oota-llvm.git
11 years agoUpdate comments for SSAUpdater to use the modern doxygen comment
Chandler Carruth [Sun, 28 Jul 2013 22:00:33 +0000 (22:00 +0000)]
Update comments for SSAUpdater to use the modern doxygen comment
standards for LLVM. Remove duplicated comments on the interface from the
implementation file (implementation comments are left there of course).
Also clean up, re-word, and fix a few typos and errors in the commenst
spotted along the way.

This is in preparation for changes to these files and to keep the
uninteresting tidying in a separate commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187335 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove use of sprintf added to X86 disassembler tablegen code. Send message with...
Craig Topper [Sun, 28 Jul 2013 21:28:02 +0000 (21:28 +0000)]
Remove use of sprintf added to X86 disassembler tablegen code. Send message with instruction name to errs() instead and use a generic message for the llvm_unreachable. Consistent with other places in this file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187333 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPartial revert of 187310; it seems MSVC 10 still spits out this warning, but MSVC...
Aaron Ballman [Sun, 28 Jul 2013 18:04:26 +0000 (18:04 +0000)]
Partial revert of 187310; it seems MSVC 10 still spits out this warning, but MSVC 11 does not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187331 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTemporarily revert r187323 until I update SSAUpdater to match mem2reg.
Chandler Carruth [Sun, 28 Jul 2013 09:05:49 +0000 (09:05 +0000)]
Temporarily revert r187323 until I update SSAUpdater to match mem2reg.
I forgot that we had two totally independent things here. :: sigh ::

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187327 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agofixed compilation issue
Elena Demikhovsky [Sun, 28 Jul 2013 08:45:12 +0000 (08:45 +0000)]
fixed compilation issue

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187325 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded encoding prefixes for KNL instructions (EVEX).
Elena Demikhovsky [Sun, 28 Jul 2013 08:28:38 +0000 (08:28 +0000)]
Added encoding prefixes for KNL instructions (EVEX).
Added 512-bit operands printing.
Added instruction formats for KNL instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187324 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoNow that mem2reg understands how to cope with a slightly wider set of
Chandler Carruth [Sun, 28 Jul 2013 08:27:12 +0000 (08:27 +0000)]
Now that mem2reg understands how to cope with a slightly wider set of
uses of an alloca, we can pre-compute promotability while analyzing an
alloca for splitting in SROA. That lets us short-circuit the common case
of a bunch of trivially promotable allocas. This cuts 20% to 30% off the
run time of SROA for typical frontend-generated IR sequneces I'm seeing.
It gets the new SROA to within 20% of ScalarRepl for such code. My
current benchmark for these numbers is PR15412, but it fits the general
pattern of IR emitted by Clang so it should be widely applicable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187323 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThread DataLayout through the callers and into mem2reg. This will be
Chandler Carruth [Sun, 28 Jul 2013 06:43:11 +0000 (06:43 +0000)]
Thread DataLayout through the callers and into mem2reg. This will be
useful in a subsequent patch, but causes an unfortunate amount of noise,
so I pulled it out into a separate patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187322 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[PowerPC] Add comment explaining preprocessor directive.
Bill Schmidt [Sun, 28 Jul 2013 03:23:32 +0000 (03:23 +0000)]
[PowerPC] Add comment explaining preprocessor directive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187320 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert 187318
Bill Schmidt [Sun, 28 Jul 2013 02:13:24 +0000 (02:13 +0000)]
Revert 187318

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187319 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[PowerPC] Remove unnecessary preprocessor checking.
Bill Schmidt [Sun, 28 Jul 2013 02:08:13 +0000 (02:08 +0000)]
[PowerPC] Remove unnecessary preprocessor checking.

The tests !defined(__ppc__) && !defined(__powerpc__) are not needed
or helpful when verifying that code is being compiled for a 64-bit
target.  The simpler test provided by this revision is sufficient to
tell if the target is 64-bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187318 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the comment
Nadav Rotem [Sat, 27 Jul 2013 23:28:47 +0000 (23:28 +0000)]
Update the comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187316 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[APFloat] Make all arithmetic operations with NaN produce positive NaNs.
Michael Gottesman [Sat, 27 Jul 2013 21:49:25 +0000 (21:49 +0000)]
[APFloat] Make all arithmetic operations with NaN produce positive NaNs.

IEEE-754R 1.4 Exclusions states that IEEE-754R does not specify the
interpretation of the sign of NaNs. In order to remove an irrelevant
variable that most floating point implementations do not use,
standardize add, sub, mul, div, mod so that operating anything with
NaN always yields a positive NaN.

In a later commit I am going to update the APIs for creating NaNs so
that one can not even create a negative NaN.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187314 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[APFloat] Move setting fcNormal in zeroSignificand() to calling code.
Michael Gottesman [Sat, 27 Jul 2013 21:49:21 +0000 (21:49 +0000)]
[APFloat] Move setting fcNormal in zeroSignificand() to calling code.

Zeroing the significand of a floating point number does not necessarily cause a
floating point number to become finite non zero. For instance, if one has a NaN,
zeroing the significand will cause it to become +/- infinity.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187313 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[APFloat] Removed nextafter from missing operations since it is implemented in APFloa...
Michael Gottesman [Sat, 27 Jul 2013 21:49:19 +0000 (21:49 +0000)]
[APFloat] Removed nextafter from missing operations since it is implemented in APFloat::next.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187312 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-enabling some more MSVC warnings; all of these compile cleanly with no further...
Aaron Ballman [Sat, 27 Jul 2013 20:20:28 +0000 (20:20 +0000)]
Re-enabling some more MSVC warnings; all of these compile cleanly with no further changes required.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187310 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMinor code simplification suggested by Duncan
Matt Arsenault [Sat, 27 Jul 2013 19:22:28 +0000 (19:22 +0000)]
Minor code simplification suggested by Duncan

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187309 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDwarfDebug: MD5 is always little endian, bswap on big endian platforms.
Benjamin Kramer [Sat, 27 Jul 2013 14:14:43 +0000 (14:14 +0000)]
DwarfDebug: MD5 is always little endian, bswap on big endian platforms.

This makes LLVM emit the same signature regardless of host and target endianess.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187304 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCreate a constant pool symbol for the GOT in the ARMCGBR the same way we
Chandler Carruth [Sat, 27 Jul 2013 11:58:26 +0000 (11:58 +0000)]
Create a constant pool symbol for the GOT in the ARMCGBR the same way we
do in the SDag when lowering references to the GOT: use
ARMConstantPoolSymbol rather than creating a dummy global variable. The
computation of the alignment still feels weird (it uses IR types and
datalayout) but it preserves the exact previous behavior. This change
fixes the memory leak of the global variable detected on the valgrind
leak checking bot.

Thanks to Benjamin Kramer for pointing me at ARMConstantPoolSymbol to
handle this use case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187303 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix yet another memory leak found by the vg-leak bot. Folks (including
Chandler Carruth [Sat, 27 Jul 2013 11:23:08 +0000 (11:23 +0000)]
Fix yet another memory leak found by the vg-leak bot. Folks (including
me) should start watching this bot more as its catching lots of bugs.

The fix here is to not construct the global if we aren't going to need
it. That's cheaper anyways, and globals have highly predictable types in
practice. I've added an assert to catch skew between our manual testing
of the type and the actual type just for paranoia's sake.

Note that this pattern is actually fine in most globals because when you
build a global with a module it automatically is moved to be owned by
that module. But here, we're in isel and don't really want to do that.
The solution of not creating a global is simpler anyways.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187302 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a memory leak in the debug emission by simply not allocating memory.
Chandler Carruth [Sat, 27 Jul 2013 11:09:58 +0000 (11:09 +0000)]
Fix a memory leak in the debug emission by simply not allocating memory.
There doesn't appear to be any reason to put this variable on the heap.
I'm suspicious of the LexicalScope above that we stuff in a map and then
delete afterward, but I'm just trying to get the valgrind bot clean.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187301 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a memory leak in the hexagon scheduler. We call initialize here more
Chandler Carruth [Sat, 27 Jul 2013 10:48:45 +0000 (10:48 +0000)]
Fix a memory leak in the hexagon scheduler. We call initialize here more
than once, and the second time through we leaked memory. Found thanks to
the vg-leak bot, but I can't locally reproduce it with valgrind. The
debugger confirms that it is in fact leaking here.

This whole code is totally gross. Why is initialize being called on each
runOnFunction??? Why aren't these OwningPtr<>s, and why aren't their
lifetimes better defined? Anyways, this is just a surgical change to
help out the leak checking bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187299 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't use all the #ifdefs to hide the stats counters and instead rely on
Chandler Carruth [Sat, 27 Jul 2013 10:17:49 +0000 (10:17 +0000)]
Don't use all the #ifdefs to hide the stats counters and instead rely on
their being optimized out in debug mode. Realistically, this just isn't
going to be the slow part anyways. This also fixes unused variable
warnings that are breaking LLD build bots. =/ I didn't see these at
first, and kept losing track of the fact that they were broken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187297 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge the removal of dead instructions and lifetime markers with the
Chandler Carruth [Sat, 27 Jul 2013 09:43:30 +0000 (09:43 +0000)]
Merge the removal of dead instructions and lifetime markers with the
analysis of the alloca. We don't need to visit all the users twice for
this. We build up a kill list during the analysis and then just process
it afterward. This recovers the tiny bit of performance lost by moving
to the visitor based analysis system as it removes one entire use-list
walk from mem2reg. In some cases, this is now faster than mem2reg was
previously.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187296 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-enabling a warning in MSVC mode now that r187292 fixed the only instance of the...
Aaron Ballman [Sat, 27 Jul 2013 03:35:44 +0000 (03:35 +0000)]
Re-enabling a warning in MSVC mode now that r187292 fixed the only instance of the warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187293 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplifyCFG: Add missing tests from r187278
Tom Stellard [Sat, 27 Jul 2013 02:54:44 +0000 (02:54 +0000)]
SimplifyCFG: Add missing tests from r187278

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187291 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate this CMakeLists.txt for r187283 too.
Nick Lewycky [Sat, 27 Jul 2013 01:26:30 +0000 (01:26 +0000)]
Update this CMakeLists.txt for r187283 too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187286 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info Verifier: verify SPs in llvm.dbg.sp.
Manman Ren [Sat, 27 Jul 2013 01:26:08 +0000 (01:26 +0000)]
Debug Info Verifier: verify SPs in llvm.dbg.sp.

Also always add DIType, DISubprogram and DIGlobalVariable to the list
in DebugInfoFinder without checking them, so we can verify them later
on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187285 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAlso update CMakeLists.txt for r187283.
Nick Lewycky [Sat, 27 Jul 2013 01:25:51 +0000 (01:25 +0000)]
Also update CMakeLists.txt for r187283.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187284 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReimplement isPotentiallyReachable to make nocapture deduction much stronger.
Nick Lewycky [Sat, 27 Jul 2013 01:24:00 +0000 (01:24 +0000)]
Reimplement isPotentiallyReachable to make nocapture deduction much stronger.
Adds unit tests for it too.

Split BasicBlockUtils into an analysis-half and a transforms-half, and put the
analysis bits into a new Analysis/CFG.{h,cpp}. Promote isPotentiallyReachable
into llvm::isPotentiallyReachable and move it into Analysis/CFG.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187283 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-enabling some more MSVC warnings; all of these compile cleanly with no further...
Aaron Ballman [Sat, 27 Jul 2013 00:13:11 +0000 (00:13 +0000)]
Re-enabling some more MSVC warnings; all of these compile cleanly with no further changes required.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187279 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplifyCFG: Use parallel-and and parallel-or mode to consolidate branch conditions
Tom Stellard [Sat, 27 Jul 2013 00:01:07 +0000 (00:01 +0000)]
SimplifyCFG: Use parallel-and and parallel-or mode to consolidate branch conditions

Merge consecutive if-regions if they contain identical statements.
Both transformations reduce number of branches.  The transformation
is guarded by a target-hook, and is currently enabled only for +R600,
but the correctness has been tested on X86 target using a variety of
CPU benchmarks.

Patch by: Mei Ye

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187278 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove the default back to pipefail. I accidentally reverted it before.
Rafael Espindola [Fri, 26 Jul 2013 23:22:43 +0000 (23:22 +0000)]
Move the default back to pipefail. I accidentally reverted it before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187271 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPropagate pipefail when cloning.
Rafael Espindola [Fri, 26 Jul 2013 23:17:38 +0000 (23:17 +0000)]
Propagate pipefail when cloning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187268 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSLP Vectorier: Don't vectorize really short chains because they are already handled...
Nadav Rotem [Fri, 26 Jul 2013 23:07:55 +0000 (23:07 +0000)]
SLP Vectorier:  Don't vectorize really short chains because they are already handled by the SelectionDAG store-vectorizer, which does a better job in deciding when to vectorize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187267 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSLP Vectorizer: Disable the vectorization of non power of two chains, such as <3...
Nadav Rotem [Fri, 26 Jul 2013 22:53:11 +0000 (22:53 +0000)]
SLP Vectorizer: Disable the vectorization of non power of two chains, such as <3 x float>, because we dont have a good cost model for these types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187265 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse pipefail when available.
Rafael Espindola [Fri, 26 Jul 2013 22:32:58 +0000 (22:32 +0000)]
Use pipefail when available.

This change makes test with RUN lines like
RUN: opt ... | FileCheck

fail if opt fails, even if it prints what FileCheck wants. Enabling this
found some interesting cases of broken tests that were not being noticed
because opt (or some other tool) was crashing late.

Pipefail is used when the shell supports it or when using the internal
python based tester.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187261 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agonext batch of -disable-debug-info-verifier
Rafael Espindola [Fri, 26 Jul 2013 22:31:26 +0000 (22:31 +0000)]
next batch of -disable-debug-info-verifier

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187260 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "[PowerPC] Improve consistency in use of __ppc__, __powerpc__, etc."
Rafael Espindola [Fri, 26 Jul 2013 22:13:57 +0000 (22:13 +0000)]
Revert "[PowerPC] Improve consistency in use of __ppc__, __powerpc__, etc."

This reverts commit r187248. It broke many bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187254 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix variable name.
Owen Anderson [Fri, 26 Jul 2013 22:06:21 +0000 (22:06 +0000)]
Fix variable name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187253 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-enabling the C4065 warning for MSVC builds as it no longer fires due to tablegen.
Aaron Ballman [Fri, 26 Jul 2013 22:03:03 +0000 (22:03 +0000)]
Re-enabling the C4065 warning for MSVC builds as it no longer fires due to tablegen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187252 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse a non-c'tor for converting a boolean into a StringRef.
Bill Wendling [Fri, 26 Jul 2013 21:50:30 +0000 (21:50 +0000)]
Use a non-c'tor for converting a boolean into a StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187250 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoWhen InstCombine tries to fold away (fsub x, (fneg y)) into (fadd x, y), it is
Owen Anderson [Fri, 26 Jul 2013 21:40:29 +0000 (21:40 +0000)]
When InstCombine tries to fold away (fsub x, (fneg y)) into (fadd x, y), it is
also worthwhile for it to look through FP extensions and truncations, whose
application commutes with fneg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187249 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[PowerPC] Improve consistency in use of __ppc__, __powerpc__, etc.
Bill Schmidt [Fri, 26 Jul 2013 21:39:15 +0000 (21:39 +0000)]
[PowerPC] Improve consistency in use of __ppc__, __powerpc__, etc.

Both GCC and LLVM will implicitly define __ppc__ and __powerpc__ for
all PowerPC targets, whether 32- or 64-bit.  They will both implicitly
define __ppc64__ and __powerpc64__ for 64-bit PowerPC targets, and not
for 32-bit targets.  We cannot be sure that all other possible
compilers used to compile Clang/LLVM define both __ppc__ and
__powerpc__, for example, so it is best to check for both when relying
on either inside the Clang/LLVM code base.

This patch makes sure we always check for both variants.  In addition,
it fixes one unnecessary check in lib/Target/PowerPC/PPCJITInfo.cpp.
(At least one of __ppc__ and __powerpc__ should always be defined when
compiling for a PowerPC target, no matter which compiler is used, so
testing for them is unnecessary.)

There are some places in the compiler that check for other variants,
like __POWERPC__ and _POWER, and I have left those in place.  There is
no need to add them elsewhere.  This seems to be in Apple-specific
code, and I won't take a chance on breaking it.

There is no intended change in behavior; thus, no test cases are
added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187248 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse more parens to clarify assert.
Eric Christopher [Fri, 26 Jul 2013 21:16:25 +0000 (21:16 +0000)]
Use more parens to clarify assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187247 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove addLetterToHash, no functional change.
Eric Christopher [Fri, 26 Jul 2013 21:07:18 +0000 (21:07 +0000)]
Remove addLetterToHash, no functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187245 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Implement llvm.trap intrinsic.
Akira Hatanaka [Fri, 26 Jul 2013 20:58:55 +0000 (20:58 +0000)]
[mips] Implement llvm.trap intrinsic.

Patch by Sasa Stankovic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187244 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Fix FP conditional move instructions to have explicit FP condition code
Akira Hatanaka [Fri, 26 Jul 2013 20:51:20 +0000 (20:51 +0000)]
[mips] Fix FP conditional move instructions to have explicit FP condition code
register operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187242 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd missing 'n'.
Rafael Espindola [Fri, 26 Jul 2013 20:44:45 +0000 (20:44 +0000)]
Add missing 'n'.

Thanks to Han Finkel for noticing it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187241 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Fix FP branch instructions to have explicit FP condition code register
Akira Hatanaka [Fri, 26 Jul 2013 20:13:47 +0000 (20:13 +0000)]
[mips] Fix FP branch instructions to have explicit FP condition code register
operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187238 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info Verifier: enable verification of DICompileUnit.
Manman Ren [Fri, 26 Jul 2013 20:04:30 +0000 (20:04 +0000)]
Debug Info Verifier: enable verification of DICompileUnit.

We used to call Verify before adding DICompileUnit to the list, and now we
remove the check and always add DICompileUnit to the list in DebugInfoFinder,
so we can verify them later on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187237 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Increase the number of floating point condition code registers to eight.
Akira Hatanaka [Fri, 26 Jul 2013 19:03:48 +0000 (19:03 +0000)]
[mips] Increase the number of floating point condition code registers to eight.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187234 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Fix floating point branch, comparison, and conditional move instructions
Akira Hatanaka [Fri, 26 Jul 2013 19:01:56 +0000 (19:01 +0000)]
[mips] Fix floating point branch, comparison, and conditional move instructions
to have register FCC0 (the first floating point condition code register) in
their Uses/Defs list.

No intended functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187233 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Delete register print method MipsInstPrinter::printCPURegs that is not
Akira Hatanaka [Fri, 26 Jul 2013 18:50:42 +0000 (18:50 +0000)]
[mips] Delete register print method MipsInstPrinter::printCPURegs that is not
needed. The generic method printOperand will do.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187231 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Print instructions "beq", "bne" and "or" using assembler pseudo
Akira Hatanaka [Fri, 26 Jul 2013 18:34:25 +0000 (18:34 +0000)]
[mips] Print instructions "beq", "bne" and "or" using assembler pseudo
instructions "beqz", "bnez" and "move", when possible.

beq $2, $zero, $L1 => beqz $2, $L1
bne $2, $zero, $L1 => bnez $2, $L1
or  $2, $3, $zero  => move $2, $3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187229 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove trailing whitespace.
Stephen Lin [Fri, 26 Jul 2013 18:09:50 +0000 (18:09 +0000)]
Remove trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187226 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCorrect case of m_UIToFp to m_UIToFP to match instruction name, add m_SIToFP for...
Stephen Lin [Fri, 26 Jul 2013 17:55:00 +0000 (17:55 +0000)]
Correct case of m_UIToFp to m_UIToFP to match instruction name, add m_SIToFP for consistency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187225 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCollapse conditional and add an assert for unhandled scope types.
Eric Christopher [Fri, 26 Jul 2013 17:45:19 +0000 (17:45 +0000)]
Collapse conditional and add an assert for unhandled scope types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187224 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-add DataLayout pointer size convenience functions.
Matt Arsenault [Fri, 26 Jul 2013 17:37:20 +0000 (17:37 +0000)]
Re-add DataLayout pointer size convenience functions.

These were reverted in r167222 along with the rest
of the last different address space pointer size attempt.
These will be used in later commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187223 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove this to the X86 directory, unfortunately the hashes are not
Eric Christopher [Fri, 26 Jul 2013 17:35:47 +0000 (17:35 +0000)]
Move this to the X86 directory, unfortunately the hashes are not
endian independent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187222 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRegenerate.
Eric Christopher [Fri, 26 Jul 2013 17:13:47 +0000 (17:13 +0000)]
Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187217 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd preliminary support for hashing DIEs and breaking them into
Eric Christopher [Fri, 26 Jul 2013 17:02:41 +0000 (17:02 +0000)]
Add preliminary support for hashing DIEs and breaking them into
type units.

Initially this support is used in the computation of an ODR checker
for C++. For now we're attaching it to the DIE, but in the future
it will be attached to the type unit.

This also starts breaking out types into the separation for type
units, but without actually splitting the DIEs.

In preparation for hashing the DIEs this adds a DIEString type
that contains a StringRef with the string contained at the label.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187213 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a way to get the context of any particular scope.
Eric Christopher [Fri, 26 Jul 2013 17:02:36 +0000 (17:02 +0000)]
Add a way to get the context of any particular scope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187212 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove dead or useless header checks from cmake and autoconf
Reid Kleckner [Fri, 26 Jul 2013 16:54:23 +0000 (16:54 +0000)]
Remove dead or useless header checks from cmake and autoconf

On Windows, this improves clean cmake configuration time on my
workstation from 1m58s to 1m32s, which is pretty significant.  There's
probably more that can be done here, but this is the low hanging fruit.

Eric volunteered to regenerate ./configure for me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187209 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoExtend the lifetime of the strings passed to posix_spawn_file_actions_addopen.
Rafael Espindola [Fri, 26 Jul 2013 16:21:31 +0000 (16:21 +0000)]
Extend the lifetime of the strings passed to posix_spawn_file_actions_addopen.

Thanks to Hal Finkel for finding the bug and for the initial patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187208 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoImprove our error handling on windows.
Rafael Espindola [Fri, 26 Jul 2013 14:55:36 +0000 (14:55 +0000)]
Improve our error handling on windows.

* Remove LLVM_ENABLE_CRT_REPORT. LLVM_DISABLE_CRASH_REPORT made it redundant.
* set Return to 1, so that we get a stack trace on failure.
* don't call _exit, so that we get a negative exit value and "not --crash"
  correctly differentiates crashes and regular errors.

This is a bit experimental since the documentation on this interface is sparse.
It doesn't bring up a dialog on my windows setup, but feel free to revert
if it causes problem for your setup (and let me know what it is so that I
can try to fix this patch).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187206 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoNext batch of -disable-debug-info-verifier.
Rafael Espindola [Fri, 26 Jul 2013 14:16:30 +0000 (14:16 +0000)]
Next batch of -disable-debug-info-verifier.

These tests fail without it if pipefail is enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187205 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a target legalize hook for SplitVectorOperand (again)
Justin Holewinski [Fri, 26 Jul 2013 13:28:29 +0000 (13:28 +0000)]
Add a target legalize hook for SplitVectorOperand (again)

CustomLowerNode was not being called during SplitVectorOperand,
meaning custom legalization could not be used by targets.

This also adds a test case for NVPTX that depends on this custom
legalization.

Differential Revision: http://llvm-reviews.chandlerc.com/D1195

Attempt to fix the buildbots by making the X86 test I just added platform independent

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187202 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "Add a target legalize hook for SplitVectorOperand"
Rafael Espindola [Fri, 26 Jul 2013 13:18:16 +0000 (13:18 +0000)]
Revert "Add a target legalize hook for SplitVectorOperand"

This reverts commit 187198. It broke the bots.

The soft float test probably needs a -triple because of name differences.
On the hard float test I am getting a "roundss $1, %xmm0, %xmm0", instead of
"vroundss $1, %xmm0, %xmm0, %xmm0".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187201 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a target legalize hook for SplitVectorOperand
Justin Holewinski [Fri, 26 Jul 2013 12:46:39 +0000 (12:46 +0000)]
Add a target legalize hook for SplitVectorOperand

CustomLowerNode was not being called during SplitVectorOperand,
meaning custom legalization could not be used by targets.

This also adds a test case for NVPTX that depends on this custom
legalization.

Differential Revision: http://llvm-reviews.chandlerc.com/D1195

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187198 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agotest commit
Richard Osborne [Fri, 26 Jul 2013 10:19:02 +0000 (10:19 +0000)]
test commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187195 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[XCore] Add TODO regarding byval structs
Richard Osborne [Fri, 26 Jul 2013 09:50:55 +0000 (09:50 +0000)]
[XCore] Add TODO regarding byval structs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187193 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-implement the analysis of uses in mem2reg to be significantly more
Chandler Carruth [Fri, 26 Jul 2013 08:20:39 +0000 (08:20 +0000)]
Re-implement the analysis of uses in mem2reg to be significantly more
robust. It now uses an InstVisitor and worklist to actually walk the
uses of the Alloca transitively and detect the pattern which we can
directly promote: loads & stores of the whole alloca and instructions we
can completely ignore.

Also, with this new implementation teach both the predicate for testing
whether we can promote and the promotion engine itself to use the same
code so we no longer have strange divergence between the two code paths.

I've added some silly test cases to demonstrate that we can handle
slightly more degenerate code patterns now. See the below for why this
is even interesting.

Performance impact: roughly 1% regression in the performance of SROA or
ScalarRepl on a large C++-ish test case where most of the allocas are
basically ready for promotion. The reason is because of silly redundant
work that I've left FIXMEs for and which I'll address in the next
commit. I wanted to separate this commit as it changes the behavior.
Once the redundant work in removing the dead uses of the alloca is
fixed, this code appears to be faster than the old version. =]

So why is this useful? Because the previous requirement for promotion
required a *specific* visit pattern of the uses of the alloca to verify:
we *had* to look for no more than 1 intervening use. The end goal is to
have SROA automatically detect when an alloca is already promotable and
directly hand it to the mem2reg machinery rather than trying to
partition and rewrite it. This is a 25% or more performance improvement
for SROA, and a significant chunk of the delta between it and
ScalarRepl. To get there, we need to make mem2reg actually capable of
promoting allocas which *look* promotable to SROA without have SROA do
tons of work to massage the code into just the right form.

This is actually the tip of the iceberg. There are tremendous potential
savings we can realize here by de-duplicating work between mem2reg and
SROA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187191 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd test cases for the various instruction alias and Intel syntax fixes that have...
Craig Topper [Fri, 26 Jul 2013 05:39:33 +0000 (05:39 +0000)]
Add test cases for the various instruction alias and Intel syntax fixes that have gone in lately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187188 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix more Intel syntax issues with FP instruction aliases. Test cases coming in a...
Craig Topper [Fri, 26 Jul 2013 05:37:46 +0000 (05:37 +0000)]
Fix more Intel syntax issues with FP instruction aliases. Test cases coming in a subsequent patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187187 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake .bc en/decoding of AttrKind stable
Tobias Grosser [Fri, 26 Jul 2013 04:16:55 +0000 (04:16 +0000)]
Make .bc en/decoding of AttrKind stable

The bitcode representation attribute kinds are encoded into / decoded from
should be independent of the current set of LLVM attributes and their position
in the AttrKind enum. This patch explicitly encodes attributes to fixed bitcode
values.

With this patch applied, LLVM does not silently misread attributes written by
LLVM 3.3. We also enhance the decoding slightly such that an error message is
printed if an unknown AttrKind encoding was dected.

Bonus: Dropping bitcode attributes from AttrKind is now easy, as old AttrKinds
       do not need to be kept to support the Bitcode reader.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187186 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTake advantage of the register enums being in order to remove a couple static tables.
Craig Topper [Fri, 26 Jul 2013 02:02:47 +0000 (02:02 +0000)]
Take advantage of the register enums being in order to remove a couple static tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187182 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[PowerPC] Support powerpc64le as a syntax-checking target.
Bill Schmidt [Fri, 26 Jul 2013 01:35:43 +0000 (01:35 +0000)]
[PowerPC] Support powerpc64le as a syntax-checking target.

This patch provides basic support for powerpc64le as an LLVM target.
However, use of this target will not actually generate little-endian
code.  Instead, use of the target will cause the correct little-endian
built-in defines to be generated, so that code that tests for
__LITTLE_ENDIAN__, for example, will be correctly parsed for
syntax-only testing.  Code generation will otherwise be the same as
powerpc64 (big-endian), for now.

The patch leaves open the possibility of creating a little-endian
PowerPC64 back end, but there is no immediate intent to create such a
thing.

The LLVM portions of this patch simply add ppc64le coverage everywhere
that ppc64 coverage currently exists.  There is nothing of any import
worth testing until such time as little-endian code generation is
implemented.  In the corresponding Clang patch, there is a new test
case variant to ensure that correct built-in defines for little-endian
code are generated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187179 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a bool->StringRef c'tor to StringRef.
Bill Wendling [Thu, 25 Jul 2013 23:06:39 +0000 (23:06 +0000)]
Add a bool->StringRef c'tor to StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187166 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPhabricator.rst: tiny fix
Hans Wennborg [Thu, 25 Jul 2013 22:58:31 +0000 (22:58 +0000)]
Phabricator.rst: tiny fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187164 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUsing a different loop induction variable than the enclosing scope. No functional...
Aaron Ballman [Thu, 25 Jul 2013 22:09:31 +0000 (22:09 +0000)]
Using a different loop induction variable than the enclosing scope.  No functional changes intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187159 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPPC32 va_list is an actual structure so va_copy needs to copy the whole
Roman Divacky [Thu, 25 Jul 2013 21:36:47 +0000 (21:36 +0000)]
PPC32 va_list is an actual structure so va_copy needs to copy the whole
structure not just a pointer. This implements that and thus fixes va_copy
on PPC32. Fixes #15286. Both bug and patch by Florian Zeitz!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187158 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info: update comments and add a FIXME.
Manman Ren [Thu, 25 Jul 2013 21:19:31 +0000 (21:19 +0000)]
Debug Info: update comments and add a FIXME.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187157 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove dead code from the makefile build system.
Rafael Espindola [Thu, 25 Jul 2013 20:25:31 +0000 (20:25 +0000)]
Remove dead code from the makefile build system.

Back in r140220 we removed the autoconf code that would set LLVMCC_OPTION
since it was only used by the test-suite. This patch now removes code
that would only be used if LLVMCC_OPTION was set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187154 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info: improve the verifier to check field types.
Manman Ren [Thu, 25 Jul 2013 19:33:30 +0000 (19:33 +0000)]
Debug Info: improve the verifier to check field types.

Make sure the context field of DIType is MDNode.
Fix testing cases to make them pass the verifier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187150 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove empty directories
Rafael Espindola [Thu, 25 Jul 2013 18:58:58 +0000 (18:58 +0000)]
Remove empty directories

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187146 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove the mblaze backend from llvm.
Rafael Espindola [Thu, 25 Jul 2013 18:55:05 +0000 (18:55 +0000)]
Remove the mblaze backend from llvm.

Approval in here http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-July/064169.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187145 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRegAllocGreedy comment.
Andrew Trick [Thu, 25 Jul 2013 18:35:22 +0000 (18:35 +0000)]
RegAllocGreedy comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187141 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEvict local live ranges if they can be reassigned.
Andrew Trick [Thu, 25 Jul 2013 18:35:19 +0000 (18:35 +0000)]
Evict local live ranges if they can be reassigned.

The previous change to local live range allocation also suppressed
eviction of local ranges. In rare cases, this could result in more
expensive register choices. This commit actually revives a feature
that I added long ago: check if live ranges can be reassigned before
eviction. But now it only happens in rare cases of evicting a local
live range because another local live range wants a cheaper register.

The benefit is improved code size for some benchmarks on x86 and armv7.

I measured no significant compile time increase and performance
changes are noise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187140 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAllocate local registers in order for optimal coloring.
Andrew Trick [Thu, 25 Jul 2013 18:35:14 +0000 (18:35 +0000)]
Allocate local registers in order for optimal coloring.

Also avoid locals evicting locals just because they want a cheaper register.

Problem: MI Sched knows exactly how many registers we have and assumes
they can be colored. In cases where we have large blocks, usually from
unrolled loops, greedy coloring fails. This is a source of
"regressions" from the MI Scheduler on x86. I noticed this issue on
x86 where we have long chains of two-address defs in the same live
range. It's easy to see this in matrix multiplication benchmarks like
IRSmk and even the unit test misched-matmul.ll.

A fundamental difference between the LLVM register allocator and
conventional graph coloring is that in our model a live range can't
discover its neighbors, it can only verify its neighbors. That's why
we initially went for greedy coloring and added eviction to deal with
the hard cases. However, for singly defined and two-address live
ranges, we can optimally color without visiting neighbors simply by
processing the live ranges in instruction order.

Other beneficial side effects:

It is much easier to understand and debug regalloc for large blocks
when the live ranges are allocated in order. Yes, global allocation is
still very confusing, but it's nice to be able to comprehend what
happened locally.

Heuristics could be added to bias register assignment based on
instruction locality (think late register pairing, banks...).

Intuituvely this will make some test cases that are on the threshold
of register pressure more stable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187139 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a way to add a kind-value string pair to an attribute.
Bill Wendling [Thu, 25 Jul 2013 18:34:24 +0000 (18:34 +0000)]
Add a way to add a kind-value string pair to an attribute.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187138 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agotypo.
Adrian Prantl [Thu, 25 Jul 2013 17:52:30 +0000 (17:52 +0000)]
typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187135 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCurrent batch of -disable-debug-info-verifier.
Rafael Espindola [Thu, 25 Jul 2013 17:16:05 +0000 (17:16 +0000)]
Current batch of -disable-debug-info-verifier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187130 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: add llc-based tests for previous commit.
Tim Northover [Thu, 25 Jul 2013 16:23:55 +0000 (16:23 +0000)]
AArch64: add llc-based tests for previous commit.

Better to have tests run even on non-AArch64 platforms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187128 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: fix even more JIT failures
Tim Northover [Thu, 25 Jul 2013 16:03:54 +0000 (16:03 +0000)]
AArch64: fix even more JIT failures

The last patch corrected some issues, but constant-pool entries had actual
codegen bugs in the large memory model (which MCJIT uses).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187126 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't end a file name with a dot. It looks odd.
Rafael Espindola [Thu, 25 Jul 2013 15:00:17 +0000 (15:00 +0000)]
Don't end a file name with a dot. It looks odd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187124 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: don't mask off shift bits when processing JIT relocations.
Tim Northover [Thu, 25 Jul 2013 12:42:52 +0000 (12:42 +0000)]
AArch64: don't mask off shift bits when processing JIT relocations.

This should actually make the MCJIT tests pass again on AArch64. I don't know
how I missed their failure before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187120 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a bug in TableGen where the intrinsic function name recognizer could mis-identify...
Justin Holewinski [Thu, 25 Jul 2013 12:32:00 +0000 (12:32 +0000)]
Fix a bug in TableGen where the intrinsic function name recognizer could mis-identify names if one was a prefix substring of the other

For two intrinsics 'llvm.nvvm.texsurf.handle' and 'llvm.nvvm.texsurf.handle.internal',
TableGen was emitting matching code like:

  if (Name.startswith("llvm.nvvm.texsurf.handle")) ...
  if (Name.startswith("llvm.nvvm.texsurf.handle.internal")) ...

We can never match "llvm.nvvm.texsurf.handle.internal" here because it will
always be erroneously matched by the first condition.

The fix is to sort the intrinsic names and emit them in reverse order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187119 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a comment cut-&-pasto.
Richard Sandiford [Thu, 25 Jul 2013 10:53:02 +0000 (10:53 +0000)]
Fix a comment cut-&-pasto.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187117 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[SystemZ] Rework compare and branch support
Richard Sandiford [Thu, 25 Jul 2013 09:34:38 +0000 (09:34 +0000)]
[SystemZ] Rework compare and branch support

Before the patch we took advantage of the fact that the compare and
branch are glued together in the selection DAG and fused them together
(where possible) while emitting them.  This seemed to work well in practice.
However, fusing the compare so early makes it harder to remove redundant
compares in cases where CC already has a suitable value.  This patch
therefore uses the peephole analyzeCompare/optimizeCompareInstr pair of
functions instead.

No behavioral change intended, but it paves the way for a later patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187116 91177308-0d34-0410-b5e6-96231b3b80d8