[clang] [llvm] [ARM] musttail fixes (PR #102896)

via cfe-commits cfe-commits at lists.llvm.org
Tue Aug 27 02:53:20 PDT 2024


https://github.com/kiran-isaac updated https://github.com/llvm/llvm-project/pull/102896

>From 24806b179d34b4afb21832e4a2150c13995b59e0 Mon Sep 17 00:00:00 2001
From: Kiran <kiran.sturt at arm.com>
Date: Thu, 8 Aug 2024 13:07:24 +0100
Subject: [PATCH 1/2] [ARM] musttail fixes

Backend:
- Caller and callee arguments no longer have to match, just to take up the same space, as they can be changed before the call
- Allowed tail calls if callee and callee both (or neither) use sret, wheras before it would be dissalowed if either used sret
- Allowed tail calls if byval args are used
- Added debug trace for IsEligibleForTailCallOptimisation

Frontend (clang):
- Do not generate extra alloca if sret is used with musttail, as the space for the sret is allocated already

Change-Id: Ic7f246a7eca43c06874922d642d7dc44bdfc98ec
---
 clang/lib/CodeGen/CGCall.cpp                  |   2 +-
 llvm/include/llvm/CodeGen/CallingConvLower.h  |   2 +
 llvm/lib/CodeGen/CallingConvLower.cpp         |  61 +++
 llvm/lib/Target/ARM/ARMISelLowering.cpp       | 141 ++----
 .../ARM/2013-05-13-AAPCS-byval-padding.ll     |  16 +-
 .../ARM/2013-05-13-AAPCS-byval-padding2.ll    |  13 +-
 llvm/test/CodeGen/ARM/fp-arg-shuffle.ll       |  22 +
 llvm/test/CodeGen/ARM/fp16-vector-argument.ll |  41 +-
 llvm/test/CodeGen/ARM/struct_byval.ll         | 455 ++++++++++++++++--
 llvm/test/CodeGen/ARM/tail-call-float.ll      |  99 +++-
 10 files changed, 661 insertions(+), 191 deletions(-)

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 070001a180ab88..1144de0b6ba7b5 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -5085,7 +5085,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
   RawAddress SRetAlloca = RawAddress::invalid();
   llvm::Value *UnusedReturnSizePtr = nullptr;
   if (RetAI.isIndirect() || RetAI.isInAlloca() || RetAI.isCoerceAndExpand()) {
-    if (IsVirtualFunctionPointerThunk && RetAI.isIndirect()) {
+    if ((IsVirtualFunctionPointerThunk || IsMustTail) && RetAI.isIndirect()) {
       SRetPtr = makeNaturalAddressForPointer(CurFn->arg_begin() +
                                                  IRFunctionArgs.getSRetArgNo(),
                                              RetTy, CharUnits::fromQuantity(1));
diff --git a/llvm/include/llvm/CodeGen/CallingConvLower.h b/llvm/include/llvm/CodeGen/CallingConvLower.h
index 932a2a94ab1f1a..fdb5982cb2042b 100644
--- a/llvm/include/llvm/CodeGen/CallingConvLower.h
+++ b/llvm/include/llvm/CodeGen/CallingConvLower.h
@@ -540,6 +540,8 @@ class CCState {
                });
   }
 
+  void dump() const;
+
 private:
   /// MarkAllocated - Mark a register and all of its aliases as allocated.
   void MarkAllocated(MCPhysReg Reg);
diff --git a/llvm/lib/CodeGen/CallingConvLower.cpp b/llvm/lib/CodeGen/CallingConvLower.cpp
index b7152587a9fa05..7ba3ea83115db2 100644
--- a/llvm/lib/CodeGen/CallingConvLower.cpp
+++ b/llvm/lib/CodeGen/CallingConvLower.cpp
@@ -290,3 +290,64 @@ bool CCState::resultsCompatible(CallingConv::ID CalleeCC,
   return std::equal(RVLocs1.begin(), RVLocs1.end(), RVLocs2.begin(),
                     RVLocs2.end(), AreCompatible);
 }
+
+void CCState::dump() const {
+  dbgs() << "CCState:\n";
+  for (const CCValAssign &Loc : Locs) {
+    if (Loc.isRegLoc()) {
+      dbgs() << "  Reg " << TRI.getName(Loc.getLocReg());
+    } else if (Loc.isMemLoc()) {
+      dbgs() << "  Mem " << Loc.getLocMemOffset();
+    } else {
+      assert(Loc.isPendingLoc());
+      dbgs() << "  Pend " << Loc.getExtraInfo();
+    }
+
+    dbgs() << " ValVT:" << Loc.getValVT();
+    dbgs() << " LocVT:" << Loc.getLocVT();
+
+    if (Loc.needsCustom())
+      dbgs() << " custom";
+
+    switch (Loc.getLocInfo()) {
+    case CCValAssign::Full:
+      dbgs() << " Full";
+      break;
+    case CCValAssign::SExt:
+      dbgs() << " SExt";
+      break;
+    case CCValAssign::ZExt:
+      dbgs() << " ZExt";
+      break;
+    case CCValAssign::AExt:
+      dbgs() << " AExt";
+      break;
+    case CCValAssign::SExtUpper:
+      dbgs() << " SExtUpper";
+      break;
+    case CCValAssign::ZExtUpper:
+      dbgs() << " ZExtUpper";
+      break;
+    case CCValAssign::AExtUpper:
+      dbgs() << " AExtUpper";
+      break;
+    case CCValAssign::BCvt:
+      dbgs() << " BCvt";
+      break;
+    case CCValAssign::Trunc:
+      dbgs() << " Trunc";
+      break;
+    case CCValAssign::VExt:
+      dbgs() << " VExt";
+      break;
+    case CCValAssign::FPExt:
+      dbgs() << " FPExt";
+      break;
+    case CCValAssign::Indirect:
+      dbgs() << " Indirect";
+      break;
+    }
+
+    dbgs() << "\n";
+  }
+}
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 476b7b349294ab..aeba673f9a2f77 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -2407,8 +2407,8 @@ ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     isTailCall = false;
 
   // For both the non-secure calls and the returns from a CMSE entry function,
-  // the function needs to do some extra work afte r the call, or before the
-  // return, respectively, thus it cannot end with atail call
+  // the function needs to do some extra work after the call, or before the
+  // return, respectively, thus it cannot end with a tail call
   if (isCmseNSCall || AFI->isCmseNSEntryFunction())
     isTailCall = false;
 
@@ -2959,50 +2959,6 @@ void ARMTargetLowering::HandleByVal(CCState *State, unsigned &Size,
   Size = std::max<int>(Size - Excess, 0);
 }
 
-/// MatchingStackOffset - Return true if the given stack call argument is
-/// already available in the same position (relatively) of the caller's
-/// incoming argument stack.
-static
-bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
-                         MachineFrameInfo &MFI, const MachineRegisterInfo *MRI,
-                         const TargetInstrInfo *TII) {
-  unsigned Bytes = Arg.getValueSizeInBits() / 8;
-  int FI = std::numeric_limits<int>::max();
-  if (Arg.getOpcode() == ISD::CopyFromReg) {
-    Register VR = cast<RegisterSDNode>(Arg.getOperand(1))->getReg();
-    if (!VR.isVirtual())
-      return false;
-    MachineInstr *Def = MRI->getVRegDef(VR);
-    if (!Def)
-      return false;
-    if (!Flags.isByVal()) {
-      if (!TII->isLoadFromStackSlot(*Def, FI))
-        return false;
-    } else {
-      return false;
-    }
-  } else if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Arg)) {
-    if (Flags.isByVal())
-      // ByVal argument is passed in as a pointer but it's now being
-      // dereferenced. e.g.
-      // define @foo(%struct.X* %A) {
-      //   tail call @bar(%struct.X* byval %A)
-      // }
-      return false;
-    SDValue Ptr = Ld->getBasePtr();
-    FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(Ptr);
-    if (!FINode)
-      return false;
-    FI = FINode->getIndex();
-  } else
-    return false;
-
-  assert(FI != std::numeric_limits<int>::max());
-  if (!MFI.isFixedObjectIndex(FI))
-    return false;
-  return Offset == MFI.getObjectOffset(FI) && Bytes == MFI.getObjectSize(FI);
-}
-
 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
 /// for tail call optimization. Targets which want to do tail call
 /// optimization should implement this function. Note that this function also
@@ -3044,8 +3000,10 @@ bool ARMTargetLowering::IsEligibleForTailCallOptimization(
     for (const CCValAssign &AL : ArgLocs)
       if (AL.isRegLoc())
         AddressRegisters.erase(AL.getLocReg());
-    if (AddressRegisters.empty())
+    if (AddressRegisters.empty()) {
+      LLVM_DEBUG(dbgs() << "false (no space for target address)\n");
       return false;
+    }
   }
 
   // Look for obvious safe cases to perform tail call optimization that do not
@@ -3054,18 +3012,26 @@ bool ARMTargetLowering::IsEligibleForTailCallOptimization(
   // Exception-handling functions need a special set of instructions to indicate
   // a return to the hardware. Tail-calling another function would probably
   // break this.
-  if (CallerF.hasFnAttribute("interrupt"))
+  if (CallerF.hasFnAttribute("interrupt")) {
+    LLVM_DEBUG(dbgs() << "false (interrupt attribute)\n");
     return false;
+  }
 
-  if (canGuaranteeTCO(CalleeCC, getTargetMachine().Options.GuaranteedTailCallOpt))
+  if (canGuaranteeTCO(CalleeCC,
+                      getTargetMachine().Options.GuaranteedTailCallOpt)) {
+    LLVM_DEBUG(dbgs() << (CalleeCC == CallerCC ? "true" : "false")
+                      << " (guaranteed tail-call CC)\n");
     return CalleeCC == CallerCC;
+  }
 
-  // Also avoid sibcall optimization if either caller or callee uses struct
-  // return semantics.
+  // Also avoid sibcall optimization if only one of caller or callee uses
+  // struct return semantics.
   bool isCalleeStructRet = Outs.empty() ? false : Outs[0].Flags.isSRet();
   bool isCallerStructRet = MF.getFunction().hasStructRetAttr();
-  if (isCalleeStructRet || isCallerStructRet)
+  if (isCalleeStructRet != isCallerStructRet) {
+    LLVM_DEBUG(dbgs() << "false (struct-ret)\n");
     return false;
+  }
 
   // Externally-defined functions with weak linkage should not be
   // tail-called on ARM when the OS does not support dynamic
@@ -3078,8 +3044,11 @@ bool ARMTargetLowering::IsEligibleForTailCallOptimization(
     const GlobalValue *GV = G->getGlobal();
     const Triple &TT = getTargetMachine().getTargetTriple();
     if (GV->hasExternalWeakLinkage() &&
-        (!TT.isOSWindows() || TT.isOSBinFormatELF() || TT.isOSBinFormatMachO()))
+        (!TT.isOSWindows() || TT.isOSBinFormatELF() ||
+         TT.isOSBinFormatMachO())) {
+      LLVM_DEBUG(dbgs() << "false (external weak linkage)\n");
       return false;
+    }
   }
 
   // Check that the call results are passed in the same way.
@@ -3088,70 +3057,44 @@ bool ARMTargetLowering::IsEligibleForTailCallOptimization(
           getEffectiveCallingConv(CalleeCC, isVarArg),
           getEffectiveCallingConv(CallerCC, CallerF.isVarArg()), MF, C, Ins,
           CCAssignFnForReturn(CalleeCC, isVarArg),
-          CCAssignFnForReturn(CallerCC, CallerF.isVarArg())))
+          CCAssignFnForReturn(CallerCC, CallerF.isVarArg()))) {
+    LLVM_DEBUG(dbgs() << "false (incompatible results)\n");
     return false;
+  }
   // The callee has to preserve all registers the caller needs to preserve.
   const ARMBaseRegisterInfo *TRI = Subtarget->getRegisterInfo();
   const uint32_t *CallerPreserved = TRI->getCallPreservedMask(MF, CallerCC);
   if (CalleeCC != CallerCC) {
     const uint32_t *CalleePreserved = TRI->getCallPreservedMask(MF, CalleeCC);
-    if (!TRI->regmaskSubsetEqual(CallerPreserved, CalleePreserved))
+    if (!TRI->regmaskSubsetEqual(CallerPreserved, CalleePreserved)) {
+      LLVM_DEBUG(dbgs() << "false (not all registers preserved)\n");
       return false;
+    }
   }
 
-  // If Caller's vararg or byval argument has been split between registers and
+  // If Caller's vararg argument has been split between registers and
   // stack, do not perform tail call, since part of the argument is in caller's
   // local frame.
   const ARMFunctionInfo *AFI_Caller = MF.getInfo<ARMFunctionInfo>();
-  if (AFI_Caller->getArgRegsSaveSize())
+  if (CLI.IsVarArg && AFI_Caller->getArgRegsSaveSize()) {
+    LLVM_DEBUG(dbgs() << "false (vararg arg reg save area)\n");
     return false;
+  }
 
   // If the callee takes no arguments then go on to check the results of the
   // call.
-  if (!Outs.empty()) {
-    if (CCInfo.getStackSize()) {
-      // Check if the arguments are already laid out in the right way as
-      // the caller's fixed stack objects.
-      MachineFrameInfo &MFI = MF.getFrameInfo();
-      const MachineRegisterInfo *MRI = &MF.getRegInfo();
-      const TargetInstrInfo *TII = Subtarget->getInstrInfo();
-      for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
-           i != e;
-           ++i, ++realArgIdx) {
-        CCValAssign &VA = ArgLocs[i];
-        EVT RegVT = VA.getLocVT();
-        SDValue Arg = OutVals[realArgIdx];
-        ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
-        if (VA.getLocInfo() == CCValAssign::Indirect)
-          return false;
-        if (VA.needsCustom() && (RegVT == MVT::f64 || RegVT == MVT::v2f64)) {
-          // f64 and vector types are split into multiple registers or
-          // register/stack-slot combinations.  The types will not match
-          // the registers; give up on memory f64 refs until we figure
-          // out what to do about this.
-          if (!VA.isRegLoc())
-            return false;
-          if (!ArgLocs[++i].isRegLoc())
-            return false;
-          if (RegVT == MVT::v2f64) {
-            if (!ArgLocs[++i].isRegLoc())
-              return false;
-            if (!ArgLocs[++i].isRegLoc())
-              return false;
-          }
-        } else if (!VA.isRegLoc()) {
-          if (!MatchingStackOffset(Arg, VA.getLocMemOffset(), Flags,
-                                   MFI, MRI, TII))
-            return false;
-        }
-      }
-    }
-
-    const MachineRegisterInfo &MRI = MF.getRegInfo();
-    if (!parametersInCSRMatch(MRI, CallerPreserved, ArgLocs, OutVals))
-      return false;
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  if (!parametersInCSRMatch(MRI, CallerPreserved, ArgLocs, OutVals)) {
+    LLVM_DEBUG(dbgs() << "false (parameters in CSRs do not match)\n");
+    return false;
   }
 
+  // If the stack arguments for this call do not fit into our own save area then
+  // the call cannot be made tail.
+  if (CCInfo.getStackSize() > AFI_Caller->getArgumentStackSize())
+    return false;
+
+  LLVM_DEBUG(dbgs() << "true\n");
   return true;
 }
 
diff --git a/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding.ll b/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding.ll
index d8e22f4f5312ae..e186ae3a961502 100644
--- a/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding.ll
+++ b/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding.ll
@@ -12,17 +12,11 @@ define void @check227(
 ; arg1 --> SP+188
 
 entry:
-
-;CHECK:  sub   sp, sp, #12
-;CHECK:  push  {r11, lr}
-;CHECK:  sub   sp, sp, #4
-;CHECK:  add   r0, sp, #12
-;CHECK:  stm   r0, {r1, r2, r3}
-;CHECK:  ldr   r0, [sp, #212]
-;CHECK:  bl    useInt
-;CHECK:  add   sp, sp, #4
-;CHECK:  pop   {r11, lr}
-;CHECK:  add   sp, sp, #12
+; CHECK: sub     sp, sp, #12
+; CHECK: stm     sp, {r1, r2, r3}
+; CHECK: ldr     r0, [sp, #200]
+; CHECK: add     sp, sp, #12
+; CHECK: b       useInt
 
   %0 = ptrtoint ptr %arg1 to i32
   tail call void @useInt(i32 %0)
diff --git a/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding2.ll b/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding2.ll
index 0c5d22984b99e1..efdecce9ae723a 100644
--- a/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding2.ll
+++ b/llvm/test/CodeGen/ARM/2013-05-13-AAPCS-byval-padding2.ll
@@ -7,14 +7,11 @@
 define void @foo(ptr byval(%struct4bytes) %p0, ; --> R0
                  ptr byval(%struct20bytes) %p1 ; --> R1,R2,R3, [SP+0 .. SP+8)
 ) {
-;CHECK:  sub  sp, sp, #16
-;CHECK:  push  {r11, lr}
-;CHECK:  add  r12, sp, #8
-;CHECK:  stm  r12, {r0, r1, r2, r3}
-;CHECK:  add  r0, sp, #12
-;CHECK:  bl  useInt
-;CHECK:  pop  {r11, lr}
-;CHECK:  add  sp, sp, #16
+;CHECK:  sub     sp, sp, #16
+;CHECK:  stm     sp, {r0, r1, r2, r3}
+;CHECK:  add     r0, sp, #4
+;CHECK:  add     sp, sp, #16
+;CHECK:  b       useInt
 
   %1 = ptrtoint ptr %p1 to i32
   tail call void @useInt(i32 %1)
diff --git a/llvm/test/CodeGen/ARM/fp-arg-shuffle.ll b/llvm/test/CodeGen/ARM/fp-arg-shuffle.ll
index 4996cc8ecbf022..2ceb7a7b97a1fe 100644
--- a/llvm/test/CodeGen/ARM/fp-arg-shuffle.ll
+++ b/llvm/test/CodeGen/ARM/fp-arg-shuffle.ll
@@ -3,6 +3,28 @@
 ; CHECK: function1
 ; CHECK-NOT: vmov
 define double @function1(double %a, double %b, double %c, double %d, double %e, double %f) nounwind noinline ssp {
+; CHECK-LABEL: function1:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    .save {r4, r5, r11, lr}
+; CHECK-NEXT:    push {r4, r5, r11, lr}
+; CHECK-NEXT:    vldr d16, [sp, #40]
+; CHECK-NEXT:    vldr d17, [sp, #32]
+; CHECK-NEXT:    vmov r12, lr, d16
+; CHECK-NEXT:    vldr d16, [sp, #16]
+; CHECK-NEXT:    vmov r4, r5, d17
+; CHECK-NEXT:    vldr d17, [sp, #24]
+; CHECK-NEXT:    str r3, [sp, #36]
+; CHECK-NEXT:    str r2, [sp, #32]
+; CHECK-NEXT:    str r1, [sp, #44]
+; CHECK-NEXT:    str r0, [sp, #40]
+; CHECK-NEXT:    vstr d17, [sp, #16]
+; CHECK-NEXT:    vstr d16, [sp, #24]
+; CHECK-NEXT:    mov r0, r12
+; CHECK-NEXT:    mov r1, lr
+; CHECK-NEXT:    mov r2, r4
+; CHECK-NEXT:    mov r3, r5
+; CHECK-NEXT:    pop {r4, r5, r11, lr}
+; CHECK-NEXT:    b function2
 entry:
   %call = tail call double @function2(double %f, double %e, double %d, double %c, double %b, double %a) nounwind
   ret double %call
diff --git a/llvm/test/CodeGen/ARM/fp16-vector-argument.ll b/llvm/test/CodeGen/ARM/fp16-vector-argument.ll
index 6fc56967bc7aa9..65aff46658fd1d 100644
--- a/llvm/test/CodeGen/ARM/fp16-vector-argument.ll
+++ b/llvm/test/CodeGen/ARM/fp16-vector-argument.ll
@@ -145,26 +145,21 @@ entry:
 define void @many_args_test(double, float, i16, <4 x half>, <8 x half>, <8 x half>, <8 x half>) {
 ; SOFT-LABEL: many_args_test:
 ; SOFT:       @ %bb.0: @ %entry
-; SOFT-NEXT:    push {r11, lr}
-; SOFT-NEXT:    sub sp, sp, #32
-; SOFT-NEXT:    add r12, sp, #80
+; SOFT-NEXT:    add r12, sp, #40
 ; SOFT-NEXT:    vld1.64 {d16, d17}, [r12]
-; SOFT-NEXT:    add r12, sp, #48
+; SOFT-NEXT:    add r12, sp, #8
 ; SOFT-NEXT:    vabs.f16 q8, q8
 ; SOFT-NEXT:    vld1.64 {d18, d19}, [r12]
-; SOFT-NEXT:    add r12, sp, #64
+; SOFT-NEXT:    add r12, sp, #24
 ; SOFT-NEXT:    vadd.f16 q8, q8, q9
 ; SOFT-NEXT:    vld1.64 {d18, d19}, [r12]
 ; SOFT-NEXT:    add r12, sp, #16
 ; SOFT-NEXT:    vmul.f16 q8, q9, q8
 ; SOFT-NEXT:    vst1.64 {d16, d17}, [r12]
-; SOFT-NEXT:    mov r12, sp
-; SOFT-NEXT:    vldr d16, [sp, #40]
-; SOFT-NEXT:    vst1.16 {d16}, [r12:64]!
-; SOFT-NEXT:    str r3, [r12]
-; SOFT-NEXT:    bl use
-; SOFT-NEXT:    add sp, sp, #32
-; SOFT-NEXT:    pop {r11, pc}
+; SOFT-NEXT:    vldr d16, [sp]
+; SOFT-NEXT:    vstr d16, [sp]
+; SOFT-NEXT:    str r3, [sp, #8]
+; SOFT-NEXT:    b use
 ;
 ; HARD-LABEL: many_args_test:
 ; HARD:       @ %bb.0: @ %entry
@@ -177,33 +172,25 @@ define void @many_args_test(double, float, i16, <4 x half>, <8 x half>, <8 x hal
 ;
 ; SOFTEB-LABEL: many_args_test:
 ; SOFTEB:       @ %bb.0: @ %entry
-; SOFTEB-NEXT:    .save {r11, lr}
-; SOFTEB-NEXT:    push {r11, lr}
-; SOFTEB-NEXT:    .pad #32
-; SOFTEB-NEXT:    sub sp, sp, #32
-; SOFTEB-NEXT:    add r12, sp, #80
-; SOFTEB-NEXT:    mov lr, sp
+; SOFTEB-NEXT:    add r12, sp, #40
 ; SOFTEB-NEXT:    vld1.64 {d16, d17}, [r12]
-; SOFTEB-NEXT:    add r12, sp, #48
+; SOFTEB-NEXT:    add r12, sp, #8
 ; SOFTEB-NEXT:    vrev64.16 q8, q8
 ; SOFTEB-NEXT:    vabs.f16 q8, q8
 ; SOFTEB-NEXT:    vld1.64 {d18, d19}, [r12]
-; SOFTEB-NEXT:    add r12, sp, #64
+; SOFTEB-NEXT:    add r12, sp, #24
 ; SOFTEB-NEXT:    vrev64.16 q9, q9
 ; SOFTEB-NEXT:    vadd.f16 q8, q8, q9
 ; SOFTEB-NEXT:    vld1.64 {d18, d19}, [r12]
 ; SOFTEB-NEXT:    add r12, sp, #16
 ; SOFTEB-NEXT:    vrev64.16 q9, q9
 ; SOFTEB-NEXT:    vmul.f16 q8, q9, q8
-; SOFTEB-NEXT:    vldr d18, [sp, #40]
-; SOFTEB-NEXT:    vrev64.16 d18, d18
-; SOFTEB-NEXT:    vst1.16 {d18}, [lr:64]!
-; SOFTEB-NEXT:    str r3, [lr]
+; SOFTEB-NEXT:    vldr d18, [sp]
 ; SOFTEB-NEXT:    vrev64.16 q8, q8
 ; SOFTEB-NEXT:    vst1.64 {d16, d17}, [r12]
-; SOFTEB-NEXT:    bl use
-; SOFTEB-NEXT:    add sp, sp, #32
-; SOFTEB-NEXT:    pop {r11, pc}
+; SOFTEB-NEXT:    vstr d18, [sp]
+; SOFTEB-NEXT:    str r3, [sp, #8]
+; SOFTEB-NEXT:    b use
 ;
 ; HARDEB-LABEL: many_args_test:
 ; HARDEB:       @ %bb.0: @ %entry
diff --git a/llvm/test/CodeGen/ARM/struct_byval.ll b/llvm/test/CodeGen/ARM/struct_byval.ll
index 73a1b5ee33bca9..5564f254c9e74d 100644
--- a/llvm/test/CodeGen/ARM/struct_byval.ll
+++ b/llvm/test/CodeGen/ARM/struct_byval.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc < %s -mtriple=armv7-apple-ios6.0 | FileCheck %s
 ; RUN: llc < %s -mtriple=thumbv7-apple-ios6.0 | FileCheck %s
 ; RUN: llc < %s -mtriple=armv7-unknown-nacl-gnueabi | FileCheck %s -check-prefix=NACL
@@ -10,11 +11,122 @@
 %struct.LargeStruct = type { i32, [1001 x i8], [300 x i32] }
 
 define i32 @f() nounwind ssp {
+; NACL-LABEL: f:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .save {r4, lr}
+; NACL-NEXT:    push {r4, lr}
+; NACL-NEXT:    .pad #152
+; NACL-NEXT:    sub sp, sp, #152
+; NACL-NEXT:    movw r0, :lower16:__stack_chk_guard
+; NACL-NEXT:    add r3, sp, #72
+; NACL-NEXT:    movt r0, :upper16:__stack_chk_guard
+; NACL-NEXT:    mov lr, sp
+; NACL-NEXT:    ldr r0, [r0]
+; NACL-NEXT:    str r0, [sp, #148]
+; NACL-NEXT:    add r0, sp, #72
+; NACL-NEXT:    add r12, r0, #16
+; NACL-NEXT:    ldm r3, {r0, r1, r2, r3}
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    ldr r4, [r12], #4
+; NACL-NEXT:    str r4, [lr], #4
+; NACL-NEXT:    bl e1
+; NACL-NEXT:    movw r1, :lower16:__stack_chk_guard
+; NACL-NEXT:    ldr r0, [sp, #148]
+; NACL-NEXT:    movt r1, :upper16:__stack_chk_guard
+; NACL-NEXT:    ldr r1, [r1]
+; NACL-NEXT:    cmp r1, r0
+; NACL-NEXT:    moveq r0, #0
+; NACL-NEXT:    addeq sp, sp, #152
+; NACL-NEXT:    popeq {r4, pc}
+; NACL-NEXT:  .LBB0_1: @ %entry
+; NACL-NEXT:    bl __stack_chk_fail
+;
+; NOMOVT-LABEL: f:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .save {r11, lr}
+; NOMOVT-NEXT:    push {r11, lr}
+; NOMOVT-NEXT:    .pad #144
+; NOMOVT-NEXT:    sub sp, sp, #144
+; NOMOVT-NEXT:    ldr r0, .LCPI0_0
+; NOMOVT-NEXT:    mov r1, sp
+; NOMOVT-NEXT:    add r3, sp, #64
+; NOMOVT-NEXT:    ldr r0, [r0]
+; NOMOVT-NEXT:    str r0, [sp, #140]
+; NOMOVT-NEXT:    add r0, sp, #64
+; NOMOVT-NEXT:    add r0, r0, #16
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldr r2, [r0], #4
+; NOMOVT-NEXT:    str r2, [r1], #4
+; NOMOVT-NEXT:    ldm r3, {r0, r1, r2, r3}
+; NOMOVT-NEXT:    bl e1
+; NOMOVT-NEXT:    ldr r0, [sp, #140]
+; NOMOVT-NEXT:    ldr r1, .LCPI0_0
+; NOMOVT-NEXT:    ldr r1, [r1]
+; NOMOVT-NEXT:    cmp r1, r0
+; NOMOVT-NEXT:    moveq r0, #0
+; NOMOVT-NEXT:    addeq sp, sp, #144
+; NOMOVT-NEXT:    popeq {r11, pc}
+; NOMOVT-NEXT:  .LBB0_1: @ %entry
+; NOMOVT-NEXT:    bl __stack_chk_fail
+; NOMOVT-NEXT:    .p2align 2
+; NOMOVT-NEXT:  @ %bb.2:
+; NOMOVT-NEXT:  .LCPI0_0:
+; NOMOVT-NEXT:    .long __stack_chk_guard
 entry:
-; CHECK-LABEL: f:
-; CHECK: ldr
-; CHECK: str
-; CHECK-NOT:bne
   %st = alloca %struct.SmallStruct, align 4
   %call = call i32 @e1(ptr byval(%struct.SmallStruct) %st)
   ret i32 0
@@ -22,20 +134,95 @@ entry:
 
 ; Generate a loop for large struct byval
 define i32 @g() nounwind ssp {
-entry:
-; CHECK-LABEL: g:
-; CHECK: ldr
-; CHECK: sub
-; CHECK: str
-; CHECK: bne
 ; NACL-LABEL: g:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .save {r4, r5, r11, lr}
+; NACL-NEXT:    push {r4, r5, r11, lr}
+; NACL-NEXT:    .pad #2224
+; NACL-NEXT:    sub sp, sp, #2224
+; NACL-NEXT:    movw r0, :lower16:__stack_chk_guard
+; NACL-NEXT:    movt r0, :upper16:__stack_chk_guard
+; NACL-NEXT:    ldr r0, [r0]
+; NACL-NEXT:    str r0, [sp, #2220]
+; NACL-NEXT:    sub sp, sp, #2192
+; NACL-NEXT:    add lr, sp, #2048
+; NACL-NEXT:    ldr r1, [sp, #2208]
+; NACL-NEXT:    add r0, lr, #156
+; NACL-NEXT:    ldr r2, [sp, #2212]
+; NACL-NEXT:    add r12, r0, #16
+; NACL-NEXT:    ldr r0, [sp, #2204]
+; NACL-NEXT:    ldr r3, [sp, #2216]
+; NACL-NEXT:    movw lr, #2192
+; NACL-NEXT:    mov r4, sp
+; NACL-NEXT:  .LBB1_1: @ %entry
+; NACL-NEXT:    @ =>This Inner Loop Header: Depth=1
+; NACL-NEXT:    ldr r5, [r12], #4
+; NACL-NEXT:    subs lr, lr, #4
+; NACL-NEXT:    str r5, [r4], #4
+; NACL-NEXT:    bne .LBB1_1
+; NACL-NEXT:  @ %bb.2: @ %entry
+; NACL-NEXT:    bl e2
+; NACL-NEXT:    add sp, sp, #2192
+; NACL-NEXT:    movw r1, :lower16:__stack_chk_guard
+; NACL-NEXT:    ldr r0, [sp, #2220]
+; NACL-NEXT:    movt r1, :upper16:__stack_chk_guard
+; NACL-NEXT:    ldr r1, [r1]
+; NACL-NEXT:    cmp r1, r0
+; NACL-NEXT:    moveq r0, #0
+; NACL-NEXT:    addeq sp, sp, #2224
+; NACL-NEXT:    popeq {r4, r5, r11, pc}
+; NACL-NEXT:  .LBB1_3: @ %entry
+; NACL-NEXT:    bl __stack_chk_fail
+;
+; NOMOVT-LABEL: g:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .save {r11, lr}
+; NOMOVT-NEXT:    push {r11, lr}
+; NOMOVT-NEXT:    .pad #168
+; NOMOVT-NEXT:    sub sp, sp, #168
+; NOMOVT-NEXT:    .pad #2048
+; NOMOVT-NEXT:    sub sp, sp, #2048
+; NOMOVT-NEXT:    ldr r0, .LCPI1_1
+; NOMOVT-NEXT:    ldr r0, [r0]
+; NOMOVT-NEXT:    str r0, [sp, #2212]
+; NOMOVT-NEXT:    sub sp, sp, #2192
+; NOMOVT-NEXT:    add lr, sp, #2048
+; NOMOVT-NEXT:    ldr r1, .LCPI1_0
+; NOMOVT-NEXT:    add r0, lr, #148
+; NOMOVT-NEXT:    mov r2, sp
+; NOMOVT-NEXT:    add r0, r0, #16
+; NOMOVT-NEXT:  .LBB1_1: @ %entry
+; NOMOVT-NEXT:    @ =>This Inner Loop Header: Depth=1
+; NOMOVT-NEXT:    ldr r3, [r0], #4
+; NOMOVT-NEXT:    subs r1, r1, #4
+; NOMOVT-NEXT:    str r3, [r2], #4
+; NOMOVT-NEXT:    bne .LBB1_1
+; NOMOVT-NEXT:  @ %bb.2: @ %entry
+; NOMOVT-NEXT:    ldr r0, [sp, #2196]
+; NOMOVT-NEXT:    ldr r1, [sp, #2200]
+; NOMOVT-NEXT:    ldr r2, [sp, #2204]
+; NOMOVT-NEXT:    ldr r3, [sp, #2208]
+; NOMOVT-NEXT:    bl e2
+; NOMOVT-NEXT:    add sp, sp, #2192
+; NOMOVT-NEXT:    ldr r0, [sp, #2212]
+; NOMOVT-NEXT:    ldr r1, .LCPI1_1
+; NOMOVT-NEXT:    ldr r1, [r1]
+; NOMOVT-NEXT:    cmp r1, r0
+; NOMOVT-NEXT:    moveq r0, #0
+; NOMOVT-NEXT:    addeq sp, sp, #168
+; NOMOVT-NEXT:    addeq sp, sp, #2048
+; NOMOVT-NEXT:    popeq {r11, pc}
+; NOMOVT-NEXT:  .LBB1_3: @ %entry
+; NOMOVT-NEXT:    bl __stack_chk_fail
+; NOMOVT-NEXT:    .p2align 2
+; NOMOVT-NEXT:  @ %bb.4:
+; NOMOVT-NEXT:  .LCPI1_0:
+; NOMOVT-NEXT:    .long 2192 @ 0x890
+; NOMOVT-NEXT:  .LCPI1_1:
+; NOMOVT-NEXT:    .long __stack_chk_guard
+entry:
 ; Ensure that use movw instead of constpool for the loop trip count. But don't
 ; match the __stack_chk_guard movw
-; NACL: movw {{r[0-9]+|lr}}, #
-; NACL: ldr
-; NACL: sub
-; NACL: str
-; NACL: bne
   %st = alloca %struct.LargeStruct, align 4
   %call = call i32 @e2(ptr byval(%struct.LargeStruct) %st)
   ret i32 0
@@ -43,17 +230,90 @@ entry:
 
 ; Generate a loop using NEON instructions
 define i32 @h() nounwind ssp {
+; NACL-LABEL: h:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .save {r4, r5, r6, r7, r8, lr}
+; NACL-NEXT:    push {r4, r5, r6, r7, r8, lr}
+; NACL-NEXT:    .pad #168
+; NACL-NEXT:    sub sp, sp, #168
+; NACL-NEXT:    .pad #2048
+; NACL-NEXT:    sub sp, sp, #2048
+; NACL-NEXT:    movw r0, :lower16:__stack_chk_guard
+; NACL-NEXT:    movt r0, :upper16:__stack_chk_guard
+; NACL-NEXT:    ldr r0, [r0]
+; NACL-NEXT:    str r0, [sp, #2212]
+; NACL-NEXT:    sub sp, sp, #2192
+; NACL-NEXT:    add r3, sp, #2192
+; NACL-NEXT:    add r0, sp, #2192
+; NACL-NEXT:    add r12, r0, #16
+; NACL-NEXT:    movw lr, #2192
+; NACL-NEXT:    ldm r3, {r0, r1, r2, r3}
+; NACL-NEXT:    mov r4, sp
+; NACL-NEXT:  .LBB2_1: @ %entry
+; NACL-NEXT:    @ =>This Inner Loop Header: Depth=1
+; NACL-NEXT:    vld1.32 {d16, d17}, [r12]!
+; NACL-NEXT:    subs lr, lr, #16
+; NACL-NEXT:    vst1.32 {d16, d17}, [r4]!
+; NACL-NEXT:    bne .LBB2_1
+; NACL-NEXT:  @ %bb.2: @ %entry
+; NACL-NEXT:    bl e3
+; NACL-NEXT:    add sp, sp, #2192
+; NACL-NEXT:    movw r1, :lower16:__stack_chk_guard
+; NACL-NEXT:    ldr r0, [sp, #2212]
+; NACL-NEXT:    movt r1, :upper16:__stack_chk_guard
+; NACL-NEXT:    ldr r1, [r1]
+; NACL-NEXT:    cmp r1, r0
+; NACL-NEXT:    moveq r0, #0
+; NACL-NEXT:    addeq sp, sp, #168
+; NACL-NEXT:    addeq sp, sp, #2048
+; NACL-NEXT:    popeq {r4, r5, r6, r7, r8, pc}
+; NACL-NEXT:  .LBB2_3: @ %entry
+; NACL-NEXT:    bl __stack_chk_fail
+;
+; NOMOVT-LABEL: h:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .save {r6, r10, r11, lr}
+; NOMOVT-NEXT:    push {r6, r10, r11, lr}
+; NOMOVT-NEXT:    .setfp r11, sp, #8
+; NOMOVT-NEXT:    add r11, sp, #8
+; NOMOVT-NEXT:    .pad #2224
+; NOMOVT-NEXT:    sub sp, sp, #2224
+; NOMOVT-NEXT:    bic sp, sp, #15
+; NOMOVT-NEXT:    ldr r0, .LCPI2_1
+; NOMOVT-NEXT:    mov r6, sp
+; NOMOVT-NEXT:    ldr r0, [r0]
+; NOMOVT-NEXT:    str r0, [r6, #2220]
+; NOMOVT-NEXT:    sub sp, sp, #2192
+; NOMOVT-NEXT:    mov r0, r6
+; NOMOVT-NEXT:    ldr r1, .LCPI2_0
+; NOMOVT-NEXT:    add r0, r0, #16
+; NOMOVT-NEXT:    mov r2, sp
+; NOMOVT-NEXT:  .LBB2_1: @ %entry
+; NOMOVT-NEXT:    @ =>This Inner Loop Header: Depth=1
+; NOMOVT-NEXT:    ldr r3, [r0], #4
+; NOMOVT-NEXT:    subs r1, r1, #4
+; NOMOVT-NEXT:    str r3, [r2], #4
+; NOMOVT-NEXT:    bne .LBB2_1
+; NOMOVT-NEXT:  @ %bb.2: @ %entry
+; NOMOVT-NEXT:    ldm r6, {r0, r1, r2, r3}
+; NOMOVT-NEXT:    bl e3
+; NOMOVT-NEXT:    add sp, sp, #2192
+; NOMOVT-NEXT:    ldr r0, [r6, #2220]
+; NOMOVT-NEXT:    ldr r1, .LCPI2_1
+; NOMOVT-NEXT:    ldr r1, [r1]
+; NOMOVT-NEXT:    cmp r1, r0
+; NOMOVT-NEXT:    moveq r0, #0
+; NOMOVT-NEXT:    subeq sp, r11, #8
+; NOMOVT-NEXT:    popeq {r6, r10, r11, pc}
+; NOMOVT-NEXT:  .LBB2_3: @ %entry
+; NOMOVT-NEXT:    bl __stack_chk_fail
+; NOMOVT-NEXT:    .p2align 2
+; NOMOVT-NEXT:  @ %bb.4:
+; NOMOVT-NEXT:  .LCPI2_0:
+; NOMOVT-NEXT:    .long 2192 @ 0x890
+; NOMOVT-NEXT:  .LCPI2_1:
+; NOMOVT-NEXT:    .long __stack_chk_guard
 entry:
-; CHECK-LABEL: h:
-; CHECK: vld1
-; CHECK: sub
-; CHECK: vst1
-; CHECK: bne
-; NACL: movw {{r[0-9]+|lr}}, #
-; NACL: vld1
-; NACL: sub
-; NACL: vst1
-; NACL: bne
   %st = alloca %struct.LargeStruct, align 16
   %call = call i32 @e3(ptr byval(%struct.LargeStruct) align 16 %st)
   ret i32 0
@@ -67,16 +327,50 @@ declare i32 @e3(ptr nocapture byval(%struct.LargeStruct) align 16 %in) nounwind
 ; We can't do tail call since address of s is passed to the callee and part of
 ; s is in caller's local frame.
 define void @f3(ptr nocapture byval(%struct.SmallStruct) %s) nounwind optsize {
-; CHECK-LABEL: f3
-; CHECK: bl _consumestruct
+; NACL-LABEL: f3:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .pad #16
+; NACL-NEXT:    sub sp, sp, #16
+; NACL-NEXT:    stm sp, {r0, r1, r2, r3}
+; NACL-NEXT:    mov r0, sp
+; NACL-NEXT:    mov r1, #80
+; NACL-NEXT:    add sp, sp, #16
+; NACL-NEXT:    b consumestruct
+;
+; NOMOVT-LABEL: f3:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .pad #16
+; NOMOVT-NEXT:    sub sp, sp, #16
+; NOMOVT-NEXT:    stm sp, {r0, r1, r2, r3}
+; NOMOVT-NEXT:    mov r0, sp
+; NOMOVT-NEXT:    mov r1, #80
+; NOMOVT-NEXT:    add sp, sp, #16
+; NOMOVT-NEXT:    b consumestruct
 entry:
   tail call void @consumestruct(ptr %s, i32 80) optsize
   ret void
 }
 
 define void @f4(ptr nocapture byval(%struct.SmallStruct) %s) nounwind optsize {
-; CHECK-LABEL: f4
-; CHECK: bl _consumestruct
+; NACL-LABEL: f4:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .pad #16
+; NACL-NEXT:    sub sp, sp, #16
+; NACL-NEXT:    stm sp, {r0, r1, r2, r3}
+; NACL-NEXT:    mov r0, sp
+; NACL-NEXT:    mov r1, #80
+; NACL-NEXT:    add sp, sp, #16
+; NACL-NEXT:    b consumestruct
+;
+; NOMOVT-LABEL: f4:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .pad #16
+; NOMOVT-NEXT:    sub sp, sp, #16
+; NOMOVT-NEXT:    stm sp, {r0, r1, r2, r3}
+; NOMOVT-NEXT:    mov r0, sp
+; NOMOVT-NEXT:    mov r1, #80
+; NOMOVT-NEXT:    add sp, sp, #16
+; NOMOVT-NEXT:    b consumestruct
 entry:
   tail call void @consumestruct(ptr %s, i32 80) optsize
   ret void
@@ -84,16 +378,34 @@ entry:
 
 ; We can do tail call here since s is in the incoming argument area.
 define void @f5(i32 %a, i32 %b, i32 %c, i32 %d, ptr nocapture byval(%struct.SmallStruct) %s) nounwind optsize {
-; CHECK-LABEL: f5
-; CHECK: b{{(\.w)?}} _consumestruct
+; NACL-LABEL: f5:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    mov r0, sp
+; NACL-NEXT:    mov r1, #80
+; NACL-NEXT:    b consumestruct
+;
+; NOMOVT-LABEL: f5:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    mov r0, sp
+; NOMOVT-NEXT:    mov r1, #80
+; NOMOVT-NEXT:    b consumestruct
 entry:
   tail call void @consumestruct(ptr %s, i32 80) optsize
   ret void
 }
 
 define void @f6(i32 %a, i32 %b, i32 %c, i32 %d, ptr nocapture byval(%struct.SmallStruct) %s) nounwind optsize {
-; CHECK-LABEL: f6
-; CHECK: b{{(\.w)?}} _consumestruct
+; NACL-LABEL: f6:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    mov r0, sp
+; NACL-NEXT:    mov r1, #80
+; NACL-NEXT:    b consumestruct
+;
+; NOMOVT-LABEL: f6:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    mov r0, sp
+; NOMOVT-NEXT:    mov r1, #80
+; NOMOVT-NEXT:    b consumestruct
 entry:
   tail call void @consumestruct(ptr %s, i32 80) optsize
   ret void
@@ -106,10 +418,81 @@ declare void @consumestruct(ptr nocapture %structp, i32 %structsize) nounwind
 
 declare void @use_I(ptr byval(%struct.I.8))
 define void @test_I_16() {
-; CHECK-LABEL: test_I_16
-; CHECK: ldrb
-; CHECK: strb
+; NACL-LABEL: test_I_16:
+; NACL:       @ %bb.0: @ %entry
+; NACL-NEXT:    .save {r11, lr}
+; NACL-NEXT:    push {r11, lr}
+; NACL-NEXT:    .pad #40
+; NACL-NEXT:    sub sp, sp, #40
+; NACL-NEXT:    ldr r0, [r0]
+; NACL-NEXT:    mov r1, sp
+; NACL-NEXT:    vld1.32 {d16, d17}, [r2]!
+; NACL-NEXT:    vst1.32 {d16, d17}, [r1]!
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    ldrb r3, [r2], #1
+; NACL-NEXT:    strb r3, [r1], #1
+; NACL-NEXT:    mov r2, r0
+; NACL-NEXT:    mov r1, r0
+; NACL-NEXT:    mov r3, r0
+; NACL-NEXT:    bl use_I
+; NACL-NEXT:    add sp, sp, #40
+; NACL-NEXT:    pop {r11, pc}
+;
+; NOMOVT-LABEL: test_I_16:
+; NOMOVT:       @ %bb.0: @ %entry
+; NOMOVT-NEXT:    .save {r11, lr}
+; NOMOVT-NEXT:    push {r11, lr}
+; NOMOVT-NEXT:    .setfp r11, sp
+; NOMOVT-NEXT:    mov r11, sp
+; NOMOVT-NEXT:    .pad #40
+; NOMOVT-NEXT:    sub sp, sp, #40
+; NOMOVT-NEXT:    bic sp, sp, #15
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    mov r2, sp
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r1], #4
+; NOMOVT-NEXT:    str r0, [r2], #4
+; NOMOVT-NEXT:    ldr r0, [r0]
+; NOMOVT-NEXT:    mov r1, r0
+; NOMOVT-NEXT:    mov r2, r0
+; NOMOVT-NEXT:    mov r3, r0
+; NOMOVT-NEXT:    bl use_I
+; NOMOVT-NEXT:    mov sp, r11
+; NOMOVT-NEXT:    pop {r11, pc}
 entry:
   call void @use_I(ptr byval(%struct.I.8) align 16 undef)
   ret void
 }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}
diff --git a/llvm/test/CodeGen/ARM/tail-call-float.ll b/llvm/test/CodeGen/ARM/tail-call-float.ll
index 8cca7e0f70683e..6fa4b9229f64c8 100644
--- a/llvm/test/CodeGen/ARM/tail-call-float.ll
+++ b/llvm/test/CodeGen/ARM/tail-call-float.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc -mtriple armv7 -target-abi aapcs -float-abi soft -O0 -o - < %s \
 ; RUN:   | FileCheck %s -check-prefix CHECK-SOFT -check-prefix CHECK
 ; RUN: llc -mtriple armv7 -target-abi aapcs -float-abi hard -O0 -o - < %s \
@@ -12,16 +13,42 @@ declare i1 @non_variadic_big(float, float, float, float, float, float)
 declare i1 @variadic(float, ...)
 
 define void @non_variadic_fp(float %x, float %y) {
-; CHECK-LABEL: non_variadic_fp:
-; CHECK: b non_variadic
+; CHECK-SOFT-LABEL: non_variadic_fp:
+; CHECK-SOFT:       @ %bb.0: @ %entry
+; CHECK-SOFT-NEXT:    mov r3, r1
+; CHECK-SOFT-NEXT:    mov r2, r0
+; CHECK-SOFT-NEXT:    mov r0, r3
+; CHECK-SOFT-NEXT:    mov r1, r2
+; CHECK-SOFT-NEXT:    b non_variadic
+;
+; CHECK-HARD-LABEL: non_variadic_fp:
+; CHECK-HARD:       @ %bb.0: @ %entry
+; CHECK-HARD-NEXT:    vmov.f32 s3, s1
+; CHECK-HARD-NEXT:    vmov.f32 s2, s0
+; CHECK-HARD-NEXT:    vmov.f32 s0, s3
+; CHECK-HARD-NEXT:    vmov.f32 s1, s2
+; CHECK-HARD-NEXT:    b non_variadic
 entry:
   %call = tail call i1 (float, float, float, float) @non_variadic(float %y, float %x, float %x, float %y)
   ret void
 }
 
 define void @variadic_fp(float %x, float %y) {
-; CHECK-LABEL: variadic_fp:
-; CHECK: b variadic
+; CHECK-SOFT-LABEL: variadic_fp:
+; CHECK-SOFT:       @ %bb.0: @ %entry
+; CHECK-SOFT-NEXT:    mov r3, r1
+; CHECK-SOFT-NEXT:    mov r2, r0
+; CHECK-SOFT-NEXT:    mov r0, r3
+; CHECK-SOFT-NEXT:    mov r1, r2
+; CHECK-SOFT-NEXT:    b variadic
+;
+; CHECK-HARD-LABEL: variadic_fp:
+; CHECK-HARD:       @ %bb.0: @ %entry
+; CHECK-HARD-NEXT:    vmov r2, s0
+; CHECK-HARD-NEXT:    vmov r3, s1
+; CHECK-HARD-NEXT:    mov r0, r3
+; CHECK-HARD-NEXT:    mov r1, r2
+; CHECK-HARD-NEXT:    b variadic
 entry:
   %call = tail call i1 (float, ...) @variadic(float %y, float %x, float %x, float %y)
   ret void
@@ -31,9 +58,32 @@ entry:
 ; of them to handle the 6 arguments. With hard-float, we have plenty of regs
 ; (s0-s15) to pass FP arguments.
 define void @non_variadic_fp_big(float %x, float %y) {
-; CHECK-LABEL: non_variadic_fp_big:
-; CHECK-SOFT: bl non_variadic_big
-; CHECK-HARD: b non_variadic_big
+; CHECK-SOFT-LABEL: non_variadic_fp_big:
+; CHECK-SOFT:       @ %bb.0: @ %entry
+; CHECK-SOFT-NEXT:    push {r11, lr}
+; CHECK-SOFT-NEXT:    sub sp, sp, #8
+; CHECK-SOFT-NEXT:    mov r3, r1
+; CHECK-SOFT-NEXT:    mov r2, r0
+; CHECK-SOFT-NEXT:    vmov s0, r3
+; CHECK-SOFT-NEXT:    vmov s0, r2
+; CHECK-SOFT-NEXT:    mov r0, sp
+; CHECK-SOFT-NEXT:    str r3, [r0, #4]
+; CHECK-SOFT-NEXT:    str r2, [r0]
+; CHECK-SOFT-NEXT:    mov r0, r3
+; CHECK-SOFT-NEXT:    mov r1, r2
+; CHECK-SOFT-NEXT:    bl non_variadic_big
+; CHECK-SOFT-NEXT:    add sp, sp, #8
+; CHECK-SOFT-NEXT:    pop {r11, pc}
+;
+; CHECK-HARD-LABEL: non_variadic_fp_big:
+; CHECK-HARD:       @ %bb.0: @ %entry
+; CHECK-HARD-NEXT:    vmov.f32 s5, s1
+; CHECK-HARD-NEXT:    vmov.f32 s4, s0
+; CHECK-HARD-NEXT:    vmov.f32 s0, s5
+; CHECK-HARD-NEXT:    vmov.f32 s1, s4
+; CHECK-HARD-NEXT:    vmov.f32 s2, s4
+; CHECK-HARD-NEXT:    vmov.f32 s3, s5
+; CHECK-HARD-NEXT:    b non_variadic_big
 entry:
   %call = tail call i1 (float, float, float, float, float, float) @non_variadic_big(float %y, float %x, float %x, float %y, float %x, float %y)
   ret void
@@ -41,9 +91,40 @@ entry:
 
 ; Variadic functions cannot use FP regs to pass arguments; only GP regs.
 define void @variadic_fp_big(float %x, float %y) {
-; CHECK-LABEL: variadic_fp_big:
-; CHECK: bl variadic
+; CHECK-SOFT-LABEL: variadic_fp_big:
+; CHECK-SOFT:       @ %bb.0: @ %entry
+; CHECK-SOFT-NEXT:    push {r11, lr}
+; CHECK-SOFT-NEXT:    sub sp, sp, #8
+; CHECK-SOFT-NEXT:    mov r3, r1
+; CHECK-SOFT-NEXT:    mov r2, r0
+; CHECK-SOFT-NEXT:    vmov s0, r3
+; CHECK-SOFT-NEXT:    vmov s0, r2
+; CHECK-SOFT-NEXT:    mov r0, sp
+; CHECK-SOFT-NEXT:    str r3, [r0, #4]
+; CHECK-SOFT-NEXT:    str r2, [r0]
+; CHECK-SOFT-NEXT:    mov r0, r3
+; CHECK-SOFT-NEXT:    mov r1, r2
+; CHECK-SOFT-NEXT:    bl variadic
+; CHECK-SOFT-NEXT:    add sp, sp, #8
+; CHECK-SOFT-NEXT:    pop {r11, pc}
+;
+; CHECK-HARD-LABEL: variadic_fp_big:
+; CHECK-HARD:       @ %bb.0: @ %entry
+; CHECK-HARD-NEXT:    push {r11, lr}
+; CHECK-HARD-NEXT:    sub sp, sp, #8
+; CHECK-HARD-NEXT:    mov r0, sp
+; CHECK-HARD-NEXT:    vstr s1, [r0, #4]
+; CHECK-HARD-NEXT:    vstr s0, [r0]
+; CHECK-HARD-NEXT:    vmov r2, s0
+; CHECK-HARD-NEXT:    vmov r3, s1
+; CHECK-HARD-NEXT:    mov r0, r3
+; CHECK-HARD-NEXT:    mov r1, r2
+; CHECK-HARD-NEXT:    bl variadic
+; CHECK-HARD-NEXT:    add sp, sp, #8
+; CHECK-HARD-NEXT:    pop {r11, pc}
 entry:
   %call = tail call i1 (float, ...) @variadic(float %y, float %x, float %x, float %y, float %x, float %y)
   ret void
 }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}

>From 39371e0582ff6bbcc558970a82d007b132e6ed10 Mon Sep 17 00:00:00 2001
From: Kiran <kiran.sturt at arm.com>
Date: Tue, 27 Aug 2024 10:43:59 +0100
Subject: [PATCH 2/2] Seperate frontend changes, add debug directives, remove
 redundant stuff from tests

---
 clang/lib/CodeGen/CGCall.cpp                 | 2 +-
 llvm/include/llvm/CodeGen/CallingConvLower.h | 2 ++
 llvm/lib/CodeGen/CallingConvLower.cpp        | 2 ++
 llvm/test/CodeGen/ARM/tail-call-float.ll     | 4 ++--
 4 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 05773f91f986ba..ca2c79b51ac96b 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -5086,7 +5086,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
   RawAddress SRetAlloca = RawAddress::invalid();
   llvm::Value *UnusedReturnSizePtr = nullptr;
   if (RetAI.isIndirect() || RetAI.isInAlloca() || RetAI.isCoerceAndExpand()) {
-    if ((IsVirtualFunctionPointerThunk || IsMustTail) && RetAI.isIndirect()) {
+    if (IsVirtualFunctionPointerThunk && RetAI.isIndirect()) {
       SRetPtr = makeNaturalAddressForPointer(CurFn->arg_begin() +
                                                  IRFunctionArgs.getSRetArgNo(),
                                              RetTy, CharUnits::fromQuantity(1));
diff --git a/llvm/include/llvm/CodeGen/CallingConvLower.h b/llvm/include/llvm/CodeGen/CallingConvLower.h
index 12a6df16e279b4..f94cb628965faa 100644
--- a/llvm/include/llvm/CodeGen/CallingConvLower.h
+++ b/llvm/include/llvm/CodeGen/CallingConvLower.h
@@ -540,7 +540,9 @@ class CCState {
                });
   }
 
+#ifndef NDEBUG
   void dump() const;
+#endif
 
 private:
   /// MarkAllocated - Mark a register and all of its aliases as allocated.
diff --git a/llvm/lib/CodeGen/CallingConvLower.cpp b/llvm/lib/CodeGen/CallingConvLower.cpp
index 7ba3ea83115db2..38884762944319 100644
--- a/llvm/lib/CodeGen/CallingConvLower.cpp
+++ b/llvm/lib/CodeGen/CallingConvLower.cpp
@@ -291,6 +291,7 @@ bool CCState::resultsCompatible(CallingConv::ID CalleeCC,
                     RVLocs2.end(), AreCompatible);
 }
 
+#ifndef NDEBUG
 void CCState::dump() const {
   dbgs() << "CCState:\n";
   for (const CCValAssign &Loc : Locs) {
@@ -351,3 +352,4 @@ void CCState::dump() const {
     dbgs() << "\n";
   }
 }
+#endif
\ No newline at end of file
diff --git a/llvm/test/CodeGen/ARM/tail-call-float.ll b/llvm/test/CodeGen/ARM/tail-call-float.ll
index 6fa4b9229f64c8..2f834976dd403c 100644
--- a/llvm/test/CodeGen/ARM/tail-call-float.ll
+++ b/llvm/test/CodeGen/ARM/tail-call-float.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc -mtriple armv7 -target-abi aapcs -float-abi soft -O0 -o - < %s \
-; RUN:   | FileCheck %s -check-prefix CHECK-SOFT -check-prefix CHECK
+; RUN:   | FileCheck %s -check-prefix CHECK-SOFT
 ; RUN: llc -mtriple armv7 -target-abi aapcs -float-abi hard -O0 -o - < %s \
-; RUN:   | FileCheck %s -check-prefix CHECK-HARD -check-prefix CHECK
+; RUN:   | FileCheck %s -check-prefix CHECK-HARD
 
 ; Tests for passing floating-point regs. Variadic functions will always use
 ; general-purpose registers. Standard functions will use the floating-point



More information about the cfe-commits mailing list