[GlobalISel] Replace hard coded dynamic alloca handling with G_DYN_STACKALLOC

[aemerson]

Summary

This change moves the actual stack pointer manipulation into the legalizer, available to targets via lower(). The codegen is slightly different because we're using explicit masks instead of G_PTRMASK, and using G_SUB rather than adding a negative amount via G_GEP.

[arsenm]

llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp

1795–1796

SDNodeFlags?

1798–1800

buildAdd?

1803

Why not use PtrMask?

[aemerson]

llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp

1803

Here's we're masking the allocation amount, which is an integer type, not a pointer, hence we can't use ptrmask.

[aemerson]

Delete unused code and use buildAdd().

[Petar.Avramovic]

One nit, inlined, other then that looks good.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

3940

uint64_t or int64_t instead of unsigned?

3957

Or use explicit APInt for -Align here just to be sure we make bitmask of appropriate size/value.

llvm/test/CodeGen/AArch64/GlobalISel/legalize-dyn-alloca.mir

98

This bitmask seems strange, it looks like zeroextended 32 bit value -32 to a 64 bit value.

[aemerson]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

3957

You're right, I'll fix this with APInt and commit.

llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

public:
   LegalizeResult lowerU64ToF32BitOps(MachineInstr &MI);
   LegalizeResult lowerUITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerSITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerMinMax(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFCopySign(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFMinNumMaxNum(MachineInstr &MI);
   LegalizeResult lowerUnmergeValues(MachineInstr &MI);
   LegalizeResult lowerShuffleVector(MachineInstr &MI);
+  LegalizeResult lowerDynStackAlloc(MachineInstr &MI);
 
 private:
   MachineRegisterInfo &MRI;
   const LegalizerInfo &LI;
   /// To keep track of changes made by the LegalizerHelper.
   GISelChangeObserver &Observer;
 };
 

llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h

public:
   MachineInstrBuilder buildConstDbgValue(const Constant &C,
                                          const MDNode *Variable,
                                          const MDNode *Expr);
 
   /// Build and insert a DBG_LABEL instructions specifying that \p Label is
   /// given. Convert "llvm.dbg.label Label" to "DBG_LABEL Label".
   MachineInstrBuilder buildDbgLabel(const MDNode *Label);
 
+  /// Build and insert \p Res = G_DYN_STACKALLOC \p Size, \p Align
+  ///
+  /// G_DYN_STACKALLOC does a dynamic stack allocation and writes the address of
+  /// the allocated memory into \p Res.
+  /// \pre setBasicBlock or setMI must have been called.
+  /// \pre \p Res must be a generic virtual register with pointer type.
+  ///
+  /// \return a MachineInstrBuilder for the newly created instruction.
+  MachineInstrBuilder buildDynStackAlloc(const DstOp &Res, const SrcOp &Size,
+                                         unsigned Align);
+
   /// Build and insert \p Res = G_FRAME_INDEX \p Idx
   ///
   /// G_FRAME_INDEX materializes the address of an alloca value or other
   /// stack-based object.
   ///
   /// \pre setBasicBlock or setMI must have been called.
   /// \pre \p Res must be a generic virtual register with pointer type.
   ///

llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp

bool IRTranslator::translateAlloca(const User &U,
   if (MRI->getType(NumElts) != IntPtrTy) {
     Register ExtElts = MRI->createGenericVirtualRegister(IntPtrTy);
     MIRBuilder.buildZExtOrTrunc(ExtElts, NumElts);
     NumElts = ExtElts;
   }
 
   Register AllocSize = MRI->createGenericVirtualRegister(IntPtrTy);
   Register TySize =
-      getOrCreateVReg(*ConstantInt::get(IntPtrIRTy, -DL->getTypeAllocSize(Ty)));
+      getOrCreateVReg(*ConstantInt::get(IntPtrIRTy, DL->getTypeAllocSize(Ty)));
   MIRBuilder.buildMul(AllocSize, NumElts, TySize);
 
-  LLT PtrTy = getLLTForType(*AI.getType(), *DL);
-  auto &TLI = *MF->getSubtarget().getTargetLowering();
-  Register SPReg = TLI.getStackPointerRegisterToSaveRestore();
-
-  Register SPTmp = MRI->createGenericVirtualRegister(PtrTy);
-  MIRBuilder.buildCopy(SPTmp, SPReg);
-
-  Register AllocTmp = MRI->createGenericVirtualRegister(PtrTy);
-  MIRBuilder.buildGEP(AllocTmp, SPTmp, AllocSize);
-
-  // Handle alignment. We have to realign if the allocation granule was smaller
-  // than stack alignment, or the specific alloca requires more than stack
-  // alignment.
   unsigned StackAlign =
       MF->getSubtarget().getFrameLowering()->getStackAlignment();
-  Align = std::max(Align, StackAlign);
-  if (Align > StackAlign || DL->getTypeAllocSize(Ty) % StackAlign != 0) {
-    // Round the size of the allocation up to the stack alignment size
-    // by add SA-1 to the size. This doesn't overflow because we're computing
-    // an address inside an alloca.
-    Register AlignedAlloc = MRI->createGenericVirtualRegister(PtrTy);
-    MIRBuilder.buildPtrMask(AlignedAlloc, AllocTmp, Log2_32(Align));
-    AllocTmp = AlignedAlloc;
-  }
+  if (Align <= StackAlign)
+    Align = 0;
+
+  SDNodeFlags Flags;
+  Flags.setNoUnsignedWrap(true);
+  auto SAMinusOne = MIRBuilder.buildConstant(IntPtrTy, StackAlign - 1);
+  auto AllocAdd =
+      MIRBuilder.buildInstr(TargetOpcode::G_ADD, {IntPtrTy},
+                            {AllocSize, SAMinusOne}, MachineInstr::NoUWrap);
+  auto AlignCst =
+      MIRBuilder.buildConstant(IntPtrTy, ~(uint64_t)(StackAlign - 1));
+  auto AlignedAlloc = MIRBuilder.buildAnd(IntPtrTy, AllocAdd, AlignCst);
 
-  MIRBuilder.buildCopy(SPReg, AllocTmp);
-  MIRBuilder.buildCopy(getOrCreateVReg(AI), AllocTmp);
+  MIRBuilder.buildDynStackAlloc(getOrCreateVReg(AI), AlignedAlloc, Align);
 
   MF->getFrameInfo().CreateVariableSizedObject(Align ? Align : 1, &AI);
   assert(MF->getFrameInfo().hasVarSizedObjects());
   return true;
 }
 
 bool IRTranslator::translateVAArg(const User &U, MachineIRBuilder &MIRBuilder) {
   // FIXME: We may need more info about the type. Because of how LLT works,

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CallLowering.h"
 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
 #include "llvm/CodeGen/GlobalISel/LegalizerInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/TargetFrameLowering.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 
 #define DEBUG_TYPE "legalizer"
case TargetOpcode::G_SEXT_INREG: {
     auto MIBSz = MIRBuilder.buildConstant(DstTy, DstTy.getScalarSizeInBits() - SizeInBits);
     MIRBuilder.buildInstr(TargetOpcode::G_SHL, {TmpRes}, {SrcReg, MIBSz->getOperand(0).getReg()});
     MIRBuilder.buildInstr(TargetOpcode::G_ASHR, {DstReg}, {TmpRes, MIBSz->getOperand(0).getReg()});
     MI.eraseFromParent();
     return Legalized;
   }
   case G_SHUFFLE_VECTOR:
     return lowerShuffleVector(MI);
+  case G_DYN_STACKALLOC:
+    return lowerDynStackAlloc(MI);
   }
 }
 
 LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorImplicitDef(
     MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy) {
   SmallVector<Register, 2> DstRegs;
 
   unsigned NarrowSize = NarrowTy.getSizeInBits();
if (Src0Ty.isScalar()) {
       BuildVec.push_back(Extract.getReg(0));
     }
   }
 
   MIRBuilder.buildBuildVector(DstReg, BuildVec);
   MI.eraseFromParent();
   return Legalized;
 }
+
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerDynStackAlloc(MachineInstr &MI) {
+  Register Dst = MI.getOperand(0).getReg();
+  Register AllocSize = MI.getOperand(1).getReg();
+  unsigned Align = MI.getOperand(2).getImm();
+
+  const auto &MF = *MI.getMF();
+  const auto &TLI = *MF.getSubtarget().getTargetLowering();
+
+  LLT PtrTy = MRI.getType(Dst);
+  LLT IntPtrTy = LLT::scalar(PtrTy.getSizeInBits());
+
+  Register SPReg = TLI.getStackPointerRegisterToSaveRestore();
+  auto SPTmp = MIRBuilder.buildCopy(PtrTy, SPReg);
+  SPTmp = MIRBuilder.buildCast(IntPtrTy, SPTmp);
+
+  // Subtract the final alloc from the SP. We use G_PTRTOINT here so we don't
+  // have to generate an extra instruction to negate the alloc and then use
+  // G_GEP to add the negative offset.
+  auto Alloc = MIRBuilder.buildSub(IntPtrTy, SPTmp, AllocSize);
+  if (Align) {
+    auto AlignCst = MIRBuilder.buildConstant(IntPtrTy, -Align);
+    Alloc = MIRBuilder.buildAnd(IntPtrTy, Alloc, AlignCst);
+  }
+
+  SPTmp = MIRBuilder.buildCast(PtrTy, Alloc);
+  MIRBuilder.buildCopy(SPReg, SPTmp);
+  MIRBuilder.buildCopy(Dst, SPTmp);
+
+  MI.eraseFromParent();
+  return Legalized;
+}

llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

MachineInstrBuilder MachineIRBuilder::buildDbgLabel(const MDNode *Label) {
   assert(isa<DILabel>(Label) && "not a label");
   assert(cast<DILabel>(Label)->isValidLocationForIntrinsic(State.DL) &&
          "Expected inlined-at fields to agree");
   auto MIB = buildInstr(TargetOpcode::DBG_LABEL);
 
   return MIB.addMetadata(Label);
 }
 
+MachineInstrBuilder MachineIRBuilder::buildDynStackAlloc(const DstOp &Res,
+                                                         const SrcOp &Size,
+                                                         unsigned Align) {
+  assert(Res.getLLTTy(*getMRI()).isPointer() && "expected ptr dst type");
+  auto MIB = buildInstr(TargetOpcode::G_DYN_STACKALLOC);
+  Res.addDefToMIB(*getMRI(), MIB);
+  Size.addSrcToMIB(MIB);
+  MIB.addImm(Align);
+  return MIB;
+}
+
 MachineInstrBuilder MachineIRBuilder::buildFrameIndex(const DstOp &Res,
                                                       int Idx) {
   assert(Res.getLLTTy(*getMRI()).isPointer() && "invalid operand type");
   auto MIB = buildInstr(TargetOpcode::G_FRAME_INDEX);
   Res.addDefToMIB(*getMRI(), MIB);
   MIB.addFrameIndex(Idx);
   return MIB;
 }

llvm/lib/Target/AArch64/AArch64LegalizerInfo.cpp

AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST) {
 
   getActionDefinitionsBuilder(G_JUMP_TABLE)
     .legalFor({{p0}, {s64}});
 
   getActionDefinitionsBuilder(G_BRJT).legalIf([=](const LegalityQuery &Query) {
     return Query.Types[0] == p0 && Query.Types[1] == s64;
   });
 
+  getActionDefinitionsBuilder(G_DYN_STACKALLOC).lower();
+
   computeTables();
   verify(*ST.getInstrInfo());
 }
 
 bool AArch64LegalizerInfo::legalizeCustom(MachineInstr &MI,
                                           MachineRegisterInfo &MRI,
                                           MachineIRBuilder &MIRBuilder,
                                           GISelChangeObserver &Observer) const {

llvm/test/CodeGen/AArch64/GlobalISel/dynamic-alloca.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 ; RUN: llc -mtriple=aarch64 -global-isel %s -o - -stop-after=irtranslator | FileCheck %s
 
-; CHECK-LABEL: name: test_simple_alloca
-; CHECK: [[NUMELTS:%[0-9]+]]:_(s32) = COPY $w0
-; CHECK: [[TYPE_SIZE:%[0-9]+]]:_(s64) = G_CONSTANT i64 -1
-; CHECK: [[NUMELTS_64:%[0-9]+]]:_(s64) = G_ZEXT [[NUMELTS]](s32)
-; CHECK: [[NUMBYTES:%[0-9]+]]:_(s64) = G_MUL [[NUMELTS_64]], [[TYPE_SIZE]]
-; CHECK: [[SP_TMP:%[0-9]+]]:_(p0) = COPY $sp
-; CHECK: [[ALLOC:%[0-9]+]]:_(p0) = G_GEP [[SP_TMP]], [[NUMBYTES]]
-; CHECK: [[ALIGNED_ALLOC:%[0-9]+]]:_(p0) = G_PTR_MASK [[ALLOC]], 4
-; CHECK: $sp = COPY [[ALIGNED_ALLOC]]
-; CHECK: [[ALLOC:%[0-9]+]]:_(p0) = COPY [[ALIGNED_ALLOC]]
-; CHECK: $x0 = COPY [[ALLOC]]
 define i8* @test_simple_alloca(i32 %numelts) {
+  ; CHECK-LABEL: name: test_simple_alloca
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $w0
+  ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+  ; CHECK:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+  ; CHECK:   [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+  ; CHECK:   [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+  ; CHECK:   [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+  ; CHECK:   [[ADD:%[0-9]+]]:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+  ; CHECK:   [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+  ; CHECK:   [[AND:%[0-9]+]]:_(s64) = G_AND [[ADD]], [[C2]]
+  ; CHECK:   [[DYN_STACKALLOC:%[0-9]+]]:_(p0) = G_DYN_STACKALLOC [[AND]](s64), 0
+  ; CHECK:   $x0 = COPY [[DYN_STACKALLOC]](p0)
+  ; CHECK:   RET_ReallyLR implicit $x0
   %addr = alloca i8, i32 %numelts
   ret i8* %addr
 }
 
-; CHECK-LABEL: name: test_aligned_alloca
-; CHECK: [[NUMELTS:%[0-9]+]]:_(s32) = COPY $w0
-; CHECK: [[TYPE_SIZE:%[0-9]+]]:_(s64) = G_CONSTANT i64 -1
-; CHECK: [[NUMELTS_64:%[0-9]+]]:_(s64) = G_ZEXT [[NUMELTS]](s32)
-; CHECK: [[NUMBYTES:%[0-9]+]]:_(s64) = G_MUL [[NUMELTS_64]], [[TYPE_SIZE]]
-; CHECK: [[SP_TMP:%[0-9]+]]:_(p0) = COPY $sp
-; CHECK: [[ALLOC:%[0-9]+]]:_(p0) = G_GEP [[SP_TMP]], [[NUMBYTES]]
-; CHECK: [[ALIGNED_ALLOC:%[0-9]+]]:_(p0) = G_PTR_MASK [[ALLOC]], 5
-; CHECK: $sp = COPY [[ALIGNED_ALLOC]]
-; CHECK: [[ALLOC:%[0-9]+]]:_(p0) = COPY [[ALIGNED_ALLOC]]
-; CHECK: $x0 = COPY [[ALLOC]]
 define i8* @test_aligned_alloca(i32 %numelts) {
+  ; CHECK-LABEL: name: test_aligned_alloca
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $w0
+  ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+  ; CHECK:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+  ; CHECK:   [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+  ; CHECK:   [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+  ; CHECK:   [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+  ; CHECK:   [[ADD:%[0-9]+]]:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+  ; CHECK:   [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+  ; CHECK:   [[AND:%[0-9]+]]:_(s64) = G_AND [[ADD]], [[C2]]
+  ; CHECK:   [[DYN_STACKALLOC:%[0-9]+]]:_(p0) = G_DYN_STACKALLOC [[AND]](s64), 32
+  ; CHECK:   $x0 = COPY [[DYN_STACKALLOC]](p0)
+  ; CHECK:   RET_ReallyLR implicit $x0
   %addr = alloca i8, i32 %numelts, align 32
   ret i8* %addr
 }
 
-; CHECK-LABEL: name: test_natural_alloca
-; CHECK: [[NUMELTS:%[0-9]+]]:_(s32) = COPY $w0
-; CHECK: [[TYPE_SIZE:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
-; CHECK: [[NUMELTS_64:%[0-9]+]]:_(s64) = G_ZEXT [[NUMELTS]](s32)
-; CHECK: [[NUMBYTES:%[0-9]+]]:_(s64) = G_MUL [[NUMELTS_64]], [[TYPE_SIZE]]
-; CHECK: [[SP_TMP:%[0-9]+]]:_(p0) = COPY $sp
-; CHECK: [[ALLOC:%[0-9]+]]:_(p0) = G_GEP [[SP_TMP]], [[NUMBYTES]]
-; CHECK: $sp = COPY [[ALLOC]]
-; CHECK: [[ALLOC_TMP:%[0-9]+]]:_(p0) = COPY [[ALLOC]]
-; CHECK: $x0 = COPY [[ALLOC_TMP]]
 define i128* @test_natural_alloca(i32 %numelts) {
+  ; CHECK-LABEL: name: test_natural_alloca
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $w0
+  ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+  ; CHECK:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
+  ; CHECK:   [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+  ; CHECK:   [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+  ; CHECK:   [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+  ; CHECK:   [[ADD:%[0-9]+]]:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+  ; CHECK:   [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+  ; CHECK:   [[AND:%[0-9]+]]:_(s64) = G_AND [[ADD]], [[C2]]
+  ; CHECK:   [[DYN_STACKALLOC:%[0-9]+]]:_(p0) = G_DYN_STACKALLOC [[AND]](s64), 0
+  ; CHECK:   $x0 = COPY [[DYN_STACKALLOC]](p0)
+  ; CHECK:   RET_ReallyLR implicit $x0
   %addr = alloca i128, i32 %numelts
   ret i128* %addr
 }

llvm/test/CodeGen/AArch64/GlobalISel/legalize-dyn-alloca.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=arm64-unknown-unknown -global-isel -global-isel-abort=1 -O0 -run-pass=legalizer %s -o - | FileCheck %s
+--- |
+  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64"
+
+  define i8* @test_simple_alloca(i32 %numelts) {
+    %addr = alloca i8, i32 %numelts
+    ret i8* %addr
+  }
+
+  define i8* @test_aligned_alloca(i32 %numelts) {
+    %addr = alloca i8, i32 %numelts, align 32
+    ret i8* %addr
+  }
+
+  define i128* @test_natural_alloca(i32 %numelts) {
+    %addr = alloca i128, i32 %numelts
+    ret i128* %addr
+  }
+
+...
+---
+name:            test_simple_alloca
+alignment:       2
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+frameInfo:
+  maxAlignment:    1
+stack:
+  - { id: 0, name: addr, type: variable-sized, alignment: 1 }
+machineFunctionInfo: {}
+body:             |
+  bb.1 (%ir-block.0):
+    liveins: $w0
+
+    ; CHECK-LABEL: name: test_simple_alloca
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+    ; CHECK: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+    ; CHECK: %5:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+    ; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND %5, [[C2]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(p0) = COPY $sp
+    ; CHECK: [[PTRTOINT:%[0-9]+]]:_(s64) = G_PTRTOINT [[COPY1]](p0)
+    ; CHECK: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[PTRTOINT]], [[AND]]
+    ; CHECK: [[INTTOPTR:%[0-9]+]]:_(p0) = G_INTTOPTR [[SUB]](s64)
+    ; CHECK: $sp = COPY [[INTTOPTR]](p0)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(p0) = COPY [[INTTOPTR]](p0)
+    ; CHECK: $x0 = COPY [[COPY2]](p0)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:_(s32) = COPY $w0
+    %3:_(s64) = G_CONSTANT i64 1
+    %1:_(s64) = G_ZEXT %0(s32)
+    %2:_(s64) = G_MUL %1, %3
+    %4:_(s64) = G_CONSTANT i64 15
+    %5:_(s64) = nuw G_ADD %2, %4
+    %6:_(s64) = G_CONSTANT i64 -16
+    %7:_(s64) = G_AND %5, %6
+    %8:_(p0) = G_DYN_STACKALLOC %7(s64), 0
+    $x0 = COPY %8(p0)
+    RET_ReallyLR implicit $x0
+
+...
+---
+name:            test_aligned_alloca
+alignment:       2
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+frameInfo:
+  maxAlignment:    32
+stack:
+  - { id: 0, name: addr, type: variable-sized, alignment: 32 }
+machineFunctionInfo: {}
+body:             |
+  bb.1 (%ir-block.0):
+    liveins: $w0
+
+    ; CHECK-LABEL: name: test_aligned_alloca
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+    ; CHECK: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+    ; CHECK: %5:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+    ; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND %5, [[C2]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(p0) = COPY $sp
+    ; CHECK: [[PTRTOINT:%[0-9]+]]:_(s64) = G_PTRTOINT [[COPY1]](p0)
+    ; CHECK: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[PTRTOINT]], [[AND]]
+    ; CHECK: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967264
+    ; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[SUB]], [[C3]]
+    ; CHECK: [[INTTOPTR:%[0-9]+]]:_(p0) = G_INTTOPTR [[AND1]](s64)
+    ; CHECK: $sp = COPY [[INTTOPTR]](p0)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(p0) = COPY [[INTTOPTR]](p0)
+    ; CHECK: $x0 = COPY [[COPY2]](p0)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:_(s32) = COPY $w0
+    %3:_(s64) = G_CONSTANT i64 1
+    %1:_(s64) = G_ZEXT %0(s32)
+    %2:_(s64) = G_MUL %1, %3
+    %4:_(s64) = G_CONSTANT i64 15
+    %5:_(s64) = nuw G_ADD %2, %4
+    %6:_(s64) = G_CONSTANT i64 -16
+    %7:_(s64) = G_AND %5, %6
+    %8:_(p0) = G_DYN_STACKALLOC %7(s64), 32
+    $x0 = COPY %8(p0)
+    RET_ReallyLR implicit $x0
+
+...
+---
+name:            test_natural_alloca
+alignment:       2
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+frameInfo:
+  maxAlignment:    1
+stack:
+  - { id: 0, name: addr, type: variable-sized, alignment: 1 }
+machineFunctionInfo: {}
+body:             |
+  bb.1 (%ir-block.0):
+    liveins: $w0
+
+    ; CHECK-LABEL: name: test_natural_alloca
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s32)
+    ; CHECK: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ZEXT]], [[C]]
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+    ; CHECK: %5:_(s64) = nuw G_ADD [[MUL]], [[C1]]
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 -16
+    ; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND %5, [[C2]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(p0) = COPY $sp
+    ; CHECK: [[PTRTOINT:%[0-9]+]]:_(s64) = G_PTRTOINT [[COPY1]](p0)
+    ; CHECK: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[PTRTOINT]], [[AND]]
+    ; CHECK: [[INTTOPTR:%[0-9]+]]:_(p0) = G_INTTOPTR [[SUB]](s64)
+    ; CHECK: $sp = COPY [[INTTOPTR]](p0)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(p0) = COPY [[INTTOPTR]](p0)
+    ; CHECK: $x0 = COPY [[COPY2]](p0)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:_(s32) = COPY $w0
+    %3:_(s64) = G_CONSTANT i64 16
+    %1:_(s64) = G_ZEXT %0(s32)
+    %2:_(s64) = G_MUL %1, %3
+    %4:_(s64) = G_CONSTANT i64 15
+    %5:_(s64) = nuw G_ADD %2, %4
+    %6:_(s64) = G_CONSTANT i64 -16
+    %7:_(s64) = G_AND %5, %6
+    %8:_(p0) = G_DYN_STACKALLOC %7(s64), 0
+    $x0 = COPY %8(p0)
+    RET_ReallyLR implicit $x0
+
+...