[AMDGPU][NFC] Fix typos

[sebastian-ne]

Summary

Fix some typos in the amdgpu backend.

[sebastian-ne]

Two more typos

[foad]

llvm/lib/Target/AMDGPU/MIMGInstructions.td

229 ↗

(On Diff #406778)

This ones seems like a legitimate use of "that that".

llvm/lib/Target/AMDGPU/R600ClauseMergePass.cpp

10

Should be "manner".

llvm/lib/Target/AMDGPU/R600InstrInfo.cpp

853 ↗

(On Diff #406778)

No, we're talking about predication so "unpredicable" is correct.

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

2958

This one doesn't seem particularly wrong.

llvm/lib/Target/AMDGPU/SIMachineScheduler.cpp

94

There is no typo fix here?

llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp

1143

No typo fixes here?

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

972

Should probably be "inactive"? Or perhaps "deactivated"?

[sebastian-ne]

Thanks! Fixed your comments

llvm/lib/Target/AMDGPU/R600InstrInfo.cpp

853 ↗

(On Diff #406778)

Ups, I didn’t predict that.

llvm/lib/Target/AMDGPU/SIMachineScheduler.cpp

94

True, I guess clang-format just formatted the whole comment, not only the changed line. (The lines were longer than 80 chars)
Is it fine to leave these changes in the patch?

[foad]

llvm/lib/Target/AMDGPU/SIMachineScheduler.cpp

94

Yes, OK.

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

980

On second thoughts maybe "deactivated" was intended? And it needs to be fixed here too. @critson I think you wrote these comments.

[critson]

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

980

I am not entirely sure what I meant by these comments, but protocol the first word is not meant to be a verb.
The two comments are probably meant to read:

// Demote - deactivate quads with only helper lanes

// Kill - deactivate lanes no longer in live mask

[sebastian-ne]

Improve WQM comment

[sebastian-ne]

friendly ping for review

[foad]

LGTM, thanks.

llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp

static std::pair<Value*, Value*> getMul64(IRBuilder<> &Builder,
   Hi = Builder.CreateTrunc(Hi, I32Ty);
   return std::make_pair(Lo, Hi);
 }
 
 static Value* getMulHu(IRBuilder<> &Builder, Value *LHS, Value *RHS) {
   return getMul64(Builder, LHS, RHS).second;
 }
 
-/// Figure out how many bits are really needed for this ddivision. \p AtLeast is
+/// Figure out how many bits are really needed for this division. \p AtLeast is
 /// an optimization hint to bypass the second ComputeNumSignBits call if we the
 /// first one is insufficient. Returns -1 on failure.
 int AMDGPUCodeGenPrepare::getDivNumBits(BinaryOperator &I,
                                         Value *Num, Value *Den,
                                         unsigned AtLeast, bool IsSigned) const {
   const DataLayout &DL = Mod->getDataLayout();
   unsigned LHSSignBits = ComputeNumSignBits(Num, DL, 0, AC, &I);
   if (LHSSignBits < AtLeast)

llvm/lib/Target/AMDGPU/AMDGPUGISel.td

def gi_smrd_buffer_imm :
     GIComplexPatternEquiv<SMRDBufferImm>;
 
 def gi_smrd_buffer_imm32 :
     GIComplexOperandMatcher<s64, "selectSMRDBufferImm32">,
     GIComplexPatternEquiv<SMRDBufferImm32>;
 
 // Separate load nodes are defined to glue m0 initialization in
 // SelectionDAG. The GISel selector can just insert m0 initialization
-// directly before before selecting a glue-less load, so hide this
+// directly before selecting a glue-less load, so hide this
 // distinction.
 
 def : GINodeEquiv<G_LOAD, AMDGPUld_glue> {
   let CheckMMOIsNonAtomic = 1;
   let IfSignExtend = G_SEXTLOAD;
   let IfZeroExtend = G_ZEXTLOAD;
 }
 

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

if (RC != &AMDGPU::VS_32RegClass) {
           if (SII->findCommutedOpIndices(Desc, OpIdx, CommuteIdx1)) {
             unsigned CommutedOpNo = CommuteIdx1 - Desc.getNumDefs();
             const TargetRegisterClass *CommutedRC = getOperandRegClass(*U, CommutedOpNo);
             if (CommutedRC == &AMDGPU::VS_32RegClass)
               AllUsesAcceptSReg = true;
           }
         }
       }
-      // If "AllUsesAcceptSReg == false" so far we haven't suceeded
+      // If "AllUsesAcceptSReg == false" so far we haven't succeeded
       // commuting current user. This means have at least one use
       // that strictly require VGPR. Thus, we will not attempt to commute
       // other user instructions.
       if (!AllUsesAcceptSReg)
         break;
     }
   }
   return !AllUsesAcceptSReg && (Limit < 10);

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

for (unsigned I = Intr->VAddrStart; I < Intr->VAddrEnd; I++) {
     if (!Addr)
       break;
 
     ++NumVAddrRegs;
     NumVAddrDwords += (MRI->getType(Addr).getSizeInBits() + 31) / 32;
   }
 
   // The legalizer preprocessed the intrinsic arguments. If we aren't using
-  // NSA, these should have beeen packed into a single value in the first
+  // NSA, these should have been packed into a single value in the first
   // address register
   const bool UseNSA = NumVAddrRegs != 1 && NumVAddrDwords == NumVAddrRegs;
   if (UseNSA && !STI.hasFeature(AMDGPU::FeatureNSAEncoding)) {
     LLVM_DEBUG(dbgs() << "Trying to use NSA on non-NSA target\n");
     return false;
   }
 
   if (IsTexFail)

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

 
   // FIXME: Clamp maximum size
   getActionDefinitionsBuilder(G_CONCAT_VECTORS)
     .legalIf(all(isRegisterType(0), isRegisterType(1)))
     .clampMaxNumElements(0, S32, 32)
     .clampMaxNumElements(1, S16, 2) // TODO: Make 4?
     .clampMaxNumElements(0, S16, 64);
 
-  // TODO: Don't fully scalarize v2s16 pieces? Or combine out thosse
+  // TODO: Don't fully scalarize v2s16 pieces? Or combine out those
   // pre-legalize.
   if (ST.hasVOP3PInsts()) {
     getActionDefinitionsBuilder(G_SHUFFLE_VECTOR)
       .customFor({V2S16, V2S16})
       .lower();
   } else
     getActionDefinitionsBuilder(G_SHUFFLE_VECTOR).lower();
 

 ///
 /// Depending on the subtarget, load/store with 16-bit element data need to be
 /// rewritten to use the low half of 32-bit registers, or directly use a packed
 /// layout. 16-bit addresses should also sometimes be packed into 32-bit
 /// registers.
 ///
 /// We don't want to directly select image instructions just yet, but also want
 /// to exposes all register repacking to the legalizer/combiners. We also don't
-/// want a selected instrution entering RegBankSelect. In order to avoid
+/// want a selected instruction entering RegBankSelect. In order to avoid
 /// defining a multitude of intermediate image instructions, directly hack on
 /// the intrinsic's arguments. In cases like a16 addresses, this requires
 /// padding now unnecessary arguments with $noreg.
 bool AMDGPULegalizerInfo::legalizeImageIntrinsic(
     MachineInstr &MI, MachineIRBuilder &B, GISelChangeObserver &Observer,
     const AMDGPU::ImageDimIntrinsicInfo *Intr) const {
 
   const unsigned NumDefs = MI.getNumExplicitDefs();

llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp

bool AMDGPULowerKernelArguments::runOnFunction(Function &F) {
   const DataLayout &DL = F.getParent()->getDataLayout();
   BasicBlock &EntryBlock = *F.begin();
   IRBuilder<> Builder(&*getInsertPt(EntryBlock));
 
   const Align KernArgBaseAlign(16); // FIXME: Increase if necessary
   const uint64_t BaseOffset = ST.getExplicitKernelArgOffset(F);
 
   Align MaxAlign;
-  // FIXME: Alignment is broken broken with explicit arg offset.;
+  // FIXME: Alignment is broken with explicit arg offset.;
   const uint64_t TotalKernArgSize = ST.getKernArgSegmentSize(F, MaxAlign);
   if (TotalKernArgSize == 0)
     return false;
 
   CallInst *KernArgSegment =
       Builder.CreateIntrinsic(Intrinsic::amdgcn_kernarg_segment_ptr, {}, {},
                               nullptr, F.getName() + ".kernarg.segment");
 

llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp

 //===-- AMDGPULowerModuleLDSPass.cpp ------------------------------*- C++ -*-=//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This pass eliminates LDS uses from non-kernel functions.
 //
 // The strategy is to create a new struct with a field for each LDS variable
 // and allocate that struct at the same address for every kernel. Uses of the
 // original LDS variables are then replaced with compile time offsets from that
 // known address. AMDGPUMachineFunction allocates the LDS global.
 //
 // Local variables with constant annotation or non-undef initializer are passed
-// through unchanged for simplication or error diagnostics in later passes.
+// through unchanged for simplification or error diagnostics in later passes.
 //
 // To reduce the memory overhead variables that are only used by kernels are
 // excluded from this transform. The analysis to determine whether a variable
 // is only used by a kernel is cheap and conservative so this may allocate
 // a variable in every kernel when it was not strictly necessary to do so.
 //
 // A possible future refinement is to specialise the structure per-kernel, so
 // that fields can be elided based on more expensive analysis.

llvm/lib/Target/AMDGPU/AMDGPUMachineCFGStructurizer.cpp

for (auto &UI : TI.uses()) {
         if (Target != InternalSucc && Target != ExternalSucc) {
           UI.setMBB(ExternalSucc);
         }
       }
     }
   }
 }
 
-// If a region region is just a sequence of regions (and the exit
+// If a region is just a sequence of regions (and the exit
 // block in the case of the top level region), we can simply skip
 // linearizing it, because it is already linear
 bool regionIsSequence(RegionMRT *Region) {
   auto Children = Region->getChildren();
   for (auto CI : *Children) {
     if (!CI->isRegion()) {
       if (CI->getMBBMRT()->getMBB()->succ_size() > 1) {
         return false;

llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp

 
 INITIALIZE_PASS(AMDGPUOpenCLEnqueuedBlockLowering, DEBUG_TYPE,
                 "Lower OpenCL enqueued blocks", false, false)
 
 ModulePass* llvm::createAMDGPUOpenCLEnqueuedBlockLoweringPass() {
   return new AMDGPUOpenCLEnqueuedBlockLowering();
 }
 
-/// Collect direct or indrect callers of \p F and save them
+/// Collect direct or indirect callers of \p F and save them
 /// to \p Callers.
 static void collectCallers(Function *F, DenseSet<Function *> &Callers) {
   for (auto U : F->users()) {
     if (auto *CI = dyn_cast<CallInst>(&*U)) {
       auto *Caller = CI->getParent()->getParent();
       if (Callers.insert(Caller).second)
         collectCallers(Caller, Callers);
     }

llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

bool AMDGPUPromoteAllocaImpl::handleAlloca(AllocaInst &I, bool SufficientLDS) {
 
   Align Alignment =
       DL.getValueOrABITypeAlignment(I.getAlign(), I.getAllocatedType());
 
   // FIXME: This computed padding is likely wrong since it depends on inverse
   // usage order.
   //
   // FIXME: It is also possible that if we're allowed to use all of the memory
-  // could could end up using more than the maximum due to alignment padding.
+  // could end up using more than the maximum due to alignment padding.
 
   uint32_t NewSize = alignTo(CurrentLocalMemUsage, Alignment);
   uint32_t AllocSize = WorkGroupSize * DL.getTypeAllocSize(AllocaTy);
   NewSize += AllocSize;
 
   if (NewSize > LocalMemLimit) {
     LLVM_DEBUG(dbgs() << "  " << AllocSize
                       << " bytes of local memory not available to promote\n");

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

B.buildInstr(AMDGPU::REG_SEQUENCE)
     .addUse(TmpReg1)
     .addImm(AMDGPU::sub1);
 
   return constrainGenericRegister(SrcReg, AMDGPU::SReg_64RegClass, MRI) &&
          constrainGenericRegister(DstReg, AMDGPU::VReg_64RegClass, MRI);
 }
 
 /// Utility function for pushing dynamic vector indexes with a constant offset
-/// into waterwall loops.
+/// into waterfall loops.
 static void reinsertVectorIndexAdd(MachineIRBuilder &B,
                                    MachineInstr &IdxUseInstr,
                                    unsigned OpIdx,
                                    unsigned ConstOffset) {
   MachineRegisterInfo &MRI = *B.getMRI();
   const LLT S32 = LLT::scalar(32);
   Register WaterfallIdx = IdxUseInstr.getOperand(OpIdx).getReg();
   B.setInsertPt(*IdxUseInstr.getParent(), IdxUseInstr.getIterator());

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h

 //===-- AMDGPUTargetMachine.h - AMDGPU TargetMachine Interface --*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
-/// The AMDGPU TargetMachine interface definition for hw codgen targets.
+/// The AMDGPU TargetMachine interface definition for hw codegen targets.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUTARGETMACHINE_H
 #define LLVM_LIB_TARGET_AMDGPU_AMDGPUTARGETMACHINE_H
 
 #include "GCNSubtarget.h"
 #include "llvm/CodeGen/TargetPassConfig.h"

llvm/lib/Target/AMDGPU/AMDILCFGStructurizer.cpp

int AMDGPUCFGStructurizer::cloneOnSideEntryTo(MachineBasicBlock *PreMBB,
   }
 
   return Cloned;
 }
 
 MachineBasicBlock *
 AMDGPUCFGStructurizer::cloneBlockForPredecessor(MachineBasicBlock *MBB,
     MachineBasicBlock *PredMBB) {
-  assert(PredMBB->isSuccessor(MBB) &&
-         "succBlk is not a prececessor of curBlk");
+  assert(PredMBB->isSuccessor(MBB) && "succBlk is not a predecessor of curBlk");
 
   MachineBasicBlock *CloneMBB = clone(MBB);  //clone instructions
   replaceInstrUseOfBlockWith(PredMBB, MBB, CloneMBB);
   //srcBlk, oldBlk, newBlk
 
   PredMBB->replaceSuccessor(MBB, CloneMBB);
 
   // add all successor to cloneBlk

llvm/lib/Target/AMDGPU/AMDKernelCodeT.h

enum amd_code_property_mask_t {
   AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS_WIDTH = 1,
   AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS = ((1 << AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS_WIDTH) - 1) << AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS_SHIFT,
 
   /// The interleave (swizzle) element size in bytes required by the
   /// code for private memory. This must be 2, 4, 8 or 16. This value
   /// is provided to the finalizer when it is invoked and is recorded
   /// here. The hardware will interleave the memory requests of each
   /// lane of a wavefront by this element size to ensure each
-  /// work-item gets a distinct memory memory location. Therefore, the
+  /// work-item gets a distinct memory location. Therefore, the
   /// finalizer ensures that all load and store operations done to
   /// private memory do not exceed this size. For example, if the
   /// element size is 4 (32-bits or dword) and a 64-bit value must be
   /// loaded, the finalizer will generate two 32-bit loads. This
   /// ensures that the interleaving will get the work-item
   /// specific dword for both halves of the 64-bit value. If it just
   /// did a 64-bit load then it would get one dword which belonged to
   /// its own work-item, but the second dword would belong to the

llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

bool AMDGPUOperand::isLiteralImm(MVT type) const {
   if (type == MVT::i64) { // Expected 64-bit int operand
     // We don't allow fp literals in 64-bit integer instructions. It is
     // unclear how we should encode them.
     return false;
   }
 
   // We allow fp literals with f16x2 operands assuming that the specified
   // literal goes into the lower half and the upper half is zero. We also
-  // require that the literal may be losslesly converted to f16.
+  // require that the literal may be losslessly converted to f16.
   MVT ExpectedType = (type == MVT::v2f16)? MVT::f16 :
                      (type == MVT::v2i16)? MVT::i16 :
                      (type == MVT::v2f32)? MVT::f32 : type;
 
   APFloat FPLiteral(APFloat::IEEEdouble(), APInt(64, Imm.Val));
   return canLosslesslyConvertToFPType(FPLiteral, ExpectedType);
 }
 

 // than a floating-point NEG modifier applied to N.
 // Beside being contr-intuitive, such use of floating-point
 // NEG modifier would have resulted in different meaning
 // of integer literals used with VOP1/2/C and VOP3,
 // for example:
 //    v_exp_f32_e32 v5, -1 // VOP1: src0 = 0xFFFFFFFF
 //    v_exp_f32_e64 v5, -1 // VOP3: src0 = 0x80000001
 // Negative fp literals with preceding "-" are
-// handled likewise for unifomtity
+// handled likewise for uniformity
 //
 bool
 AMDGPUAsmParser::parseSP3NegModifier() {
 
   AsmToken NextToken[2];
   peekTokens(NextToken);
 
   if (isToken(AsmToken::Minus) &&

 
 bool
 AMDGPUAsmParser::validateSendMsg(const OperandInfoTy &Msg,
                                  const OperandInfoTy &Op,
                                  const OperandInfoTy &Stream) {
   using namespace llvm::AMDGPU::SendMsg;
 
   // Validation strictness depends on whether message is specified
-  // in a symbolc or in a numeric form. In the latter case
+  // in a symbolic or in a numeric form. In the latter case
   // only encoding possibility is checked.
   bool Strict = Msg.IsSymbolic;
 
   if (!isValidMsgId(Msg.Id, getSTI(), Strict)) {
     Error(Msg.Loc, "invalid message id");
     return false;
   }
   if (Strict && (msgRequiresOp(Msg.Id) != Op.IsDefined)) {

 }
 
 #define GET_REGISTER_MATCHER
 #define GET_MATCHER_IMPLEMENTATION
 #define GET_MNEMONIC_SPELL_CHECKER
 #define GET_MNEMONIC_CHECKER
 #include "AMDGPUGenAsmMatcher.inc"
 
-// This fuction should be defined after auto-generated include so that we have
+// This function should be defined after auto-generated include so that we have
 // MatchClassKind enum defined
 unsigned AMDGPUAsmParser::validateTargetOperandClass(MCParsedAsmOperand &Op,
                                                      unsigned Kind) {
   // Tokens like "glc" would be parsed as immediate operands in ParseOperand().
   // But MatchInstructionImpl() expects to meet token and fails to validate
   // operand. This method checks if we are given immediate operand but expect to
   // get corresponding token.
   AMDGPUOperand &Operand = (AMDGPUOperand&)Op;

llvm/lib/Target/AMDGPU/BUFInstructions.td

class MTBUF_Real <MTBUF_Pseudo ps> :
   bits<7>  srsrc;
   bits<1>  tfe;
   bits<8>  soffset;
 
   bits<4> dfmt = format{3-0};
   bits<3> nfmt = format{6-4};
 
   // GFX90A+ only: instruction uses AccVGPR for data
-  // Bit superceedes tfe.
+  // Bit supersedes tfe.
   bits<1> acc = !if(ps.has_vdata, vdata{9}, 0);
 }
 
 class getMTBUFInsDA<list<RegisterClass> vdataList,
                     list<RegisterClass> vaddrList=[]> {
   RegisterClass vdataClass = !if(!empty(vdataList), ?, !head(vdataList));
   RegisterClass vaddrClass = !if(!empty(vaddrList), ?, !head(vaddrList));
   RegisterOperand vdata_op = getLdStRegisterOperand<vdataClass>.ret;
class MUBUF_Real <MUBUF_Pseudo ps> :
   bits<5>  cpol;
   bits<8>  vaddr;
   bits<10> vdata;
   bits<7>  srsrc;
   bits<1>  tfe;
   bits<8>  soffset;
 
   // GFX90A+ only: instruction uses AccVGPR for data
-  // Bit superceedes tfe.
+  // Bit supersedes tfe.
   bits<1> acc = !if(ps.has_vdata, vdata{9}, 0);
 }
 
 
 // For cache invalidation instructions.
 class MUBUF_Invalidate <string opName, SDPatternOperator node = null_frag> :
   MUBUF_Pseudo<opName, (outs), (ins), "", [(node)]> {
 

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/ScheduleDAG.h"
 #include "llvm/Support/TargetParser.h"
 
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
-// Hazard Recoginizer Implementation
+// Hazard Recognizer Implementation
 //===----------------------------------------------------------------------===//
 
 static bool shouldRunLdsBranchVmemWARHazardFixup(const MachineFunction &MF,
                                                  const GCNSubtarget &ST);
 
 GCNHazardRecognizer::GCNHazardRecognizer(const MachineFunction &MF) :
   IsHazardRecognizerMode(false),
   CurrCycleInstr(nullptr),
int GCNHazardRecognizer::checkSoftClauseHazards(MachineInstr *MEM) {
 
   // A soft-clause is any group of consecutive SMEM instructions.  The
   // instructions in this group may return out of order and/or may be
   // replayed (i.e. the same instruction issued more than once).
   //
   // In order to handle these situations correctly we need to make sure that
   // when a clause has more than one instruction, no instruction in the clause
   // writes to a register that is read by another instruction in the clause
-  // (including itself). If we encounter this situaion, we need to break the
+  // (including itself). If we encounter this situation, we need to break the
   // clause by inserting a non SMEM instruction.
 
   for (MachineInstr *MI : EmittedInstrs) {
     // When we hit a non-SMEM instruction then we have passed the start of the
     // clause and we can stop.
     if (!MI)
       break;
 

llvm/lib/Target/AMDGPU/GCNNSAReassign.cpp

-//===-- GCNNSAReassign.cpp - Reassign registers in NSA unstructions -------===//
+//===-- GCNNSAReassign.cpp - Reassign registers in NSA instructions -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
 /// \brief Try to reassign registers on GFX10+ from non-sequential to sequential
-/// in NSA image instructions. Later SIShrinkInstructions pass will relace NSA
+/// in NSA image instructions. Later SIShrinkInstructions pass will replace NSA
 /// with sequential versions where possible.
 ///
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/ADT/Statistic.h"

llvm/lib/Target/AMDGPU/GCNRegPressure.h

 //===- GCNRegPressure.h -----------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 ///
 /// \file
 /// This file defines the GCNRegPressure class, which tracks registry pressure
 /// by bookkeeping number of SGPR/VGPRs used, weights for large SGPR/VGPRs. It
 /// also implements a compare function, which compares different register
-/// pressures, and declares one with max occupance as winner.
+/// pressures, and declares one with max occupancy as winner.
 ///
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
 #define LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
 
 #include "GCNSubtarget.h"
 #include "llvm/CodeGen/LiveIntervals.h"

llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp

EmitNote(ElfNote::NoteNameV2, MCConstantExpr::create(DescSZ, getContext()),
              OS.emitInt16(VendorNameSize);
              OS.emitInt16(ArchNameSize);
              OS.emitInt32(Major);
              OS.emitInt32(Minor);
              OS.emitInt32(Stepping);
              OS.emitBytes(VendorName);
              OS.emitInt8(0); // NULL terminate VendorName
              OS.emitBytes(ArchName);
-             OS.emitInt8(0); // NULL terminte ArchName
+             OS.emitInt8(0); // NULL terminate ArchName
            });
 }
 
 void
 AMDGPUTargetELFStreamer::EmitAMDKernelCodeT(const amd_kernel_code_t &Header) {
 
   MCStreamer &OS = getStreamer();
   OS.PushSection();

llvm/lib/Target/AMDGPU/R600ClauseMergePass.cpp

 //===-- R600ClauseMergePass - Merge consecutive CF_ALU -------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
-/// R600EmitClauseMarker pass emits CFAlu instruction in a conservative maneer.
+/// R600EmitClauseMarker pass emits CFAlu instruction in a conservative manner.
 /// This pass is merging consecutive CFAlus where applicable.
 /// It needs to be called after IfCvt for best results.
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/R600MCTargetDesc.h"
 #include "R600.h"
 #include "R600Subtarget.h"
 

llvm/lib/Target/AMDGPU/R600EmitClauseMarkers.cpp

public:
   }
 };
 
 char R600EmitClauseMarkers::ID = 0;
 
 } // end anonymous namespace
 
 INITIALIZE_PASS_BEGIN(R600EmitClauseMarkers, "emitclausemarkers",
-                      "R600 Emit Clause Markters", false, false)
+                      "R600 Emit Clause Markers", false, false)
 INITIALIZE_PASS_END(R600EmitClauseMarkers, "emitclausemarkers",
-                      "R600 Emit Clause Markters", false, false)
+                    "R600 Emit Clause Markers", false, false)
 
 FunctionPass *llvm::createR600EmitClauseMarkers() {
   return new R600EmitClauseMarkers();
 }

llvm/lib/Target/AMDGPU/R600ISelLowering.cpp

return DAG.getNode(ISD::SELECT_CC, DL, VT,
       Cond, HWFalse,
       True, False,
       DAG.getCondCode(ISD::SETNE));
 }
 
 /// LLVM generates byte-addressed pointers.  For indirect addressing, we need to
 /// convert these pointers to a register index.  Each register holds
 /// 16 bytes, (4 x 32bit sub-register), but we need to take into account the
-/// \p StackWidth, which tells us how many of the 4 sub-registrers will be used
+/// \p StackWidth, which tells us how many of the 4 sub-registers will be used
 /// for indirect addressing.
 SDValue R600TargetLowering::stackPtrToRegIndex(SDValue Ptr,
                                                unsigned StackWidth,
                                                SelectionDAG &DAG) const {
   unsigned SRLPad;
   switch(StackWidth) {
   case 1:
     SRLPad = 2;
SDValue R600TargetLowering::lowerPrivateTruncStore(StoreSDNode *Store,
   // Get offset in dword
   SDValue ByteIdx = DAG.getNode(ISD::AND, DL, MVT::i32, LoadPtr,
                                 DAG.getConstant(0x3, DL, MVT::i32));
 
   // Convert byte offset to bit shift
   SDValue ShiftAmt = DAG.getNode(ISD::SHL, DL, MVT::i32, ByteIdx,
                                  DAG.getConstant(3, DL, MVT::i32));
 
-  // TODO: Contrary to the name of the functiom,
+  // TODO: Contrary to the name of the function,
   // it also handles sub i32 non-truncating stores (like i1)
   SDValue SExtValue = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i32,
                                   Store->getValue());
 
   // Mask the value to the right type
   SDValue MaskedValue = DAG.getZeroExtendInReg(SExtValue, DL, MemVT);
 
   // Shift the value in place
static SDValue CompactSwizzlableVector(
   for (unsigned i = 0; i < 4; i++)
     NewBldVec[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltTy, VectorEntry,
                                DAG.getIntPtrConstant(i, DL));
 
   for (unsigned i = 0; i < 4; i++) {
     if (NewBldVec[i].isUndef())
       // We mask write here to teach later passes that the ith element of this
       // vector is undef. Thus we can use it to reduce 128 bits reg usage,
-      // break false dependencies and additionnaly make assembly easier to read.
+      // break false dependencies and additionally make assembly easier to read.
       RemapSwizzle[i] = 7; // SEL_MASK_WRITE
     if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(NewBldVec[i])) {
       if (C->isZero()) {
         RemapSwizzle[i] = 4; // SEL_0
         NewBldVec[i] = DAG.getUNDEF(MVT::f32);
       } else if (C->isExactlyValue(1.0)) {
         RemapSwizzle[i] = 5; // SEL_1
         NewBldVec[i] = DAG.getUNDEF(MVT::f32);

llvm/lib/Target/AMDGPU/R600Packetizer.cpp

bool isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) override {
     bool ARDef =
         TII->definesAddressRegister(*MII) || TII->definesAddressRegister(*MIJ);
     bool ARUse =
         TII->usesAddressRegister(*MII) || TII->usesAddressRegister(*MIJ);
 
     return !ARDef || !ARUse;
   }
 
-  // isLegalToPruneDependencies - Is it legal to prune dependece between SUI
+  // isLegalToPruneDependencies - Is it legal to prune dependency between SUI
   // and SUJ.
   bool isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ) override {
     return false;
   }
 
   void setIsLastBit(MachineInstr *MI, unsigned Bit) const {
     unsigned LastOp = TII->getOperandIdx(MI->getOpcode(), R600::OpName::last);
     MI->getOperand(LastOp).setImm(Bit);

llvm/lib/Target/AMDGPU/R600TargetMachine.h

 //===-- R600TargetMachine.h - AMDGPU TargetMachine Interface ----*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
-/// The AMDGPU TargetMachine interface definition for hw codgen targets.
+/// The AMDGPU TargetMachine interface definition for hw codegen targets.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_R600TARGETMACHINE_H
 #define LLVM_LIB_TARGET_AMDGPU_R600TARGETMACHINE_H
 
 #include "AMDGPUTargetMachine.h"
 #include "R600Subtarget.h"

llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp

for (const MachineOperand &MO : MI.operands()) {
     } else {
       Loc->second.first |= State;
       Loc->second.second |= Mask;
     }
   }
 }
 
 // Check register def/use conflicts, occupancy limits and collect def/use maps.
-// Return true if instruction can be bundled with previous. It it cannot
+// Return true if instruction can be bundled with previous. If it cannot
 // def/use maps are not updated.
 bool SIFormMemoryClauses::processRegUses(const MachineInstr &MI,
                                          RegUse &Defs, RegUse &Uses,
                                          GCNDownwardRPTracker &RPT) {
   if (!canBundle(MI, Defs, Uses))
     return false;
 
   if (!checkPressure(MI, RPT))

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

for (MachineBasicBlock &MBB : MF) {
           }
         } else if (TII->isStoreToStackSlot(MI, FrameIndex) ||
                    TII->isLoadFromStackSlot(MI, FrameIndex))
           if (!MFI.isFixedObjectIndex(FrameIndex))
             NonVGPRSpillFIs.set(FrameIndex);
       }
     }
 
-    // Stack slot coloring may assign different objets to the same stack slot.
+    // Stack slot coloring may assign different objects to the same stack slot.
     // If not, then the VGPR to AGPR spill slot is dead.
     for (unsigned FI : SpillFIs.set_bits())
       if (!NonVGPRSpillFIs.test(FI))
         FuncInfo->setVGPRToAGPRSpillDead(FI);
 
     for (MachineBasicBlock &MBB : MF) {
       for (MCPhysReg Reg : FuncInfo->getVGPRSpillAGPRs())
         MBB.addLiveIn(Reg);

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

if (IsFast)
       *IsFast = AlignedBy4;
 
     return AlignedBy4;
   }
 
   if (Subtarget->hasUnalignedBufferAccessEnabled() &&
       !(AddrSpace == AMDGPUAS::LOCAL_ADDRESS ||
         AddrSpace == AMDGPUAS::REGION_ADDRESS)) {
-    // If we have an uniform constant load, it still requires using a slow
+    // If we have a uniform constant load, it still requires using a slow
     // buffer instruction if unaligned.
     if (IsFast) {
       // Accesses can really be issued as 1-byte aligned or 4-byte aligned, so
-      // 2-byte alignment is worse than 1 unless doing a 2-byte accesss.
+      // 2-byte alignment is worse than 1 unless doing a 2-byte access.
       *IsFast = (AddrSpace == AMDGPUAS::CONSTANT_ADDRESS ||
                  AddrSpace == AMDGPUAS::CONSTANT_ADDRESS_32BIT) ?
         Alignment >= Align(4) : Alignment != Align(2);
     }
 
     return true;
   }
 
bool SITargetLowering::isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
   switch (VT.getSimpleVT().SimpleTy) {
   case MVT::f32: {
     // If mad is not available this depends only on if f32 fma is full rate.
     if (!Subtarget->hasMadMacF32Insts())
       return Subtarget->hasFastFMAF32();
 
     // Otherwise f32 mad is always full rate and returns the same result as
     // the separate operations so should be preferred over fma.
-    // However does not support denomals.
+    // However does not support denormals.
     if (hasFP32Denormals(MF))
       return Subtarget->hasFastFMAF32() || Subtarget->hasDLInsts();
 
     // If the subtarget has v_fmac_f32, that's just as good as v_mac_f32.
     return Subtarget->hasFastFMAF32() && Subtarget->hasDLInsts();
   }
   case MVT::f64:
     return true;
SDValue SITargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
   if (Subtarget->hasLDSMisalignedBug() &&
       AS == AMDGPUAS::FLAT_ADDRESS &&
       Alignment < MemVT.getStoreSize() && MemVT.getSizeInBits() > 32) {
     return SplitVectorLoad(Op, DAG);
   }
 
   MachineFunction &MF = DAG.getMachineFunction();
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
-  // If there is a possibilty that flat instruction access scratch memory
+  // If there is a possibility that flat instruction access scratch memory
   // then we need to use the same legalization rules we use for private.
   if (AS == AMDGPUAS::FLAT_ADDRESS &&
       !Subtarget->hasMultiDwordFlatScratchAddressing())
     AS = MFI->hasFlatScratchInit() ?
          AMDGPUAS::PRIVATE_ADDRESS : AMDGPUAS::GLOBAL_ADDRESS;
 
   unsigned NumElements = MemVT.getVectorNumElements();
 
if (Subtarget->hasDS96AndDS128() &&
          MemVT.getStoreSize() == 12) &&
         allowsMisalignedMemoryAccessesImpl(MemVT.getSizeInBits(), AS,
                                            Load->getAlign()))
       return SDValue();
 
     if (NumElements > 2)
       return SplitVectorLoad(Op, DAG);
 
-    // SI has a hardware bug in the LDS / GDS boounds checking: if the base
+    // SI has a hardware bug in the LDS / GDS bounds checking: if the base
     // address is negative, then the instruction is incorrectly treated as
     // out-of-bounds even if base + offsets is in bounds. Split vectorized
     // loads here to avoid emitting ds_read2_b32. We may re-combine the
     // load later in the SILoadStoreOptimizer.
     if (Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS &&
         NumElements == 2 && MemVT.getStoreSize() == 8 &&
         Load->getAlignment() < 8) {
       return SplitVectorLoad(Op, DAG);
SDValue SITargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   if (Subtarget->hasLDSMisalignedBug() &&
       AS == AMDGPUAS::FLAT_ADDRESS &&
       Store->getAlignment() < VT.getStoreSize() && VT.getSizeInBits() > 32) {
     return SplitVectorStore(Op, DAG);
   }
 
   MachineFunction &MF = DAG.getMachineFunction();
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
-  // If there is a possibilty that flat instruction access scratch memory
+  // If there is a possibility that flat instruction access scratch memory
   // then we need to use the same legalization rules we use for private.
   if (AS == AMDGPUAS::FLAT_ADDRESS &&
       !Subtarget->hasMultiDwordFlatScratchAddressing())
     AS = MFI->hasFlatScratchInit() ?
          AMDGPUAS::PRIVATE_ADDRESS : AMDGPUAS::GLOBAL_ADDRESS;
 
   unsigned NumElements = VT.getVectorNumElements();
   if (AS == AMDGPUAS::GLOBAL_ADDRESS ||
if (Subtarget->hasDS96AndDS128() &&
          (VT.getStoreSize() == 12)) &&
         allowsMisalignedMemoryAccessesImpl(VT.getSizeInBits(), AS,
                                            Store->getAlign()))
       return SDValue();
 
     if (NumElements > 2)
       return SplitVectorStore(Op, DAG);
 
-    // SI has a hardware bug in the LDS / GDS boounds checking: if the base
+    // SI has a hardware bug in the LDS / GDS bounds checking: if the base
     // address is negative, then the instruction is incorrectly treated as
     // out-of-bounds even if base + offsets is in bounds. Split vectorized
     // stores here to avoid emitting ds_write2_b32. We may re-combine the
     // store later in the SILoadStoreOptimizer.
     if (!Subtarget->hasUsableDSOffset() &&
         NumElements == 2 && VT.getStoreSize() == 8 &&
         Store->getAlignment() < 8) {
       return SplitVectorStore(Op, DAG);
if (vectorEltWillFoldAway(Lo) || vectorEltWillFoldAway(Hi)) {
         } else if (Op.isUndef()) {
           // Handled below based on what the other operand is.
           NewElts[I] = Op;
         } else {
           NewElts[I] = DAG.getNode(ISD::FCANONICALIZE, SL, EltVT, Op);
         }
       }
 
-      // If one half is undef, and one is constant, perfer a splat vector rather
+      // If one half is undef, and one is constant, prefer a splat vector rather
       // than the normal qNaN. If it's a register, prefer 0.0 since that's
       // cheaper to use and may be free with a packed operation.
       if (NewElts[0].isUndef()) {
         if (isa<ConstantFPSDNode>(NewElts[1]))
           NewElts[0] = isa<ConstantFPSDNode>(NewElts[1]) ?
             NewElts[1]: DAG.getConstantFP(0.0f, SL, EltVT);
       }
 
SDValue SITargetLowering::performFAddCombine(SDNode *N,
   SelectionDAG &DAG = DCI.DAG;
   EVT VT = N->getValueType(0);
 
   SDLoc SL(N);
   SDValue LHS = N->getOperand(0);
   SDValue RHS = N->getOperand(1);
 
   // These should really be instruction patterns, but writing patterns with
-  // source modiifiers is a pain.
+  // source modifiers is a pain.
 
   // fadd (fadd (a, a), b) -> mad 2.0, a, b
   if (LHS.getOpcode() == ISD::FADD) {
     SDValue A = LHS.getOperand(0);
     if (A == LHS.getOperand(1)) {
       unsigned FusedOp = getFusedOpcode(DAG, N, LHS.getNode());
       if (FusedOp != 0) {
         const SDValue Two = DAG.getConstantFP(2.0, SL, VT);
SDValue SITargetLowering::performFMACombine(SDNode *N,
   SDValue FMA = N->getOperand(2);
 
   if (FMA.getOpcode() != ISD::FMA ||
       Op1.getOpcode() != ISD::FP_EXTEND ||
       Op2.getOpcode() != ISD::FP_EXTEND)
     return SDValue();
 
   // fdot2_f32_f16 always flushes fp32 denormal operand and output to zero,
-  // regardless of the denorm mode setting. Therefore, unsafe-fp-math/fp-contract
-  // is sufficient to allow generaing fdot2.
+  // regardless of the denorm mode setting. Therefore,
+  // unsafe-fp-math/fp-contract is sufficient to allow generating fdot2.
   const TargetOptions &Options = DAG.getTarget().Options;
   if (Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath ||
       (N->getFlags().hasAllowContract() &&
        FMA->getFlags().hasAllowContract())) {
     Op1 = Op1.getOperand(0);
     Op2 = Op2.getOperand(0);
     if (Op1.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
         Op2.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
void SITargetLowering::AddIMGInit(MachineInstr &MI) const {
 
   // Abandon attempt if the dst size isn't large enough
   // - this is in fact an error but this is picked up elsewhere and
   // reported correctly.
   uint32_t DstSize = TRI.getRegSizeInBits(*TII->getOpRegClass(MI, DstIdx)) / 32;
   if (DstSize < InitIdx)
     return;
 
-  // Create a register for the intialization value.
+  // Create a register for the initialization value.
   Register PrevDst = MRI.createVirtualRegister(TII->getOpRegClass(MI, DstIdx));
   unsigned NewDst = 0; // Final initialized value will be in here
 
   // If PRTStrictNull feature is enabled (the default) then initialize
   // all the result registers to 0, otherwise just the error indication
   // register (VGPRn+1)
   unsigned SizeLeft = Subtarget->usePRTStrictNull() ? InitIdx : 1;
   unsigned CurrIdx = Subtarget->usePRTStrictNull() ? 0 : (InitIdx - 1);
void SITargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
 
   MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
 
   if (TII->isVOP3(MI.getOpcode())) {
     // Make sure constant bus requirements are respected.
     TII->legalizeOperandsVOP3(MRI, MI);
 
     // Prefer VGPRs over AGPRs in mAI instructions where possible.
-    // This saves a chain-copy of registers and better ballance register
+    // This saves a chain-copy of registers and better balance register
     // use between vgpr and agpr as agpr tuples tend to be big.
     if (MI.getDesc().OpInfo) {
       unsigned Opc = MI.getOpcode();
       const SIRegisterInfo *TRI = Subtarget->getRegisterInfo();
       for (auto I : { AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src0),
                       AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src1) }) {
         if (I == -1)
           break;
if ((AS == AMDGPUAS::GLOBAL_ADDRESS || AS == AMDGPUAS::FLAT_ADDRESS) &&
 
       if (AS == AMDGPUAS::FLAT_ADDRESS)
         return AtomicExpansionKind::CmpXChg;
 
       return RMW->use_empty() ? ReportUnsafeHWInst(AtomicExpansionKind::None)
                               : AtomicExpansionKind::CmpXChg;
     }
 
-    // DS FP atomics do repect the denormal mode, but the rounding mode is fixed
-    // to round-to-nearest-even.
+    // DS FP atomics do respect the denormal mode, but the rounding mode is
+    // fixed to round-to-nearest-even.
     // The only exception is DS_ADD_F64 which never flushes regardless of mode.
     if (AS == AMDGPUAS::LOCAL_ADDRESS && Subtarget->hasLDSFPAtomicAdd()) {
       if (!Ty->isDoubleTy())
         return AtomicExpansionKind::None;
 
       if (fpModeMatchesGlobalFPAtomicMode(RMW))
         return AtomicExpansionKind::None;
 

 }
 
 // FIXME: This is a workaround for DivergenceAnalysis not understanding always
 // uniform values (as produced by the mask results of control flow intrinsics)
 // used outside of divergent blocks. The phi users need to also be treated as
 // always uniform.
 static bool hasCFUser(const Value *V, SmallPtrSet<const Value *, 16> &Visited,
                       unsigned WaveSize) {
-  // FIXME: We asssume we never cast the mask results of a control flow
+  // FIXME: We assume we never cast the mask results of a control flow
   // intrinsic.
   // Early exit if the type won't be consistent as a compile time hack.
   IntegerType *IT = dyn_cast<IntegerType>(V->getType());
   if (!IT || IT->getBitWidth() != WaveSize)
     return false;
 
   if (!isa<Instruction>(V))
     return false;
bool SITargetLowering::hasMemSDNodeUser(SDNode *N) const {
   }
   return false;
 }
 
 bool SITargetLowering::isReassocProfitable(SelectionDAG &DAG, SDValue N0,
                                            SDValue N1) const {
   if (!N0.hasOneUse())
     return false;
-  // Take care of the oportunity to keep N0 uniform
+  // Take care of the opportunity to keep N0 uniform
   if (N0->isDivergent() || !N1->isDivergent())
     return true;
   // Check if we have a good chance to form the memory access pattern with the
   // base and offset
   return (DAG.isBaseWithConstantOffset(N0) &&
           hasMemSDNodeUser(*N0->use_begin()));
 }
 
 MachineMemOperand::Flags
 SITargetLowering::getTargetMMOFlags(const Instruction &I) const {
   // Propagate metadata set by AMDGPUAnnotateUniformValues to the MMO of a load.
   if (I.getMetadata("amdgpu.noclobber"))
     return MONoClobber;
   return MachineMemOperand::MONone;
 }

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

if (MI.modifiesRegister(AMDGPU::EXEC, TRI)) {
           ScoreBrackets.hasPendingEvent(GDS_GPR_LOCK)) {
         Wait.ExpCnt = 0;
       }
     }
 
     if (MI.isCall() && callWaitsOnFunctionEntry(MI)) {
       // The function is going to insert a wait on everything in its prolog.
       // This still needs to be careful if the call target is a load (e.g. a GOT
-      // load). We also need to check WAW depenancy with saved PC.
+      // load). We also need to check WAW dependency with saved PC.
       Wait = AMDGPU::Waitcnt();
 
       int CallAddrOpIdx =
           AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::src0);
 
       if (MI.getOperand(CallAddrOpIdx).isReg()) {
         RegInterval CallAddrOpInterval =
           ScoreBrackets.getRegInterval(&MI, TII, MRI, TRI, CallAddrOpIdx);

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

if (isDS(Opc0) && isDS(Opc1)) {
     // Skip read2 / write2 variants for simplicity.
     // TODO: We should report true if the used offsets are adjacent (excluded
     // st64 versions).
     int Offset0Idx = AMDGPU::getNamedOperandIdx(Opc0, AMDGPU::OpName::offset);
     int Offset1Idx = AMDGPU::getNamedOperandIdx(Opc1, AMDGPU::OpName::offset);
     if (Offset0Idx == -1 || Offset1Idx == -1)
       return false;
 
-    // XXX - be careful of datalesss loads
+    // XXX - be careful of dataless loads
     // getNamedOperandIdx returns the index for MachineInstrs.  Since they
     // include the output in the operand list, but SDNodes don't, we need to
     // subtract the index by one.
     Offset0Idx -= get(Opc0).NumDefs;
     Offset1Idx -= get(Opc1).NumDefs;
     Offset0 = cast<ConstantSDNode>(Load0->getOperand(Offset0Idx))->getZExtValue();
     Offset1 = cast<ConstantSDNode>(Load1->getOperand(Offset1Idx))->getZExtValue();
     return true;
if (!BaseOps1.empty() && !BaseOps2.empty()) {
     const MachineInstr &SecondLdSt = *BaseOps2.front()->getParent();
     if (!memOpsHaveSameBasePtr(FirstLdSt, BaseOps1, SecondLdSt, BaseOps2))
       return false;
   } else if (!BaseOps1.empty() || !BaseOps2.empty()) {
     // If only one base op is empty, they do not have the same base ptr
     return false;
   }
 
-  // In order to avoid regester pressure, on an average, the number of DWORDS
+  // In order to avoid register pressure, on an average, the number of DWORDS
   // loaded together by all clustered mem ops should not exceed 8. This is an
   // empirical value based on certain observations and performance related
   // experiments.
   // The good thing about this heuristic is - it avoids clustering of too many
   // sub-word loads, and also avoids clustering of wide loads. Below is the
   // brief summary of how the heuristic behaves for various `LoadSize`.
   // (1) 1 <= LoadSize <= 4: cluster at max 8 mem ops
   // (2) 5 <= LoadSize <= 8: cluster at max 4 mem ops
bool SIInstrInfo::FoldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
                                 Register Reg, MachineRegisterInfo *MRI) const {
   if (!MRI->hasOneNonDBGUse(Reg))
     return false;
 
   switch (DefMI.getOpcode()) {
   default:
     return false;
   case AMDGPU::S_MOV_B64:
-    // TODO: We could fold 64-bit immediates, but this get compilicated
+    // TODO: We could fold 64-bit immediates, but this get complicated
     // when there are sub-registers.
     return false;
 
   case AMDGPU::V_MOV_B32_e32:
   case AMDGPU::S_MOV_B32:
   case AMDGPU::V_ACCVGPR_WRITE_B32_e64:
     break;
   }
if (Opc == AMDGPU::V_MAD_F32_e64 || Opc == AMDGPU::V_MAC_F32_e64 ||
     bool IsF32 = Opc == AMDGPU::V_MAD_F32_e64 || Opc == AMDGPU::V_MAC_F32_e64 ||
                  Opc == AMDGPU::V_FMA_F32_e64 || Opc == AMDGPU::V_FMAC_F32_e64;
     bool IsFMA = Opc == AMDGPU::V_FMA_F32_e64 || Opc == AMDGPU::V_FMAC_F32_e64 ||
                  Opc == AMDGPU::V_FMA_F16_e64 || Opc == AMDGPU::V_FMAC_F16_e64;
     MachineOperand *Src1 = getNamedOperand(UseMI, AMDGPU::OpName::src1);
     MachineOperand *Src2 = getNamedOperand(UseMI, AMDGPU::OpName::src2);
 
     // Multiplied part is the constant: Use v_madmk_{f16, f32}.
-    // We should only expect these to be on src0 due to canonicalizations.
+    // We should only expect these to be on src0 due to canonicalization.
     if (Src0->isReg() && Src0->getReg() == Reg) {
       if (!Src1->isReg() || RI.isSGPRClass(MRI->getRegClass(Src1->getReg())))
         return false;
 
       if (!Src2->isReg() || RI.isSGPRClass(MRI->getRegClass(Src2->getReg())))
         return false;
 
       unsigned NewOpc =
bool SIInstrInfo::verifyInstruction(const MachineInstr &MI,
   if (isSDWA(MI)) {
     if (!ST.hasSDWA()) {
       ErrInfo = "SDWA is not supported on this target";
       return false;
     }
 
     int DstIdx = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::vdst);
 
-    const int OpIndicies[] = { DstIdx, Src0Idx, Src1Idx, Src2Idx };
+    const int OpIndices[] = {DstIdx, Src0Idx, Src1Idx, Src2Idx};
 
-    for (int OpIdx: OpIndicies) {
+    for (int OpIdx : OpIndices) {
       if (OpIdx == -1)
         continue;
       const MachineOperand &MO = MI.getOperand(OpIdx);
 
       if (!ST.hasSDWAScalar()) {
         // Only VGPRS on VI
         if (!MO.isReg() || !RI.hasVGPRs(RI.getRegClassForReg(MRI, MO.getReg()))) {
           ErrInfo = "Only VGPRs allowed as operands in SDWA instructions on VI";
for (int OpIdx : OpIndices) {
             return false;
           }
         }
       }
     }
 
     SGPRUsed = findImplicitSGPRRead(MI);
     if (SGPRUsed != AMDGPU::NoRegister) {
-      // Implicit uses may safely overlap true overands
+      // Implicit uses may safely overlap true operands
       if (llvm::all_of(SGPRsUsed, [this, SGPRUsed](unsigned SGPR) {
             return !RI.regsOverlap(SGPRUsed, SGPR);
           })) {
         ++ConstantBusCount;
         SGPRsUsed.push_back(SGPRUsed);
       }
     }
 
const TargetRegisterClass *SIInstrInfo::getRegClass(const MCInstrDesc &TID,
     const MachineFunction &MF)
   const {
   if (OpNum >= TID.getNumOperands())
     return nullptr;
   auto RegClass = TID.OpInfo[OpNum].RegClass;
   bool IsAllocatable = false;
   if (TID.TSFlags & (SIInstrFlags::DS | SIInstrFlags::FLAT)) {
     // vdst and vdata should be both VGPR or AGPR, same for the DS instructions
-    // with two data operands. Request register class constainted to VGPR only
+    // with two data operands. Request register class constrained to VGPR only
     // of both operands present as Machine Copy Propagation can not check this
     // constraint and possibly other passes too.
     //
     // The check is limited to FLAT and DS because atomics in non-flat encoding
     // have their vdst and vdata tied to be the same register.
     const int VDstIdx = AMDGPU::getNamedOperandIdx(TID.Opcode,
                                                    AMDGPU::OpName::vdst);
     const int DataIdx = AMDGPU::getNamedOperandIdx(TID.Opcode,
if (!VAddrDef || VAddrDef->getOpcode() != AMDGPU::V_MOV_B32_e32 ||
         !VAddrDef->getOperand(1).isImm() ||
         VAddrDef->getOperand(1).getImm() != 0)
       return false;
   }
 
   const MCInstrDesc &NewDesc = get(NewOpc);
   Inst.setDesc(NewDesc);
 
-  // Callers expect interator to be valid after this call, so modify the
+  // Callers expect iterator to be valid after this call, so modify the
   // instruction in place.
   if (OldVAddrIdx == NewVAddrIdx) {
     MachineOperand &NewVAddr = Inst.getOperand(NewVAddrIdx);
     // Clear use list from the old vaddr holding a zero register.
     MRI.removeRegOperandFromUseList(&NewVAddr);
     MRI.moveOperands(&NewVAddr, &SAddr, 1);
     Inst.RemoveOperand(OldSAddrIdx);
     // Update the use list with the pointer we have just moved from vaddr to
-    // saddr poisition. Otherwise new vaddr will be missing from the use list.
+    // saddr position. Otherwise new vaddr will be missing from the use list.
     MRI.removeRegOperandFromUseList(&NewVAddr);
     MRI.addRegOperandToUseList(&NewVAddr);
   } else {
     assert(OldSAddrIdx == NewVAddrIdx);
 
     if (OldVAddrIdx >= 0) {
       int NewVDstIn = AMDGPU::getNamedOperandIdx(NewOpc,
                                                  AMDGPU::OpName::vdst_in);
for (unsigned Idx = 0; Idx < NumSubRegs; Idx += 2) {
     auto Cmp =
         BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_CMP_EQ_U64_e64), NewCondReg)
             .addReg(CurReg);
     if (NumSubRegs <= 2)
       Cmp.addReg(VRsrc);
     else
       Cmp.addReg(VRsrc, VRsrcUndef, TRI->getSubRegFromChannel(Idx, 2));
 
-    // Combine the comparision results with AND.
+    // Combine the comparison results with AND.
     if (CondReg == AMDGPU::NoRegister) // First.
       CondReg = NewCondReg;
     else { // If not the first, we create an AND.
       Register AndReg = MRI.createVirtualRegister(BoolXExecRC);
       BuildMI(LoopBB, I, DL, TII.get(AndOpc), AndReg)
               .addReg(CondReg)
               .addReg(NewCondReg);
       CondReg = AndReg;
int RsrcIdx =
       AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::srsrc);
   if (RsrcIdx != -1) {
     // We have an MUBUF instruction
     MachineOperand *Rsrc = &MI.getOperand(RsrcIdx);
     unsigned RsrcRC = get(MI.getOpcode()).OpInfo[RsrcIdx].RegClass;
     if (RI.getCommonSubClass(MRI.getRegClass(Rsrc->getReg()),
                              RI.getRegClass(RsrcRC))) {
       // The operands are legal.
-      // FIXME: We may need to legalize operands besided srsrc.
+      // FIXME: We may need to legalize operands besides srsrc.
       return CreatedBB;
     }
 
     // Legalize a VGPR Rsrc.
     //
     // If the instruction is _ADDR64, we can avoid a waterfall by extracting
     // the base pointer from the VGPR Rsrc, adding it to the VAddr, then using
     // a zero-value SRsrc.
if (VAddr && AMDGPU::getIfAddr64Inst(MI.getOpcode()) != -1) {
       std::tie(RsrcPtr, NewSRsrc) = extractRsrcPtr(*this, MI, *Rsrc);
 
       Register NewVAddr = MRI.createVirtualRegister(&AMDGPU::VReg_64RegClass);
       MachineOperand *VData = getNamedOperand(MI, AMDGPU::OpName::vdata);
       MachineOperand *Offset = getNamedOperand(MI, AMDGPU::OpName::offset);
       MachineOperand *SOffset = getNamedOperand(MI, AMDGPU::OpName::soffset);
       unsigned Addr64Opcode = AMDGPU::getAddr64Inst(MI.getOpcode());
 
-      // Atomics rith return have have an additional tied operand and are
+      // Atomics with return have an additional tied operand and are
       // missing some of the special bits.
       MachineOperand *VDataIn = getNamedOperand(MI, AMDGPU::OpName::vdata_in);
       MachineInstr *Addr64;
 
       if (!VDataIn) {
         // Regular buffer load / store.
         MachineInstrBuilder MIB =
             BuildMI(MBB, MI, MI.getDebugLoc(), get(Addr64Opcode))
BuildMI(MBB, MII, DL, get(AMDGPU::V_XNOR_B32_e64), NewDest)
       .add(Src1);
 
     MRI.replaceRegWith(Dest.getReg(), NewDest);
     addUsersToMoveToVALUWorklist(NewDest, MRI, Worklist);
   } else {
     // Using the identity !(x ^ y) == (!x ^ y) == (x ^ !y), we can
     // invert either source and then perform the XOR. If either source is a
     // scalar register, then we can leave the inversion on the scalar unit to
-    // acheive a better distrubution of scalar and vector instructions.
+    // achieve a better distribution of scalar and vector instructions.
     bool Src0IsSGPR = Src0.isReg() &&
                       RI.isSGPRClass(MRI.getRegClass(Src0.getReg()));
     bool Src1IsSGPR = Src1.isReg() &&
                       RI.isSGPRClass(MRI.getRegClass(Src1.getReg()));
     MachineInstr *Xor;
     Register Temp = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
     Register NewDest = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
 
void SIInstrInfo::splitScalar64BitAddSub(SetVectorType &Worklist,
 
   MRI.replaceRegWith(Dest.getReg(), FullDestReg);
 
   // Try to legalize the operands in case we need to swap the order to keep it
   // valid.
   legalizeOperands(*LoHalf, MDT);
   legalizeOperands(*HiHalf, MDT);
 
-  // Move all users of this moved vlaue.
+  // Move all users of this moved value.
   addUsersToMoveToVALUWorklist(FullDestReg, MRI, Worklist);
 }
 
 void SIInstrInfo::splitScalar64BitBinaryOp(SetVectorType &Worklist,
                                            MachineInstr &Inst, unsigned Opcode,
                                            MachineDominatorTree *MDT) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
BuildMI(MBB, MII, DL, get(TargetOpcode::REG_SEQUENCE), FullDestReg)
     .addReg(DestSub1)
     .addImm(AMDGPU::sub1);
 
   MRI.replaceRegWith(Dest.getReg(), FullDestReg);
 
   Worklist.insert(&LoHalf);
   Worklist.insert(&HiHalf);
 
-  // Move all users of this moved vlaue.
+  // Move all users of this moved value.
   addUsersToMoveToVALUWorklist(FullDestReg, MRI, Worklist);
 }
 
 void SIInstrInfo::splitScalar64BitXnor(SetVectorType &Worklist,
                                        MachineInstr &Inst,
                                        MachineDominatorTree *MDT) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
MachineOperand SrcRegSub1 = buildExtractSubRegOrImm(MII, MRI, Src, SrcRC,
                                                       AMDGPU::sub1, SrcSubRC);
 
   BuildMI(MBB, MII, DL, InstDesc, MidReg).add(SrcRegSub0).addImm(0);
 
   BuildMI(MBB, MII, DL, InstDesc, ResultReg).add(SrcRegSub1).addReg(MidReg);
 
   MRI.replaceRegWith(Dest.getReg(), ResultReg);
 
-  // We don't need to legalize operands here. src0 for etiher instruction can be
+  // We don't need to legalize operands here. src0 for either instruction can be
   // an SGPR, and the second input is unused or determined here.
   addUsersToMoveToVALUWorklist(ResultReg, MRI, Worklist);
 }
 
 void SIInstrInfo::splitScalar64BitBFE(SetVectorType &Worklist,
                                       MachineInstr &Inst) const {
   MachineBasicBlock &MBB = *Inst.getParent();
   MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();

 // SCC must be changed to an instruction that defines VCC. This function makes
 // sure that the instruction that defines SCC is added to the moveToVALU
 // worklist.
 void SIInstrInfo::addSCCDefsToVALUWorklist(MachineOperand &Op,
                                            SetVectorType &Worklist) const {
   assert(Op.isReg() && Op.getReg() == AMDGPU::SCC && Op.isUse());
 
   MachineInstr *SCCUseInst = Op.getParent();
-  // Look for a preceeding instruction that either defines VCC or SCC. If VCC
+  // Look for a preceding instruction that either defines VCC or SCC. If VCC
   // then there is nothing to do because the defining instruction has been
   // converted to a VALU already. If SCC then that instruction needs to be
   // converted to a VALU.
   for (MachineInstr &MI :
        make_range(std::next(MachineBasicBlock::reverse_iterator(SCCUseInst)),
                   SCCUseInst->getParent()->rend())) {
     if (MI.modifiesRegister(AMDGPU::VCC, &RI))
       break;
bool SIInstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, Register SrcReg,
   if (!SrcReg || SrcReg.isPhysical())
     return false;
 
   if (SrcReg2 && !getFoldableImm(SrcReg2, *MRI, CmpValue))
     return false;
 
   const auto optimizeCmpAnd = [&CmpInstr, SrcReg, CmpValue, MRI,
                                this](int64_t ExpectedValue, unsigned SrcSize,
-                                     bool IsReversable, bool IsSigned) -> bool {
+                                     bool IsReversible, bool IsSigned) -> bool {
     // s_cmp_eq_u32 (s_and_b32 $src, 1 << n), 1 << n => s_and_b32 $src, 1 << n
     // s_cmp_eq_i32 (s_and_b32 $src, 1 << n), 1 << n => s_and_b32 $src, 1 << n
     // s_cmp_ge_u32 (s_and_b32 $src, 1 << n), 1 << n => s_and_b32 $src, 1 << n
     // s_cmp_ge_i32 (s_and_b32 $src, 1 << n), 1 << n => s_and_b32 $src, 1 << n
     // s_cmp_eq_u64 (s_and_b64 $src, 1 << n), 1 << n => s_and_b64 $src, 1 << n
     // s_cmp_lg_u32 (s_and_b32 $src, 1 << n), 0 => s_and_b32 $src, 1 << n
     // s_cmp_lg_i32 (s_and_b32 $src, 1 << n), 0 => s_and_b32 $src, 1 << n
     // s_cmp_gt_u32 (s_and_b32 $src, 1 << n), 0 => s_and_b32 $src, 1 << n
const auto optimizeCmpAnd = [&CmpInstr, SrcReg, CmpValue, MRI,
     unsigned BitNo = countTrailingZeros((uint64_t)Mask);
     if (IsSigned && BitNo == SrcSize - 1)
       return false;
 
     ExpectedValue <<= BitNo;
 
     bool IsReversedCC = false;
     if (CmpValue != ExpectedValue) {
-      if (!IsReversable)
+      if (!IsReversible)
         return false;
       IsReversedCC = CmpValue == (ExpectedValue ^ Mask);
       if (!IsReversedCC)
         return false;
     }
 
     Register DefReg = Def->getOperand(0).getReg();
     if (IsReversedCC && !MRI->hasOneNonDBGUse(DefReg))

llvm/lib/Target/AMDGPU/SIInstrInfo.td

class getAsmDPP8 <bit HasDst, int NumSrcArgs, bit HasModifiers, ValueType DstVT = i32>
   : getAsmDPP<HasDst, NumSrcArgs, HasModifiers, DstVT> {
   let ret = dst#args#" $dpp8$fi";
 }
 
 
 class getAsmSDWA <bit HasDst, int NumSrcArgs, ValueType DstVT = i32> {
   string dst = !if(HasDst,
                    !if(!eq(DstVT.Size, 1),
-                       " vcc", // use vcc token as dst for VOPC instructioins
+                       " vcc", // use vcc token as dst for VOPC instructions
                        "$vdst"),
                     "");
   string src0 = "$src0_modifiers";
   string src1 = "$src1_modifiers";
   string args = !if(!eq(NumSrcArgs, 0),
                     "",
                     !if(!eq(NumSrcArgs, 1),
                         ", "#src0#"$clamp",

llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

bool SILoadStoreOptimizer::promoteConstantOffsetToImm(
   //   addr3 = &a + 8192;   load3 = load(addr3,  0)
   //   addr4 = &a + 10240;  load4 = load(addr4,  0)
   //   addr5 = &a + 12288;  load5 = load(addr5,  0)
   //
   // Starting from the first load, the optimization will try to find a new base
   // from which (&a + 4096) has 13 bit distance. Both &a + 6144 and &a + 8192
   // has 13bit distance from &a + 4096. The heuristic considers &a + 8192
   // as the new-base(anchor) because of the maximum distance which can
-  // accomodate more intermediate bases presumeably.
+  // accommodate more intermediate bases presumably.
   //
   // Step3: move (&a + 8192) above load1. Compute and promote offsets from
   // (&a + 8192) for load1, load2, load4.
   //   addr = &a + 8192
   //   load1 = load(addr,       -4096)
   //   load2 = load(addr,       -2048)
   //   load3 = load(addr,       0)
   //   load4 = load(addr,       2048)

llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp

void SILowerControlFlow::findMaskOperands(MachineInstr &MI, unsigned OpNo,
   }
 
   MachineInstr *Def = MRI->getUniqueVRegDef(Op.getReg());
   if (!Def || Def->getParent() != MI.getParent() ||
       !(Def->isFullCopy() || (Def->getOpcode() == MI.getOpcode())))
     return;
 
   // Make sure we do not modify exec between def and use.
-  // A copy with implcitly defined exec inserted earlier is an exclusion, it
+  // A copy with implicitly defined exec inserted earlier is an exclusion, it
   // does not really modify exec.
   for (auto I = Def->getIterator(); I != MI.getIterator(); ++I)
     if (I->modifiesRegister(AMDGPU::EXEC, TRI) &&
         !(I->isCopy() && I->getOperand(0).getReg() != Exec))
       return;
 
   for (const auto &SrcOp : Def->explicit_operands())
     if (SrcOp.isReg() && SrcOp.isUse() &&
void SILowerControlFlow::combineMasks(MachineInstr &MI) {
   Register Reg = MI.getOperand(OpToReplace).getReg();
   MI.RemoveOperand(OpToReplace);
   MI.addOperand(Ops[UniqueOpndIdx]);
   if (MRI->use_empty(Reg))
     MRI->getUniqueVRegDef(Reg)->eraseFromParent();
 }
 
 void SILowerControlFlow::optimizeEndCf() {
-  // If the only instruction immediately following this END_CF is an another
+  // If the only instruction immediately following this END_CF is another
   // END_CF in the only successor we can avoid emitting exec mask restore here.
   if (!EnableOptimizeEndCf)
     return;
 
   for (MachineInstr *MI : reverse(LoweredEndCf)) {
     MachineBasicBlock &MBB = *MI->getParent();
     auto Next =
       skipIgnoreExecInstsTrivialSucc(MBB, std::next(MI->getIterator()));

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

#endif
       // We need to preserve inactive lanes, so always save, even caller-save
       // registers.
       if (!isEntryFunction()) {
         SpillFI = FrameInfo.CreateSpillStackObject(4, Align(4));
       }
 
       SpillVGPRs.push_back(SGPRSpillVGPR(LaneVGPR, SpillFI));
 
-      // Add this register as live-in to all blocks to avoid machine verifer
+      // Add this register as live-in to all blocks to avoid machine verifier
       // complaining about use of an undefined physical register.
       for (MachineBasicBlock &BB : MF)
         BB.addLiveIn(LaneVGPR);
     } else {
       LaneVGPR = SpillVGPRs.back().VGPR;
     }
 
     SpillLanes.push_back(SpilledReg(LaneVGPR, VGPRIndex));

llvm/lib/Target/AMDGPU/SIMachineScheduler.cpp

 // locally inside the blocks where it takes care of low latencies, and then
 // chooses the order of the blocks by taking care of high latencies.
 // Dividing the instructions into blocks helps control keeping register
 // usage low.
 //
 // First the instructions are put into blocks.
 //   We want the blocks help control register usage and hide high latencies
 //   later. To help control register usage, we typically want all local
-//   computations, when for example you create a result that can be comsummed
+//   computations, when for example you create a result that can be consumed
 //   right away, to be contained in a block. Block inputs and outputs would
 //   typically be important results that are needed in several locations of
 //   the shader. Since we do want blocks to help hide high latencies, we want
 //   the instructions inside the block to have a minimal set of dependencies
 //   on high latencies. It will make it easy to pick blocks to hide specific
 //   high latencies.
 //   The block creation algorithm is divided into several steps, and several
 //   variants can be tried during the scheduling process.

 // latencies.
 //
 // Actually when talking about 'low latency' or 'high latency' it includes
 // both the latency to get the cache (or global mem) data go to the register,
 // and the bandwidth limitations.
 // Increasing the number of active wavefronts helps hide the former, but it
 // doesn't solve the latter, thus why even if wavefront count is high, we have
 // to try have as many instructions hiding high latencies as possible.
-// The OpenCL doc says for example latency of 400 cycles for a global mem access,
-// which is hidden by 10 instructions if the wavefront count is 10.
+// The OpenCL doc says for example latency of 400 cycles for a global mem
+// access, which is hidden by 10 instructions if the wavefront count is 10.
 
 // Some figures taken from AMD docs:
 // Both texture and constant L1 caches are 4-way associative with 64 bytes
 // lines.
 // Constant cache is shared with 4 CUs.
 // For texture sampling, the address generation unit receives 4 texture
 // addresses per cycle, thus we could expect texture sampling latency to be
 // equivalent to 4 instructions in the very best case (a VGPR is 64 work items,
void SIScheduleBlock::initRegPressure(MachineBasicBlock::iterator BeginBlock,
   // We want LiveOutRegs to contain only Regs whose content will be read after
   // in another block, and whose content was written in the current block,
   // that is we want it to get 1, 3, 5, 7
   // Since we made the MIs of a block to be packed all together before
   // scheduling, then the LiveIntervals were correct, and the RPTracker was
   // able to correctly handle 5 vs 6, 2 vs 3.
   // (Note: This is not sufficient for RPTracker to not do mistakes for case 4)
   // The RPTracker's LiveOutRegs has 1, 3, (some correct or incorrect)4, 5, 7
-  // Comparing to LiveInRegs is not sufficient to differenciate 4 vs 5, 7
+  // Comparing to LiveInRegs is not sufficient to differentiate 4 vs 5, 7
   // The use of findDefBetween removes the case 4.
   for (const auto &RegMaskPair : RPTracker.getPressure().LiveOutRegs) {
     Register Reg = RegMaskPair.RegUnit;
     if (Reg.isVirtual() &&
         isDefBetween(Reg, LIS->getInstructionIndex(*BeginBlock).getRegSlot(),
                      LIS->getInstructionIndex(*EndBlock).getRegSlot(), MRI,
                      LIS)) {
       LiveOutRegs.insert(Reg);
void SIScheduleBlock::schedule(MachineBasicBlock::iterator BeginBlock,
   while (!TopReadySUs.empty()) {
     SUnit *SU = pickNode();
     ScheduledSUnits.push_back(SU);
     TopRPTracker.setPos(SU->getInstr());
     TopRPTracker.advance();
     nodeScheduled(SU);
   }
 
-  // TODO: compute InternalAdditionnalPressure.
+  // TODO: compute InternalAdditionalPressure.
   InternalAdditionalPressure.resize(TopPressure.MaxSetPressure.size());
 
   // Check everything is right.
 #ifndef NDEBUG
   assert(SUnits.size() == ScheduledSUnits.size() &&
             TopReadySUs.empty());
   for (SUnit* SU : SUnits) {
     assert(SU->isScheduled &&
if (DAG->IsHighLatencySU[SU.NodeNum]) {
       // enforced (via links).
       // Instead, look at the dependencies between the
       // high latency instructions and deduce if it is
       // a data dependency or not.
       for (unsigned j : FormingGroup) {
         bool HasSubGraph;
         std::vector<int> SubGraph;
         // By construction (topological order), if SU and
-        // DAG->SUnits[j] are linked, DAG->SUnits[j] is neccessary
+        // DAG->SUnits[j] are linked, DAG->SUnits[j] is necessary
         // in the parent graph of SU.
 #ifndef NDEBUG
         SubGraph = DAG->GetTopo()->GetSubGraph(SU, DAG->SUnits[j],
                                                HasSubGraph);
         assert(!HasSubGraph);
 #endif
         SubGraph = DAG->GetTopo()->GetSubGraph(DAG->SUnits[j], SU,
                                                HasSubGraph);
if (SIInstrInfo::isEXP(*SU.getInstr())) {
       // Check the EXP can be added to the group safely,
       // ie without needing any other instruction.
       // The EXP is allowed to depend on other EXP
       // (they will be in the same group).
       for (unsigned j : ExpGroup) {
         bool HasSubGraph;
         std::vector<int> SubGraph;
         // By construction (topological order), if SU and
-        // DAG->SUnits[j] are linked, DAG->SUnits[j] is neccessary
+        // DAG->SUnits[j] are linked, DAG->SUnits[j] is necessary
         // in the parent graph of SU.
 #ifndef NDEBUG
         SubGraph = DAG->GetTopo()->GetSubGraph(SU, DAG->SUnits[j],
                                                HasSubGraph);
         assert(!HasSubGraph);
 #endif
         SubGraph = DAG->GetTopo()->GetSubGraph(DAG->SUnits[j], SU,
                                                HasSubGraph);
         if (!HasSubGraph)
           continue; // No dependencies between each other
 
         // SubGraph contains all the instructions required
         // between EXP SUnits[j] and EXP SU.
         for (unsigned k : SubGraph) {
           if (!SIInstrInfo::isEXP(*DAG->SUnits[k].getInstr()))
             // Other instructions than EXP would be required in the group.
-            // Abort the groupping.
+            // Abort the grouping.
             return;
         }
       }
 
       ExpGroup.push_back(SUNum);
     }
   }
 

llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp

enum class SIAtomicScope {
   SINGLETHREAD,
   WAVEFRONT,
   WORKGROUP,
   AGENT,
   SYSTEM
 };
 
 /// The distinct address spaces supported by the AMDGPU target for
-/// atomic memory operation. Can be ORed toether.
+/// atomic memory operation. Can be ORed together.
 enum class SIAtomicAddrSpace {
   NONE = 0u,
   GLOBAL = 1u << 0,
   LDS = 1u << 1,
   SCRATCH = 1u << 2,
   GDS = 1u << 3,
   OTHER = 1u << 4,
 
case SIAtomicScope::WORKGROUP:
       // synchronizing with global/GDS memory as LDS operations could be
       // reordered with respect to later global/GDS memory operations of the
       // same wave.
       LGKMCnt |= IsCrossAddrSpaceOrdering;
       break;
     case SIAtomicScope::WAVEFRONT:
     case SIAtomicScope::SINGLETHREAD:
       // The LDS keeps all memory operations in order for
-      // the same wavesfront.
+      // the same wavefront.
       break;
     default:
       llvm_unreachable("Unsupported synchronization scope");
     }
   }
 
   if ((AddrSpace & SIAtomicAddrSpace::GDS) != SIAtomicAddrSpace::NONE) {
     switch (Scope) {
case SIAtomicScope::WORKGROUP:
       // synchronizing with global/GDS memory as LDS operations could be
       // reordered with respect to later global/GDS memory operations of the
       // same wave.
       LGKMCnt |= IsCrossAddrSpaceOrdering;
       break;
     case SIAtomicScope::WAVEFRONT:
     case SIAtomicScope::SINGLETHREAD:
       // The LDS keeps all memory operations in order for
-      // the same wavesfront.
+      // the same wavefront.
       break;
     default:
       llvm_unreachable("Unsupported synchronization scope");
     }
   }
 
   if ((AddrSpace & SIAtomicAddrSpace::GDS) != SIAtomicAddrSpace::NONE) {
     switch (Scope) {

llvm/lib/Target/AMDGPU/SIOptimizeExecMaskingPreRA.cpp

if (MBB.succ_empty()) {
           if (Pred->succ_size() == 1)
             Blocks.push_back(Pred);
         }
       }
       continue;
     }
 
     // If the only user of a logical operation is move to exec, fold it now
-    // to prevent forming of saveexec. I.e:
+    // to prevent forming of saveexec. I.e.:
     //
     //    %0:sreg_64 = COPY $exec
     //    %1:sreg_64 = S_AND_B64 %0:sreg_64, %2:sreg_64
     // =>
     //    %1 = S_AND_B64 $exec, %2:sreg_64
     unsigned ScanThreshold = 10;
     for (auto I = MBB.rbegin(), E = MBB.rend(); I != E
          && ScanThreshold--; ++I) {

llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp

if (TII->getNamedOperand(*MI, AMDGPU::OpName::src0) == SrcOp) {
     }
   } else if (TII->getNamedOperand(*MI, AMDGPU::OpName::src1) == SrcOp) {
     if (auto *Mod = TII->getNamedOperand(*MI, AMDGPU::OpName::src1_modifiers)) {
       Mods = Mod->getImm();
     }
   }
   if (Abs || Neg) {
     assert(!Sext &&
-           "Float and integer src modifiers can't be set simulteniously");
+           "Float and integer src modifiers can't be set simultaneously");
     Mods |= Abs ? SISrcMods::ABS : 0u;
     Mods ^= Neg ? SISrcMods::NEG : 0u;
   } else if (Sext) {
     Mods |= SISrcMods::SEXT;
   }
 
   return Mods;
 }
if (DstUnused &&
     SDWAInst.add(Tied);
     SDWAInst->tieOperands(PreserveDstIdx, SDWAInst->getNumOperands() - 1);
   }
 
   // Apply all sdwa operand patterns.
   bool Converted = false;
   for (auto &Operand : SDWAOperands) {
     LLVM_DEBUG(dbgs() << *SDWAInst << "\nOperand: " << *Operand);
-    // There should be no intesection between SDWA operands and potential MIs
+    // There should be no intersection between SDWA operands and potential MIs
     // e.g.:
     // v_and_b32 v0, 0xff, v1 -> src:v1 sel:BYTE_0
     // v_and_b32 v2, 0xff, v0 -> src:v0 sel:BYTE_0
     // v_add_u32 v3, v4, v2
     //
-    // In that example it is possible that we would fold 2nd instruction into 3rd
-    // (v_add_u32_sdwa) and then try to fold 1st instruction into 2nd (that was
-    // already destroyed). So if SDWAOperand is also a potential MI then do not
-    // apply it.
+    // In that example it is possible that we would fold 2nd instruction into
+    // 3rd (v_add_u32_sdwa) and then try to fold 1st instruction into 2nd (that
+    // was already destroyed). So if SDWAOperand is also a potential MI then do
+    // not apply it.
     if (PotentialMatches.count(Operand->getParentInst()) == 0)
       Converted |= Operand->convertToSDWA(*SDWAInst, TII);
   }
   if (Converted) {
     ConvertedInstructions.push_back(SDWAInst);
   } else {
     SDWAInst->eraseFromParent();
     return false;

llvm/lib/Target/AMDGPU/SIPreEmitPeephole.cpp

for (; M != E; ++M) {
         return Changed;
       ReadsSreg |= M->readsRegister(SReg, TRI);
     }
     if (M == E || !M->isMoveImmediate() || !M->getOperand(1).isImm() ||
         (M->getOperand(1).getImm() != -1 && M->getOperand(1).getImm() != 0))
       return Changed;
     MaskValue = M->getOperand(1).getImm();
     // First if sreg is only used in the AND instruction fold the immediate
-    // into into the AND.
+    // into the AND.
     if (!ReadsSreg && Op2.isKill()) {
       A->getOperand(2).ChangeToImmediate(MaskValue);
       M->eraseFromParent();
     }
   } else if (Op2.isImm()) {
     MaskValue = Op2.getImm();
   } else {
     llvm_unreachable("Op2 must be register or immediate");

llvm/lib/Target/AMDGPU/SIRegisterInfo.td

class RegSeqNames<int last_reg, int stride, int size, string prefix,
   int end_reg = !add(start, size, -1);
   list<string> ret =
     !if(!le(end_reg, last_reg),
         !listconcat([prefix # "[" # start # ":" # end_reg # "]"],
                     RegSeqNames<last_reg, stride, size, prefix, next>.ret),
                     []);
 }
 
-// Generates list of dags for register tupless.
+// Generates list of dags for register tuples.
 class RegSeqDags<RegisterClass RC, int last_reg, int stride, int size,
                 int start = 0> {
   dag trunc_rc = (trunc RC,
                   !if(!and(!eq(stride, 1), !eq(start, 0)),
                       !sub(!add(last_reg, 2), size),
                       !add(last_reg, 1)));
   list<dag> ret =
     !if(!lt(start, size),

llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp

 // =>
 //
 // mov t, x (t is potentially dead and move eliminated)
 // v_swap_b32 x, y
 //
 // Returns next valid instruction pointer if was able to create v_swap_b32.
 //
 // This shall not be done too early not to prevent possible folding which may
-// remove matched moves, and this should prefereably be done before RA to
+// remove matched moves, and this should preferably be done before RA to
 // release saved registers and also possibly after RA which can insert copies
 // too.
 //
-// This is really just a generic peephole that is not a canocical shrinking,
+// This is really just a generic peephole that is not a canonical shrinking,
 // although requirements match the pass placement and it reduces code size too.
 static MachineInstr* matchSwap(MachineInstr &MovT, MachineRegisterInfo &MRI,
                                const SIInstrInfo *TII) {
   assert(MovT.getOpcode() == AMDGPU::V_MOV_B32_e32 ||
          MovT.getOpcode() == AMDGPU::COPY);
 
   Register T = MovT.getOperand(0).getReg();
   unsigned Tsub = MovT.getOperand(0).getSubReg();

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

MachineInstr *EarlyTermMI =
       BuildMI(MBB, MI, DL, TII->get(AMDGPU::SI_EARLY_TERMINATE_SCC0));
 
   // In the case we got this far some lanes are still live,
   // update EXEC to deactivate lanes as appropriate.
   MachineInstr *NewTerm;
   MachineInstr *WQMMaskMI = nullptr;
   Register LiveMaskWQM;
   if (IsDemote) {
-    // Demotes deactive quads with only helper lanes
+    // Demote - deactivate quads with only helper lanes
     LiveMaskWQM = MRI->createVirtualRegister(TRI->getBoolRC());
     WQMMaskMI =
         BuildMI(MBB, MI, DL, TII->get(WQMOpc), LiveMaskWQM).addReg(LiveMaskReg);
     NewTerm = BuildMI(MBB, MI, DL, TII->get(AndOpc), Exec)
                   .addReg(Exec)
                   .addReg(LiveMaskWQM);
   } else {
-    // Kills deactivate lanes
+    // Kill - deactivate lanes no longer in live mask
     if (Op.isImm()) {
       unsigned MovOpc = ST->isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
       NewTerm = BuildMI(MBB, &MI, DL, TII->get(MovOpc), Exec).addImm(0);
     } else if (!IsWQM) {
       NewTerm = BuildMI(MBB, &MI, DL, TII->get(AndOpc), Exec)
                     .addReg(Exec)
                     .addReg(LiveMaskReg);
     } else {

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

 LLVM_READNONE
 inline bool isLegal64BitDPPControl(unsigned DC) {
   return DC >= DPP::ROW_NEWBCAST_FIRST && DC <= DPP::ROW_NEWBCAST_LAST;
 }
 
 /// \returns true if the intrinsic is divergent
 bool isIntrinsicSourceOfDivergence(unsigned IntrID);
 
-// Track defaults for fields in the MODE registser.
+// Track defaults for fields in the MODE register.
 struct SIModeRegisterDefaults {
   /// Floating point opcodes that support exception flag gathering quiet and
   /// propagate signaling NaN inputs per IEEE 754-2008. Min_dx10 and max_dx10
   /// become IEEE 754- 2008 compliant due to signaling NaN propagation and
   /// quieting.
   bool IEEE : 1;
 
   /// Used by the vector ALU to force DX10-style treatment of NaNs: when set,

llvm/lib/Target/AMDGPU/Utils/AMDGPUMemoryUtils.h

 Align getAlign(DataLayout const &DL, const GlobalVariable *GV);
 
 std::vector<GlobalVariable *> findVariablesToLower(Module &M,
                                                    const Function *F = nullptr);
 
 /// Replace all uses of constant \p C with instructions in \p F.
 void replaceConstantUsesInFunction(ConstantExpr *C, const Function *F);
 
-/// Given a \p Def clobbering a load from \p Ptr accroding to the MSSA check
-/// if this is actually a memory update or an artifical clobber to facilitate
+/// Given a \p Def clobbering a load from \p Ptr according to the MSSA check
+/// if this is actually a memory update or an artificial clobber to facilitate
 /// ordering constraints.
 bool isReallyAClobber(const Value *Ptr, MemoryDef *Def, AAResults *AA);
 
 /// Check is a \p Load is clobbered in its function.
 bool isClobberedInFunction(const LoadInst *Load, MemorySSA *MSSA,
                            AAResults *AA);
 
 } // end namespace AMDGPU
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_AMDGPU_UTILS_AMDGPUMEMORYUTILS_H

llvm/lib/Target/AMDGPU/VOP3PInstructions.td

class VOPProfileMAI<VOPProfile P, RegisterOperand _SrcRC, RegisterOperand _DstRC,
   let HasClamp = 0;
   let HasIntClamp = 0;
   let HasOMod = 0;
   let HasModifiers = 0;
   let Asm64 = "$vdst, $src0, $src1, $src2$cbsz$abid$blgp";
   let Ins64 = (ins Src0RC64:$src0, Src1RC64:$src1, Src2RC64:$src2, cbsz:$cbsz, abid:$abid, blgp:$blgp);
   // Dst and SrcC cannot partially overlap if SrcC/Dst is bigger than 4 VGPRs.
   // We then create two versions of the instruction: with tied dst and src2
-  // and with the eralyclobber flag on the dst. This is strciter than the
+  // and with the earlyclobber flag on the dst. This is stricter than the
   // actual HW restriction. In particular earlyclobber also affects src0 and
   // src1 allocation which is not required.
   bit NoDstOverlap = !gt(DstVT.Size, 128);
 }
 
 def VOPProfileMAI_F32_F32_X4    : VOPProfileMAI<VOP_V4F32_F32_F32_V4F32,       AISrc_128_f32,  ADst_128>;
 def VOPProfileMAI_F32_F32_X16   : VOPProfileMAI<VOP_V16F32_F32_F32_V16F32,     AISrc_512_f32,  ADst_512>;
 def VOPProfileMAI_F32_F32_X32   : VOPProfileMAI<VOP_V32F32_F32_F32_V32F32,     AISrc_1024_f32, ADst_1024>;