OSDN Git Service

android-x86/external-llvm.git
6 years ago[llvm-cov] Use portable output redirection in a test
Vedant Kumar [Fri, 8 Sep 2017 20:24:23 +0000 (20:24 +0000)]
[llvm-cov] Use portable output redirection in a test

A follow-up to a test fix (r312825).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312826 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Try to appease a Windows bot
Vedant Kumar [Fri, 8 Sep 2017 20:18:17 +0000 (20:18 +0000)]
[llvm-cov] Try to appease a Windows bot

On a Windows bot, I see a FileCheck error where the source being matched
over no longer exists, i.e it seems like it's FileCheck'ing some stale
output:

http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win/builds/4747

You can see "// CHECK: [[@LINE]]|{{ +}Marker at 19:3 = 1" in the
FileCheck stderr, but that CHECK line doesn't exist.

Remove the input file to FileCheck before running the test, to try and
appease the bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312825 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Start using !con operator
Matt Arsenault [Fri, 8 Sep 2017 19:09:13 +0000 (19:09 +0000)]
AMDGPU: Start using !con operator

We have a lot of operand definition work essentially producing
every valid permutation of operands to workaround builiding
operand lists based on the instruction features. Apparently tablegen
already has a mostly undocumented operator to concat dags which
simplies this.

Convert one simple place to use this. The BUF instruction definitions
have much more complicated logic that can be totally rewritten now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312822 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Disable name-compression in a test binary
Vedant Kumar [Fri, 8 Sep 2017 19:08:39 +0000 (19:08 +0000)]
[llvm-cov] Disable name-compression in a test binary

This should fix the lld bot:

The Buildbot has detected a new failure on builder llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast while building cfe.
Full details are available at:
http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/16993

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312821 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Recompute scc liveness
Matt Arsenault [Fri, 8 Sep 2017 18:51:26 +0000 (18:51 +0000)]
AMDGPU: Recompute scc liveness

The various scalar bit operations set SCC,
so one is erased or moved it needs to be recomputed.
Not sure why the existing tests don't fail on this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312819 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Coverage] Build sorted and unique segments
Vedant Kumar [Fri, 8 Sep 2017 18:44:50 +0000 (18:44 +0000)]
[Coverage] Build sorted and unique segments

A coverage segment contains a starting line and column, an execution
count, and some other metadata. Clients of the coverage library use
segments to prepare line-oriented reports.

Users of the coverage library depend on segments being unique and sorted
in source order. Currently this is not guaranteed (this is why the clang
change which introduced deferred regions was reverted).

This commit documents the "unique and sorted" condition and asserts that
it holds. It also fixes the SegmentBuilder so that it produces correct
output in some edge cases.

Testing: I've added unit tests for some edge cases. I've also checked
that the new SegmentBuilder implementation is fully covered. Apart from
running check-profile and the llvm-cov tests, I've successfully used a
stage1 llvm-cov to prepare a coverage report for an instrumented clang
binary.

Differential Revision: https://reviews.llvm.org/D36813

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312817 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Fix a lifetime issue
Vedant Kumar [Fri, 8 Sep 2017 18:44:49 +0000 (18:44 +0000)]
[llvm-cov] Fix a lifetime issue

This fixes an issue where a std::string was moved to a constructor
which accepted a StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312816 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Coverage] Define LineColPair for convenience. NFC.
Vedant Kumar [Fri, 8 Sep 2017 18:44:48 +0000 (18:44 +0000)]
[Coverage] Define LineColPair for convenience. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312815 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Coverage] Report errors when reading malformed source regions
Vedant Kumar [Fri, 8 Sep 2017 18:44:47 +0000 (18:44 +0000)]
[Coverage] Report errors when reading malformed source regions

Each source region has a start and end location. Report an error when
the end location does not precede the begin location.

The old lineExecutionCounts.covmapping test actually had a buggy source
region in it. This commit introduces a regenerated copy of the coverage
and moves the old copy to malformedRegions.covmapping, for a test.

Differential Revision: https://reviews.llvm.org/D37387

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312814 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Unify region marker placement between text/html modes
Vedant Kumar [Fri, 8 Sep 2017 18:44:46 +0000 (18:44 +0000)]
[llvm-cov] Unify region marker placement between text/html modes

Make sure that the text and html emitters always emit the same set of
region markers, and avoid emitting redundant markers for line segments
which don't end on the line they start on.

This is related to D35925, and depends on D36014

Differential Revision: https://reviews.llvm.org/D36020

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312813 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] Fix GCC pedantic warnings about default arguments for lambdas.
Chandler Carruth [Fri, 8 Sep 2017 18:23:42 +0000 (18:23 +0000)]
[x86] Fix GCC pedantic warnings about default arguments for lambdas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312809 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Simplify the slow-incdec test and add test cases with optsize.
Craig Topper [Fri, 8 Sep 2017 17:33:54 +0000 (17:33 +0000)]
[X86] Simplify the slow-incdec test and add test cases with optsize.

I think we want to consider using inc/dec with optsize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312804 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLPVectorizer] Add struct InstructionsState that holds information about analysis...
Dinar Temirbulatov [Fri, 8 Sep 2017 17:08:17 +0000 (17:08 +0000)]
[SLPVectorizer] Add struct InstructionsState that holds information about analysis of vector to be vectorized.

Reviewers: spatel, mzolotukhin, mkuper, hfinkel, RKSimon, filcab, ABataev, davide

Subscribers: llvm-commits, rengolin

Differential Revision: https://reviews.llvm.org/D37212

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312802 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a bug for rL312641.
Wei Mi [Fri, 8 Sep 2017 16:44:52 +0000 (16:44 +0000)]
Fix a bug for rL312641.

rL312641 Allowed llvm.memcpy/memset/memmove to be tail calls when parent
function return the intrinsics's first argument. However on arm-none-eabi
platform, llvm.memcpy will be expanded to __aeabi_memcpy which doesn't
have return value. The fix is to check the libcall name after expansion
to match "memcpy/memset/memmove" before allowing those intrinsic to be
tail calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312799 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoPreserve existing regs when adding pristines to LivePhysRegs/LiveRegUnits
Krzysztof Parzyszek [Fri, 8 Sep 2017 16:29:50 +0000 (16:29 +0000)]
Preserve existing regs when adding pristines to LivePhysRegs/LiveRegUnits

Differential Revision: https://reviews.llvm.org/D37600

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312797 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLP] Fix the warning about paths not returning the value, NFC.
Alexey Bataev [Fri, 8 Sep 2017 14:32:20 +0000 (14:32 +0000)]
[SLP] Fix the warning about paths not returning the value, NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312793 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLP] Support for horizontal min/max reduction.
Alexey Bataev [Fri, 8 Sep 2017 13:49:36 +0000 (13:49 +0000)]
[SLP] Support for horizontal min/max reduction.

SLP vectorizer supports horizontal reductions for Add/FAdd binary
operations. Patch adds support for horizontal min/max reductions.
Function getReductionCost() is split to getArithmeticReductionCost() for
binary operation reductions and getMinMaxReductionCost() for min/max
reductions.
Patch fixes PR26956.

Differential revision: https://reviews.llvm.org/D27846

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312791 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Added PR31045 test case
Simon Pilgrim [Fri, 8 Sep 2017 10:49:11 +0000 (10:49 +0000)]
[X86] Added PR31045 test case

Reduced version of 'addr-calc-crash.ll' that was included in D27044, that had been fixed already by D31286/rL298633

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312786 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-enable "[IRCE] Identify loops with latch comparison against current IV value"
Max Kazantsev [Fri, 8 Sep 2017 10:15:05 +0000 (10:15 +0000)]
Re-enable "[IRCE] Identify loops with latch comparison against current IV value"

Re-applying after the found bug was fixed.

Differential Revision: https://reviews.llvm.org/D36215

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312783 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump] Verify line table prologue
Jonas Devlieghere [Fri, 8 Sep 2017 09:48:51 +0000 (09:48 +0000)]
[dwarfdump] Verify line table prologue

This patch adds prologue verification, which is already present in
Apple's dwarfdump. It checks for invalid directory indices and warns
about duplicate file paths.

Differential revision: https://reviews.llvm.org/D37511

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312782 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Adding a test point for PR34149 'Suboptimal codegen for "fast" minnum and maxnum'
Jatin Bhateja [Fri, 8 Sep 2017 09:15:36 +0000 (09:15 +0000)]
[X86] Adding a test point for PR34149 'Suboptimal codegen for "fast" minnum and maxnum'

Differential Revision: https://reviews.llvm.org/D37614

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312778 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-dlltool] Mention arm64 in the lists of architecture alternatives
Martin Storsjo [Fri, 8 Sep 2017 06:49:46 +0000 (06:49 +0000)]
[llvm-dlltool] Mention arm64 in the lists of architecture alternatives

This was missed in SVN r310223 when arm64 support was added.

Differential Revision: https://reviews.llvm.org/D37588

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312776 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agodiff --git a/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp b/lib/Transform...
Max Kazantsev [Fri, 8 Sep 2017 04:26:41 +0000 (04:26 +0000)]
diff --git a/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp b/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp
index f72a808..9fa49fd 100644
--- a/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp
+++ b/lib/Transforms/Scalar/InductiveRangeCheckElimination.cpp
@@ -450,20 +450,10 @@ struct LoopStructure {
   // equivalent to:
   //
   // intN_ty inc = IndVarIncreasing ? 1 : -1;
-  // pred_ty predicate = IndVarIncreasing
-  //                         ? IsSignedPredicate ? ICMP_SLT : ICMP_ULT
-  //                         : IsSignedPredicate ? ICMP_SGT : ICMP_UGT;
+  // pred_ty predicate = IndVarIncreasing ? ICMP_SLT : ICMP_SGT;
   //
-  //
-  // for (intN_ty iv = IndVarStart; predicate(IndVarBase, LoopExitAt);
-  //      iv = IndVarNext)
+  // for (intN_ty iv = IndVarStart; predicate(iv, LoopExitAt); iv = IndVarBase)
   //   ... body ...
-  //
-  // Here IndVarBase is either current or next value of the induction variable.
-  // in the former case, IsIndVarNext = false and IndVarBase points to the
-  // Phi node of the induction variable. Otherwise, IsIndVarNext = true and
-  // IndVarBase points to IV increment instruction.
-  //

   Value *IndVarBase;
   Value *IndVarStart;
@@ -471,13 +461,12 @@ struct LoopStructure {
   Value *LoopExitAt;
   bool IndVarIncreasing;
   bool IsSignedPredicate;
-  bool IsIndVarNext;

   LoopStructure()
       : Tag(""), Header(nullptr), Latch(nullptr), LatchBr(nullptr),
         LatchExit(nullptr), LatchBrExitIdx(-1), IndVarBase(nullptr),
         IndVarStart(nullptr), IndVarStep(nullptr), LoopExitAt(nullptr),
-        IndVarIncreasing(false), IsSignedPredicate(true), IsIndVarNext(false) {}
+        IndVarIncreasing(false), IsSignedPredicate(true) {}

   template <typename M> LoopStructure map(M Map) const {
     LoopStructure Result;
@@ -493,7 +482,6 @@ struct LoopStructure {
     Result.LoopExitAt = Map(LoopExitAt);
     Result.IndVarIncreasing = IndVarIncreasing;
     Result.IsSignedPredicate = IsSignedPredicate;
-    Result.IsIndVarNext = IsIndVarNext;
     return Result;
   }

@@ -841,42 +829,21 @@ LoopStructure::parseLoopStructure(ScalarEvolution &SE,
     return false;
   };

-  // `ICI` can either be a comparison against IV or a comparison of IV.next.
-  // Depending on the interpretation, we calculate the start value differently.
+  // `ICI` is interpreted as taking the backedge if the *next* value of the
+  // induction variable satisfies some constraint.

-  // Pair {IndVarBase; IsIndVarNext} semantically designates whether the latch
-  // comparisons happens against the IV before or after its value is
-  // incremented. Two valid combinations for them are:
-  //
-  // 1) { phi [ iv.start, preheader ], [ iv.next, latch ]; false },
-  // 2) { iv.next; true }.
-  //
-  // The latch comparison happens against IndVarBase which can be either current
-  // or next value of the induction variable.
   const SCEVAddRecExpr *IndVarBase = cast<SCEVAddRecExpr>(LeftSCEV);
   bool IsIncreasing = false;
   bool IsSignedPredicate = true;
-  bool IsIndVarNext = false;
   ConstantInt *StepCI;
   if (!IsInductionVar(IndVarBase, IsIncreasing, StepCI)) {
     FailureReason = "LHS in icmp not induction variable";
     return None;
   }

-  const SCEV *IndVarStart = nullptr;
-  // TODO: Currently we only handle comparison against IV, but we can extend
-  // this analysis to be able to deal with comparison against sext(iv) and such.
-  if (isa<PHINode>(LeftValue) &&
-      cast<PHINode>(LeftValue)->getParent() == Header)
-    // The comparison is made against current IV value.
-    IndVarStart = IndVarBase->getStart();
-  else {
-    // Assume that the comparison is made against next IV value.
-    const SCEV *StartNext = IndVarBase->getStart();
-    const SCEV *Addend = SE.getNegativeSCEV(IndVarBase->getStepRecurrence(SE));
-    IndVarStart = SE.getAddExpr(StartNext, Addend);
-    IsIndVarNext = true;
-  }
+  const SCEV *StartNext = IndVarBase->getStart();
+  const SCEV *Addend = SE.getNegativeSCEV(IndVarBase->getStepRecurrence(SE));
+  const SCEV *IndVarStart = SE.getAddExpr(StartNext, Addend);
   const SCEV *Step = SE.getSCEV(StepCI);

   ConstantInt *One = ConstantInt::get(IndVarTy, 1);
@@ -1060,7 +1027,6 @@ LoopStructure::parseLoopStructure(ScalarEvolution &SE,
   Result.IndVarIncreasing = IsIncreasing;
   Result.LoopExitAt = RightValue;
   Result.IsSignedPredicate = IsSignedPredicate;
-  Result.IsIndVarNext = IsIndVarNext;

   FailureReason = nullptr;

@@ -1350,9 +1316,8 @@ LoopConstrainer::RewrittenRangeInfo LoopConstrainer::changeIterationSpaceEnd(
                                       BranchToContinuation);

     NewPHI->addIncoming(PN->getIncomingValueForBlock(Preheader), Preheader);
-    auto *FixupValue =
-        LS.IsIndVarNext ? PN->getIncomingValueForBlock(LS.Latch) : PN;
-    NewPHI->addIncoming(FixupValue, RRI.ExitSelector);
+    NewPHI->addIncoming(PN->getIncomingValueForBlock(LS.Latch),
+                        RRI.ExitSelector);
     RRI.PHIValuesAtPseudoExit.push_back(NewPHI);
   }

@@ -1735,10 +1700,7 @@ bool InductiveRangeCheckElimination::runOnLoop(Loop *L, LPPassManager &LPM) {
   }
   LoopStructure LS = MaybeLoopStructure.getValue();
   const SCEVAddRecExpr *IndVar =
-      cast<SCEVAddRecExpr>(SE.getSCEV(LS.IndVarBase));
-  if (LS.IsIndVarNext)
-    IndVar = cast<SCEVAddRecExpr>(SE.getMinusSCEV(IndVar,
-                                                  SE.getSCEV(LS.IndVarStep)));
+      cast<SCEVAddRecExpr>(SE.getMinusSCEV(SE.getSCEV(LS.IndVarBase), SE.getSCEV(LS.IndVarStep)));

   Optional<InductiveRangeCheck::Range> SafeIterRange;
   Instruction *ExprInsertPt = Preheader->getTerminator();
diff --git a/test/Transforms/IRCE/latch-comparison-against-current-value.ll b/test/Transforms/IRCE/latch-comparison-against-current-value.ll
deleted file mode 100644
index afea0e6..0000000
--- a/test/Transforms/IRCE/latch-comparison-against-current-value.ll
+++ /dev/null
@@ -1,182 +0,0 @@
-; RUN: opt -verify-loop-info -irce-print-changed-loops -irce -S < %s 2>&1 | FileCheck %s
-
-; Check that IRCE is able to deal with loops where the latch comparison is
-; done against current value of the IV, not the IV.next.
-
-; CHECK: irce: in function test_01: constrained Loop at depth 1 containing: %loop<header><exiting>,%in.bounds<latch><exiting>
-; CHECK: irce: in function test_02: constrained Loop at depth 1 containing: %loop<header><exiting>,%in.bounds<latch><exiting>
-; CHECK-NOT: irce: in function test_03: constrained Loop at depth 1 containing: %loop<header><exiting>,%in.bounds<latch><exiting>
-; CHECK-NOT: irce: in function test_04: constrained Loop at depth 1 containing: %loop<header><exiting>,%in.bounds<latch><exiting>
-
-; SLT condition for increasing loop from 0 to 100.
-define void @test_01(i32* %arr, i32* %a_len_ptr) #0 {
-
-; CHECK:      test_01
-; CHECK:        entry:
-; CHECK-NEXT:     %exit.mainloop.at = load i32, i32* %a_len_ptr, !range !0
-; CHECK-NEXT:     [[COND2:%[^ ]+]] = icmp slt i32 0, %exit.mainloop.at
-; CHECK-NEXT:     br i1 [[COND2]], label %loop.preheader, label %main.pseudo.exit
-; CHECK:        loop:
-; CHECK-NEXT:     %idx = phi i32 [ %idx.next, %in.bounds ], [ 0, %loop.preheader ]
-; CHECK-NEXT:     %idx.next = add nuw nsw i32 %idx, 1
-; CHECK-NEXT:     %abc = icmp slt i32 %idx, %exit.mainloop.at
-; CHECK-NEXT:     br i1 true, label %in.bounds, label %out.of.bounds.loopexit1
-; CHECK:        in.bounds:
-; CHECK-NEXT:     %addr = getelementptr i32, i32* %arr, i32 %idx
-; CHECK-NEXT:     store i32 0, i32* %addr
-; CHECK-NEXT:     %next = icmp slt i32 %idx, 100
-; CHECK-NEXT:     [[COND3:%[^ ]+]] = icmp slt i32 %idx, %exit.mainloop.at
-; CHECK-NEXT:     br i1 [[COND3]], label %loop, label %main.exit.selector
-; CHECK:        main.exit.selector:
-; CHECK-NEXT:     %idx.lcssa = phi i32 [ %idx, %in.bounds ]
-; CHECK-NEXT:     [[COND4:%[^ ]+]] = icmp slt i32 %idx.lcssa, 100
-; CHECK-NEXT:     br i1 [[COND4]], label %main.pseudo.exit, label %exit
-; CHECK-NOT: loop.preloop:
-; CHECK:        loop.postloop:
-; CHECK-NEXT:    %idx.postloop = phi i32 [ %idx.copy, %postloop ], [ %idx.next.postloop, %in.bounds.postloop ]
-; CHECK-NEXT:     %idx.next.postloop = add nuw nsw i32 %idx.postloop, 1
-; CHECK-NEXT:     %abc.postloop = icmp slt i32 %idx.postloop, %exit.mainloop.at
-; CHECK-NEXT:     br i1 %abc.postloop, label %in.bounds.postloop, label %out.of.bounds.loopexit
-
-entry:
-  %len = load i32, i32* %a_len_ptr, !range !0
-  br label %loop
-
-loop:
-  %idx = phi i32 [ 0, %entry ], [ %idx.next, %in.bounds ]
-  %idx.next = add nsw nuw i32 %idx, 1
-  %abc = icmp slt i32 %idx, %len
-  br i1 %abc, label %in.bounds, label %out.of.bounds
-
-in.bounds:
-  %addr = getelementptr i32, i32* %arr, i32 %idx
-  store i32 0, i32* %addr
-  %next = icmp slt i32 %idx, 100
-  br i1 %next, label %loop, label %exit
-
-out.of.bounds:
-  ret void
-
-exit:
-  ret void
-}
-
-; ULT condition for increasing loop from 0 to 100.
-define void @test_02(i32* %arr, i32* %a_len_ptr) #0 {
-
-; CHECK:      test_02
-; CHECK:        entry:
-; CHECK-NEXT:     %exit.mainloop.at = load i32, i32* %a_len_ptr, !range !0
-; CHECK-NEXT:     [[COND2:%[^ ]+]] = icmp ult i32 0, %exit.mainloop.at
-; CHECK-NEXT:     br i1 [[COND2]], label %loop.preheader, label %main.pseudo.exit
-; CHECK:        loop:
-; CHECK-NEXT:     %idx = phi i32 [ %idx.next, %in.bounds ], [ 0, %loop.preheader ]
-; CHECK-NEXT:     %idx.next = add nuw nsw i32 %idx, 1
-; CHECK-NEXT:     %abc = icmp ult i32 %idx, %exit.mainloop.at
-; CHECK-NEXT:     br i1 true, label %in.bounds, label %out.of.bounds.loopexit1
-; CHECK:        in.bounds:
-; CHECK-NEXT:     %addr = getelementptr i32, i32* %arr, i32 %idx
-; CHECK-NEXT:     store i32 0, i32* %addr
-; CHECK-NEXT:     %next = icmp ult i32 %idx, 100
-; CHECK-NEXT:     [[COND3:%[^ ]+]] = icmp ult i32 %idx, %exit.mainloop.at
-; CHECK-NEXT:     br i1 [[COND3]], label %loop, label %main.exit.selector
-; CHECK:        main.exit.selector:
-; CHECK-NEXT:     %idx.lcssa = phi i32 [ %idx, %in.bounds ]
-; CHECK-NEXT:     [[COND4:%[^ ]+]] = icmp ult i32 %idx.lcssa, 100
-; CHECK-NEXT:     br i1 [[COND4]], label %main.pseudo.exit, label %exit
-; CHECK-NOT: loop.preloop:
-; CHECK:        loop.postloop:
-; CHECK-NEXT:    %idx.postloop = phi i32 [ %idx.copy, %postloop ], [ %idx.next.postloop, %in.bounds.postloop ]
-; CHECK-NEXT:     %idx.next.postloop = add nuw nsw i32 %idx.postloop, 1
-; CHECK-NEXT:     %abc.postloop = icmp ult i32 %idx.postloop, %exit.mainloop.at
-; CHECK-NEXT:     br i1 %abc.postloop, label %in.bounds.postloop, label %out.of.bounds.loopexit
-
-entry:
-  %len = load i32, i32* %a_len_ptr, !range !0
-  br label %loop
-
-loop:
-  %idx = phi i32 [ 0, %entry ], [ %idx.next, %in.bounds ]
-  %idx.next = add nsw nuw i32 %idx, 1
-  %abc = icmp ult i32 %idx, %len
-  br i1 %abc, label %in.bounds, label %out.of.bounds
-
-in.bounds:
-  %addr = getelementptr i32, i32* %arr, i32 %idx
-  store i32 0, i32* %addr
-  %next = icmp ult i32 %idx, 100
-  br i1 %next, label %loop, label %exit
-
-out.of.bounds:
-  ret void
-
-exit:
-  ret void
-}
-
-; Same as test_01, but comparison happens against IV extended to a wider type.
-; This test ensures that IRCE rejects it and does not falsely assume that it was
-; a comparison against iv.next.
-; TODO: We can actually extend the recognition to cover this case.
-define void @test_03(i32* %arr, i64* %a_len_ptr) #0 {
-
-; CHECK:      test_03
-
-entry:
-  %len = load i64, i64* %a_len_ptr, !range !1
-  br label %loop
-
-loop:
-  %idx = phi i32 [ 0, %entry ], [ %idx.next, %in.bounds ]
-  %idx.next = add nsw nuw i32 %idx, 1
-  %idx.ext = sext i32 %idx to i64
-  %abc = icmp slt i64 %idx.ext, %len
-  br i1 %abc, label %in.bounds, label %out.of.bounds
-
-in.bounds:
-  %addr = getelementptr i32, i32* %arr, i32 %idx
-  store i32 0, i32* %addr
-  %next = icmp slt i32 %idx, 100
-  br i1 %next, label %loop, label %exit
-
-out.of.bounds:
-  ret void
-
-exit:
-  ret void
-}
-
-; Same as test_02, but comparison happens against IV extended to a wider type.
-; This test ensures that IRCE rejects it and does not falsely assume that it was
-; a comparison against iv.next.
-; TODO: We can actually extend the recognition to cover this case.
-define void @test_04(i32* %arr, i64* %a_len_ptr) #0 {
-
-; CHECK:      test_04
-
-entry:
-  %len = load i64, i64* %a_len_ptr, !range !1
-  br label %loop
-
-loop:
-  %idx = phi i32 [ 0, %entry ], [ %idx.next, %in.bounds ]
-  %idx.next = add nsw nuw i32 %idx, 1
-  %idx.ext = sext i32 %idx to i64
-  %abc = icmp ult i64 %idx.ext, %len
-  br i1 %abc, label %in.bounds, label %out.of.bounds
-
-in.bounds:
-  %addr = getelementptr i32, i32* %arr, i32 %idx
-  store i32 0, i32* %addr
-  %next = icmp ult i32 %idx, 100
-  br i1 %next, label %loop, label %exit
-
-out.of.bounds:
-  ret void
-
-exit:
-  ret void
-}
-
-!0 = !{i32 0, i32 50}
-!1 = !{i64 0, i64 50}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312775 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a crash when emitting debug info for multi-reg function arguments
Adrian Prantl [Fri, 8 Sep 2017 02:31:37 +0000 (02:31 +0000)]
Fix a crash when emitting debug info for multi-reg function arguments
by reusing more of the existing machinery

This is a follow-up to r312169.
Thanks to Björn Pettersson for the testcase!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312773 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[XRay][CodeGen][PowerPC] Fix tail exit codegen for XRay in PPC
Dean Michael Berris [Fri, 8 Sep 2017 01:47:56 +0000 (01:47 +0000)]
[XRay][CodeGen][PowerPC] Fix tail exit codegen for XRay in PPC

Summary:
This fixes code-gen for XRay in PPC. The regression wasn't caught by
codegen tests  which we add in this change.

What happened was the following:

- For tail exits, we used to unconditionally prepend the returns/exits
  with a pseudo-instruction that gets lowered to the instrumentation
  sled (and leave the actual return/exit instruction as-is).
- Changes to the XRay instrumentation pass caused the tail exits to
  suddenly also emit the tail exit pseudo-instruction, since the check
  for whether a return instruction was also a call instruction meant it
  was a tail exit instruction.
- None of the tests caught the regression either due to non-existent
  tests, or the tests being disabled/removed for continuous breakage.

This change re-introduces some of the basic tests and verifies that
we're back to a state that allows the back-end to generate appropriate
XRay instrumented binaries for PPC in the presence of tail exits.

Reviewers: echristo, timshen

Subscribers: nemanjai, kbarton, llvm-commits

Differential Revision: https://reviews.llvm.org/D37570

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312772 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] Flesh out the custom ISel for RMW aritmetic ops with used flags to
Chandler Carruth [Fri, 8 Sep 2017 00:17:12 +0000 (00:17 +0000)]
[x86] Flesh out the custom ISel for RMW aritmetic ops with used flags to
cover the bitwise operators.

Nothing really exciting here, this just stamps out the rest of the core
operations that can RMW memory and set flags.

Still not implemented here: ADC, SBB. Those will require more
interesting logic to channel the flags *in*, and I'm not currently
planning to try to tackle that. It might be interesting for someone who
wants to improve our code generation for bignum implementations.

Differential Revision: https://reviews.llvm.org/D37141

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312768 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoWholeProgramDevirt: When promoting for single-impl devirt, also rename the comdat.
Peter Collingbourne [Fri, 8 Sep 2017 00:10:53 +0000 (00:10 +0000)]
WholeProgramDevirt: When promoting for single-impl devirt, also rename the comdat.

This is required when targeting COFF, as the comdat name must match
one of the names of the symbols in the comdat.

Differential Revision: https://reviews.llvm.org/D37550

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312767 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] Extend the manual ISel of `add` and `sub` with both RMW memory
Chandler Carruth [Thu, 7 Sep 2017 23:54:24 +0000 (23:54 +0000)]
[x86] Extend the manual ISel of `add` and `sub` with both RMW memory
operands and used flags to support matching immediate operands.

This is a bit trickier than register operands, and we still want to fall
back on a register operands even for things that appear to be
"immediates" when they won't actually select into the operation's
immediate operand. This also requires us to handle things like selecting
`sub` vs. `add` to minimize the number of bits needed to represent the
immediate, and picking the shortest immediate encoding. In order to
that, we in turn need to scan to make sure that CF isn't used as it will
get inverted.

The end result seems very nice though, and we're now generating
optimal instruction sequences for these patterns IMO.

A follow-up patch will further expand this to other operations with RMW
memory operands. But handing `add` and `sub` are useful starting points
to flesh out the machinery and make sure interesting and complex cases
can be handled.

Thanks to Craig Topper who provided a few fixes and improvements to this
patch in addition to the review!

Differential Revision: https://reviews.llvm.org/D37139

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312764 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDon't call exit from cl::PrintHelpMessage.
Rafael Espindola [Thu, 7 Sep 2017 23:30:48 +0000 (23:30 +0000)]
Don't call exit from cl::PrintHelpMessage.

Most callers were not expecting the exit(0) and trying to exit with a
different value.

This also adds back the call to cl::PrintHelpMessage in llvm-ar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312761 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Bitcode] Fix some Clang-tidy modernize-use-using and Include What You Use warnings...
Eugene Zelenko [Thu, 7 Sep 2017 23:28:24 +0000 (23:28 +0000)]
[Bitcode] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312760 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSink some IntrinsicInst.h and Intrinsics.h out of llvm/include
Reid Kleckner [Thu, 7 Sep 2017 23:27:44 +0000 (23:27 +0000)]
Sink some IntrinsicInst.h and Intrinsics.h out of llvm/include

Many of these uses can get by with forward declarations. Hopefully this
speeds up compilation after adding a single intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312759 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r312318, r312325, r312424, r312489
Richard Trieu [Thu, 7 Sep 2017 23:20:35 +0000 (23:20 +0000)]
Revert r312318, r312325, r312424, r312489

r312318 - Debug info for variables whose type is shrinked to bool
r312325, r312424, r312489 - Test case for r312318

Revision 312318 introduced a null dereference bug.
Details in https://bugs.llvm.org/show_bug.cgi?id=34490

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312758 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Add support for special section indexes in symbol table greater than...
Petr Hosek [Thu, 7 Sep 2017 23:02:50 +0000 (23:02 +0000)]
[llvm-objcopy] Add support for special section indexes in symbol table greater than SHN_LORESERVE

As is indexes above SHN_LORESERVE will not be handled correctly because
they'll be treated as indexes of sections rather than special values
that should just be copied. This change adds support to copy them
though.

Patch by Jake Ehrlich

Differential Revision: https://reviews.llvm.org/D37393

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312756 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMove duplicate helpers from DbgValueInst / DbgDeclareInst to DbgInfoIntrinsic
Reid Kleckner [Thu, 7 Sep 2017 22:46:24 +0000 (22:46 +0000)]
Move duplicate helpers from DbgValueInst / DbgDeclareInst to DbgInfoIntrinsic

NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312754 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-ar: exit with 1 if there is an error.
Rafael Espindola [Thu, 7 Sep 2017 22:20:38 +0000 (22:20 +0000)]
llvm-ar: exit with 1 if there is an error.

This is pr34396.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312752 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DWARF] Line 0 should not have a discriminator.
Paul Robinson [Thu, 7 Sep 2017 22:15:44 +0000 (22:15 +0000)]
[DWARF] Line 0 should not have a discriminator.
It's meaningless and takes up extra space in the line table.

Differential Revision: https://reviews.llvm.org/D37364

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312751 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix llvm-xray tests to avoid subshells
Reid Kleckner [Thu, 7 Sep 2017 21:28:09 +0000 (21:28 +0000)]
Fix llvm-xray tests to avoid subshells

We already uses pipefail to detect failure of a redirected command, so
the "|| echo failure" construct was unnecessary.

These tests run and pass on Windows now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312747 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Add ErrorSuccess and void specializations to AsyncHandlerTraits.
Lang Hames [Thu, 7 Sep 2017 21:04:00 +0000 (21:04 +0000)]
[ORC] Add ErrorSuccess and void specializations to AsyncHandlerTraits.

This will allow async handlers to be added that return void or Error::success().
Such handlers are expected to be common, since one of the primary uses of
addAsyncHandler is to run the body of the handler in a detached thread, in which
case the main handler returns immediately and does not need to provide an Error
value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312746 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[yaml2obj][ELF] Add support for symbol indexes greater than SHN_LORESERVE
Petr Hosek [Thu, 7 Sep 2017 20:44:16 +0000 (20:44 +0000)]
[yaml2obj][ELF] Add support for symbol indexes greater than SHN_LORESERVE

Right now Symbols must be either undefined or defined in a specific
section. Some symbols have section indexes like SHN_ABS however. This
change adds support for outputting symbols that have such section
indexes.

Patch by Jake Ehrlich

Differential Revision: https://reviews.llvm.org/D37391

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312745 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoCOFF: PDB: Allow multiple modules with the same name.
Peter Collingbourne [Thu, 7 Sep 2017 20:39:46 +0000 (20:39 +0000)]
COFF: PDB: Allow multiple modules with the same name.

It is possible for two modules to have the same name if they are
archive members with the same name, or if we are doing LTO (in which
case all modules will have the name "lto.tmp").

Differential Revision: https://reviews.llvm.org/D37589

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312744 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove dead code. NFCI.
Peter Collingbourne [Thu, 7 Sep 2017 19:17:30 +0000 (19:17 +0000)]
Remove dead code. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312740 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[XRay][tools] Disable windows for tests that use an unsupported shell redirect.
Keith Wyss [Thu, 7 Sep 2017 19:10:34 +0000 (19:10 +0000)]
[XRay][tools] Disable windows for tests that use an unsupported shell redirect.

The tests are filechecking against stderr and use some magic to make stdout go
away and pipe stderr to FileCheck. This broke bots on windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312739 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CUDA] Added rudimentary support for CUDA-9 and sm_70.
Artem Belevich [Thu, 7 Sep 2017 18:14:32 +0000 (18:14 +0000)]
[CUDA] Added rudimentary support for CUDA-9 and sm_70.

For now CUDA-9 is not included in the list of CUDA versions clang
searches for, so the path to CUDA-9 must be explicitly passed
via --cuda-path=.

On LLVM side NVPTX added sm_70 GPU type which bumps required
PTX version to 6.0, but otherwise is equivalent to sm_62 at the moment.

Differential Revision: https://reviews.llvm.org/D37576

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312734 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[XRay][tools] Function call stack based analysis tooling for XRay traces
Keith Wyss [Thu, 7 Sep 2017 18:07:48 +0000 (18:07 +0000)]
[XRay][tools] Function call stack based analysis tooling for XRay traces

Second try after fixing a code san problem with iterator reference types.

This change introduces a subcommand to the llvm-xray tool called
"stacks" which allows for analysing XRay traces provided as inputs and
accounting time to stacks instead of just individual functions. This
gives us a more precise view of where in a program the latency is
actually attributed.

The tool uses a trie data structure to keep track of the caller-callee
relationships as we process the XRay traces. In particular, we keep
track of the function call stack as we enter functions. While we're
doing this we're adding nodes in a trie and indicating a "calls"
relatinship between the caller (current top of the stack) and the callee
(the new top of the stack). When we push function ids onto the stack, we
keep track of the timestamp (TSC) for the enter event.

When exiting functions, we are able to account the duration by getting
the difference between the timestamp of the exit event and the
corresponding entry event in the stack. This works even if we somehow
miss the exit events for intermediary functions (i.e. if the exit event
is not cleanly associated with the enter event at the top of the stack).

The output of the tool currently provides just the top N leaf functions
that contribute the most latency, and the top N stacks that have the
most frequency. In the future we can provide more sophisticated query
mechanisms and potentially an export to database feature to make offline
analysis of the stack traces possible with existing tools.

Differential revision: D34863

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312733 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Start selecting v_mad_mix_f32
Matt Arsenault [Thu, 7 Sep 2017 18:05:07 +0000 (18:05 +0000)]
AMDGPU: Start selecting v_mad_mix_f32

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312732 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDAG: Allow creating extract_vector_elt post-legalize
Matt Arsenault [Thu, 7 Sep 2017 17:24:43 +0000 (17:24 +0000)]
DAG: Allow creating extract_vector_elt post-legalize

Fixes some combine issues for AMDGPU where we weren't
getting the many extract_vector_elt combines expected
in a future patch.

This should really be checking isOperationLegalOrCustom on
the extract. That improves a number of x86 lit tests, but
a few get stuck in an infinite loop from one place
where a similar looking extract is created. I have a
different workaround in the backend for that which
keeps many of those improvements, but also adds a few
regressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312730 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Handle non-temporal loads and stores
Konstantin Zhuravlyov [Thu, 7 Sep 2017 17:14:54 +0000 (17:14 +0000)]
AMDGPU: Handle non-temporal loads and stores

Differential Revision: https://reviews.llvm.org/D36862

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312729 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Handle more than one memory operand in SIMemoryLegalizer
Konstantin Zhuravlyov [Thu, 7 Sep 2017 16:14:21 +0000 (16:14 +0000)]
AMDGPU: Handle more than one memory operand in SIMemoryLegalizer

Differential Revision: https://reviews.llvm.org/D37397

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312725 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Remove redundant vcvt patterns.
Benjamin Kramer [Thu, 7 Sep 2017 14:52:26 +0000 (14:52 +0000)]
[ARM] Remove redundant vcvt patterns.

These don't add any value as they're just compositions of existing
patterns. However, they can confuse the cost logic in ISel, leading to
duplicated vcvt instructions like in PR33199.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312724 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][LLVM]Expanding Supports lowerInterleavedLoad() in X86InterleavedAccess (VF...
Michael Zuckerman [Thu, 7 Sep 2017 14:02:13 +0000 (14:02 +0000)]
[X86][LLVM]Expanding Supports lowerInterleavedLoad() in X86InterleavedAccess (VF{8|16|32} stride 3).

This patch expands the support of lowerInterleavedload to {8|16|32}x8i stride 3.

LLVM creates suboptimal shuffle code-gen for AVX2. In overall, this patch is a specific fix for the pattern (Strid=3 VF={8|16|32}) and we plan to include the store (deinterleved side).

The patch goal is to optimize the following sequence:
a0 b0 c0 a1 b1 c1 a2 b2
c2 a3 b3 c3 a4 b4 c4 a5
b5 c5 a6 b6 c6 a7 b7 c7

into

a0 a1 a2 a3 a4 a5 a6 a7
b0 b1 b2 b3 b4 b5 b6 b7
c0 c1 c2 c3 c4 c5 c6 c7

Reviewers
1. zvi
2. igor
3. guyblank
4. dorit
5. Ayal

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312722 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Use RegisterMCAsmBackend to register all MIPS asm backends. NFC
Simon Atanasyan [Thu, 7 Sep 2017 12:54:26 +0000 (12:54 +0000)]
[mips] Use RegisterMCAsmBackend to register all MIPS asm backends. NFC

This change converts the `MipsAsmBackend` constructor to the "standard"
form. It makes possible to use `RegisterMCAsmBackend` for the backends
registrations. Now we pass `Triple` instance to the `MipsAsmBackend`
ctor and deduce all required options like endianness and bitness from
the triple. We still need to implement explicit ABI checking for
providing correct options to backends.

Differential revision: https://reviews.llvm.org/D37519

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312720 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachineCombiner] Update instruction depths incrementally for large BBs.
Florian Hahn [Thu, 7 Sep 2017 12:49:39 +0000 (12:49 +0000)]
[MachineCombiner] Update instruction depths incrementally for large BBs.

Summary:
For large basic blocks with lots of combinable instructions, the
MachineTraceMetrics computations in MachineCombiner can dominate the compile
time, as computing the trace information is quadratic in the number of
instructions in a BB and it's relevant successors/predecessors.

In most cases, knowing the instruction depth should be enough to make
combination decisions. As we already iterate over all instructions in a basic
block, the instruction depth can be computed incrementally. This reduces the
cost of machine-combine drastically in cases where lots of instructions
are combined. The major drawback is that AFAIK, computing the critical path
length cannot be done incrementally. Therefore we only compute
instruction depths incrementally, for basic blocks with more
instructions than inc_threshold. The -machine-combiner-inc-threshold
option can be used to set the threshold and allows for easier
experimenting and checking if using incremental updates for all basic
blocks has any impact on the performance.

Reviewers: sanjoy, Gerolf, MatzeB, efriedma, fhahn

Reviewed By: fhahn

Subscribers: kiranchandramohan, javed.absar, efriedma, llvm-commits

Differential Revision: https://reviews.llvm.org/D36619

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312719 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachineTraceMetrics] Add computeDepth function (NFCI).
Florian Hahn [Thu, 7 Sep 2017 11:51:30 +0000 (11:51 +0000)]
[MachineTraceMetrics] Add computeDepth function (NFCI).

Summary:
This function is used in D36619 to update the instruction depths
incrementally.

Reviewers: efriedma, Gerolf, MatzeB, fhahn

Reviewed By: fhahn

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D36696

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312714 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Sparc][NFC] Clean up SelectCC lowering
Alex Bradbury [Thu, 7 Sep 2017 11:30:55 +0000 (11:30 +0000)]
[Sparc][NFC] Clean up SelectCC lowering
The ARM, BPF, MSP430, Sparc and Mips backends all use a similar code sequence
for lowering SelectCC. As pointed out by @reames in D29937, this code isn't
particularly clear and in most of these backends doesn't actually match the
comments. This patch makes the code sequence clearer for the Sparc backend
through better variable naming and more accurate comments (e.g. we are
inserting triangle control flow, _not_ diamond). There is no functional
change.

Differential Revision: https://reviews.llvm.org/D37194

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312713 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFixing incorrectly capitalised regexps.
Benjamin Kramer [Thu, 7 Sep 2017 09:54:03 +0000 (09:54 +0000)]
Fixing incorrectly capitalised regexps.

Patch by Sam Allen!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312709 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[RegAlloc] Make sure live-ranges reflect the state of the IR when removing...
Jonas Paulsson [Thu, 7 Sep 2017 09:13:17 +0000 (09:13 +0000)]
Revert "[RegAlloc] Make sure live-ranges reflect the state of the IR when removing them"

This temporarily reverts commit 463fa38 (r311401).

See https://bugs.llvm.org/show_bug.cgi?id=34502

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312708 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] Update to cmov promotion tests for D36711; NFC
Alexander Ivchenko [Thu, 7 Sep 2017 08:59:05 +0000 (08:59 +0000)]
[x86] Update to cmov promotion tests for D36711; NFC

Adding i8 -> [i16, i32, i64] and i32 -> i64 cases.
This way we can see what the current codegen looks like.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312707 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoX86: Improve AVX512 fptoui lowering
Zvi Rackover [Thu, 7 Sep 2017 07:40:34 +0000 (07:40 +0000)]
X86: Improve AVX512 fptoui lowering

Summary:
Add patterns for
  fptoui <16 x float> to <16 x i8>
  fptoui <16 x float> to <16 x i16>

Reviewers: igorb, delena, craig.topper

Reviewed By: craig.topper

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37505

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312704 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Force shuffle lowering to only create X86ISD::VPERM2X128 with 64-bit element...
Craig Topper [Thu, 7 Sep 2017 06:11:10 +0000 (06:11 +0000)]
[X86] Force shuffle lowering to only create X86ISD::VPERM2X128 with 64-bit element types so we can remove some patterns from isel.

Intrinsic handling is still creating these nodes with 32-bit elements as well. But at least this gets rid of 8 and 16.

Ideally, someday we'll convert the intrinsics to generic vector shuffles and remove the intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312702 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Don't legalize i16 extloads to i32 with legal i16
Matt Arsenault [Thu, 7 Sep 2017 05:37:34 +0000 (05:37 +0000)]
AMDGPU: Don't legalize i16 extloads to i32 with legal i16

Keeping non-i16 extloads makes it easier to match some new
gfx9 load instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312699 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoModuleSummaryAnalysis: Correctly handle all function operand references.
Peter Collingbourne [Thu, 7 Sep 2017 05:35:35 +0000 (05:35 +0000)]
ModuleSummaryAnalysis: Correctly handle all function operand references.

The current code that handles personality functions when creating a
module summary does not correctly handle the case where a function's
personality function operand refers to the function indirectly
(e.g. via a bitcast). This patch handles such cases by treating
personality function references like any other reference, i.e. by
adding them to the function's reference list. This has the minor side
benefit of allowing personality functions to participate in early
dead stripping.

We do this by calling findRefEdges on the function itself. This way
we also end up handling other function operands (specifically prefix
data and prologue data) for free.

Differential Revision: https://reviews.llvm.org/D37553

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312698 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove patterns for selecting a v8f32 X86ISD::MOVSS or v4f64 X86ISD::MOVSD.
Craig Topper [Thu, 7 Sep 2017 05:08:16 +0000 (05:08 +0000)]
[X86] Remove patterns for selecting a v8f32 X86ISD::MOVSS or v4f64 X86ISD::MOVSD.

I don't think we ever generate these. If we did, I would expect we would also be able to generate v16f32 and v8f64, but we don't have those patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312694 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoARM: track globals promoted to coalesced const pool entries
Saleem Abdulrasool [Thu, 7 Sep 2017 04:00:13 +0000 (04:00 +0000)]
ARM: track globals promoted to coalesced const pool entries

Globals that are promoted to an ARM constant pool may alias with another
existing constant pool entry. We need to keep a reference to all globals
that were promoted to each constant pool value so that we can emit a
distinct label for each promoted global. These labels are necessary so
that debug info can refer to the promoted global without an undefined
reference during linking.

Patch by Stephen Crane!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312692 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoObject: Downgrade invalid weak externals from an assert fail to an llvm::Error when...
Peter Collingbourne [Thu, 7 Sep 2017 01:33:52 +0000 (01:33 +0000)]
Object: Downgrade invalid weak externals from an assert fail to an llvm::Error when creating an irsymtab.

This fixes bitcode emission for modules containing invalid weak externals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312686 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoInstSimplify: canonicalize is idempotent
Matt Arsenault [Thu, 7 Sep 2017 01:21:43 +0000 (01:21 +0000)]
InstSimplify: canonicalize is idempotent

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312685 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoLTO: Remove unnecessary Windows support code.
Peter Collingbourne [Thu, 7 Sep 2017 00:55:00 +0000 (00:55 +0000)]
LTO: Remove unnecessary Windows support code.

I empirically verified that open files can in fact be renamed on
Windows with sys::fs::rename, so remove the incorrect code and comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312683 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReland "[llvm-objcopy] Add support for relocations"
Petr Hosek [Wed, 6 Sep 2017 23:41:02 +0000 (23:41 +0000)]
Reland "[llvm-objcopy] Add support for relocations"

This change adds support for SHT_REL and SHT_RELA sections in
llvm-objcopy.

Patch by Jake Ehrlich

Differential Revision: https://reviews.llvm.org/D36554

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312680 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Pass] Fix some Clang-tidy modernize and Include What You Use warnings; other minor...
Eugene Zelenko [Wed, 6 Sep 2017 23:05:38 +0000 (23:05 +0000)]
[Pass] Fix some Clang-tidy modernize and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312679 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUtil: Improve update_llc_test_checks to scrub macosx-style assembly annotations
Zvi Rackover [Wed, 6 Sep 2017 23:04:28 +0000 (23:04 +0000)]
Util: Improve update_llc_test_checks to scrub macosx-style assembly annotations

Summary:
In D37523 Sanjay pointed out that the tool does not scrub macosx-style 'End of Function' annotations,
where the comments begin with a double-#.

I tested this patch by verifying all existing occurences of 'End function' are scrubbed:
find ./test/CodeGen/X86 -name '*.ll' | xargs grep -l "End function" | xargs utils/update_llc_test_checks.py --llc-binary build/bin/llc

Reviewers: spatel, chandlerc, craig.topper

Reviewed By: spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37532

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312678 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Use v_pk_max_f16 for fcanonicalize
Stanislav Mekhanoshin [Wed, 6 Sep 2017 22:27:29 +0000 (22:27 +0000)]
[AMDGPU] Use v_pk_max_f16 for fcanonicalize

Differential Revision: https://reviews.llvm.org/D37325

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312676 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Only treat imports/exports as symbols when reading relocatable object...
Sam Clegg [Wed, 6 Sep 2017 22:05:41 +0000 (22:05 +0000)]
[WebAssembly] Only treat imports/exports as symbols when reading relocatable object files

This change only treats imported and exports functions and globals
as symbol table entries the object has a "linking" section (i.e. it is
relocatable object file).

In this case all globals must be of type I32 and initialized with
i32.const.  This was previously being assumed but not checked for and
was causing a failure on big endian machines due to using the wrong
value of then union.

See: https://bugs.llvm.org/show_bug.cgi?id=34487

Differential Revision: https://reviews.llvm.org/D37497

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312674 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemoves redundant `llvm::`, add comments and simplify a return type of a function.
Rui Ueyama [Wed, 6 Sep 2017 22:05:32 +0000 (22:05 +0000)]
Removes redundant `llvm::`, add comments and simplify a return type of a function.

No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312673 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoInsert IMPLICIT_DEFS for undef uses in tail merging
Matthias Braun [Wed, 6 Sep 2017 20:45:24 +0000 (20:45 +0000)]
Insert IMPLICIT_DEFS for undef uses in tail merging

Tail merging can convert an undef use into a normal one when creating a
common tail. Doing so can make the register live out from a block which
previously contained the undef use. To keep the liveness up-to-date,
insert IMPLICIT_DEFs in such blocks when necessary.

To enable this patch the computeLiveIns() function which used to
compute live-ins for a block and set them immediately is split into new
functions:
- computeLiveIns() just computes the live-ins in a LivePhysRegs set.
- addLiveIns() applies the live-ins to a block live-in list.
- computeAndAddLiveIns() is a convenience function combining the other
  two functions and behaving like computeLiveIns() before this patch.

Based on a patch by Krzysztof Parzyszek <kparzysz@codeaurora.org>

Differential Revision: https://reviews.llvm.org/D37034

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312668 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[docs] Add a note on iteration of unordered containers to coding standards
Mandeep Singh Grang [Wed, 6 Sep 2017 20:19:10 +0000 (20:19 +0000)]
[docs] Add a note on iteration of unordered containers to coding standards

Summary: Beware of non-determinism due to ordering of pointers

Reviewers: dblaikie, dexonsmith

Reviewed By: dblaikie

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37525

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312667 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDisable jump threading into loop headers
Krzysztof Parzyszek [Wed, 6 Sep 2017 19:36:58 +0000 (19:36 +0000)]
Disable jump threading into loop headers

Consider this type of a loop:
    for (...) {
      ...
      if (...) continue;
      ...
    }
Normally, the "continue" would branch to the loop control code that
checks whether the loop should continue iterating and which contains
the (often) unique loop latch branch. In certain cases jump threading
can "thread" the inner branch directly to the loop header, creating
a second loop latch. Loop canonicalization would then transform this
loop into a loop nest. The problem with this is that in such a loop
nest neither loop is countable even if the original loop was. This
may inhibit subsequent loop optimizations and be detrimental to
performance.

Differential Revision: https://reviews.llvm.org/D36404

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312664 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] fix triple and regenerate checks for psubus; NFC
Sanjay Patel [Wed, 6 Sep 2017 19:05:20 +0000 (19:05 +0000)]
[x86] fix triple and regenerate checks for psubus; NFC

Patch by Yulia Koval!

Differential Revision: https://reviews.llvm.org/D37523

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312662 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Move more isel patterns to X86InstrVecCompiler.td. NFC
Craig Topper [Wed, 6 Sep 2017 19:03:55 +0000 (19:03 +0000)]
[X86] Move more isel patterns to X86InstrVecCompiler.td. NFC

This moves more of our subvector insert/extract tricks to X86InstrVecCompiler.td and refactors them into multiclasses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312661 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Fixed encoding of v_pk_mul_f16 in fcanonicalize
Stanislav Mekhanoshin [Wed, 6 Sep 2017 18:29:51 +0000 (18:29 +0000)]
[AMDGPU] Fixed encoding of v_pk_mul_f16 in fcanonicalize

Differential Revision: https://reviews.llvm.org/D37522

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312660 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IfConversion] Remove kill flags from common instructions as well
Krzysztof Parzyszek [Wed, 6 Sep 2017 17:57:13 +0000 (17:57 +0000)]
[IfConversion] Remove kill flags from common instructions as well

When if-converting a diamond, two separate blocks will be placed back
to back to form a straight line code. To ensure correctness of the
liveness information, any registers that are live in the second block
should not be killed in the first block, even if they were in the
original code.
Additionally, when the two blocks share common instructions at the
beginning, these instructions will not be duplicated, but only placed
once, before both of the blocks. Since the function "isIdenticalTo"
(as used here) ignores kill flags, the common initial code in one
block may have a kill flag for a register that is live in the other
block.
Because the code that removes kill flags only runs for the non-common
parts of the predicated blocks, a kill flag mismatch in the common
code could still lead to a live register being killed prematurely.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312654 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Actually add the new file that was supposed to go with r312649.
Craig Topper [Wed, 6 Sep 2017 17:06:40 +0000 (17:06 +0000)]
[X86] Actually add the new file that was supposed to go with r312649.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312650 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Introduce a new td file to hold patterns some of the non instruction patterns...
Craig Topper [Wed, 6 Sep 2017 16:56:52 +0000 (16:56 +0000)]
[X86] Introduce a new td file to hold patterns some of the non instruction patterns from SSE and AVX512

This patch moves some of similar non-instruction patterns from X86InstrSSE.td and X86InstrAVX512.td to a common file.

This is intended as a starting point. There are many other optimization patterns that exist in both files that we could move here.

Differential Revision: https://reviews.llvm.org/D37455

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312649 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix PR33878: BasicAA incorrectly assumes different address spaces don't alias
Nuno Lopes [Wed, 6 Sep 2017 16:55:31 +0000 (16:55 +0000)]
Fix PR33878: BasicAA incorrectly assumes different address spaces don't alias
Remove code that assumed that a nullptr of address space != 0 couldnt alias with a non-null pointer. This is incorrect, since nothing can be concluded about a null pointer in an address space != 0.
This code was written before address spaces were introduced

Differential Revision: https://reviews.llvm.org/D37518

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312648 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMinor style fixes in lib/Support/**/Program.(inc|cpp).
Alexander Kornienko [Wed, 6 Sep 2017 16:28:33 +0000 (16:28 +0000)]
Minor style fixes in lib/Support/**/Program.(inc|cpp).

No functional changes intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312646 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[llvm-objcopy] Add support for relocations"
Petr Hosek [Wed, 6 Sep 2017 16:23:15 +0000 (16:23 +0000)]
Revert "[llvm-objcopy] Add support for relocations"

This reverts r312643 because it's failing on llvm-i686-linux-RA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312645 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Add option to generate calls to "abort" for "unreachable"
Krzysztof Parzyszek [Wed, 6 Sep 2017 16:22:55 +0000 (16:22 +0000)]
[Hexagon] Add option to generate calls to "abort" for "unreachable"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312644 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Add support for relocations
Petr Hosek [Wed, 6 Sep 2017 16:19:48 +0000 (16:19 +0000)]
[llvm-objcopy] Add support for relocations

This change adds support for SHT_REL and SHT_RELA sections in
llvm-objcopy.

Patch by Jake Ehrlich

Differential Revision: https://reviews.llvm.org/D36554

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312643 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TailCall] Allow llvm.memcpy/memset/memmove to be tail calls when parent
Wei Mi [Wed, 6 Sep 2017 16:05:17 +0000 (16:05 +0000)]
[TailCall] Allow llvm.memcpy/memset/memmove to be tail calls when parent
function return the intrinsics's first argument.

llvm.memcpy/memset/memmove return void but they will return the first
argument after they are expanded as libcalls. Now if the parent function
has any return value, llvm.memcpy cannot be turned into tail call after
expansion.

The patch is to handle that case in SelectionDAGBuilder so when caller
function return the same value as the first argument of llvm.memcpy,
tail call is allowed.

Differential Revision: https://reviews.llvm.org/D37406

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312641 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Fix shouldClusterMemOps to process flat loads
Stanislav Mekhanoshin [Wed, 6 Sep 2017 15:31:30 +0000 (15:31 +0000)]
[AMDGPU] Fix shouldClusterMemOps to process flat loads

Flat loads do not have vdata operand but have vdst instead.

Differential Revision: https://reviews.llvm.org/D37502

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312640 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Make worst-case assumption about the wait states in inline assembly
Nicolai Haehnle [Wed, 6 Sep 2017 13:50:13 +0000 (13:50 +0000)]
AMDGPU: Make worst-case assumption about the wait states in inline assembly

Summary:
Mesa still uses a hack where empty inline assembly is used as a kind of
optimization barrier. This exposed a problem where not enough wait states
were inserted, because the hazard recognizer implicitly assumed that each
inline assembly "instruction" has at least one wait state.

Reviewers: arsenm

Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye

Differential Revision: https://reviews.llvm.org/D37205

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312635 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][X87] Ensure x87 instructions are tagged as altering the FPSW reg
Simon Pilgrim [Wed, 6 Sep 2017 10:23:12 +0000 (10:23 +0000)]
[X86][X87] Ensure x87 instructions are tagged as altering the FPSW reg

As noted in PR34080, a lot of x87 instructions alter the FPSW status register (or leave it in an undefined state) but aren't tagged as such in the tablegen.

This patch tags the control word, stack, wait and math instructions as altering FPSW, which matches what the AMD APMs suggests happens.

Differential Revision: https://reviews.llvm.org/D36414

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312629 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV][NFC] Fix sorting of includes in lib/Target/RISCV
Alex Bradbury [Wed, 6 Sep 2017 09:21:21 +0000 (09:21 +0000)]
[RISCV][NFC] Fix sorting of includes in lib/Target/RISCV

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312624 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombiner] When combining EXTRACT_SUBVECTOR of a BUILD_VECTOR, make sure we don...
Craig Topper [Wed, 6 Sep 2017 06:50:03 +0000 (06:50 +0000)]
[DAGCombiner] When combining EXTRACT_SUBVECTOR of a BUILD_VECTOR, make sure we don't create a BUILD_VECTOR with an illegal type after type legalization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312621 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] Fix PR34377 by disabling cmov conversion when we relied on it
Chandler Carruth [Wed, 6 Sep 2017 06:28:08 +0000 (06:28 +0000)]
[x86] Fix PR34377 by disabling cmov conversion when we relied on it
performing a zext of a register.

On the PR there is discussion of how to more effectively handle this,
but this patch prevents us from miscompiling code.

Differential Revision: https://reviews.llvm.org/D37504

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312620 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoX86 Tests: Tidy up AVX512 conversion tests. NFC.
Zvi Rackover [Wed, 6 Sep 2017 05:33:04 +0000 (05:33 +0000)]
X86 Tests: Tidy up AVX512 conversion tests. NFC.

Rename functions to a consistent format to make it easier to track coverage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312619 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUpdating a test reference for rL312608.
Jatin Bhateja [Wed, 6 Sep 2017 03:58:14 +0000 (03:58 +0000)]
Updating a test reference for  rL312608.

Differential Revision: https://reviews.llvm.org/D37501

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312614 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add more FMA3 patterns to cover a load in all 3 possible positions.
Craig Topper [Wed, 6 Sep 2017 03:35:58 +0000 (03:35 +0000)]
[X86] Add more FMA3 patterns to cover a load in all 3 possible positions.

This matches what we already do for AVX512. The peephole pass makes up for this in most if not all cases. But this makes isel behavior for these consistent with every other instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312613 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] Don't use xscvdpspn on the P7
Hal Finkel [Wed, 6 Sep 2017 03:08:26 +0000 (03:08 +0000)]
[PowerPC] Don't use xscvdpspn on the P7

xscvdpspn was not introduced until the P8, so don't use it on the P7. Fixes a
regression introduced in r288152.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312612 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Allow cross-lane permutations for sub targets supporting AVX2.
Jatin Bhateja [Wed, 6 Sep 2017 02:58:47 +0000 (02:58 +0000)]
[X86] Allow cross-lane permutations for sub targets supporting AVX2.

Summary:
Most instructions in AVX work “in-lane”, that is, each source element is applied only to other
elements of the same lane, thus a cross lane permutation is costly and needs more than one instrution.
AVX2 includes instructions to perform any-to-any permutation of words over a 256-bit register
and vectorized table lookup.

This should also Fix PR34369

Differential Revision: https://reviews.llvm.org/D37388

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312608 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Fix some comments in JITSymbol.
Lang Hames [Wed, 6 Sep 2017 02:53:37 +0000 (02:53 +0000)]
[ORC] Fix some comments in JITSymbol.

Patch by Breckin Loggins. Thanks Breckin!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312607 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix crbug 759265 by suppressing llvm mt warnings.
Eric Beckmann [Wed, 6 Sep 2017 01:50:36 +0000 (01:50 +0000)]
Fix crbug 759265 by suppressing llvm mt warnings.

Summary:
Previous would throw warning whenever libxml2 is not installed.  Now
only give this warning if merging manifest fails.

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37240

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312604 91177308-0d34-0410-b5e6-96231b3b80d8