]
Further generalize this scheduler test.
The order of copies depends on queue order, which is not very stable.
llvm-svn: 179456
Andrew Trick [Sat, 13 Apr 2013 07:29:21 +0000 (07:29 +0000)]
Fix a dislexic regex.
llvm-svn: 179455
Simon Atanasyan [Sat, 13 Apr 2013 06:43:15 +0000 (06:43 +0000)]
Remove duplicated comment.
llvm-svn: 179454
Andrew Trick [Sat, 13 Apr 2013 06:12:46 +0000 (06:12 +0000)]
Add a missing REQUIRES: asserts
llvm-svn: 179453
Andrew Trick [Sat, 13 Apr 2013 06:07:49 +0000 (06:07 +0000)]
MI-Sched: DEBUG formatting.
llvm-svn: 179452
Andrew Trick [Sat, 13 Apr 2013 06:07:45 +0000 (06:07 +0000)]
MI-Sched cleanup. If an instruction has no valid sched class, do not attempt to check for a variant.
llvm-svn: 179451
Andrew Trick [Sat, 13 Apr 2013 06:07:43 +0000 (06:07 +0000)]
X86 machine model: reduce SandyBridge and Haswell ILPWindow.
The initial values were arbitrary. I want them to be more
conservative. This represents the number of latency cycles hidden by
OOO execution. In practice, I think it should be within a small factor
of the complex floating point operation latency so the scheduler can
make some attempt to hide latency even for smallish blocks.
These are by no means the best values, just a starting point for
tuning heuristics. Some benchmarks such as TSVC run faster with this
lower value for SandyBridge. I haven't run anything on Haswell, but
it's shouldn't be 2x SB.
llvm-svn: 179450
Andrew Trick [Sat, 13 Apr 2013 06:07:40 +0000 (06:07 +0000)]
MI-Sched: schedule physreg copies.
The register allocator expects minimal physreg live ranges. Schedule
physreg copies accordingly. This is slightly tricky when they occur in
the middle of the scheduling region. For now, this is handled by
rescheduling the copy when its associated instruction is
scheduled. Eventually we may instead bundle them, but only if we can
preserve the bundles as parallel copies during regalloc.
llvm-svn: 179449