Merge clang, llvm, lld, lldb, compiler-rt and libc++ 5.0.0 release.

MFC r309126 (by emaste):

  Correct lld llvm-tblgen dependency file name

MFC r309169:

  Get rid of separate Subversion mergeinfo properties for llvm-dwarfdump
  and llvm-lto.  The mergeinfo confuses Subversion enormously, and these
  directories will just use the mergeinfo for llvm itself.

MFC r312765:

  Pull in r276136 from upstream llvm trunk (by Wei Mi):

    Use ValueOffsetPair to enhance value reuse during SCEV expansion.

    In D12090, the ExprValueMap was added to reuse existing value during
    SCEV expansion. However, const folding and sext/zext distribution can
    make the reuse still difficult.

    A simplified case is: suppose we know S1 expands to V1 in
    ExprValueMap, and
      S1 = S2 + C_a
      S3 = S2 + C_b
    where C_a and C_b are different SCEVConstants. Then we'd like to
    expand S3 as V1 - C_a + C_b instead of expanding S2 literally. It is
    helpful when S2 is a complex SCEV expr and S2 has no entry in
    ExprValueMap, which is usually caused by the fact that S3 is
    generated from S1 after const folding.

    In order to do that, we represent ExprValueMap as a mapping from SCEV
    to ValueOffsetPair. We will save both S1->{V1, 0} and S2->{V1, C_a}
    into the ExprValueMap when we create SCEV for V1. When S3 is
    expanded, it will first expand S2 to V1 - C_a because of S2->{V1,
    C_a} in the map, then expand S3 to V1 - C_a + C_b.

    Differential Revision: https://reviews.llvm.org/D21313

  This should fix assertion failures when building OpenCV >= 3.1.

  PR:		215649

MFC r312831:

  Revert r312765 for now, since it causes assertions when building
  lang/spidermonkey24.

  Reported by:	antoine
  PR:		215649

MFC r316511 (by jhb):

  Add an implementation of __ffssi2() derived from __ffsdi2().

  Newer versions of GCC include an __ffssi2() symbol in libgcc and the
  compiler can emit calls to it in generated code.  This is true for at
  least GCC 6.2 when compiling world for mips and mips64.

  Reviewed by:	jmallett, dim
  Sponsored by:	DARPA / AFRL
  Differential Revision:	https://reviews.freebsd.org/D10086

MFC r318601 (by adrian):

  [libcompiler-rt] add bswapdi2/bswapsi2

  This is required for mips gcc 6.3 userland to build/run.

  Reviewed by:	emaste, dim
  Approved by:	emaste
  Differential Revision:	https://reviews.freebsd.org/D10838

MFC r318884 (by emaste):

  lldb: map TRAP_CAP to a trace trap

  In the absense of a more specific handler for TRAP_CAP (generated by
  ENOTCAPABLE or ECAPMODE while in capability mode) treat it as a trace
  trap.

  Example usage (testing the bug in PR219173):

  % proccontrol -m trapcap lldb usr.bin/hexdump/obj/hexdump -- -Cv -s 1 /bin/ls
  ...
  (lldb) run
  Process 12980 launching
  Process 12980 launched: '.../usr.bin/hexdump/obj/hexdump' (x86_64)
  Process 12980 stopped
  * thread #1, stop reason = trace
      frame #0: 0x0000004b80c65f1a libc.so.7`__sys_lseek + 10
  ...

  In the future we should have LLDB control the trapcap procctl itself
  (as it does with ASLR), as well as report a specific stop reason.
  This change eliminates an assertion failure from LLDB for now.

MFC r319796:

  Remove a few unneeded files from libllvm, libclang and liblldb.

MFC r319885 (by emaste):

  lld: ELF: Fix ICF crash on absolute symbol relocations.

  If two sections contained relocations to absolute symbols with the same
  value we would crash when trying to access their sections. Add a check that
  both symbols point to sections before accessing their sections, and treat
  absolute symbols as equal if their values are equal.

  Obtained from:	LLD commit r292578

MFC r319918:

  Revert r319796 for now, it can cause undefined references when linking
  in some circumstances.

  Reported by:	Shawn Webb <shawn.webb@hardenedbsd.org>

MFC r319957 (by emaste):

  lld: Add armelf emulation mode

  Obtained from:	LLD r305375

MFC r321369:

  Upgrade our copies of clang, llvm, lld, lldb, compiler-rt and libc++ to
  5.0.0 (trunk r308421).  Upstream has branched for the 5.0.0 release,
  which should be in about a month.  Please report bugs and regressions,
  so we can get them into the release.

  Please note that from 3.5.0 onwards, clang, llvm and lldb require C++11
  support to build; see UPDATING for more information.

MFC r321420:

  Add a few more object files to liblldb, which should solve errors when
  linking the lldb executable in some cases.  In particular, when the
  -ffunction-sections -fdata-sections options are turned off, or
  ineffective.

  Reported by:	Shawn Webb, Mark Millard

MFC r321433:

  Cleanup stale Options.inc files from the previous libllvm build for
  clang 4.0.0.  Otherwise, these can get included before the two newly
  generated ones (which are different) for clang 5.0.0.

  Reported by:	Mark Millard

MFC r321439 (by bdrewery):

  Move llvm Options.inc hack from r321433 for NO_CLEAN to lib/clang/libllvm.

  The files are only ever generated to .OBJDIR, not to WORLDTMP (as a
  sysroot) and are only ever included from a compilation.  So using
  a beforebuild target here removes the file before the compilation
  tries to include it.

MFC r321664:

  Pull in r308891 from upstream llvm trunk (by Benjamin Kramer):

    [CodeGenPrepare] Cut off FindAllMemoryUses if there are too many uses.

    This avoids excessive compile time. The case I'm looking at is
    Function.cpp from an old version of LLVM that still had the giant
    memcmp string matcher in it. Before r308322 this compiled in about 2
    minutes, after it, clang takes infinite* time to compile it. With
    this patch we're at 5 min, which is still bad but this is a
    pathological case.

    The cut off at 20 uses was chosen by looking at other cut-offs in LLVM
    for user scanning. It's probably too high, but does the job and is
    very unlikely to regress anything.

    Fixes PR33900.

    * I'm impatient and aborted after 15 minutes, on the bug report it was
      killed after 2h.

  Pull in r308986 from upstream llvm trunk (by Simon Pilgrim):

    [X86][CGP] Reduce memcmp() expansion to 2 load pairs (PR33914)

    D35067/rL308322 attempted to support up to 4 load pairs for memcmp
    inlining which resulted in regressions for some optimized libc memcmp
    implementations (PR33914).

    Until we can match these more optimal cases, this patch reduces the
    memcmp expansion to a maximum of 2 load pairs (which matches what we
    do for -Os).

    This patch should be considered for the 5.0.0 release branch as well

    Differential Revision: https://reviews.llvm.org/D35830

  These fix a hang (or extremely long compile time) when building older
  LLVM ports.

  Reported by:    antoine
  PR:             219139

MFC r321719:

  Pull in r309503 from upstream clang trunk (by Richard Smith):

    PR33902: Invalidate line number cache when adding more text to
    existing buffer.

    This led to crashes as the line number cache would report a bogus
    line number for a line of code, and we'd try to find a nonexistent
    column within the line when printing diagnostics.

  This fixes an assertion when building the graphics/champlain port.

  Reported by:	antoine, kwm
  PR:		219139

MFC r321723:

  Upgrade our copies of clang, llvm, lld and lldb to r309439 from the
  upstream release_50 branch.  This is just after upstream's 5.0.0-rc1.

MFC r322320:

  Upgrade our copies of clang, llvm and libc++ to r310316 from the
  upstream release_50 branch.

MFC r322326 (by emaste):

  lldb: Make i386-*-freebsd expression work on JIT path

  * Enable i386 ABI creation for freebsd
  * Added an extra argument in ABISysV_i386::PrepareTrivialCall for mmap
    syscall
  * Unlike linux, the last argument of mmap is actually 64-bit(off_t).
    This requires us to push an additional word for the higher order bits.
  * Prior to this change, ktrace dump will show mmap failures due to
    invalid argument coming from the 6th mmap argument.

  Submitted by:	Karnajit Wangkhem
  Differential Revision:	https://reviews.llvm.org/D34776

MFC r322360 (by emaste):

  lldb: Report inferior signals as signals, not exceptions, on FreeBSD

  This is the FreeBSD equivalent of LLVM r238549.

  This serves 2 purposes:

  * LLDB should handle inferior process signals SIGSEGV/SIGILL/SIGBUS/
    SIGFPE the way it is suppose to be handled. Prior to this fix these
    signals will neither create a coredump, nor exit from the debugger
    or work for signal handling scenario.
  * eInvalidCrashReason need not report "unknown crash reason" if we have
    a valid si_signo

  llvm.org/pr23699

  Patch by Karnajit Wangkhem

  Differential Revision:  https://reviews.llvm.org/D35223

  Submitted by:	Karnajit Wangkhem
  Obtained from:	LLVM r310591

MFC r322474 (by emaste):

  lld: Add `-z muldefs` option.

  Obtained from:	LLVM r310757

MFC r322740:

  Upgrade our copies of clang, llvm, lld and libc++ to r311219 from the
  upstream release_50 branch.

MFC r322855:

  Upgrade our copies of clang, llvm, lldb and compiler-rt to r311606 from
  the upstream release_50 branch.

  As of this version, lib/msun's trig test should also work correctly
  again (see bug 220989 for more information).

  PR:		220989

MFC r323112:

  Upgrade our copies of clang, llvm, lldb and compiler-rt to r312293 from
  the upstream release_50 branch.  This corresponds to 5.0.0 rc4.

  As of this version, the cad/stepcode port should now compile in a more
  reasonable time on i386 (see bug 221836 for more information).

  PR:		221836

MFC r323245:

  Upgrade our copies of clang, llvm, lld, lldb, compiler-rt and libc++ to
  5.0.0 release (upstream r312559).

  Release notes for llvm, clang and lld will be available here soon:
  <http://releases.llvm.org/5.0.0/docs/ReleaseNotes.html>
  <http://releases.llvm.org/5.0.0/tools/clang/docs/ReleaseNotes.html>
  <http://releases.llvm.org/5.0.0/tools/lld/docs/ReleaseNotes.html>

  Relnotes:	yes

(cherry picked from commit 12cd91cf4c6b96a24427c0de5374916f2808d263)

1 files changed, 225 insertions, 164 deletions

diff --git a/contrib/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/contrib/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index baa28de..625c9b7 100644
--- a/contrib/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/contrib/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -14,12 +14,50 @@
 
 #include "AMDGPU.h"
 #include "AMDGPUSubtarget.h"
+#include "Utils/AMDGPUBaseInfo.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/ADT/None.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/Analysis/CaptureTracking.h"
 #include "llvm/Analysis/ValueTracking.h"
+#include "llvm/CodeGen/TargetPassConfig.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Constant.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
-#include "llvm/IR/MDBuilder.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Metadata.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
+#include "llvm/IR/User.h"
+#include "llvm/IR/Value.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
+#include <algorithm>
+#include <cassert>
+#include <cstdint>
+#include <map>
+#include <tuple>
+#include <utility>
+#include <vector>
 
 #define DEBUG_TYPE "amdgpu-promote-alloca"
 
@@ -31,16 +69,16 @@ namespace {
 class AMDGPUPromoteAlloca : public FunctionPass {
 private:
   const TargetMachine *TM;
-  Module *Mod;
-  const DataLayout *DL;
-  MDNode *MaxWorkGroupSizeRange;
+  Module *Mod = nullptr;
+  const DataLayout *DL = nullptr;
+  AMDGPUAS AS;
 
   // FIXME: This should be per-kernel.
-  uint32_t LocalMemLimit;
-  uint32_t CurrentLocalMemUsage;
+  uint32_t LocalMemLimit = 0;
+  uint32_t CurrentLocalMemUsage = 0;
 
-  bool IsAMDGCN;
-  bool IsAMDHSA;
+  bool IsAMDGCN = false;
+  bool IsAMDHSA = false;
 
   std::pair<Value *, Value *> getLocalSizeYZ(IRBuilder<> &Builder);
   Value *getWorkitemID(IRBuilder<> &Builder, unsigned N);
@@ -59,26 +97,20 @@ private:
                                        Instruction *UseInst,
                                        int OpIdx0, int OpIdx1) const;
 
+  /// Check whether we have enough local memory for promotion.
+  bool hasSufficientLocalMem(const Function &F);
+
 public:
   static char ID;
 
-  AMDGPUPromoteAlloca(const TargetMachine *TM_ = nullptr) :
-    FunctionPass(ID),
-    TM(TM_),
-    Mod(nullptr),
-    DL(nullptr),
-    MaxWorkGroupSizeRange(nullptr),
-    LocalMemLimit(0),
-    CurrentLocalMemUsage(0),
-    IsAMDGCN(false),
-    IsAMDHSA(false) { }
+  AMDGPUPromoteAlloca() : FunctionPass(ID) {}
 
   bool doInitialization(Module &M) override;
   bool runOnFunction(Function &F) override;
 
   StringRef getPassName() const override { return "AMDGPU Promote Alloca"; }
 
-  void handleAlloca(AllocaInst &I);
+  bool handleAlloca(AllocaInst &I, bool SufficientLDS);
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesCFG();
@@ -86,146 +118,60 @@ public:
   }
 };
 
-} // End anonymous namespace
+} // end anonymous namespace
 
 char AMDGPUPromoteAlloca::ID = 0;
 
-INITIALIZE_TM_PASS(AMDGPUPromoteAlloca, DEBUG_TYPE,
-                   "AMDGPU promote alloca to vector or LDS", false, false)
+INITIALIZE_PASS(AMDGPUPromoteAlloca, DEBUG_TYPE,
+                "AMDGPU promote alloca to vector or LDS", false, false)
 
 char &llvm::AMDGPUPromoteAllocaID = AMDGPUPromoteAlloca::ID;
 
-
 bool AMDGPUPromoteAlloca::doInitialization(Module &M) {
-  if (!TM)
-    return false;
-
   Mod = &M;
   DL = &Mod->getDataLayout();
 
-  // The maximum workitem id.
-  //
-  // FIXME: Should get as subtarget property. Usually runtime enforced max is
-  // 256.
-  MDBuilder MDB(Mod->getContext());
-  MaxWorkGroupSizeRange = MDB.createRange(APInt(32, 0), APInt(32, 2048));
-
-  const Triple &TT = TM->getTargetTriple();
-
-  IsAMDGCN = TT.getArch() == Triple::amdgcn;
-  IsAMDHSA = TT.getOS() == Triple::AMDHSA;
-
   return false;
 }
 
 bool AMDGPUPromoteAlloca::runOnFunction(Function &F) {
-  if (!TM || skipFunction(F))
+  if (skipFunction(F))
     return false;
 
-  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(F);
-  if (!ST.isPromoteAllocaEnabled())
-    return false;
-
-  FunctionType *FTy = F.getFunctionType();
-
-  // If the function has any arguments in the local address space, then it's
-  // possible these arguments require the entire local memory space, so
-  // we cannot use local memory in the pass.
-  for (Type *ParamTy : FTy->params()) {
-    PointerType *PtrTy = dyn_cast<PointerType>(ParamTy);
-    if (PtrTy && PtrTy->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS) {
-      LocalMemLimit = 0;
-      DEBUG(dbgs() << "Function has local memory argument. Promoting to "
-                      "local memory disabled.\n");
-      return false;
-    }
-  }
-
-  LocalMemLimit = ST.getLocalMemorySize();
-  if (LocalMemLimit == 0)
+  if (auto *TPC = getAnalysisIfAvailable<TargetPassConfig>())
+    TM = &TPC->getTM<TargetMachine>();
+  else
     return false;
 
-  const DataLayout &DL = Mod->getDataLayout();
-
-  // Check how much local memory is being used by global objects
-  CurrentLocalMemUsage = 0;
-  for (GlobalVariable &GV : Mod->globals()) {
-    if (GV.getType()->getAddressSpace() != AMDGPUAS::LOCAL_ADDRESS)
-      continue;
-
-    for (const User *U : GV.users()) {
-      const Instruction *Use = dyn_cast<Instruction>(U);
-      if (!Use)
-        continue;
-
-      if (Use->getParent()->getParent() == &F) {
-        unsigned Align = GV.getAlignment();
-        if (Align == 0)
-          Align = DL.getABITypeAlignment(GV.getValueType());
-
-        // FIXME: Try to account for padding here. The padding is currently
-        // determined from the inverse order of uses in the function. I'm not
-        // sure if the use list order is in any way connected to this, so the
-        // total reported size is likely incorrect.
-        uint64_t AllocSize = DL.getTypeAllocSize(GV.getValueType());
-        CurrentLocalMemUsage = alignTo(CurrentLocalMemUsage, Align);
-        CurrentLocalMemUsage += AllocSize;
-        break;
-      }
-    }
-  }
-
-  unsigned MaxOccupancy = ST.getOccupancyWithLocalMemSize(CurrentLocalMemUsage);
-
-  // Restrict local memory usage so that we don't drastically reduce occupancy,
-  // unless it is already significantly reduced.
-
-  // TODO: Have some sort of hint or other heuristics to guess occupancy based
-  // on other factors..
-  unsigned OccupancyHint = ST.getWavesPerEU(F).second;
-  if (OccupancyHint == 0)
-    OccupancyHint = 7;
-
-  // Clamp to max value.
-  OccupancyHint = std::min(OccupancyHint, ST.getMaxWavesPerEU());
-
-  // Check the hint but ignore it if it's obviously wrong from the existing LDS
-  // usage.
-  MaxOccupancy = std::min(OccupancyHint, MaxOccupancy);
-
-
-  // Round up to the next tier of usage.
-  unsigned MaxSizeWithWaveCount
-    = ST.getMaxLocalMemSizeWithWaveCount(MaxOccupancy);
+  const Triple &TT = TM->getTargetTriple();
+  IsAMDGCN = TT.getArch() == Triple::amdgcn;
+  IsAMDHSA = TT.getOS() == Triple::AMDHSA;
 
-  // Program is possibly broken by using more local mem than available.
-  if (CurrentLocalMemUsage > MaxSizeWithWaveCount)
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(F);
+  if (!ST.isPromoteAllocaEnabled())
     return false;
 
-  LocalMemLimit = MaxSizeWithWaveCount;
-
-  DEBUG(
-    dbgs() << F.getName() << " uses " << CurrentLocalMemUsage << " bytes of LDS\n"
-    << "  Rounding size to " << MaxSizeWithWaveCount
-    << " with a maximum occupancy of " << MaxOccupancy << '\n'
-    << " and " << (LocalMemLimit - CurrentLocalMemUsage)
-    << " available for promotion\n"
-  );
+  AS = AMDGPU::getAMDGPUAS(*F.getParent());
 
+  bool SufficientLDS = hasSufficientLocalMem(F);
+  bool Changed = false;
   BasicBlock &EntryBB = *F.begin();
   for (auto I = EntryBB.begin(), E = EntryBB.end(); I != E; ) {
     AllocaInst *AI = dyn_cast<AllocaInst>(I);
 
     ++I;
     if (AI)
-      handleAlloca(*AI);
+      Changed |= handleAlloca(*AI, SufficientLDS);
   }
 
-  return true;
+  return Changed;
 }
 
 std::pair<Value *, Value *>
 AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(
+                                *Builder.GetInsertBlock()->getParent());
+
   if (!IsAMDHSA) {
     Function *LocalSizeYFn
       = Intrinsic::getDeclaration(Mod, Intrinsic::r600_read_local_size_y);
@@ -235,8 +181,8 @@ AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
     CallInst *LocalSizeY = Builder.CreateCall(LocalSizeYFn, {});
     CallInst *LocalSizeZ = Builder.CreateCall(LocalSizeZFn, {});
 
-    LocalSizeY->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
-    LocalSizeZ->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
+    ST.makeLIDRangeMetadata(LocalSizeY);
+    ST.makeLIDRangeMetadata(LocalSizeZ);
 
     return std::make_pair(LocalSizeY, LocalSizeZ);
   }
@@ -279,15 +225,15 @@ AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
     = Intrinsic::getDeclaration(Mod, Intrinsic::amdgcn_dispatch_ptr);
 
   CallInst *DispatchPtr = Builder.CreateCall(DispatchPtrFn, {});
-  DispatchPtr->addAttribute(AttributeSet::ReturnIndex, Attribute::NoAlias);
-  DispatchPtr->addAttribute(AttributeSet::ReturnIndex, Attribute::NonNull);
+  DispatchPtr->addAttribute(AttributeList::ReturnIndex, Attribute::NoAlias);
+  DispatchPtr->addAttribute(AttributeList::ReturnIndex, Attribute::NonNull);
 
   // Size of the dispatch packet struct.
-  DispatchPtr->addDereferenceableAttr(AttributeSet::ReturnIndex, 64);
+  DispatchPtr->addDereferenceableAttr(AttributeList::ReturnIndex, 64);
 
   Type *I32Ty = Type::getInt32Ty(Mod->getContext());
   Value *CastDispatchPtr = Builder.CreateBitCast(
-    DispatchPtr, PointerType::get(I32Ty, AMDGPUAS::CONSTANT_ADDRESS));
+    DispatchPtr, PointerType::get(I32Ty, AS.CONSTANT_ADDRESS));
 
   // We could do a single 64-bit load here, but it's likely that the basic
   // 32-bit and extract sequence is already present, and it is probably easier
@@ -298,10 +244,10 @@ AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
   Value *GEPZU = Builder.CreateConstInBoundsGEP1_64(CastDispatchPtr, 2);
   LoadInst *LoadZU = Builder.CreateAlignedLoad(GEPZU, 4);
 
-  MDNode *MD = llvm::MDNode::get(Mod->getContext(), None);
+  MDNode *MD = MDNode::get(Mod->getContext(), None);
   LoadXY->setMetadata(LLVMContext::MD_invariant_load, MD);
   LoadZU->setMetadata(LLVMContext::MD_invariant_load, MD);
-  LoadZU->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
+  ST.makeLIDRangeMetadata(LoadZU);
 
   // Extract y component. Upper half of LoadZU should be zero already.
   Value *Y = Builder.CreateLShr(LoadXY, 16);
@@ -310,6 +256,8 @@ AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
 }
 
 Value *AMDGPUPromoteAlloca::getWorkitemID(IRBuilder<> &Builder, unsigned N) {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(
+                                *Builder.GetInsertBlock()->getParent());
   Intrinsic::ID IntrID = Intrinsic::ID::not_intrinsic;
 
   switch (N) {
@@ -332,7 +280,7 @@ Value *AMDGPUPromoteAlloca::getWorkitemID(IRBuilder<> &Builder, unsigned N) {
 
   Function *WorkitemIdFn = Intrinsic::getDeclaration(Mod, IntrID);
   CallInst *CI = Builder.CreateCall(WorkitemIdFn);
-  CI->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
+  ST.makeLIDRangeMetadata(CI);
 
   return CI;
 }
@@ -369,29 +317,37 @@ static Value* GEPToVectorIndex(GetElementPtrInst *GEP) {
 // instructions.
 static bool canVectorizeInst(Instruction *Inst, User *User) {
   switch (Inst->getOpcode()) {
-  case Instruction::Load:
+  case Instruction::Load: {
+    LoadInst *LI = cast<LoadInst>(Inst);
+    // Currently only handle the case where the Pointer Operand is a GEP so check for that case.
+    return isa<GetElementPtrInst>(LI->getPointerOperand()) && !LI->isVolatile();
+  }
   case Instruction::BitCast:
   case Instruction::AddrSpaceCast:
     return true;
   case Instruction::Store: {
-    // Must be the stored pointer operand, not a stored value.
+    // Must be the stored pointer operand, not a stored value, plus
+    // since it should be canonical form, the User should be a GEP.
     StoreInst *SI = cast<StoreInst>(Inst);
-    return SI->getPointerOperand() == User;
+    return (SI->getPointerOperand() == User) && isa<GetElementPtrInst>(User) && !SI->isVolatile();
   }
   default:
     return false;
   }
 }
 
-static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
+static bool tryPromoteAllocaToVector(AllocaInst *Alloca, AMDGPUAS AS) {
   ArrayType *AllocaTy = dyn_cast<ArrayType>(Alloca->getAllocatedType());
 
   DEBUG(dbgs() << "Alloca candidate for vectorization\n");
 
   // FIXME: There is no reason why we can't support larger arrays, we
   // are just being conservative for now.
+  // FIXME: We also reject alloca's of the form [ 2 x [ 2 x i32 ]] or equivalent. Potentially these
+  // could also be promoted but we don't currently handle this case
   if (!AllocaTy ||
       AllocaTy->getElementType()->isVectorTy() ||
+      AllocaTy->getElementType()->isArrayTy() ||
       AllocaTy->getNumElements() > 4 ||
       AllocaTy->getNumElements() < 2) {
     DEBUG(dbgs() << "  Cannot convert type to vector\n");
@@ -438,8 +394,8 @@ static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
     IRBuilder<> Builder(Inst);
     switch (Inst->getOpcode()) {
     case Instruction::Load: {
-      Type *VecPtrTy = VectorTy->getPointerTo(AMDGPUAS::PRIVATE_ADDRESS);
-      Value *Ptr = Inst->getOperand(0);
+      Type *VecPtrTy = VectorTy->getPointerTo(AS.PRIVATE_ADDRESS);
+      Value *Ptr = cast<LoadInst>(Inst)->getPointerOperand();
       Value *Index = calculateVectorIndex(Ptr, GEPVectorIdx);
 
       Value *BitCast = Builder.CreateBitCast(Alloca, VecPtrTy);
@@ -450,14 +406,15 @@ static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
       break;
     }
     case Instruction::Store: {
-      Type *VecPtrTy = VectorTy->getPointerTo(AMDGPUAS::PRIVATE_ADDRESS);
+      Type *VecPtrTy = VectorTy->getPointerTo(AS.PRIVATE_ADDRESS);
 
-      Value *Ptr = Inst->getOperand(1);
+      StoreInst *SI = cast<StoreInst>(Inst);
+      Value *Ptr = SI->getPointerOperand();
       Value *Index = calculateVectorIndex(Ptr, GEPVectorIdx);
       Value *BitCast = Builder.CreateBitCast(Alloca, VecPtrTy);
       Value *VecValue = Builder.CreateLoad(BitCast);
       Value *NewVecValue = Builder.CreateInsertElement(VecValue,
-                                                       Inst->getOperand(0),
+                                                       SI->getValueOperand(),
                                                        Index);
       Builder.CreateStore(NewVecValue, BitCast);
       Inst->eraseFromParent();
@@ -580,6 +537,9 @@ bool AMDGPUPromoteAlloca::collectUsesWithPtrTypes(
     }
 
     if (UseInst->getOpcode() == Instruction::AddrSpaceCast) {
+      // Give up if the pointer may be captured.
+      if (PointerMayBeCaptured(UseInst, true, true))
+        return false;
       // Don't collect the users of this.
       WorkList.push_back(User);
       continue;
@@ -626,12 +586,105 @@ bool AMDGPUPromoteAlloca::collectUsesWithPtrTypes(
   return true;
 }
 
+bool AMDGPUPromoteAlloca::hasSufficientLocalMem(const Function &F) {
+
+  FunctionType *FTy = F.getFunctionType();
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>(F);
+
+  // If the function has any arguments in the local address space, then it's
+  // possible these arguments require the entire local memory space, so
+  // we cannot use local memory in the pass.
+  for (Type *ParamTy : FTy->params()) {
+    PointerType *PtrTy = dyn_cast<PointerType>(ParamTy);
+    if (PtrTy && PtrTy->getAddressSpace() == AS.LOCAL_ADDRESS) {
+      LocalMemLimit = 0;
+      DEBUG(dbgs() << "Function has local memory argument. Promoting to "
+                      "local memory disabled.\n");
+      return false;
+    }
+  }
+
+  LocalMemLimit = ST.getLocalMemorySize();
+  if (LocalMemLimit == 0)
+    return false;
+
+  const DataLayout &DL = Mod->getDataLayout();
+
+  // Check how much local memory is being used by global objects
+  CurrentLocalMemUsage = 0;
+  for (GlobalVariable &GV : Mod->globals()) {
+    if (GV.getType()->getAddressSpace() != AS.LOCAL_ADDRESS)
+      continue;
+
+    for (const User *U : GV.users()) {
+      const Instruction *Use = dyn_cast<Instruction>(U);
+      if (!Use)
+        continue;
+
+      if (Use->getParent()->getParent() == &F) {
+        unsigned Align = GV.getAlignment();
+        if (Align == 0)
+          Align = DL.getABITypeAlignment(GV.getValueType());
+
+        // FIXME: Try to account for padding here. The padding is currently
+        // determined from the inverse order of uses in the function. I'm not
+        // sure if the use list order is in any way connected to this, so the
+        // total reported size is likely incorrect.
+        uint64_t AllocSize = DL.getTypeAllocSize(GV.getValueType());
+        CurrentLocalMemUsage = alignTo(CurrentLocalMemUsage, Align);
+        CurrentLocalMemUsage += AllocSize;
+        break;
+      }
+    }
+  }
+
+  unsigned MaxOccupancy = ST.getOccupancyWithLocalMemSize(CurrentLocalMemUsage,
+                                                          F);
+
+  // Restrict local memory usage so that we don't drastically reduce occupancy,
+  // unless it is already significantly reduced.
+
+  // TODO: Have some sort of hint or other heuristics to guess occupancy based
+  // on other factors..
+  unsigned OccupancyHint = ST.getWavesPerEU(F).second;
+  if (OccupancyHint == 0)
+    OccupancyHint = 7;
+
+  // Clamp to max value.
+  OccupancyHint = std::min(OccupancyHint, ST.getMaxWavesPerEU());
+
+  // Check the hint but ignore it if it's obviously wrong from the existing LDS
+  // usage.
+  MaxOccupancy = std::min(OccupancyHint, MaxOccupancy);
+
+
+  // Round up to the next tier of usage.
+  unsigned MaxSizeWithWaveCount
+    = ST.getMaxLocalMemSizeWithWaveCount(MaxOccupancy, F);
+
+  // Program is possibly broken by using more local mem than available.
+  if (CurrentLocalMemUsage > MaxSizeWithWaveCount)
+    return false;
+
+  LocalMemLimit = MaxSizeWithWaveCount;
+
+  DEBUG(
+    dbgs() << F.getName() << " uses " << CurrentLocalMemUsage << " bytes of LDS\n"
+    << "  Rounding size to " << MaxSizeWithWaveCount
+    << " with a maximum occupancy of " << MaxOccupancy << '\n'
+    << " and " << (LocalMemLimit - CurrentLocalMemUsage)
+    << " available for promotion\n"
+  );
+
+  return true;
+}
+
 // FIXME: Should try to pick the most likely to be profitable allocas first.
-void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
+bool AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I, bool SufficientLDS) {
   // Array allocations are probably not worth handling, since an allocation of
   // the array type is the canonical form.
   if (!I.isStaticAlloca() || I.isArrayAllocation())
-    return;
+    return false;
 
   IRBuilder<> Builder(&I);
 
@@ -640,23 +693,30 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
 
   DEBUG(dbgs() << "Trying to promote " << I << '\n');
 
-  if (tryPromoteAllocaToVector(&I)) {
-    DEBUG(dbgs() << " alloca is not a candidate for vectorization.\n");
-    return;
-  }
+  if (tryPromoteAllocaToVector(&I, AS))
+    return true; // Promoted to vector.
 
   const Function &ContainingFunction = *I.getParent()->getParent();
+  CallingConv::ID CC = ContainingFunction.getCallingConv();
 
   // Don't promote the alloca to LDS for shader calling conventions as the work
   // item ID intrinsics are not supported for these calling conventions.
   // Furthermore not all LDS is available for some of the stages.
-  if (AMDGPU::isShader(ContainingFunction.getCallingConv()))
-    return;
+  switch (CC) {
+  case CallingConv::AMDGPU_KERNEL:
+  case CallingConv::SPIR_KERNEL:
+    break;
+  default:
+    DEBUG(dbgs() << " promote alloca to LDS not supported with calling convention.\n");
+    return false;
+  }
+
+  // Not likely to have sufficient local memory for promotion.
+  if (!SufficientLDS)
+    return false;
 
   const AMDGPUSubtarget &ST =
     TM->getSubtarget<AMDGPUSubtarget>(ContainingFunction);
-  // FIXME: We should also try to get this value from the reqd_work_group_size
-  // function attribute if it is available.
   unsigned WorkGroupSize = ST.getFlatWorkGroupSizes(ContainingFunction).second;
 
   const DataLayout &DL = Mod->getDataLayout();
@@ -678,7 +738,7 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
   if (NewSize > LocalMemLimit) {
     DEBUG(dbgs() << "  " << AllocSize
           << " bytes of local memory not available to promote\n");
-    return;
+    return false;
   }
 
   CurrentLocalMemUsage = NewSize;
@@ -687,7 +747,7 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
 
   if (!collectUsesWithPtrTypes(&I, &I, WorkList)) {
     DEBUG(dbgs() << " Do not know how to convert all uses\n");
-    return;
+    return false;
   }
 
   DEBUG(dbgs() << "Promoting alloca to local memory\n");
@@ -701,7 +761,7 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
       Twine(F->getName()) + Twine('.') + I.getName(),
       nullptr,
       GlobalVariable::NotThreadLocal,
-      AMDGPUAS::LOCAL_ADDRESS);
+      AS.LOCAL_ADDRESS);
   GV->setUnnamedAddr(GlobalValue::UnnamedAddr::Global);
   GV->setAlignment(I.getAlignment());
 
@@ -734,7 +794,7 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
       if (ICmpInst *CI = dyn_cast<ICmpInst>(V)) {
         Value *Src0 = CI->getOperand(0);
         Type *EltTy = Src0->getType()->getPointerElementType();
-        PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);
+        PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);
 
         if (isa<ConstantPointerNull>(CI->getOperand(0)))
           CI->setOperand(0, ConstantPointerNull::get(NewTy));
@@ -751,7 +811,7 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
         continue;
 
       Type *EltTy = V->getType()->getPointerElementType();
-      PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);
+      PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);
 
       // FIXME: It doesn't really make sense to try to do this for all
       // instructions.
@@ -819,22 +879,23 @@ void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
       Type *SrcTy = Src->getType()->getPointerElementType();
       Function *ObjectSize = Intrinsic::getDeclaration(Mod,
         Intrinsic::objectsize,
-        { Intr->getType(), PointerType::get(SrcTy, AMDGPUAS::LOCAL_ADDRESS) }
+        { Intr->getType(), PointerType::get(SrcTy, AS.LOCAL_ADDRESS) }
       );
 
-      CallInst *NewCall
-        = Builder.CreateCall(ObjectSize, { Src, Intr->getOperand(1) });
+      CallInst *NewCall = Builder.CreateCall(
+          ObjectSize, {Src, Intr->getOperand(1), Intr->getOperand(2)});
       Intr->replaceAllUsesWith(NewCall);
       Intr->eraseFromParent();
       continue;
     }
     default:
-      Intr->dump();
+      Intr->print(errs());
       llvm_unreachable("Don't know how to promote alloca intrinsic use.");
     }
   }
+  return true;
 }
 
-FunctionPass *llvm::createAMDGPUPromoteAlloca(const TargetMachine *TM) {
-  return new AMDGPUPromoteAlloca(TM);
+FunctionPass *llvm::createAMDGPUPromoteAlloca() {
+  return new AMDGPUPromoteAlloca();
 }