oota-llvm.git
11 years agoX86 cost model: Adjust cost for custom lowered vector multiplies
Arnold Schwaighofer [Sat, 2 Mar 2013 04:02:52 +0000 (04:02 +0000)]
X86 cost model: Adjust cost for custom lowered vector multiplies

This matters for example in following matrix multiply:

int **mmult(int rows, int cols, int **m1, int **m2, int **m3) {
  int i, j, k, val;
  for (i=0; i<rows; i++) {
    for (j=0; j<cols; j++) {
      val = 0;
      for (k=0; k<cols; k++) {
        val += m1[i][k] * m2[k][j];
      }
      m3[i][j] = val;
    }
  }
  return(m3);
}

Taken from the test-suite benchmark Shootout.

We estimate the cost of the multiply to be 2 while we generate 9 instructions
for it and end up being quite a bit slower than the scalar version (48% on my
machine).

Also, properly differentiate between avx1 and avx2. On avx-1 we still split the
vector into 2 128bits and handle the subvector muls like above with 9
instructions.
Only on avx-2 will we have a cost of 9 for v4i64.

I changed the test case in test/Transforms/LoopVectorize/X86/avx1.ll to use an
add instead of a mul because with a mul we now no longer vectorize. I did
verify that the mul would be indeed more expensive when vectorized with 3
kernels:

for (i ...)
   r += a[i] * 3;
for (i ...)
  m1[i] = m1[i] * 3; // This matches the test case in avx1.ll
and a matrix multiply.

In each case the vectorized version was considerably slower.

radar://13304919

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176403 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded FIXME for future Hexagon cleanup.
Andrew Trick [Sat, 2 Mar 2013 01:43:08 +0000 (01:43 +0000)]
Added FIXME for future Hexagon cleanup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176400 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPR14448 - prevent the loop vectorizer from vectorizing the same loop twice.
Nadav Rotem [Sat, 2 Mar 2013 01:33:49 +0000 (01:33 +0000)]
PR14448 - prevent the loop vectorizer from vectorizing the same loop twice.
The LoopVectorizer often runs multiple times on the same function due to inlining.
When this happens the loop vectorizer often vectorizes the same loops multiple times, increasing code size and adding unneeded branches.
With this patch, the vectorizer during vectorization puts metadata on scalar loops and marks them as 'already vectorized' so that it knows to ignore them when it sees them a second time.

PR14448.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176399 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoModify {Call,Invoke}Inst::addAttribute to take an AttrKind.
Peter Collingbourne [Sat, 2 Mar 2013 01:20:18 +0000 (01:20 +0000)]
Modify {Call,Invoke}Inst::addAttribute to take an AttrKind.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176397 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCMake: Always include the CheckCXXCompilerFlag in HandleLLVMOptions.cmake.
Jordan Rose [Sat, 2 Mar 2013 01:00:40 +0000 (01:00 +0000)]
CMake: Always include the CheckCXXCompilerFlag in HandleLLVMOptions.cmake.

Previously we relied on it being included by config-ix.cmake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176396 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "Rewrite a test to count emitted instructions without using -stats"
Michael Gottesman [Sat, 2 Mar 2013 00:53:20 +0000 (00:53 +0000)]
Revert "Rewrite a test to count emitted instructions without using -stats"

This reverts commit aac7922b8fe7ae733d3fe6697e6789fd730315dc. I am reverting the
commit since it broke the phase 1 public buildbot for a few hours.

http://lab.llvm.org:8013/builders/clang-x86_64-darwin11-nobootstrap-RA/builds/2137

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176394 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove duplicate line and move another closer to its actual use
Eli Bendersky [Fri, 1 Mar 2013 23:32:40 +0000 (23:32 +0000)]
Remove duplicate line and move another closer to its actual use

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176391 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMIsched machine model: tablegen subtarget emitter improvement.
Andrew Trick [Fri, 1 Mar 2013 23:31:26 +0000 (23:31 +0000)]
MIsched machine model: tablegen subtarget emitter improvement.

Fix the way resources are counted. I'm taking some time to cleanup the
way MachineScheduler handles in-order machine resources. Eventually
we'll need more PPC/Atom test cases in tree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176390 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIn llvm::MemoryBuffer::getFile() remove an unnecessary stat call check.
Argyrios Kyrtzidis [Fri, 1 Mar 2013 22:48:51 +0000 (22:48 +0000)]
In llvm::MemoryBuffer::getFile() remove an unnecessary stat call check.

The sys::fs::is_directory() check is unnecessary because, if the filename is
a directory, the function will fail anyway with the same error code returned.
Remove the check to avoid an unnecessary stat call.

Someone needs to review on windows and see if the check is necessary there or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176386 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix my email address in CREDITS.TXT.
Stefanus Du Toit [Fri, 1 Mar 2013 22:20:03 +0000 (22:20 +0000)]
Fix my email address in CREDITS.TXT.

Checking to see if svn notifications also use correct address now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176385 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Fix inefficient code generation.
Akira Hatanaka [Fri, 1 Mar 2013 21:52:08 +0000 (21:52 +0000)]
[mips] Fix inefficient code generation.

This patch eliminates the need to emit a constant move instruction when this
pattern is matched:

(select (setgt a, Constant), T, F)

The pattern above effectively turns into this:

(conditional-move (setlt a, Constant + 1), F, T)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176384 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemoved extraneous #include "LLVMContextImpl.h" from lib/IR/Module.cpp
Jean-Luc Duprat [Fri, 1 Mar 2013 21:37:24 +0000 (21:37 +0000)]
Removed extraneous #include "LLVMContextImpl.h" from lib/IR/Module.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176382 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRewrite a test to count emitted instructions without using -stats
Eli Bendersky [Fri, 1 Mar 2013 21:34:37 +0000 (21:34 +0000)]
Rewrite a test to count emitted instructions without using -stats

Also removed the comments of "should produce..." because they completely
don't match the actually produced output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176381 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix indentation.
Akira Hatanaka [Fri, 1 Mar 2013 21:22:21 +0000 (21:22 +0000)]
Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176380 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSet properties for f128 type.
Akira Hatanaka [Fri, 1 Mar 2013 21:11:44 +0000 (21:11 +0000)]
Set properties for f128 type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176378 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRewrite a test to check actual output rather than intermediate implementation
Eli Bendersky [Fri, 1 Mar 2013 20:54:00 +0000 (20:54 +0000)]
Rewrite a test to check actual output rather than intermediate implementation
detail.

The was this test was written, it was relying on an implementation detail
(fixups) and hence was very brittle (relying, among other things, on the
exact ordering of statistics printed by MC).

The test was rewritten to check a more observable output difference. While it
doesn't cover 100% of the things the original test covered, it's a good
practice to write regression tests this way. If we want to check that
internal details and invariants hold, such tests should be expressed as unit
tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176377 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoNo need to force-create clang-tools-extra lit.site.cfg
Edwin Vane [Fri, 1 Mar 2013 19:58:58 +0000 (19:58 +0000)]
No need to force-create clang-tools-extra lit.site.cfg

The make (all) target takes care of creating lit configs and auto-generating
tests. The problem with the original 'lit.site.cfg' target is it's not
recursive and doesn't fully create everything necessary for testing
clang-tools-extra.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176374 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd regression tests (WORKSFORME)
Michael Liao [Fri, 1 Mar 2013 19:23:37 +0000 (19:23 +0000)]
Add regression tests (WORKSFORME)

- These tests wont't crash on trunk but would be better to add them so that
  they don't break again in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176369 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGenerate an error message instead of asserting or segfaulting when we can't
Chad Rosier [Fri, 1 Mar 2013 19:12:05 +0000 (19:12 +0000)]
Generate an error message instead of asserting or segfaulting when we can't
handle indirect register inputs.
rdar://13322011

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176367 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: Don't hang forever if a PHI only has skipped PHI uses.
Benjamin Kramer [Fri, 1 Mar 2013 19:07:31 +0000 (19:07 +0000)]
LoopVectorize: Don't hang forever if a PHI only has skipped PHI uses.

Fixes PR15384.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176366 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCache the result of Function::getIntrinsicID() in a DenseMap attached to the LLVMContext.
Michael Ilseman [Fri, 1 Mar 2013 18:48:54 +0000 (18:48 +0000)]
Cache the result of Function::getIntrinsicID() in a DenseMap attached to the LLVMContext.

This reduces the time actually spent doing string to ID conversion and shows a 10% improvement in compile time for a particularly bad case that involves ARM Neon intrinsics (these have many overloads).

Patch by Jean-Luc Duprat!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176365 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix PR10475
Michael Liao [Fri, 1 Mar 2013 18:40:30 +0000 (18:40 +0000)]
Fix PR10475

- ISD::SHL/SRL/SRA must have either both scalar or both vector operands
  but TLI.getShiftAmountTy() so far only return scalar type. As a
  result, backend logic assuming that breaks.
- Rename the original TLI.getShiftAmountTy() to
  TLI.getScalarShiftAmountTy() and re-define TLI.getShiftAmountTy() to
  return target-specificed scalar type or the same vector type as the
  1st operand.
- Fix most TICG logic assuming TLI.getShiftAmountTy() a simple scalar
  type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176364 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd support for using non-pic code for arm and thumb1 when emitting the sjlj
Chad Rosier [Fri, 1 Mar 2013 18:30:38 +0000 (18:30 +0000)]
Add support for using non-pic code for arm and thumb1 when emitting the sjlj
dispatch code.  As far as I can tell the thumb2 code is behaving as expected.
I was able to compile and run the associated test case for both arm and thumb1.
rdar://13066352

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176363 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: fix sampler tests after fixing wait insertions
Christian Konig [Fri, 1 Mar 2013 17:39:05 +0000 (17:39 +0000)]
R600/SI: fix sampler tests after fixing wait insertions

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176359 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoHexagon: Add constant extender support framework.
Jyotsna Verma [Fri, 1 Mar 2013 17:37:13 +0000 (17:37 +0000)]
Hexagon: Add constant extender support framework.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176358 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agotest commit to use consistent comment notation.
Peng Cheng [Fri, 1 Mar 2013 16:49:35 +0000 (16:49 +0000)]
test commit to use consistent comment notation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176353 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGCInfoDeleter code cleanup after r175528
Yiannis Tsiouris [Fri, 1 Mar 2013 11:40:32 +0000 (11:40 +0000)]
GCInfoDeleter code cleanup after r175528

Remove GCInfoDeleter from passes and comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176347 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: handle all registers in copyPhysReg v2
Christian Konig [Fri, 1 Mar 2013 09:46:27 +0000 (09:46 +0000)]
R600/SI: handle all registers in copyPhysReg v2

v2: based on Michels patch, but now allows copying of all registers sizes.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176346 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: remove S_MOV immediate patterns
Christian Konig [Fri, 1 Mar 2013 09:46:22 +0000 (09:46 +0000)]
R600/SI: remove S_MOV immediate patterns

They won't match anyway.

Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176345 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: remove GPR*AlignEncode
Christian Konig [Fri, 1 Mar 2013 09:46:17 +0000 (09:46 +0000)]
R600/SI: remove GPR*AlignEncode

It's much easier to specify the encoding with tablegen directly.

Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176344 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: fix warning about overloaded virtual
Christian Konig [Fri, 1 Mar 2013 09:46:11 +0000 (09:46 +0000)]
R600/SI: fix warning about overloaded virtual

Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176343 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600/SI: fix inserting waits for unordered defines
Christian Konig [Fri, 1 Mar 2013 09:46:04 +0000 (09:46 +0000)]
R600/SI: fix inserting waits for unordered defines

Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176342 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGCC thinks that this variable might be used uninitialized (it isn't).
Duncan Sands [Fri, 1 Mar 2013 09:46:03 +0000 (09:46 +0000)]
GCC thinks that this variable might be used uninitialized (it isn't).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176341 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDocs for llvm-symbolizer command-line tool
Alexey Samsonov [Fri, 1 Mar 2013 07:58:27 +0000 (07:58 +0000)]
Docs for llvm-symbolizer command-line tool

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176337 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMinor coding style fix
Michael Liao [Fri, 1 Mar 2013 04:19:34 +0000 (04:19 +0000)]
Minor coding style fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176334 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Remove unused option. Fix 80-column violations.
Akira Hatanaka [Fri, 1 Mar 2013 02:17:02 +0000 (02:17 +0000)]
[mips] Remove unused option. Fix 80-column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176330 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Add the capability to search delay slot filling instructions in
Akira Hatanaka [Fri, 1 Mar 2013 02:03:51 +0000 (02:03 +0000)]
[mips] Add the capability to search delay slot filling instructions in
successor basic blocks.

Currently this is off by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176329 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Do not add SecondLastInst to list BranchInstrs if there is only one
Akira Hatanaka [Fri, 1 Mar 2013 01:22:26 +0000 (01:22 +0000)]
[mips] Do not add SecondLastInst to list BranchInstrs if there is only one
terminator.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176326 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Define an overloaded version of function MipsInstrInfo::AnalyzeBranchAdd.
Akira Hatanaka [Fri, 1 Mar 2013 01:10:17 +0000 (01:10 +0000)]
[mips] Define an overloaded version of function MipsInstrInfo::AnalyzeBranchAdd.

This function will be used later when the capability to search delay slot
filling instructions in successor blocks is added. No intended functionality
changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176325 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Add options to disable searching backward and in successor blocks.
Akira Hatanaka [Fri, 1 Mar 2013 01:02:36 +0000 (01:02 +0000)]
[mips] Add options to disable searching backward and in successor blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176321 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Add capability to search in the forward direction for instructions that
Akira Hatanaka [Fri, 1 Mar 2013 00:50:52 +0000 (00:50 +0000)]
[mips] Add capability to search in the forward direction for instructions that
can fill the delay slot.

Currently, this is off by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176320 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Define helper function searchRange
Akira Hatanaka [Fri, 1 Mar 2013 00:26:14 +0000 (00:26 +0000)]
[mips] Define helper function searchRange

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176318 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Rename function findDelayInstr to searchBackward.
Akira Hatanaka [Fri, 1 Mar 2013 00:20:16 +0000 (00:20 +0000)]
[mips] Rename function findDelayInstr to searchBackward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176317 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoScheduler diagnostics. Print the register name.
Andrew Trick [Fri, 1 Mar 2013 00:19:14 +0000 (00:19 +0000)]
Scheduler diagnostics. Print the register name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176316 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix incorrect ScheduleDAG comment and formalize Weak edges.
Andrew Trick [Fri, 1 Mar 2013 00:19:12 +0000 (00:19 +0000)]
Fix incorrect ScheduleDAG comment and formalize Weak edges.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176315 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoInstructions schedulers should report correct height/depth.
Andrew Trick [Fri, 1 Mar 2013 00:19:09 +0000 (00:19 +0000)]
Instructions schedulers should report correct height/depth.

We avoided computing DAG height/depth during Node printing because it
shouldn't depend on an otherwise valid DAG. But this has become far
too annoying for the common case of a valid DAG where we want to see
valid values. If doing the computation on-the-fly turns out to be a
problem in practice, then I'll add a mode to the diagnostics to only
force it when we're likely to have a valid DAG, otherwise explicitly
print INVALID instead of bogus numbers. For now, just go for it all
the time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176314 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Define class MemDefsUses.
Akira Hatanaka [Fri, 1 Mar 2013 00:16:31 +0000 (00:16 +0000)]
[mips] Define class MemDefsUses.

This class tracks dependence between memory instructions using underlying
objects of memory operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176313 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoConnect LLVM CMake build scripts to LLDB's CMake scripts:
Daniel Malea [Thu, 28 Feb 2013 23:15:15 +0000 (23:15 +0000)]
Connect LLVM CMake build scripts to LLDB's CMake scripts:
- if you have LLDB checked out in $llvm/tools, CMake will build it now!
- LLDB is known to build on Linux with libstdc++ and GCC 4.6/4.7 or Clang 3.3
- to run lldb tests, do "make check-lldb" after a build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176307 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoA small refactoring + adding comments.
Eli Bendersky [Thu, 28 Feb 2013 23:09:18 +0000 (23:09 +0000)]
A small refactoring + adding comments.

SelectionDAGIsel::LowerArguments needs a function, not a basic block. So it
makes sense to pass it the function instead of extracting a basic-block from
the function and then tossing it. This is also more self-documenting (functions
have arguments, BBs don't).

In addition, added comments to a couple of Select* methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176305 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't add the 'Value' string if there isn't one.
Bill Wendling [Thu, 28 Feb 2013 21:17:03 +0000 (21:17 +0000)]
Don't add the 'Value' string if there isn't one.

This was causing the folding set to fail to fold attributes, because it was
being calculated in one spot without an empty values string but here with an
empty values string.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176301 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a bug in instcombine for fmul in fast math mode.
Quentin Colombet [Thu, 28 Feb 2013 21:12:40 +0000 (21:12 +0000)]
Fix a bug in instcombine for fmul in fast math mode.
The instcombine recognized pattern looks like:
a = b * c
d = a +/- Cst
or
a = b * c
d = Cst +/- a

When creating the new operands for fadd or fsub instruction following the related fmul, the first operand was created with the second original operand (M0 was created with C1) and the second with the first (M1 with Opnd0).

The fix consists in creating the new operands with the appropriate original operand, i.e., M0 with Opnd0 and M1 with C1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176300 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove an assert earlier in a file and check that the result of
Eric Christopher [Thu, 28 Feb 2013 20:26:17 +0000 (20:26 +0000)]
Move an assert earlier in a file and check that the result of
our bitwise compare is equal to the field we're looking for.

Noticed on inspection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176296 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoBrag about function call vectorization in the docs.
Benjamin Kramer [Thu, 28 Feb 2013 19:33:46 +0000 (19:33 +0000)]
Brag about function call vectorization in the docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176292 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't add an attribute that already exists and don't remove an attribute that doesn...
Bill Wendling [Thu, 28 Feb 2013 19:17:40 +0000 (19:17 +0000)]
Don't add an attribute that already exists and don't remove an attribute that doesn't exist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176289 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTidy up; no functional change.
Chad Rosier [Thu, 28 Feb 2013 19:16:42 +0000 (19:16 +0000)]
Tidy up; no functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176288 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCost model support for lowered math builtins.
Benjamin Kramer [Thu, 28 Feb 2013 19:09:33 +0000 (19:09 +0000)]
Cost model support for lowered math builtins.

We make the cost for calling libm functions extremely high as emitting the
calls is expensive and causes spills (on x86) so performance suffers. We still
vectorize important calls like ceilf and friends on SSE4.1. and fabs.

Differential Revision: http://llvm-reviews.chandlerc.com/D466

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176287 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoStyle; no functional change.
Chad Rosier [Thu, 28 Feb 2013 18:54:27 +0000 (18:54 +0000)]
Style; no functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176285 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPut some per-instruction statistics of fast isel under NDEBUG, together with
Eli Bendersky [Thu, 28 Feb 2013 18:05:12 +0000 (18:05 +0000)]
Put some per-instruction statistics of fast isel under NDEBUG, together with
other per-instruction statistics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176273 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-format comments (and check commit access)
Yiannis Tsiouris [Thu, 28 Feb 2013 16:59:10 +0000 (16:59 +0000)]
Re-format comments (and check commit access)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176270 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: remove post-encoder method from FCMP (immediate) instructions.
Tim Northover [Thu, 28 Feb 2013 14:46:14 +0000 (14:46 +0000)]
AArch64: remove post-encoder method from FCMP (immediate) instructions.

The work done by the post-encoder (setting architecturally unused bits to 0 as
required) can be done by the existing operand that covers the "#0.0". This
removes at least one use of the discouraged PostEncoderMethod uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176261 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: be more careful resorting to inefficient addressing for weak vars.
Tim Northover [Thu, 28 Feb 2013 14:36:31 +0000 (14:36 +0000)]
AArch64: be more careful resorting to inefficient addressing for weak vars.

If an otherwise weak var is actually defined in this unit, it can't be
undefined at runtime so we can use normal global variable sequences (ADRP/ADD)
to access it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176259 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: don't drop GlobalAddress offset when handling extern_weak decls.
Tim Northover [Thu, 28 Feb 2013 14:36:24 +0000 (14:36 +0000)]
AArch64: don't drop GlobalAddress offset when handling extern_weak decls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176258 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd the -disable-opt option to LTO. This adds:
Bill Wendling [Thu, 28 Feb 2013 14:11:10 +0000 (14:11 +0000)]
Add the -disable-opt option to LTO. This adds:

- Consistency with opt (which supports the same option with the same meaning and
  description).
- Debugging gold plugin-based linking without optimizations getting in the way.
- Debugging programs linked with the gold plugin while preserving the original
  debug info.
- Fine-grained control over LTO passes using the gold plugin in combination with
  opt (or clang/dragonegg).

Patch by Cristiano Giuffrida!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176257 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: Use cbnz instead of cmp/b.ne pair for atomic operations.
Tim Northover [Thu, 28 Feb 2013 13:52:07 +0000 (13:52 +0000)]
AArch64: Use cbnz instead of cmp/b.ne pair for atomic operations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176253 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Implement sanitize_memory attribute.
Evgeniy Stepanov [Thu, 28 Feb 2013 11:25:14 +0000 (11:25 +0000)]
[msan] Implement sanitize_memory attribute.

Shadow checks are disabled and memory loads always produce fully initialized
values in functions that don't have a sanitize_memory attribute. Value and
argument shadow is propagated as usual.

This change also updates blacklist behaviour to match the above.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176247 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCorrections for XFAIL armv5 tests
Renato Golin [Thu, 28 Feb 2013 10:05:10 +0000 (10:05 +0000)]
Corrections for XFAIL armv5 tests

Most of the tests that behave differently on llvm-arm-linux buildbot
did so becase the triple wasn't set correctly to armv5, so we can
revert most of the special behaviour added previously. Some tests
still need the special treatment, though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176243 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unused leftover declarations.
Evgeniy Stepanov [Thu, 28 Feb 2013 08:42:11 +0000 (08:42 +0000)]
Remove unused leftover declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176240 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEnable syntax highlighting for reStructuredText files.
Bill Wendling [Thu, 28 Feb 2013 06:43:24 +0000 (06:43 +0000)]
Enable syntax highlighting for reStructuredText files.
Patch by Journeyer J. Joh!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176235 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[Support][ErrorOr] Add support for implicit conversion from error code/condition...
Michael J. Spencer [Thu, 28 Feb 2013 01:44:26 +0000 (01:44 +0000)]
[Support][ErrorOr] Add support for implicit conversion from error code/condition enums.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176228 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[PathV2] In llvm::sys::fs::unique_file, make sure it doesn't fall into an infinite...
Argyrios Kyrtzidis [Thu, 28 Feb 2013 00:38:19 +0000 (00:38 +0000)]
[PathV2] In llvm::sys::fs::unique_file, make sure it doesn't fall into an infinite loop by constantly trying
to create the parent path.

This can happen if the path is a relative filename and the current directory was removed.
Thanks to Daniel D. for the hint in fixing it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176226 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a problem in alias analysis. It is about the misinterpretation of "Object".
Shuxin Yang [Thu, 28 Feb 2013 00:24:45 +0000 (00:24 +0000)]
Fix a problem in alias analysis. It is about the misinterpretation of "Object".

This problem is exposed by r171325 which is already reverted. It is rather
hard to fabricate a testing case without it.

r171325 should *NOT* be resurrected as it has a potential problem although
this problem dosen't directly contribute to PR14988.

The bug is tracked by:
  - rdar://13063553, and
  - http://llvm.org/bugs/show_bug.cgi?id=14988

Thank Arnold for coming up a better solution to this problem. After
comparing this solution and my original proposal, I decided to ditch mine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176225 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unnecessary check against isGlobalVariable. We check it
Eric Christopher [Wed, 27 Feb 2013 23:49:50 +0000 (23:49 +0000)]
Remove unnecessary check against isGlobalVariable. We check it
a few lines above.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176224 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRework comment slightly and fix a few typos.
Eric Christopher [Wed, 27 Feb 2013 23:49:47 +0000 (23:49 +0000)]
Rework comment slightly and fix a few typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176223 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unnecessary cast to void.
Eric Christopher [Wed, 27 Feb 2013 23:49:45 +0000 (23:49 +0000)]
Remove unnecessary cast to void.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176222 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info: for static member variables, always put AT_MIPS_linkage_name to the
Manman Ren [Wed, 27 Feb 2013 23:21:02 +0000 (23:21 +0000)]
Debug Info: for static member variables, always put AT_MIPS_linkage_name to the
definition DIE (TAG_variable), and put AT_MIPS_linkage_name to TAG_member when
DarwinGDBCompat is true.

Darwin GDB needs AT_MIPS_linkage_name at both places to work.

Follow-up patch to r176143.
rdar://problem/13291234

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176220 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSilence the unused variable warning.
Nadav Rotem [Wed, 27 Feb 2013 22:52:54 +0000 (22:52 +0000)]
Silence the unused variable warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176218 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThe FastISEL should be fast. But when we record statistics we use atomic operations...
Nadav Rotem [Wed, 27 Feb 2013 21:59:43 +0000 (21:59 +0000)]
The FastISEL should be fast. But when we record statistics we use atomic operations to increment the counters.
This patch disables the counters on non-debug builds. This reduces the runtime of SelectionDAGISel::SelectCodeCommon by ~5%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176214 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCopy missing member in DataLayout copy ctor.
David Blaikie [Wed, 27 Feb 2013 21:44:59 +0000 (21:44 +0000)]
Copy missing member in DataLayout copy ctor.

Test case is missing due to it not being reachable through the current tools
but out of tree code such as the sample at
http://llvm.org/docs/tutorial/LangImpl4.html

Patch by Peng Cheng <gm4cheng@gmail.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176213 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoARM: FMA is legal only if VFP4 is available.
Jim Grosbach [Wed, 27 Feb 2013 21:31:12 +0000 (21:31 +0000)]
ARM: FMA is legal only if VFP4 is available.

rdar://13306723

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176212 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove this instance of dl as it's defined in a previous scope.
Chad Rosier [Wed, 27 Feb 2013 20:34:14 +0000 (20:34 +0000)]
Remove this instance of dl as it's defined in a previous scope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176208 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReverted: r176136 - Have a way for a target to opt-out of target-independent fast...
Michael Ilseman [Wed, 27 Feb 2013 19:54:00 +0000 (19:54 +0000)]
Reverted: r176136 - Have a way for a target to opt-out of target-independent fast isel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176204 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[docs] Discuss manpage output.
Sean Silva [Wed, 27 Feb 2013 18:48:42 +0000 (18:48 +0000)]
[docs] Discuss manpage output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176199 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSwitching -1ULL to UINT64_MAX to fix MSVC warnings. Patch thanks to Peng Cheng!
Aaron Ballman [Wed, 27 Feb 2013 18:38:33 +0000 (18:38 +0000)]
Switching -1ULL to UINT64_MAX to fix MSVC warnings.  Patch thanks to Peng Cheng!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176197 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[docs] Provide pointer for building Sphinx docs.
Sean Silva [Wed, 27 Feb 2013 18:33:21 +0000 (18:33 +0000)]
[docs] Provide pointer for building Sphinx docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176195 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSuppressing MSVC warnings; patch thanks to Peng Cheng!
Aaron Ballman [Wed, 27 Feb 2013 18:25:41 +0000 (18:25 +0000)]
Suppressing MSVC warnings; patch thanks to Peng Cheng!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176193 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoprint TLS segment
Shankar Easwaran [Wed, 27 Feb 2013 17:57:17 +0000 (17:57 +0000)]
print TLS segment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176192 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoARM: permit full range of valid ADR immediates.
Tim Northover [Wed, 27 Feb 2013 16:43:09 +0000 (16:43 +0000)]
ARM: permit full range of valid ADR immediates.

This fixes an issue where trying to assemlbe valid ADR instructions would cause
LLVM to hit a failed assertion.

Patch by Keith Walker.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176189 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: Vectorize math builtin calls.
Benjamin Kramer [Wed, 27 Feb 2013 15:24:19 +0000 (15:24 +0000)]
LoopVectorize: Vectorize math builtin calls.

This properly asks TargetLibraryInfo if a call is available and if it is, it
can be translated into the corresponding LLVM builtin. We don't vectorize sqrt()
yet because I'm not sure about the semantics for negative numbers. The other
intrinsic should be exact equivalents to the libm functions.

Differential Revision: http://llvm-reviews.chandlerc.com/D465

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176188 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoForgot that this file needs to be C89 not C99 thanks to MSVC. No functionality
Nick Lewycky [Wed, 27 Feb 2013 08:28:35 +0000 (08:28 +0000)]
Forgot that this file needs to be C89 not C99 thanks to MSVC. No functionality
change!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176176 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIn GCC 4.7, function names are now forbidden from .gcda files. Support this by
Nick Lewycky [Wed, 27 Feb 2013 06:22:56 +0000 (06:22 +0000)]
In GCC 4.7, function names are now forbidden from .gcda files. Support this by
passing a null pointer to the function name in to GCDAProfiling, and add another
switch onto GCOVProfiling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176173 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDoh, fix behaviour change introduced in r176168 which is tested in clang,
Nick Lewycky [Wed, 27 Feb 2013 06:21:30 +0000 (06:21 +0000)]
Doh, fix behaviour change introduced in r176168 which is tested in clang,
not llvm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176172 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert r176166 because it broke one of the lit tests.
Nadav Rotem [Wed, 27 Feb 2013 05:56:20 +0000 (05:56 +0000)]
Revert r176166 because it broke one of the lit tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176171 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFor each function that we optimize we initialize a new list of lib functions. For...
Nadav Rotem [Wed, 27 Feb 2013 05:53:43 +0000 (05:53 +0000)]
For each function that we optimize we initialize a new list of lib functions. For each function name we malloc memory. This patch changes the Libcall map to use BumpPtrAllocator. Now we malloc only once. This speeds up instcombine by a few % on a large c++ program.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176170 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIRBuilder has grown all sorts of useful utility functions. Make use of them to
Nick Lewycky [Wed, 27 Feb 2013 05:46:30 +0000 (05:46 +0000)]
IRBuilder has grown all sorts of useful utility functions. Make use of them to
clean up this code a tiny bit. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176168 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agostd::string to StringRef.
Nadav Rotem [Wed, 27 Feb 2013 05:23:56 +0000 (05:23 +0000)]
std::string to StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176166 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix cut/paste error in a comment.
Reed Kotler [Wed, 27 Feb 2013 04:20:14 +0000 (04:20 +0000)]
Fix cut/paste error in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176165 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd the skeleton for the Mips constant island pass.
Reed Kotler [Wed, 27 Feb 2013 03:33:58 +0000 (03:33 +0000)]
Add the skeleton for the Mips constant island pass.
It will only be used for Mips 16 at this time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176161 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIR: Don't constant fold GEP bitcasts between different address spaces
Meador Inge [Wed, 27 Feb 2013 02:26:42 +0000 (02:26 +0000)]
IR: Don't constant fold GEP bitcasts between different address spaces

PR15262 reported a bug where the following instruction:

  i8 getelementptr inbounds i8* bitcast ([4 x i8] addrspace(12)* @buf to i8*),
                                i32 2

was getting folded into:

  addrspace(12)* getelementptr inbounds ([4 x i8] addrspace(12)* @buf, i32 0,
                                        i32 2)

This caused instcombine to crash because the original instruction and
the folded instruction have different types.  The issue was fixed by
disallowing bitcasts between different address spaces to be folded away.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176156 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSelectionDAG: If llvm.donothing has a landingpad, we should clear
Manman Ren [Wed, 27 Feb 2013 02:11:57 +0000 (02:11 +0000)]
SelectionDAG: If llvm.donothing has a landingpad, we should clear
CurrentCallSite to avoid an assertion failure:
assert(MMI.getCurrentCallSite() == 0 && "Overlapping call sites!");

rdar://problem/13228754

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176154 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDebug Info: for static member variables, add AT_MIPS_linkage_name to the
Manman Ren [Wed, 27 Feb 2013 00:02:32 +0000 (00:02 +0000)]
Debug Info: for static member variables, add AT_MIPS_linkage_name to the
definition DIE, to make old GDB happy.

We have a regression for old GDB when Clang uses DW_TAG_member to declare
static members inside a class, instead of DW_TAG_variable. This patch will fix
this regression.

rdar://problem/13291234

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176143 91177308-0d34-0410-b5e6-96231b3b80d8