[AArch64] Make the use of FP instructions optional, but enabled by default.

This adds a new subtarget feature called FPARMv8 (implied by NEON), and
predicates the support of the FP instructions and registers on this feature.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193739 91177308-0d34-0410-b5e6-96231b3b80d8

24 files changed, 225 insertions, 40 deletions

diff --git a/lib/Target/AArch64/AArch64.td b/lib/Target/AArch64/AArch64.td
index e17052b4a5..083064b711 100644
--- a/lib/Target/AArch64/AArch64.td
+++ b/lib/Target/AArch64/AArch64.td
@@ -21,8 +21,11 @@ include "llvm/Target/Target.td"
 // AArch64 Subtarget features.
 //
 
+def FeatureFPARMv8 : SubtargetFeature<"fp-armv8", "HasFPARMv8", "true",
+  "Enable ARMv8 FP">;
+
 def FeatureNEON : SubtargetFeature<"neon", "HasNEON", "true",
-  "Enable Advanced SIMD instructions">;
+  "Enable Advanced SIMD instructions", [FeatureFPARMv8]>;
 
 def FeatureCrypto : SubtargetFeature<"crypto", "HasCrypto", "true",
   "Enable cryptographic instructions">;
@@ -33,7 +36,8 @@ def FeatureCrypto : SubtargetFeature<"crypto", "HasCrypto", "true",
 
 include "AArch64Schedule.td"
 
-def : Processor<"generic", GenericItineraries, [FeatureNEON, FeatureCrypto]>;
+def : Processor<"generic", GenericItineraries,
+                [FeatureFPARMv8, FeatureNEON, FeatureCrypto]>;
 
 //===----------------------------------------------------------------------===//
 // Register File Description
diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 87bb847993..323acdd01d 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -50,10 +50,13 @@ AArch64TargetLowering::AArch64TargetLowering(AArch64TargetMachine &TM)
   // Scalar register <-> type mapping
   addRegisterClass(MVT::i32, &AArch64::GPR32RegClass);
   addRegisterClass(MVT::i64, &AArch64::GPR64RegClass);
-  addRegisterClass(MVT::f16, &AArch64::FPR16RegClass);
-  addRegisterClass(MVT::f32, &AArch64::FPR32RegClass);
-  addRegisterClass(MVT::f64, &AArch64::FPR64RegClass);
-  addRegisterClass(MVT::f128, &AArch64::FPR128RegClass);
+
+  if (Subtarget->hasFPARMv8()) {
+    addRegisterClass(MVT::f16, &AArch64::FPR16RegClass);
+    addRegisterClass(MVT::f32, &AArch64::FPR32RegClass);
+    addRegisterClass(MVT::f64, &AArch64::FPR64RegClass);
+    addRegisterClass(MVT::f128, &AArch64::FPR128RegClass);
+  }
 
   if (Subtarget->hasNEON()) {
     // And the vectors
@@ -961,24 +964,31 @@ AArch64TargetLowering::SaveVarArgRegisters(CCState &CCInfo, SelectionDAG &DAG,
     }
   }
 
+  if (getSubtarget()->hasFPARMv8()) {
   unsigned FPRSaveSize = 16 * (NumFPRArgRegs - FirstVariadicFPR);
   int FPRIdx = 0;
-  if (FPRSaveSize != 0) {
-    FPRIdx = MFI->CreateStackObject(FPRSaveSize, 16, false);
-
-    SDValue FIN = DAG.getFrameIndex(FPRIdx, getPointerTy());
-
-    for (unsigned i = FirstVariadicFPR; i < NumFPRArgRegs; ++i) {
-      unsigned VReg = MF.addLiveIn(AArch64FPRArgRegs[i],
-                                   &AArch64::FPR128RegClass);
-      SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::f128);
-      SDValue Store = DAG.getStore(Val.getValue(1), DL, Val, FIN,
-                                   MachinePointerInfo::getStack(i * 16),
-                                   false, false, 0);
-      MemOps.push_back(Store);
-      FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
-                        DAG.getConstant(16, getPointerTy()));
+    // According to the AArch64 Procedure Call Standard, section B.1/B.3, we
+    // can omit a register save area if we know we'll never use registers of
+    // that class.
+    if (FPRSaveSize != 0) {
+      FPRIdx = MFI->CreateStackObject(FPRSaveSize, 16, false);
+
+      SDValue FIN = DAG.getFrameIndex(FPRIdx, getPointerTy());
+
+      for (unsigned i = FirstVariadicFPR; i < NumFPRArgRegs; ++i) {
+        unsigned VReg = MF.addLiveIn(AArch64FPRArgRegs[i],
+            &AArch64::FPR128RegClass);
+        SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::f128);
+        SDValue Store = DAG.getStore(Val.getValue(1), DL, Val, FIN,
+            MachinePointerInfo::getStack(i * 16),
+            false, false, 0);
+        MemOps.push_back(Store);
+        FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
+            DAG.getConstant(16, getPointerTy()));
+      }
     }
+    FuncInfo->setVariadicFPRIdx(FPRIdx);
+    FuncInfo->setVariadicFPRSize(FPRSaveSize);
   }
 
   int StackIdx = MFI->CreateFixedObject(8, CCInfo.getNextStackOffset(), true);
@@ -986,8 +996,6 @@ AArch64TargetLowering::SaveVarArgRegisters(CCState &CCInfo, SelectionDAG &DAG,
   FuncInfo->setVariadicStackIdx(StackIdx);
   FuncInfo->setVariadicGPRIdx(GPRIdx);
   FuncInfo->setVariadicGPRSize(GPRSaveSize);
-  FuncInfo->setVariadicFPRIdx(FPRIdx);
-  FuncInfo->setVariadicFPRSize(FPRSaveSize);
 
   if (!MemOps.empty()) {
     Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &MemOps[0],
diff --git a/lib/Target/AArch64/AArch64InstrFormats.td b/lib/Target/AArch64/AArch64InstrFormats.td
index d69db317fd..7a41ce05a1 100644
--- a/lib/Target/AArch64/AArch64InstrFormats.td
+++ b/lib/Target/AArch64/AArch64InstrFormats.td
@@ -383,6 +383,8 @@ class A64I_extract<bit sf, bits<3> op, bit n,
   // Inherits Rd in 4-0
 }
 
+let Predicates = [HasFPARMv8] in {
+
 // Format for floating-point compare instructions.
 class A64I_fpcmp<bit m, bit s, bits<2> type, bits<2> op, bits<5> opcode2,
                 dag outs, dag ins, string asmstr,
@@ -562,6 +564,8 @@ class A64I_fpimm<bit m, bit s, bits<2> type, bits<5> imm5,
   // Inherit Rd in 4-0
 }
 
+}
+
 // Format for load-register (literal) instructions.
 class A64I_LDRlit<bits<2> opc, bit v,
                   dag outs, dag ins, string asmstr,
diff --git a/lib/Target/AArch64/AArch64InstrInfo.td b/lib/Target/AArch64/AArch64InstrInfo.td
index 43df2b4362..ae217f9d4f 100644
--- a/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/lib/Target/AArch64/AArch64InstrInfo.td
@@ -14,6 +14,8 @@
 //===----------------------------------------------------------------------===//
 // ARM Instruction Predicate Definitions.
 //
+def HasFPARMv8       : Predicate<"Subtarget->hasFPARMv8()">,
+                               AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
 def HasNEON          : Predicate<"Subtarget->hasNEON()">,
                                  AssemblerPredicate<"FeatureNEON", "neon">;
 def HasCrypto        : Predicate<"Subtarget->hasCrypto()">,
@@ -2195,7 +2197,7 @@ def FNMSUBdddd : A64I_fpdp3Impl<"fnmsub", FPR64, f64, 0b01, 0b1, 0b1, fnmsub>;
 
 // Extra patterns for when we're allowed to optimise separate multiplication and
 // addition.
-let Predicates = [UseFusedMAC] in {
+let Predicates = [HasFPARMv8, UseFusedMAC] in {
 def : Pat<(f32 (fadd FPR32:$Ra, (f32 (fmul FPR32:$Rn, FPR32:$Rm)))),
           (FMADDssss FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
 def : Pat<(f32 (fsub FPR32:$Ra, (f32 (fmul FPR32:$Rn, FPR32:$Rm)))),
@@ -2351,6 +2353,7 @@ defm FCVTM : A64I_fptointRM<0b10, 0b0, "fcvtm">;
 defm FCVTZ : A64I_fptointRM<0b11, 0b0, "fcvtz">;
 defm FCVTA : A64I_fptointRM<0b00, 0b1, "fcvta">;
 
+let Predicates = [HasFPARMv8] in {
 def : Pat<(i32 (fp_to_sint f32:$Rn)), (FCVTZSws $Rn)>;
 def : Pat<(i64 (fp_to_sint f32:$Rn)), (FCVTZSxs $Rn)>;
 def : Pat<(i32 (fp_to_uint f32:$Rn)), (FCVTZUws $Rn)>;
@@ -2359,6 +2362,7 @@ def : Pat<(i32 (fp_to_sint f64:$Rn)), (FCVTZSwd $Rn)>;
 def : Pat<(i64 (fp_to_sint f64:$Rn)), (FCVTZSxd $Rn)>;
 def : Pat<(i32 (fp_to_uint f64:$Rn)), (FCVTZUwd $Rn)>;
 def : Pat<(i64 (fp_to_uint f64:$Rn)), (FCVTZUxd $Rn)>;
+}
 
 multiclass A64I_inttofp<bit o0, string asmop> {
   def CVTFsw : A64I_fpintI<0b0, 0b00, 0b00, {0, 1, o0}, FPR32, GPR32, asmop>;
@@ -2370,6 +2374,7 @@ multiclass A64I_inttofp<bit o0, string asmop> {
 defm S : A64I_inttofp<0b0, "scvtf">;
 defm U : A64I_inttofp<0b1, "ucvtf">;
 
+let Predicates = [HasFPARMv8] in {
 def : Pat<(f32 (sint_to_fp i32:$Rn)), (SCVTFsw $Rn)>;
 def : Pat<(f32 (sint_to_fp i64:$Rn)), (SCVTFsx $Rn)>;
 def : Pat<(f64 (sint_to_fp i32:$Rn)), (SCVTFdw $Rn)>;
@@ -2378,16 +2383,19 @@ def : Pat<(f32 (uint_to_fp i32:$Rn)), (UCVTFsw $Rn)>;
 def : Pat<(f32 (uint_to_fp i64:$Rn)), (UCVTFsx $Rn)>;
 def : Pat<(f64 (uint_to_fp i32:$Rn)), (UCVTFdw $Rn)>;
 def : Pat<(f64 (uint_to_fp i64:$Rn)), (UCVTFdx $Rn)>;
+}
 
 def FMOVws : A64I_fpintI<0b0, 0b00, 0b00, 0b110, GPR32, FPR32, "fmov">;
 def FMOVsw : A64I_fpintI<0b0, 0b00, 0b00, 0b111, FPR32, GPR32, "fmov">;
 def FMOVxd : A64I_fpintI<0b1, 0b01, 0b00, 0b110, GPR64, FPR64, "fmov">;
 def FMOVdx : A64I_fpintI<0b1, 0b01, 0b00, 0b111, FPR64, GPR64, "fmov">;
 
+let Predicates = [HasFPARMv8] in {
 def : Pat<(i32 (bitconvert f32:$Rn)), (FMOVws $Rn)>;
 def : Pat<(f32 (bitconvert i32:$Rn)), (FMOVsw $Rn)>;
 def : Pat<(i64 (bitconvert f64:$Rn)), (FMOVxd $Rn)>;
 def : Pat<(f64 (bitconvert i64:$Rn)), (FMOVdx $Rn)>;
+}
 
 def lane1_asmoperand : AsmOperandClass {
   let Name = "Lane1";
@@ -2410,11 +2418,13 @@ let DecoderMethod =  "DecodeFMOVLaneInstruction" in {
                           "fmov\t$Rd.d[$Lane], $Rn", [], NoItinerary>;
 }
 
+let Predicates = [HasFPARMv8] in {
 def : InstAlias<"fmov $Rd, $Rn.2d[$Lane]",
                 (FMOVxv GPR64:$Rd, VPR128:$Rn, lane1:$Lane), 0b0>;
 
 def : InstAlias<"fmov $Rd.2d[$Lane], $Rn",
                 (FMOVvx VPR128:$Rd, GPR64:$Rn, lane1:$Lane), 0b0>;
+}
 
 //===----------------------------------------------------------------------===//
 // Floating-point immediate instructions
@@ -2508,11 +2518,15 @@ let mayLoad = 1 in {
   def LDRx_lit : A64I_LDRlitSimple<0b01, 0b0, GPR64>;
 }
 
+let Predicates = [HasFPARMv8] in {
 def LDRs_lit  : A64I_LDRlitSimple<0b00, 0b1, FPR32>;
 def LDRd_lit  : A64I_LDRlitSimple<0b01, 0b1, FPR64>;
+}
 
 let mayLoad = 1 in {
+  let Predicates = [HasFPARMv8] in {
   def LDRq_lit : A64I_LDRlitSimple<0b10, 0b1, FPR128>;
+  }
 
 
   def LDRSWx_lit : A64I_LDRlit<0b10, 0b0,
@@ -3106,6 +3120,7 @@ defm LS32
 defm LS64
   : A64I_LDRSTR_unsigned<"LS64", 0b11, 0b0, 0b0, "", GPR64, dword_addrparams>;
 
+let Predicates = [HasFPARMv8] in {
 // STR/LDR to/from a B register
 defm LSFP8
   : A64I_LDRSTR_unsigned<"LSFP8", 0b00, 0b1, 0b0, "", FPR8, byte_addrparams>;
@@ -3124,6 +3139,7 @@ defm LSFP64
 defm LSFP128
   : A64I_LDRSTR_unsigned<"LSFP128", 0b00, 0b1, 0b1, "", FPR128,
                          qword_addrparams>;
+}
 
 //===------------------------------
 // 2.3 Signed loads
@@ -3579,10 +3595,13 @@ multiclass A64I_LSPsimple<bits<2> opc, bit v, RegisterClass SomeReg,
 
 defm LSPair32 : A64I_LSPsimple<0b00, 0b0, GPR32, word_simm7, "LSPair32">;
 defm LSPair64 : A64I_LSPsimple<0b10, 0b0, GPR64, dword_simm7, "LSPair64">;
+
+let Predicates = [HasFPARMv8] in {
 defm LSFPPair32 : A64I_LSPsimple<0b00, 0b1, FPR32, word_simm7, "LSFPPair32">;
 defm LSFPPair64 : A64I_LSPsimple<0b01, 0b1, FPR64,  dword_simm7, "LSFPPair64">;
 defm LSFPPair128 : A64I_LSPsimple<0b10, 0b1, FPR128, qword_simm7,
                                   "LSFPPair128">;
+}
 
 
 def LDPSWx : A64I_LSPoffset<0b01, 0b0, 0b1,
diff --git a/lib/Target/AArch64/AArch64Subtarget.cpp b/lib/Target/AArch64/AArch64Subtarget.cpp
index d71bb4e973..eece389302 100644
--- a/lib/Target/AArch64/AArch64Subtarget.cpp
+++ b/lib/Target/AArch64/AArch64Subtarget.cpp
@@ -26,10 +26,27 @@
 using namespace llvm;
 
 AArch64Subtarget::AArch64Subtarget(StringRef TT, StringRef CPU, StringRef FS)
-    : AArch64GenSubtargetInfo(TT, CPU, FS), HasNEON(false), HasCrypto(false),
-      TargetTriple(TT) {
+    : AArch64GenSubtargetInfo(TT, CPU, FS), HasFPARMv8(false), HasNEON(false),
+      HasCrypto(false), TargetTriple(TT), CPUString(CPU) {
 
-  ParseSubtargetFeatures(CPU, FS);
+  initializeSubtargetFeatures(CPU, FS);
+}
+
+void AArch64Subtarget::initializeSubtargetFeatures(StringRef CPU,
+                                                   StringRef FS) {
+  if (CPU.empty())
+    CPUString = "generic";
+
+  std::string FullFS = FS;
+  if (CPUString == "generic") {
+    // Enable FP by default.
+    if (FullFS.empty())
+      FullFS = "+fp-armv8";
+    else
+      FullFS = "+fp-armv8," + FullFS;
+  }
+
+  ParseSubtargetFeatures(CPU, FullFS);
 }
 
 bool AArch64Subtarget::GVIsIndirectSymbol(const GlobalValue *GV,
diff --git a/lib/Target/AArch64/AArch64Subtarget.h b/lib/Target/AArch64/AArch64Subtarget.h
index f262b94617..57eb187c9d 100644
--- a/lib/Target/AArch64/AArch64Subtarget.h
+++ b/lib/Target/AArch64/AArch64Subtarget.h
@@ -28,11 +28,19 @@ class GlobalValue;
 
 class AArch64Subtarget : public AArch64GenSubtargetInfo {
 protected:
+  bool HasFPARMv8;
   bool HasNEON;
   bool HasCrypto;
 
   /// TargetTriple - What processor and OS we're targeting.
   Triple TargetTriple;
+
+  /// CPUString - String name of used CPU.
+  std::string CPUString;
+
+private:
+  void initializeSubtargetFeatures(StringRef CPU, StringRef FS);
+
 public:
   /// This constructor initializes the data members to match that
   /// of the specified triple.
@@ -52,9 +60,11 @@ public:
   bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); }
   bool isTargetLinux() const { return TargetTriple.isOSLinux(); }
 
+  bool hasFPARMv8() const { return HasFPARMv8; }
   bool hasNEON() const { return HasNEON; }
-
   bool hasCrypto() const { return HasCrypto; }
+
+  const std::string & getCPUString() const { return CPUString; }
 };
 } // End llvm namespace
 
diff --git a/test/CodeGen/AArch64/alloca.ll b/test/CodeGen/AArch64/alloca.ll
index c2f84cdcce..1d3c0a02ac 100644
--- a/test/CodeGen/AArch64/alloca.ll
+++ b/test/CodeGen/AArch64/alloca.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -mtriple=aarch64-none-linux-gnu -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 -verify-machineinstrs < %s | FileCheck --check-prefix=CHECK-NOFP %s
 
 declare void @use_addr(i8*)
 
@@ -66,16 +67,22 @@ define i64 @test_alloca_with_local(i64 %n) {
 }
 
 define void @test_variadic_alloca(i64 %n, ...) {
-; CHECK-LABEL: test_variadic_alloca:
+; CHECK: test_variadic_alloca:
 
 ; CHECK: sub     sp, sp, #208
 ; CHECK: stp     x29, x30, [sp, #192]
 ; CHECK: add     x29, sp, #192
 ; CHECK: sub     [[TMP:x[0-9]+]], x29, #192
 ; CHECK: add     x8, [[TMP]], #0
-; CHECK: str     q7, [x8, #112]
+; CHECK-FP: str     q7, [x8, #112]
 ; [...]
-; CHECK: str     q1, [x8, #16]
+; CHECK-FP: str     q1, [x8, #16]
+
+; CHECK-NOFP: sub     sp, sp, #80
+; CHECK-NOFP: stp     x29, x30, [sp, #64]
+; CHECK-NOFP: add     x29, sp, #64
+; CHECK-NOFP: sub     [[TMP:x[0-9]+]], x29, #64
+; CHECK-NOFP: add     x8, [[TMP]], #0
 
   %addr = alloca i8, i64 %n
 
@@ -86,6 +93,10 @@ define void @test_variadic_alloca(i64 %n, ...) {
 ; CHECK: sub sp, x29, #192
 ; CHECK: ldp x29, x30, [sp, #192]
 ; CHECK: add sp, sp, #208
+
+; CHECK-NOFP: sub sp, x29, #64
+; CHECK-NOFP: ldp x29, x30, [sp, #64]
+; CHECK-NOFP: add sp, sp, #80
 }
 
 define void @test_alloca_large_frame(i64 %n) {
diff --git a/test/CodeGen/AArch64/cond-sel.ll b/test/CodeGen/AArch64/cond-sel.ll
index 3a5b8191a8..9c1dfeb3c8 100644
--- a/test/CodeGen/AArch64/cond-sel.ll
+++ b/test/CodeGen/AArch64/cond-sel.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 @var32 = global i32 0
 @var64 = global i64 0
@@ -30,6 +31,7 @@ define void @test_floatcsel(float %lhs32, float %rhs32, double %lhs64, double %r
 
   %tst1 = fcmp one float %lhs32, %rhs32
 ; CHECK: fcmp {{s[0-9]+}}, {{s[0-9]+}}
+; CHECK-NOFP-NOT: fcmp
   %val1 = select i1 %tst1, i32 42, i32 52
   store i32 %val1, i32* @var32
 ; CHECK: movz [[W52:w[0-9]+]], #52
@@ -40,6 +42,7 @@ define void @test_floatcsel(float %lhs32, float %rhs32, double %lhs64, double %r
 
   %tst2 = fcmp ueq double %lhs64, %rhs64
 ; CHECK: fcmp {{d[0-9]+}}, {{d[0-9]+}}
+; CHECK-NOFP-NOT: fcmp
   %val2 = select i1 %tst2, i64 9, i64 15
   store i64 %val2, i64* @var64
 ; CHECK: movz [[CONST15:x[0-9]+]], #15
diff --git a/test/CodeGen/AArch64/directcond.ll b/test/CodeGen/AArch64/directcond.ll
index 13f032dbec..12c7b6aed6 100644
--- a/test/CodeGen/AArch64/directcond.ll
+++ b/test/CodeGen/AArch64/directcond.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 define i32 @test_select_i32(i1 %bit, i32 %a, i32 %b) {
 ; CHECK-LABEL: test_select_i32:
@@ -26,7 +27,7 @@ define float @test_select_float(i1 %bit, float %a, float %b) {
 ; CHECK: movz [[ONE:w[0-9]+]], #1
 ; CHECK: tst w0, [[ONE]]
 ; CHECK-NEXT: fcsel s0, s0, s1, ne
-
+; CHECK-NOFP-NOT: fcsel
   ret float %val
 }
 
@@ -36,6 +37,7 @@ define double @test_select_double(i1 %bit, double %a, double %b) {
 ; CHECK: movz [[ONE:w[0-9]+]], #1
 ; CHECK: tst w0, [[ONE]]
 ; CHECK-NEXT: fcsel d0, d0, d1, ne
+; CHECK-NOFP-NOT: fcsel
 
   ret double %val
 }
@@ -56,6 +58,7 @@ define i1 @test_setcc_float(float %lhs, float %rhs) {
   %val = fcmp oeq float %lhs, %rhs
 ; CHECK: fcmp s0, s1
 ; CHECK: csinc w0, wzr, wzr, ne
+; CHECK-NOFP-NOT: fcmp
   ret i1 %val
 }
 
@@ -64,6 +67,7 @@ define i1 @test_setcc_double(double %lhs, double %rhs) {
   %val = fcmp oeq double %lhs, %rhs
 ; CHECK: fcmp d0, d1
 ; CHECK: csinc w0, wzr, wzr, ne
+; CHECK-NOFP-NOT: fcmp
   ret i1 %val
 }
 
diff --git a/test/CodeGen/AArch64/func-argpassing.ll b/test/CodeGen/AArch64/func-argpassing.ll
index 83fc0a0e50..430d77f9e9 100644
--- a/test/CodeGen/AArch64/func-argpassing.ll
+++ b/test/CodeGen/AArch64/func-argpassing.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 %myStruct = type { i64 , i8, i32 }
 
@@ -23,6 +24,7 @@ define void @add_floats(float %val1, float %val2) {
 ; CHECK-LABEL: add_floats:
     %newval = fadd float %val1, %val2
 ; CHECK: fadd [[ADDRES:s[0-9]+]], s0, s1
+; CHECK-NOFP-NOT: fadd
     store float %newval, float* @varfloat
 ; CHECK: str [[ADDRES]], [{{x[0-9]+}}, #:lo12:varfloat]
     ret void
@@ -84,6 +86,7 @@ define double @return_double() {
 ; CHECK-LABEL: return_double:
     ret double 3.14
 ; CHECK: ldr d0, [{{x[0-9]+}}, #:lo12:.LCPI
+; CHECK-NOFP-NOT: ldr d0,
 }
 
 ; This is the kind of IR clang will produce for returning a struct
@@ -139,6 +142,7 @@ define i32 @struct_on_stack(i8 %var0, i16 %var1, i32 %var2, i64 %var3, i128 %var
     store volatile double %notstacked, double* @vardouble
 ; CHECK-NOT: ldr d0
 ; CHECK: str d0, [{{x[0-9]+}}, #:lo12:vardouble
+; CHECK-NOFP-NOT: str d0,
 
     %retval = load volatile i32* %stacked
     ret i32 %retval
diff --git a/test/CodeGen/AArch64/func-calls.ll b/test/CodeGen/AArch64/func-calls.ll
index 03cd8070a6..ac188bb3bb 100644
--- a/test/CodeGen/AArch64/func-calls.ll
+++ b/test/CodeGen/AArch64/func-calls.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 %myStruct = type { i64 , i8, i32 }
 
@@ -31,6 +32,8 @@ define void @simple_args() {
 ; CHECK-DAG: ldr s1, [{{x[0-9]+}}, #:lo12:varfloat_2]
 ; CHECK-DAG: ldr s0, [{{x[0-9]+}}, #:lo12:varfloat]
 ; CHECK: bl take_floats
+; CHECK-NOFP-NOT: ldr s1,
+; CHECK-NOFP-NOT: ldr s0,
 
   ret void
 }
@@ -52,6 +55,7 @@ define void @simple_rets() {
   store double %dbl, double* @vardouble
 ; CHECK: bl return_double
 ; CHECK: str d0, [{{x[0-9]+}}, #:lo12:vardouble]
+; CHECK-NOFP-NOT: str d0,
 
   %arr = call [2 x i64] @return_smallstruct()
   store [2 x i64] %arr, [2 x i64]* @varsmallstruct
@@ -87,6 +91,7 @@ define void @check_stack_args() {
 ; CHECK-DAG: str {{w[0-9]+}}, [x[[SPREG]], #12]
 ; CHECK-DAG: fmov d0,
 ; CHECK: bl struct_on_stack
+; CHECK-NOFP-NOT: fmov
 
   call void @stacked_fpu(float -1.0, double 1.0, float 4.0, float 2.0,
                          float -2.0, float -8.0, float 16.0, float 1.0,
diff --git a/test/CodeGen/AArch64/ldst-regoffset.ll b/test/CodeGen/AArch64/ldst-regoffset.ll
index c83fb52146..db30fd915f 100644
--- a/test/CodeGen/AArch64/ldst-regoffset.ll
+++ b/test/CodeGen/AArch64/ldst-regoffset.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 @var_8bit = global i8 0
 @var_16bit = global i16 0
@@ -197,11 +198,13 @@ define void @ldst_float(float* %base, i32 %off32, i64 %off64) {
    %val_sxtwN = load volatile float* %addr_sxtwN
    store volatile float %val_sxtwN, float* @var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #2]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   %addr_lslN = getelementptr float* %base, i64 %off64
   %val_lslN = load volatile float* %addr_lslN
   store volatile float %val_lslN, float* @var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}, lsl #2]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   %addrint_uxtw = ptrtoint float* %base to i64
   %offset_uxtw = zext i32 %off32 to i64
@@ -210,6 +213,7 @@ define void @ldst_float(float* %base, i32 %off32, i64 %off64) {
   %val_uxtw = load volatile float* %addr_uxtw
   store volatile float %val_uxtw, float* @var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   %base_sxtw = ptrtoint float* %base to i64
   %offset_sxtw = sext i32 %off32 to i64
@@ -218,6 +222,7 @@ define void @ldst_float(float* %base, i32 %off32, i64 %off64) {
   %val64_sxtw = load volatile float* %addr_sxtw
   store volatile float %val64_sxtw, float* @var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   %base_lsl = ptrtoint float* %base to i64
   %addrint_lsl = add i64 %base_lsl, %off64
@@ -225,6 +230,7 @@ define void @ldst_float(float* %base, i32 %off32, i64 %off64) {
   %val64_lsl = load volatile float* %addr_lsl
   store volatile float %val64_lsl, float* @var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   %base_uxtwN = ptrtoint float* %base to i64
   %offset_uxtwN = zext i32 %off32 to i64
@@ -234,6 +240,7 @@ define void @ldst_float(float* %base, i32 %off32, i64 %off64) {
   %val64 = load volatile float* @var_float
   store volatile float %val64, float* %addr_uxtwN
 ; CHECK: str {{s[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw #2]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
    ret void
 }
 
@@ -244,11 +251,13 @@ define void @ldst_double(double* %base, i32 %off32, i64 %off64) {
    %val_sxtwN = load volatile double* %addr_sxtwN
    store volatile double %val_sxtwN, double* @var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #3]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   %addr_lslN = getelementptr double* %base, i64 %off64
   %val_lslN = load volatile double* %addr_lslN
   store volatile double %val_lslN, double* @var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}, lsl #3]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   %addrint_uxtw = ptrtoint double* %base to i64
   %offset_uxtw = zext i32 %off32 to i64
@@ -257,6 +266,7 @@ define void @ldst_double(double* %base, i32 %off32, i64 %off64) {
   %val_uxtw = load volatile double* %addr_uxtw
   store volatile double %val_uxtw, double* @var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   %base_sxtw = ptrtoint double* %base to i64
   %offset_sxtw = sext i32 %off32 to i64
@@ -265,6 +275,7 @@ define void @ldst_double(double* %base, i32 %off32, i64 %off64) {
   %val64_sxtw = load volatile double* %addr_sxtw
   store volatile double %val64_sxtw, double* @var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   %base_lsl = ptrtoint double* %base to i64
   %addrint_lsl = add i64 %base_lsl, %off64
@@ -272,6 +283,7 @@ define void @ldst_double(double* %base, i32 %off32, i64 %off64) {
   %val64_lsl = load volatile double* %addr_lsl
   store volatile double %val64_lsl, double* @var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   %base_uxtwN = ptrtoint double* %base to i64
   %offset_uxtwN = zext i32 %off32 to i64
@@ -281,6 +293,7 @@ define void @ldst_double(double* %base, i32 %off32, i64 %off64) {
   %val64 = load volatile double* @var_double
   store volatile double %val64, double* %addr_uxtwN
 ; CHECK: str {{d[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw #3]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
    ret void
 }
 
@@ -292,11 +305,13 @@ define void @ldst_128bit(fp128* %base, i32 %off32, i64 %off64) {
    %val_sxtwN = load volatile fp128* %addr_sxtwN
    store volatile fp128 %val_sxtwN, fp128* %base
 ; CHECK: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
 
   %addr_lslN = getelementptr fp128* %base, i64 %off64
   %val_lslN = load volatile fp128* %addr_lslN
   store volatile fp128 %val_lslN, fp128* %base
 ; CHECK: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}, lsl #4]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
 
   %addrint_uxtw = ptrtoint fp128* %base to i64
   %offset_uxtw = zext i32 %off32 to i64
@@ -305,6 +320,7 @@ define void @ldst_128bit(fp128* %base, i32 %off32, i64 %off64) {
   %val_uxtw = load volatile fp128* %addr_uxtw
   store volatile fp128 %val_uxtw, fp128* %base
 ; CHECK: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
 
   %base_sxtw = ptrtoint fp128* %base to i64
   %offset_sxtw = sext i32 %off32 to i64
@@ -313,6 +329,7 @@ define void @ldst_128bit(fp128* %base, i32 %off32, i64 %off64) {
   %val64_sxtw = load volatile fp128* %addr_sxtw
   store volatile fp128 %val64_sxtw, fp128* %base
 ; CHECK: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
 
   %base_lsl = ptrtoint fp128* %base to i64
   %addrint_lsl = add i64 %base_lsl, %off64
@@ -320,6 +337,7 @@ define void @ldst_128bit(fp128* %base, i32 %off32, i64 %off64) {
   %val64_lsl = load volatile fp128* %addr_lsl
   store volatile fp128 %val64_lsl, fp128* %base
 ; CHECK: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{x[0-9]+}}]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
 
   %base_uxtwN = ptrtoint fp128* %base to i64
   %offset_uxtwN = zext i32 %off32 to i64
@@ -329,5 +347,6 @@ define void @ldst_128bit(fp128* %base, i32 %off32, i64 %off64) {
   %val64 = load volatile fp128* %base
   store volatile fp128 %val64, fp128* %addr_uxtwN
 ; CHECK: str {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, uxtw #4]
+; CHECK-NOFP-NOT: ldr {{q[0-9]+}}, [{{x[0-9]+}}, {{w[0-9]+}}, sxtw #4]
    ret void
 }
diff --git a/test/CodeGen/AArch64/ldst-unscaledimm.ll b/test/CodeGen/AArch64/ldst-unscaledimm.ll
index 03dedccce2..bea5bb5d6d 100644
--- a/test/CodeGen/AArch64/ldst-unscaledimm.ll
+++ b/test/CodeGen/AArch64/ldst-unscaledimm.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 @var_8bit = global i8 0
 @var_16bit = global i16 0
@@ -194,9 +195,11 @@ define void @ldst_float() {
 
   %valfp = load volatile float* %addrfp
 ; CHECK: ldur {{s[0-9]+}}, [{{x[0-9]+}}, #-5]
+; CHECK-NOFP-NOT: ldur {{s[0-9]+}},
 
   store volatile float %valfp, float* %addrfp
 ; CHECK: stur {{s[0-9]+}}, [{{x[0-9]+}}, #-5]
+; CHECK-NOFP-NOT: stur {{s[0-9]+}},
 
   ret void
 }
@@ -210,9 +213,11 @@ define void @ldst_double() {
 
   %valfp = load volatile double* %addrfp
 ; CHECK: ldur {{d[0-9]+}}, [{{x[0-9]+}}, #4]
+; CHECK-NOFP-NOT: ldur {{d[0-9]+}},
 
   store volatile double %valfp, double* %addrfp
 ; CHECK: stur {{d[0-9]+}}, [{{x[0-9]+}}, #4]
+; CHECK-NOFP-NOT: stur {{d[0-9]+}},
 
    ret void
 }
diff --git a/test/CodeGen/AArch64/ldst-unsignedimm.ll b/test/CodeGen/AArch64/ldst-unsignedimm.ll
index 77cef4eea0..44c1586e1e 100644
--- a/test/CodeGen/AArch64/ldst-unsignedimm.ll
+++ b/test/CodeGen/AArch64/ldst-unsignedimm.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
 
 @var_8bit = global i8 0
 @var_16bit = global i16 0
@@ -230,9 +231,11 @@ define void @ldst_float() {
    %valfp = load volatile float* @var_float
 ; CHECK: adrp {{x[0-9]+}}, var_float
 ; CHECK: ldr {{s[0-9]+}}, [{{x[0-9]+}}, #:lo12:var_float]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
   store volatile float %valfp, float* @var_float
 ; CHECK: str {{s[0-9]+}}, [{{x[0-9]+}}, #:lo12:var_float]
+; CHECK-NOFP-NOT: str {{s[0-9]+}},
 
    ret void
 }
@@ -243,9 +246,11 @@ define void @ldst_double() {
    %valfp = load volatile double* @var_double
 ; CHECK: adrp {{x[0-9]+}}, var_double
 ; CHECK: ldr {{d[0-9]+}}, [{{x[0-9]+}}, #:lo12:var_double]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
 
   store volatile double %valfp, double* @var_double
 ; CHECK: str {{d[0-9]+}}, [{{x[0-9]+}}, #:lo12:var_double]
+; CHECK-NOFP-NOT: str {{d[0-9]+}},
 
    ret void
 }
diff --git a/test/CodeGen/AArch64/literal_pools.ll b/test/CodeGen/AArch64/literal_pools.ll
index fb45416ead..fc33aee10d 100644
--- a/test/CodeGen/AArch64/literal_pools.ll
+++ b/test/CodeGen/AArch64/literal_pools.ll
@@ -1,5 +1,7 @@
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s
 ; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -code-model=large | FileCheck --check-prefix=CHECK-LARGE %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
+; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -code-model=large -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP-LARGE %s
 
 @var32 = global i32 0
 @var64 = global i64 0
@@ -66,6 +68,7 @@ define void @floating_lits() {
   %newfloat = fadd float %floatval, 128.0
 ; CHECK: adrp x[[LITBASE:[0-9]+]], [[CURLIT:.LCPI1_[0-9]+]]
 ; CHECK: ldr [[LIT128:s[0-9]+]], [x[[LITBASE]], #:lo12:[[CURLIT]]]
+; CHECK-NOFP-NOT: ldr {{s[0-9]+}},
 
 ; CHECK-LARGE: movz x[[LITADDR:[0-9]+]], #:abs_g3:[[CURLIT:.LCPI1_[0-9]+]]
 ; CHECK-LARGE: movk x[[LITADDR]], #:abs_g2_nc:[[CURLIT]]
@@ -73,6 +76,8 @@ define void @floating_lits() {
 ; CHECK-LARGE: movk x[[LITADDR]], #:abs_g0_nc:[[CURLIT]]
 ; CHECK-LARGE: ldr {{s[0-9]+}}, [x[[LITADDR]]]
 ; CHECK-LARGE: fadd
+; CHECK-NOFP-LARGE-NOT: ldr {{s[0-9]+}},
+; CHECK-NOFP-LARGE-NOT: fadd
 
   store float %newfloat, float* @varfloat
 
@@ -82,12 +87,15 @@ define void @floating_lits() {
 ; CHECK: ldr [[LIT129:d[0-9]+]], [x[[LITBASE]], #:lo12:[[CURLIT]]]
 ; CHECK: fadd {{s[0-9]+}}, {{s[0-9]+}}, [[LIT128]]
 ; CHECK: fadd {{d[0-9]+}}, {{d[0-9]+}}, [[LIT129]]
+; CHECK-NOFP-NOT: ldr {{d[0-9]+}},
+; CHECK-NOFP-NOT: fadd
 
 ; CHECK-LARGE: movz x[[LITADDR:[0-9]+]], #:abs_g3:[[CURLIT:.LCPI1_[0-9]+]]
 ; CHECK-LARGE: movk x[[LITADDR]], #:abs_g2_nc:[[CURLIT]]
 ; CHECK-LARGE: movk x[[LITADDR]], #:abs_g1_nc:[[CURLIT]]
 ; CHECK-LARGE: movk x[[LITADDR]], #:abs_g0_nc:[[CURLIT]]
 ; CHECK-LARGE: ldr {{d[0-9]+}}, [x[[LITADDR]]]
+; CHECK-NOFP-LARGE-NOT: ldr {{d[0-9]+}},
 
   store double %newdouble, double* @vardouble
 
diff --git a/test/CodeGen/AArch64/variadic.ll b/test/CodeGen/AArch64/variadic.ll
index cc2a4c37ae..f3d376beeb 100644
--- a/test/CodeGen/AArch64/variadic.ll
+++ b/test/CodeGen/AArch64/variadic.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 < %s | FileCheck --check-prefix=CHECK-NOFP %s
 
 %va_list = type {i8*, i8*, i8*, i32, i32}
 
@@ -15,11 +16,20 @@ define void @test_simple(i32 %n, ...) {
 ; CHECK: add x[[GPRBASE:[0-9]+]], sp, #[[GPRFROMSP:[0-9]+]]
 ; CHECK: str x7, [x[[GPRBASE]], #48]
 
+; CHECK-NOFP: sub sp, sp, #[[STACKSIZE:[0-9]+]]
+; CHECK-NOFP: add x[[VA_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
+; CHECK-NOFP: add x[[GPRBASE:[0-9]+]], sp, #[[GPRFROMSP:[0-9]+]]
+; CHECK-NOFP: str x7, [x[[GPRBASE]], #48]
+; CHECK-NOFP-NOT: str q7,
+; CHECK-NOFP: str x1, [sp, #[[GPRFROMSP]]]
+
 ; Omit the middle ones
 
 ; CHECK: str q0, [sp]
 ; CHECK: str x1, [sp, #[[GPRFROMSP]]]
 
+; CHECK-NOFP-NOT: str q0, [sp]
+
   %addr = bitcast %va_list* @var to i8*
   call void @llvm.va_start(i8* %addr)
 ; CHECK: movn [[VR_OFFS:w[0-9]+]], #127
@@ -33,6 +43,14 @@ define void @test_simple(i32 %n, ...) {
 ; CHECK: add [[STACK:x[0-9]+]], sp, #[[STACKSIZE]]
 ; CHECK: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
 
+; CHECK-NOFP: str wzr, [x[[VA_LIST]], #28]
+; CHECK-NOFP: movn [[GR_OFFS:w[0-9]+]], #55
+; CHECK-NOFP: str [[GR_OFFS]], [x[[VA_LIST]], #24]
+; CHECK-NOFP: add [[GR_TOP:x[0-9]+]], x[[GPRBASE]], #56
+; CHECK-NOFP: str [[GR_TOP]], [x[[VA_LIST]], #8]
+; CHECK-NOFP: add [[STACK:x[0-9]+]], sp, #[[STACKSIZE]]
+; CHECK-NOFP: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
+
   ret void
 }
 
@@ -44,11 +62,19 @@ define void @test_fewargs(i32 %n, i32 %n1, i32 %n2, float %m, ...) {
 ; CHECK: add x[[GPRBASE:[0-9]+]], sp, #[[GPRFROMSP:[0-9]+]]
 ; CHECK: str x7, [x[[GPRBASE]], #32]
 
+; CHECK-NOFP: sub sp, sp, #[[STACKSIZE:[0-9]+]]
+; CHECK-NOFP-NOT: str q7,
+; CHECK-NOFP: mov x[[GPRBASE:[0-9]+]], sp
+; CHECK-NOFP: str x7, [x[[GPRBASE]], #24]
+
 ; Omit the middle ones
 
 ; CHECK: str q1, [sp]
 ; CHECK: str x3, [sp, #[[GPRFROMSP]]]
 
+; CHECK-NOFP-NOT: str q1, [sp]
+; CHECK-NOFP: str x4, [sp]
+
   %addr = bitcast %va_list* @var to i8*
   call void @llvm.va_start(i8* %addr)
 ; CHECK: add x[[VA_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
@@ -63,6 +89,15 @@ define void @test_fewargs(i32 %n, i32 %n1, i32 %n2, float %m, ...) {
 ; CHECK: add [[STACK:x[0-9]+]], sp, #[[STACKSIZE]]
 ; CHECK: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
 
+; CHECK-NOFP: add x[[VA_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
+; CHECK-NOFP: str wzr, [x[[VA_LIST]], #28]
+; CHECK-NOFP: movn [[GR_OFFS:w[0-9]+]], #31
+; CHECK-NOFP: str [[GR_OFFS]], [x[[VA_LIST]], #24]
+; CHECK-NOFP: add [[GR_TOP:x[0-9]+]], x[[GPRBASE]], #32
+; CHECK-NOFP: str [[GR_TOP]], [x[[VA_LIST]], #8]
+; CHECK-NOFP: add [[STACK:x[0-9]+]], sp, #[[STACKSIZE]]
+; CHECK-NOFP: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
+
   ret void
 }
 
@@ -75,6 +110,9 @@ define void @test_nospare([8 x i64], [8 x float], ...) {
 ; CHECK: mov [[STACK:x[0-9]+]], sp
 ; CHECK: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
 
+; CHECK-NOFP-NOT: sub sp, sp
+; CHECK-NOFP: add [[STACK:x[0-9]+]], sp, #64
+; CHECK-NOFP: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
   ret void
 }
 
@@ -87,6 +125,10 @@ define void @test_offsetstack([10 x i64], [3 x float], ...) {
 ; CHECK: str q7, [x[[FPRBASE]], #64]
 
 ; CHECK-NOT: str x{{[0-9]+}},
+
+; CHECK-NOFP-NOT: str q7,
+; CHECK-NOT: str x7,
+
 ; Omit the middle ones
 
 ; CHECK: str q3, [sp]
@@ -102,6 +144,11 @@ define void @test_offsetstack([10 x i64], [3 x float], ...) {
 ; CHECK: add [[STACK:x[0-9]+]], sp, #96
 ; CHECK: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
 
+; CHECK-NOFP: add x[[VA_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
+; CHECK-NOFP: add [[STACK:x[0-9]+]], sp, #40
+; CHECK-NOFP: str [[STACK]], [{{x[0-9]+}}, #:lo12:var]
+; CHECK-NOFP: str wzr, [x[[VA_LIST]], #28]
+; CHECK-NOFP: str wzr, [x[[VA_LIST]], #24]
   ret void
 }
 
@@ -110,12 +157,14 @@ declare void @llvm.va_end(i8*)
 define void @test_va_end() nounwind {
 ; CHECK-LABEL: test_va_end:
 ; CHECK-NEXT: BB#0
+; CHECK-NOFP: BB#0
 
   %addr = bitcast %va_list* @var to i8*
   call void @llvm.va_end(i8* %addr)
 
   ret void
 ; CHECK-NEXT: ret
+; CHECK-NOFP-NEXT: ret
 }
 
 declare void @llvm.va_copy(i8* %dest, i8* %src)
@@ -132,6 +181,8 @@ define void @test_va_copy() {
 
 ; CHECK: ldr [[BLOCK:x[0-9]+]], [{{x[0-9]+}}, #:lo12:var]
 ; CHECK: add x[[SRC_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
+; CHECK-NOFP: ldr [[BLOCK:x[0-9]+]], [{{x[0-9]+}}, #:lo12:var]
+; CHECK-NOFP: add x[[SRC_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:var
 
 ; CHECK: str [[BLOCK]], [{{x[0-9]+}}, #:lo12:second_list]
 
@@ -140,6 +191,14 @@ define void @test_va_copy() {
 
 ; CHECK: str [[BLOCK]], [x[[DEST_LIST]], #24]
 
+; CHECK-NOFP: str [[BLOCK]], [{{x[0-9]+}}, #:lo12:second_list]
+
+; CHECK-NOFP: ldr [[BLOCK:x[0-9]+]], [x[[SRC_LIST]], #24]
+; CHECK-NOFP: add x[[DEST_LIST:[0-9]+]], {{x[0-9]+}}, #:lo12:second_list
+
+; CHECK-NOFP: str [[BLOCK]], [x[[DEST_LIST]], #24]
+
   ret void
 ; CHECK: ret
+; CHECK-NOFP: ret
 }
diff --git a/test/MC/AArch64/basic-a64-instructions.s b/test/MC/AArch64/basic-a64-instructions.s
index e4f6b21892..a50efb3310 100644
--- a/test/MC/AArch64/basic-a64-instructions.s
+++ b/test/MC/AArch64/basic-a64-instructions.s
@@ -1,4 +1,4 @@
-// RUN: llvm-mc -triple aarch64-none-linux-gnu -show-encoding < %s | FileCheck %s
+// RUN: llvm-mc -triple aarch64-none-linux-gnu -show-encoding -mattr=+fp-armv8 < %s | FileCheck %s
   .globl _func
 
 // Check that the assembler can handle the documented syntax from the ARM ARM.
diff --git a/test/MC/AArch64/elf-reloc-ldstunsimm.s b/test/MC/AArch64/elf-reloc-ldstunsimm.s
index 64bf63aa97..371e7e51f2 100644
--- a/test/MC/AArch64/elf-reloc-ldstunsimm.s
+++ b/test/MC/AArch64/elf-reloc-ldstunsimm.s
@@ -1,4 +1,4 @@
-// RUN: llvm-mc -triple=aarch64-none-linux-gnu -filetype=obj %s -o - | \
+// RUN: llvm-mc -triple=aarch64-none-linux-gnu -mattr=+fp-armv8 -filetype=obj %s -o - | \
 // RUN:   llvm-readobj -r | FileCheck -check-prefix=OBJ %s
 
         ldrb w0, [sp, #:lo12:some_label]
diff --git a/test/MC/AArch64/inline-asm-modifiers.s b/test/MC/AArch64/inline-asm-modifiers.s
index caeb617337..cf34a952e9 100644
--- a/test/MC/AArch64/inline-asm-modifiers.s
+++ b/test/MC/AArch64/inline-asm-modifiers.s
@@ -1,4 +1,4 @@
-// RUN: llvm-mc -triple=aarch64-none-linux-gnu -filetype=obj < %s | llvm-objdump -r - | FileCheck %s
+// RUN: llvm-mc -triple=aarch64-none-linux-gnu -filetype=obj -mattr=+fp-armv8 < %s | llvm-objdump -r - | FileCheck %s
 
 	.file	"<stdin>"
 	.text
diff --git a/test/MC/Disassembler/AArch64/a64-ignored-fields.txt b/test/MC/Disassembler/AArch64/a64-ignored-fields.txt
index 966530d36a..799ecdfdcd 100644
--- a/test/MC/Disassembler/AArch64/a64-ignored-fields.txt
+++ b/test/MC/Disassembler/AArch64/a64-ignored-fields.txt
@@ -1,4 +1,4 @@
-# RUN: llvm-mc -triple=aarch64 -disassemble -show-encoding < %s | FileCheck %s
+# RUN: llvm-mc -triple=aarch64 -mattr=fp-armv8 -disassemble -show-encoding < %s | FileCheck %s
 
 # The "Rm" bits are ignored, but the canonical representation has them filled
 # with 0s. This is what we should produce even if the input bit-pattern had
diff --git a/test/MC/Disassembler/AArch64/basic-a64-instructions.txt b/test/MC/Disassembler/AArch64/basic-a64-instructions.txt
index 4fa2d5078b..40926b1fdd 100644
--- a/test/MC/Disassembler/AArch64/basic-a64-instructions.txt
+++ b/test/MC/Disassembler/AArch64/basic-a64-instructions.txt
@@ -1,4 +1,4 @@
-# RUN: llvm-mc -triple=aarch64 -disassemble < %s | FileCheck %s
+# RUN: llvm-mc -triple=aarch64 -mattr=+fp-armv8 -disassemble < %s | FileCheck %s
 
 #------------------------------------------------------------------------------
 # Add/sub (immediate)
diff --git a/test/MC/Disassembler/AArch64/basic-a64-unpredictable.txt b/test/MC/Disassembler/AArch64/basic-a64-unpredictable.txt
index adb8f75ed9..53638638d5 100644
--- a/test/MC/Disassembler/AArch64/basic-a64-unpredictable.txt
+++ b/test/MC/Disassembler/AArch64/basic-a64-unpredictable.txt
@@ -1,4 +1,4 @@
-# RUN: llvm-mc -triple=aarch64 -disassemble < %s 2>&1 | FileCheck %s
+# RUN: llvm-mc -triple=aarch64 -mattr=+fp-armv8 -disassemble < %s 2>&1 | FileCheck %s
 
 #------------------------------------------------------------------------------
 # Load-store exclusive
diff --git a/test/MC/Disassembler/AArch64/ldp-postind.predictable.txt b/test/MC/Disassembler/AArch64/ldp-postind.predictable.txt
index 775660bba8..637ebdbdba 100644
--- a/test/MC/Disassembler/AArch64/ldp-postind.predictable.txt
+++ b/test/MC/Disassembler/AArch64/ldp-postind.predictable.txt
@@ -1,4 +1,4 @@
-# RUN: llvm-mc -triple=aarch64 -disassemble < %s 2>&1 | FileCheck %s
+# RUN: llvm-mc -triple=aarch64 -mattr=+fp-armv8 -disassemble < %s 2>&1 | FileCheck %s
 
 # None of these instructions should be classified as unpredictable:
 
diff --git a/test/MC/Disassembler/AArch64/ldp-preind.predictable.txt b/test/MC/Disassembler/AArch64/ldp-preind.predictable.txt
index 48ea8170ba..f52d37ffda 100644
--- a/test/MC/Disassembler/AArch64/ldp-preind.predictable.txt
+++ b/test/MC/Disassembler/AArch64/ldp-preind.predictable.txt
@@ -1,4 +1,4 @@
-# RUN: llvm-mc -triple=aarch64 -disassemble < %s 2>&1 | FileCheck %s
+# RUN: llvm-mc -triple=aarch64 -mattr=+fp-armv8 -disassemble < %s 2>&1 | FileCheck %s
 
 # None of these instructions should be classified as unpredictable: