[AArch64] Let patterns for NEON instructions check runtime mode.

This helps identify any failures where the compiler might otherwise
silently emit instructions that are not valid for the given runtime
mode. We can probably do a similar thing for HasSVE predicates.

Author: sdesmalen-arm

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -22271,7 +22271,8 @@ static SDValue vectorToScalarBitmask(SDNode *N, SelectionDAG &DAG) {
   ComparisonResult = DAG.getSExtOrTrunc(ComparisonResult, DL, VecVT);
 
   SmallVector<SDValue, 16> MaskConstants;
-  if (VecVT == MVT::v16i8) {
+  if (DAG.getSubtarget<AArch64Subtarget>().isNeonAvailable() &&
+      VecVT == MVT::v16i8) {
     // v16i8 is a special case, as we have 16 entries but only 8 positional bits
     // per entry. We split it into two halves, apply the mask, zip the halves to
     // create 8x 16-bit values, and the perform the vector reduce.

llvm/lib/Target/AArch64/AArch64InstrInfo.td

@@ -107,7 +107,7 @@ def HasRCPC_IMMO      : Predicate<"Subtarget->hasRCPC_IMMO()">,
 
 def HasFPARMv8       : Predicate<"Subtarget->hasFPARMv8()">,
                                AssemblerPredicateWithAll<(all_of FeatureFPARMv8), "fp-armv8">;
-def HasNEON          : Predicate<"Subtarget->hasNEON()">,
+def HasNEON          : Predicate<"Subtarget->isNeonAvailable()">,
                                  AssemblerPredicateWithAll<(all_of FeatureNEON), "neon">;
 def HasSM4           : Predicate<"Subtarget->hasSM4()">,
                                  AssemblerPredicateWithAll<(all_of FeatureSM4), "sm4">;
@@ -234,12 +234,9 @@ def HasSMEF16F16orSMEF8F16
                 AssemblerPredicateWithAll<(any_of FeatureSMEF16F16, FeatureSMEF8F16),
                 "sme-f16f16 or sme-f8f16">;
 
-// A subset of NEON instructions are legal in Streaming SVE execution mode,
-// they should be enabled if either has been specified.
-def HasNEONorSME
-    : Predicate<"Subtarget->hasNEON() || Subtarget->hasSME()">,
-                AssemblerPredicateWithAll<(any_of FeatureNEON, FeatureSME),
-                "neon or sme">;
+// A subset of NEON instructions are legal in Streaming SVE execution mode.
+def HasNEONorSME     : Predicate<"Subtarget->hasNEON()">,
+                                 AssemblerPredicateWithAll<(any_of FeatureNEON), "neon">;
 def HasRCPC          : Predicate<"Subtarget->hasRCPC()">,
                                  AssemblerPredicateWithAll<(all_of FeatureRCPC), "rcpc">;
 def HasAltNZCV       : Predicate<"Subtarget->hasAlternativeNZCV()">,
@@ -323,8 +320,6 @@ def NoUseScalarIncVL : Predicate<"!Subtarget->useScalarIncVL()">;
 
 def UseSVEFPLD1R : Predicate<"!Subtarget->noSVEFPLD1R()">;
 
-def IsNeonAvailable : Predicate<"Subtarget->isNeonAvailable()">;
-
 def AArch64LocalRecover : SDNode<"ISD::LOCAL_RECOVER",
                                   SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>,
                                                        SDTCisInt<1>]>>;
@@ -6016,7 +6011,7 @@ def : Pat<(v2f64 (AArch64frsqrts (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
 // Some float -> int -> float conversion patterns for which we want to keep the
 // int values in FP registers using the corresponding NEON instructions to
 // avoid more costly int <-> fp register transfers.
-let Predicates = [HasNEON] in {
+let Predicates = [HasNEONorSME] in {
 def : Pat<(f64 (any_sint_to_fp (i64 (any_fp_to_sint f64:$Rn)))),
           (SCVTFv1i64 (i64 (FCVTZSv1i64 f64:$Rn)))>;
 def : Pat<(f32 (any_sint_to_fp (i32 (any_fp_to_sint f32:$Rn)))),
@@ -8379,7 +8374,7 @@ def : Ld1Lane64IdxOpPat<extloadi8, VectorIndexH, v4i16, i32, LD1i8, VectorIndexH
 
 // Same as above, but the first element is populated using
 // scalar_to_vector + insert_subvector instead of insert_vector_elt.
-let Predicates = [IsNeonAvailable] in {
+let Predicates = [HasNeonOrSME] in {
   class Ld1Lane128FirstElm<ValueType ResultTy, ValueType VecTy,
                           SDPatternOperator ExtLoad, Instruction LD1>
     : Pat<(ResultTy (scalar_to_vector (i32 (ExtLoad GPR64sp:$Rn)))),