[AArch64][GlobalISel] Clean up CTLZ vector type legalization.

Similar to other operations, s8, s16 and s32 vector elements are clamped to
legal vector sizes, but in this case s64 are scalarized to use the gpr
instructions. This allows vector types to split as opposed to scalarizing.

Author: davemgreen

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

@@ -6139,6 +6139,7 @@ LegalizerHelper::moreElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case TargetOpcode::G_FCANONICALIZE:
   case TargetOpcode::G_SEXT_INREG:
   case TargetOpcode::G_ABS:
+  case TargetOpcode::G_CTLZ:
     if (TypeIdx != 0)
       return UnableToLegalize;
     Observer.changingInstr(MI);

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

@@ -326,12 +326,23 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .maxScalarEltSameAsIf(always, 1, 0);
 
   getActionDefinitionsBuilder(G_CTLZ)
-      .legalForCartesianProduct(
-          {s32, s64, v8s8, v16s8, v4s16, v8s16, v2s32, v4s32})
-      .scalarize(1)
+      .legalFor({{s32, s32},
+                 {s64, s64},
+                 {v8s8, v8s8},
+                 {v16s8, v16s8},
+                 {v4s16, v4s16},
+                 {v8s16, v8s16},
+                 {v2s32, v2s32},
+                 {v4s32, v4s32}})
       .widenScalarToNextPow2(1, /*Min=*/32)
       .clampScalar(1, s32, s64)
+      .clampNumElements(0, v8s8, v16s8)
+      .clampNumElements(0, v4s16, v8s16)
+      .clampNumElements(0, v2s32, v4s32)
+      .moreElementsToNextPow2(0)
+      .scalarizeIf(scalarOrEltWiderThan(0, 32), 0)
       .scalarSameSizeAs(0, 1);
+
   getActionDefinitionsBuilder(G_CTLZ_ZERO_UNDEF).lower();
 
   getActionDefinitionsBuilder(G_CTTZ)