[ARM] Use MQPR not QPR for MVE registers

[dmgreen]

Summary

We should be using MQPR, and if we don't we can get COPYs and PHIs created for QPR. These get folded into instructions, failing verification checks.

[ostannard]

llvm/test/CodeGen/Thumb2/mve-crash-qpr.ll

10 ↗

(On Diff #215083)

I don't think these check lines are really adding any value, this is a complex test file so we'd expect the generated code to change as we improve the MVE code generation. Could we instead test this by stopping as soon as possible after instruction selection, and checking that the vregs are all MQPR class in the MIR? That would also allow the IR to be greatly reduced.

[dmgreen]

Sorry for the delay. This test case was originally dependant upon masked stores. I've redone it without that and simplified it a little.

llvm/test/CodeGen/Thumb2/mve-crash-qpr.ll

10 ↗

(On Diff #215083)

I'm not a huge fan of weird tests. I'd prefer it if they were all pretty boring, just simple enough to be autogenerated. They are easy to update if they do change, and you can see if they are smaller or the same. They act as useful codegen checks, to see when changes do have an effect on codegen. Perhaps this test isn't the best example of that though.

I've tried to cut it down and removed most of the check lines. Hopefully its simple enough now.

[dmgreen]

Found another one. This time in getLargestLegalSuperClass.

[dmgreen]

Updated check lines

[ostannard]

LGTM, thanks.

llvm/lib/Target/ARM/ARMBaseRegisterInfo.cpp

 
 bool ARMBaseRegisterInfo::
 isAsmClobberable(const MachineFunction &MF, unsigned PhysReg) const {
   return !getReservedRegs(MF).test(PhysReg);
 }
 
 const TargetRegisterClass *
 ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC,
-                                               const MachineFunction &) const {
+                                               const MachineFunction &MF) const {
   const TargetRegisterClass *Super = RC;
   TargetRegisterClass::sc_iterator I = RC->getSuperClasses();
   do {
     switch (Super->getID()) {
     case ARM::GPRRegClassID:
     case ARM::SPRRegClassID:
     case ARM::DPRRegClassID:
+    case ARM::GPRPairRegClassID:
+      return Super;
     case ARM::QPRRegClassID:
     case ARM::QQPRRegClassID:
     case ARM::QQQQPRRegClassID:
-    case ARM::GPRPairRegClassID:
+      if (MF.getSubtarget<ARMSubtarget>().hasNEON())
         return Super;
     }
     Super = *I++;
   } while (Super);
   return RC;
 }
 
 const TargetRegisterClass *
 ARMBaseRegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind)

llvm/lib/Target/ARM/ARMISelLowering.cpp

void ARMTargetLowering::addAllExtLoads(const MVT From, const MVT To,
   setLoadExtAction(ISD::ZEXTLOAD, From, To, Action);
   setLoadExtAction(ISD::SEXTLOAD, From, To, Action);
 }
 
 void ARMTargetLowering::addMVEVectorTypes(bool HasMVEFP) {
   const MVT IntTypes[] = { MVT::v16i8, MVT::v8i16, MVT::v4i32 };
 
   for (auto VT : IntTypes) {
-    addRegisterClass(VT, &ARM::QPRRegClass);
+    addRegisterClass(VT, &ARM::MQPRRegClass);
     setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
     setOperationAction(ISD::SHL, VT, Custom);
     setOperationAction(ISD::SRA, VT, Custom);
     setOperationAction(ISD::SRL, VT, Custom);
     setOperationAction(ISD::SMIN, VT, Legal);
for (unsigned im = (unsigned)ISD::PRE_INC;
          im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
       setIndexedLoadAction(im, VT, Legal);
       setIndexedStoreAction(im, VT, Legal);
     }
   }
 
   const MVT FloatTypes[] = { MVT::v8f16, MVT::v4f32 };
   for (auto VT : FloatTypes) {
-    addRegisterClass(VT, &ARM::QPRRegClass);
+    addRegisterClass(VT, &ARM::MQPRRegClass);
     if (!HasMVEFP)
       setAllExpand(VT);
 
     // These are legal or custom whether we have MVE.fp or not
     setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT, VT.getVectorElementType(), Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
for (auto VT : FloatTypes) {
     }
   }
 
   // We 'support' these types up to bitcast/load/store level, regardless of
   // MVE integer-only / float support. Only doing FP data processing on the FP
   // vector types is inhibited at integer-only level.
   const MVT LongTypes[] = { MVT::v2i64, MVT::v2f64 };
   for (auto VT : LongTypes) {
-    addRegisterClass(VT, &ARM::QPRRegClass);
+    addRegisterClass(VT, &ARM::MQPRRegClass);
     setAllExpand(VT);
     setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
   }
   // We can do bitwise operations on v2i64 vectors
   setOperationAction(ISD::AND, MVT::v2i64, Legal);
   setOperationAction(ISD::OR, MVT::v2i64, Legal);

llvm/lib/Target/ARM/ARMInstrMVE.td

let Predicates = [HasMVEInt] in {
   defm : MVEExtLoad<"8", "16", "8",  "B", "", t2addrmode_imm7<0>>;
   defm : MVEExtLoad<"4", "32", "16", "H", "_align2", t2addrmode_imm7<1>>;
 }
 
 
 // Bit convert patterns
 
 let Predicates = [HasMVEInt] in {
-  def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
-  def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v2i64 MQPR:$src))), (v2f64 MQPR:$src)>;
+  def : Pat<(v2i64 (bitconvert (v2f64 MQPR:$src))), (v2i64 MQPR:$src)>;
 
-  def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
-  def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
+  def : Pat<(v4i32 (bitconvert (v4f32 MQPR:$src))), (v4i32 MQPR:$src)>;
+  def : Pat<(v4f32 (bitconvert (v4i32 MQPR:$src))), (v4f32 MQPR:$src)>;
 
-  def : Pat<(v8i16 (bitconvert (v8f16 QPR:$src))), (v8i16  QPR:$src)>;
-  def : Pat<(v8f16 (bitconvert (v8i16 QPR:$src))), (v8f16  QPR:$src)>;
+  def : Pat<(v8i16 (bitconvert (v8f16 MQPR:$src))), (v8i16  MQPR:$src)>;
+  def : Pat<(v8f16 (bitconvert (v8i16 MQPR:$src))), (v8f16  MQPR:$src)>;
 }
 
 let Predicates = [IsLE,HasMVEInt] in {
-  def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
-  def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
-  def : Pat<(v2f64 (bitconvert (v8f16 QPR:$src))), (v2f64 QPR:$src)>;
-  def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
-  def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
-
-  def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
-  def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
-  def : Pat<(v2i64 (bitconvert (v8f16 QPR:$src))), (v2i64 QPR:$src)>;
-  def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
-  def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
-
-  def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
-  def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
-  def : Pat<(v4f32 (bitconvert (v8f16 QPR:$src))), (v4f32 QPR:$src)>;
-  def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
-  def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
-
-  def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
-  def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
-  def : Pat<(v4i32 (bitconvert (v8f16 QPR:$src))), (v4i32 QPR:$src)>;
-  def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
-  def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
-
-  def : Pat<(v8f16 (bitconvert (v2f64 QPR:$src))), (v8f16 QPR:$src)>;
-  def : Pat<(v8f16 (bitconvert (v2i64 QPR:$src))), (v8f16 QPR:$src)>;
-  def : Pat<(v8f16 (bitconvert (v4f32 QPR:$src))), (v8f16 QPR:$src)>;
-  def : Pat<(v8f16 (bitconvert (v4i32 QPR:$src))), (v8f16 QPR:$src)>;
-  def : Pat<(v8f16 (bitconvert (v16i8 QPR:$src))), (v8f16 QPR:$src)>;
-
-  def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
-  def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
-  def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
-  def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
-  def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
-
-  def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
-  def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
-  def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
-  def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
-  def : Pat<(v16i8 (bitconvert (v8f16 QPR:$src))), (v16i8 QPR:$src)>;
-  def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v4f32 MQPR:$src))), (v2f64 MQPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v4i32 MQPR:$src))), (v2f64 MQPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v8f16 MQPR:$src))), (v2f64 MQPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v8i16 MQPR:$src))), (v2f64 MQPR:$src)>;
+  def : Pat<(v2f64 (bitconvert (v16i8 MQPR:$src))), (v2f64 MQPR:$src)>;
+
+  def : Pat<(v2i64 (bitconvert (v4f32 MQPR:$src))), (v2i64 MQPR:$src)>;
+  def : Pat<(v2i64 (bitconvert (v4i32 MQPR:$src))), (v2i64 MQPR:$src)>;
+  def : Pat<(v2i64 (bitconvert (v8f16 MQPR:$src))), (v2i64 MQPR:$src)>;
+  def : Pat<(v2i64 (bitconvert (v8i16 MQPR:$src))), (v2i64 MQPR:$src)>;
+  def : Pat<(v2i64 (bitconvert (v16i8 MQPR:$src))), (v2i64 MQPR:$src)>;
+
+  def : Pat<(v4f32 (bitconvert (v2f64 MQPR:$src))), (v4f32 MQPR:$src)>;
+  def : Pat<(v4f32 (bitconvert (v2i64 MQPR:$src))), (v4f32 MQPR:$src)>;
+  def : Pat<(v4f32 (bitconvert (v8f16 MQPR:$src))), (v4f32 MQPR:$src)>;
+  def : Pat<(v4f32 (bitconvert (v8i16 MQPR:$src))), (v4f32 MQPR:$src)>;
+  def : Pat<(v4f32 (bitconvert (v16i8 MQPR:$src))), (v4f32 MQPR:$src)>;
+
+  def : Pat<(v4i32 (bitconvert (v2f64 MQPR:$src))), (v4i32 MQPR:$src)>;
+  def : Pat<(v4i32 (bitconvert (v2i64 MQPR:$src))), (v4i32 MQPR:$src)>;
+  def : Pat<(v4i32 (bitconvert (v8f16 MQPR:$src))), (v4i32 MQPR:$src)>;
+  def : Pat<(v4i32 (bitconvert (v8i16 MQPR:$src))), (v4i32 MQPR:$src)>;
+  def : Pat<(v4i32 (bitconvert (v16i8 MQPR:$src))), (v4i32 MQPR:$src)>;
+
+  def : Pat<(v8f16 (bitconvert (v2f64 MQPR:$src))), (v8f16 MQPR:$src)>;
+  def : Pat<(v8f16 (bitconvert (v2i64 MQPR:$src))), (v8f16 MQPR:$src)>;
+  def : Pat<(v8f16 (bitconvert (v4f32 MQPR:$src))), (v8f16 MQPR:$src)>;
+  def : Pat<(v8f16 (bitconvert (v4i32 MQPR:$src))), (v8f16 MQPR:$src)>;
+  def : Pat<(v8f16 (bitconvert (v16i8 MQPR:$src))), (v8f16 MQPR:$src)>;
+
+  def : Pat<(v8i16 (bitconvert (v2f64 MQPR:$src))), (v8i16 MQPR:$src)>;
+  def : Pat<(v8i16 (bitconvert (v2i64 MQPR:$src))), (v8i16 MQPR:$src)>;
+  def : Pat<(v8i16 (bitconvert (v4f32 MQPR:$src))), (v8i16 MQPR:$src)>;
+  def : Pat<(v8i16 (bitconvert (v4i32 MQPR:$src))), (v8i16 MQPR:$src)>;
+  def : Pat<(v8i16 (bitconvert (v16i8 MQPR:$src))), (v8i16 MQPR:$src)>;
+
+  def : Pat<(v16i8 (bitconvert (v2f64 MQPR:$src))), (v16i8 MQPR:$src)>;
+  def : Pat<(v16i8 (bitconvert (v2i64 MQPR:$src))), (v16i8 MQPR:$src)>;
+  def : Pat<(v16i8 (bitconvert (v4f32 MQPR:$src))), (v16i8 MQPR:$src)>;
+  def : Pat<(v16i8 (bitconvert (v4i32 MQPR:$src))), (v16i8 MQPR:$src)>;
+  def : Pat<(v16i8 (bitconvert (v8f16 MQPR:$src))), (v16i8 MQPR:$src)>;
+  def : Pat<(v16i8 (bitconvert (v8i16 MQPR:$src))), (v16i8 MQPR:$src)>;
 }
 
 let Predicates = [IsBE,HasMVEInt] in {
-  def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v2f64 (bitconvert (v8f16 QPR:$src))), (v2f64 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 (MVE_VREV64_8 QPR:$src))>;
-
-  def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v2i64 (bitconvert (v8f16 QPR:$src))), (v2i64 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 (MVE_VREV64_8 QPR:$src))>;
-
-  def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v4f32 (bitconvert (v8f16 QPR:$src))), (v4f32 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 (MVE_VREV32_8 QPR:$src))>;
-
-  def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 (MVE_VREV64_32 QPR:$src))>;
-  def : Pat<(v4i32 (bitconvert (v8f16 QPR:$src))), (v4i32 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 (MVE_VREV32_8 QPR:$src))>;
-
-  def : Pat<(v8f16 (bitconvert (v2f64 QPR:$src))), (v8f16 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v8f16 (bitconvert (v2i64 QPR:$src))), (v8f16 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v8f16 (bitconvert (v4f32 QPR:$src))), (v8f16 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v8f16 (bitconvert (v4i32 QPR:$src))), (v8f16 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v8f16 (bitconvert (v16i8 QPR:$src))), (v8f16 (MVE_VREV16_8 QPR:$src))>;
-
-  def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 (MVE_VREV64_16 QPR:$src))>;
-  def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 (MVE_VREV32_16 QPR:$src))>;
-  def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 (MVE_VREV16_8 QPR:$src))>;
-
-  def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 (MVE_VREV64_8 QPR:$src))>;
-  def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 (MVE_VREV64_8 QPR:$src))>;
-  def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 (MVE_VREV32_8 QPR:$src))>;
-  def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 (MVE_VREV32_8 QPR:$src))>;
-  def : Pat<(v16i8 (bitconvert (v8f16 QPR:$src))), (v16i8 (MVE_VREV16_8 QPR:$src))>;
-  def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 (MVE_VREV16_8 QPR:$src))>;
+  def : Pat<(v2f64 (bitconvert (v4f32 MQPR:$src))), (v2f64 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v2f64 (bitconvert (v4i32 MQPR:$src))), (v2f64 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v2f64 (bitconvert (v8f16 MQPR:$src))), (v2f64 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v2f64 (bitconvert (v8i16 MQPR:$src))), (v2f64 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v2f64 (bitconvert (v16i8 MQPR:$src))), (v2f64 (MVE_VREV64_8 MQPR:$src))>;
+
+  def : Pat<(v2i64 (bitconvert (v4f32 MQPR:$src))), (v2i64 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v2i64 (bitconvert (v4i32 MQPR:$src))), (v2i64 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v2i64 (bitconvert (v8f16 MQPR:$src))), (v2i64 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v2i64 (bitconvert (v8i16 MQPR:$src))), (v2i64 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v2i64 (bitconvert (v16i8 MQPR:$src))), (v2i64 (MVE_VREV64_8 MQPR:$src))>;
+
+  def : Pat<(v4f32 (bitconvert (v2f64 MQPR:$src))), (v4f32 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v4f32 (bitconvert (v2i64 MQPR:$src))), (v4f32 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v4f32 (bitconvert (v8f16 MQPR:$src))), (v4f32 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v4f32 (bitconvert (v8i16 MQPR:$src))), (v4f32 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v4f32 (bitconvert (v16i8 MQPR:$src))), (v4f32 (MVE_VREV32_8 MQPR:$src))>;
+
+  def : Pat<(v4i32 (bitconvert (v2f64 MQPR:$src))), (v4i32 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v4i32 (bitconvert (v2i64 MQPR:$src))), (v4i32 (MVE_VREV64_32 MQPR:$src))>;
+  def : Pat<(v4i32 (bitconvert (v8f16 MQPR:$src))), (v4i32 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v4i32 (bitconvert (v8i16 MQPR:$src))), (v4i32 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v4i32 (bitconvert (v16i8 MQPR:$src))), (v4i32 (MVE_VREV32_8 MQPR:$src))>;
+
+  def : Pat<(v8f16 (bitconvert (v2f64 MQPR:$src))), (v8f16 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v8f16 (bitconvert (v2i64 MQPR:$src))), (v8f16 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v8f16 (bitconvert (v4f32 MQPR:$src))), (v8f16 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v8f16 (bitconvert (v4i32 MQPR:$src))), (v8f16 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v8f16 (bitconvert (v16i8 MQPR:$src))), (v8f16 (MVE_VREV16_8 MQPR:$src))>;
+
+  def : Pat<(v8i16 (bitconvert (v2f64 MQPR:$src))), (v8i16 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v8i16 (bitconvert (v2i64 MQPR:$src))), (v8i16 (MVE_VREV64_16 MQPR:$src))>;
+  def : Pat<(v8i16 (bitconvert (v4f32 MQPR:$src))), (v8i16 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v8i16 (bitconvert (v4i32 MQPR:$src))), (v8i16 (MVE_VREV32_16 MQPR:$src))>;
+  def : Pat<(v8i16 (bitconvert (v16i8 MQPR:$src))), (v8i16 (MVE_VREV16_8 MQPR:$src))>;
+
+  def : Pat<(v16i8 (bitconvert (v2f64 MQPR:$src))), (v16i8 (MVE_VREV64_8 MQPR:$src))>;
+  def : Pat<(v16i8 (bitconvert (v2i64 MQPR:$src))), (v16i8 (MVE_VREV64_8 MQPR:$src))>;
+  def : Pat<(v16i8 (bitconvert (v4f32 MQPR:$src))), (v16i8 (MVE_VREV32_8 MQPR:$src))>;
+  def : Pat<(v16i8 (bitconvert (v4i32 MQPR:$src))), (v16i8 (MVE_VREV32_8 MQPR:$src))>;
+  def : Pat<(v16i8 (bitconvert (v8f16 MQPR:$src))), (v16i8 (MVE_VREV16_8 MQPR:$src))>;
+  def : Pat<(v16i8 (bitconvert (v8i16 MQPR:$src))), (v16i8 (MVE_VREV16_8 MQPR:$src))>;
 }

llvm/test/CodeGen/Thumb2/mve-phireg.ll

+; RUN: llc -O3 -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -verify-machineinstrs %s -o - | FileCheck %s
+
+; verify-machineinstrs previously caught the incorrect use of QPR in the stack reloads.
+
+define arm_aapcs_vfpcc void @k() {
+; CHECK-LABEL: k:
+; CHECK:    vstrw.32
+; CHECK:    vldrw.u32
+entry:
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body, %entry
+  %vec.ind = phi <8 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, %entry ], [ zeroinitializer, %vector.body ]
+  %0 = and <8 x i32> %vec.ind, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
+  %1 = icmp eq <8 x i32> %0, zeroinitializer
+  %2 = select <8 x i1> %1, <8 x i16> <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>, <8 x i16> <i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6>
+  %3 = bitcast i16* undef to <8 x i16>*
+  store <8 x i16> %2, <8 x i16>* %3, align 2
+  %4 = icmp eq i32 undef, 128
+  br i1 %4, label %for.cond4.preheader, label %vector.body
+
+for.cond4.preheader:                              ; preds = %vector.body
+  br i1 undef, label %vector.body105, label %for.body10
+
+for.cond4.loopexit:                               ; preds = %for.body10
+  %call5 = call arm_aapcs_vfpcc i32 bitcast (i32 (...)* @l to i32 ()*)()
+  br label %vector.body105
+
+for.body10:                                       ; preds = %for.body10, %for.cond4.preheader
+  %exitcond88 = icmp eq i32 undef, 7
+  br i1 %exitcond88, label %for.cond4.loopexit, label %for.body10
+
+vector.body105:                                   ; preds = %vector.body105, %for.cond4.loopexit, %for.cond4.preheader
+  %vec.ind113 = phi <8 x i32> [ %vec.ind.next114, %vector.body105 ], [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, %for.cond4.loopexit ], [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, %for.cond4.preheader ]
+  %5 = and <8 x i32> %vec.ind113, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
+  %vec.ind.next114 = add <8 x i32> %vec.ind113, <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
+  %6 = icmp eq i32 undef, 256
+  br i1 %6, label %vector.body115.ph, label %vector.body105
+
+vector.body115.ph:                                ; preds = %vector.body105
+  tail call void asm sideeffect "nop", "~{s0},~{s4},~{s8},~{s12},~{s16},~{s20},~{s24},~{s28},~{memory}"()
+  br label %vector.body115
+
+vector.body115:                                   ; preds = %vector.body115, %vector.body115.ph
+  %vec.ind123 = phi <4 x i32> [ %vec.ind.next124, %vector.body115 ], [ <i32 0, i32 1, i32 2, i32 3>, %vector.body115.ph ]
+  %7 = icmp eq <4 x i32> %vec.ind123, zeroinitializer
+  %vec.ind.next124 = add <4 x i32> %vec.ind123, <i32 4, i32 4, i32 4, i32 4>
+  br label %vector.body115
+}
+
+
+@a = external dso_local global i32, align 4
+@b = dso_local local_unnamed_addr global i32 ptrtoint (i32* @a to i32), align 4
+@c = dso_local global i32 2, align 4
+@d = dso_local global i32 2, align 4
+
+define dso_local i32 @e() #0 {
+; CHECK-LABEL: e:
+; CHECK-NEXT:    vstrw.32
+; CHECK-NEXT:    vldrw.u32
+entry:
+  %f = alloca i16, align 2
+  %g = alloca [3 x [8 x [4 x i16*]]], align 4
+  store i16 4, i16* %f, align 2
+  %0 = load i32, i32* @c, align 4
+  %1 = load i32, i32* @d, align 4
+  %arrayinit.element7 = getelementptr inbounds [3 x [8 x [4 x i16*]]], [3 x [8 x [4 x i16*]]]* %g, i32 0, i32 0, i32 1, i32 1
+  %2 = bitcast i16** %arrayinit.element7 to i32*
+  store i32 %0, i32* %2, align 4
+  %arrayinit.element8 = getelementptr inbounds [3 x [8 x [4 x i16*]]], [3 x [8 x [4 x i16*]]]* %g, i32 0, i32 0, i32 1, i32 2
+  store i16* null, i16** %arrayinit.element8, align 4
+  %3 = bitcast i16** undef to i32*
+  store i32 %1, i32* %3, align 4
+  %4 = bitcast i16** undef to i32*
+  store i32 %0, i32* %4, align 4
+  %arrayinit.element13 = getelementptr inbounds [3 x [8 x [4 x i16*]]], [3 x [8 x [4 x i16*]]]* %g, i32 0, i32 0, i32 2, i32 2
+  %5 = bitcast i16** %arrayinit.element13 to <4 x i16*>*
+  store <4 x i16*> <i16* inttoptr (i32 4 to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*), i16* null>, <4 x i16*>* %5, align 4
+  %arrayinit.element24 = getelementptr inbounds [3 x [8 x [4 x i16*]]], [3 x [8 x [4 x i16*]]]* %g, i32 0, i32 0, i32 4, i32 2
+  %6 = bitcast i16** %arrayinit.element24 to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32* @d to i16*), i16* null, i16* bitcast (i32* @d to i16*), i16* bitcast (i32 ()* @e to i16*)>, <4 x i16*>* %6, align 4
+  %7 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* inttoptr (i32 4 to i16*), i16* bitcast (i32 ()* @e to i16*), i16* bitcast (i32* @c to i16*), i16* null>, <4 x i16*>* %7, align 4
+  %8 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32* @c to i16*), i16* bitcast (i32 ()* @e to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*)>, <4 x i16*>* %8, align 4
+  %9 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32 ()* @e to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*)>, <4 x i16*>* %9, align 4
+  %10 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*), i16* null, i16* bitcast (i32 ()* @e to i16*)>, <4 x i16*>* %10, align 4
+  call void @llvm.memset.p0i8.i32(i8* nonnull align 4 dereferenceable(64) undef, i8 0, i32 64, i1 false)
+  %11 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32* @d to i16*), i16* bitcast (i32 ()* @e to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @d to i16*)>, <4 x i16*>* %11, align 4
+  %12 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* null, i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @c to i16*)>, <4 x i16*>* %12, align 4
+  %13 = bitcast i16** undef to <4 x i16*>*
+  store <4 x i16*> <i16* bitcast (i32* @c to i16*), i16* bitcast (i32* @d to i16*), i16* bitcast (i32* @c to i16*), i16* null>, <4 x i16*>* %13, align 4
+  %arrayinit.begin78 = getelementptr inbounds [3 x [8 x [4 x i16*]]], [3 x [8 x [4 x i16*]]]* %g, i32 0, i32 2, i32 3, i32 0
+  store i16* inttoptr (i32 4 to i16*), i16** %arrayinit.begin78, align 4
+  store i32 0, i32* @b, align 4
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.cond, %entry
+  br label %for.cond
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.memset.p0i8.i32(i8* nocapture writeonly, i8, i32, i1 immarg) #1
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #1
+
+
+declare arm_aapcs_vfpcc i32 @l(...)