AMDGPU: Partially clean up canonicalized predicates in tablegen

llvm/lib/Target/AMDGPU/SIInstrInfo.td

@@ -835,6 +835,19 @@ def fp16_zeros_high_16bits : PatLeaf<(f16 VGPR_32:$src), [{
   return fp16SrcZerosHighBits(N->getOpcode());
 }]>;
 
+def is_canonicalized : PatLeaf<(fAny srcvalue:$src), [{
+  const SITargetLowering &Lowering =
+      *static_cast<const SITargetLowering *>(getTargetLowering());
+  return Lowering.isCanonicalized(*CurDAG, SDValue(N, 0));
+}]> {
+  let GISelPredicateCode = [{
+    const SITargetLowering *TLI = static_cast<const SITargetLowering *>(
+        MF.getSubtarget().getTargetLowering());
+    const MachineOperand &Dst = MI.getOperand(0);
+    assert(Dst.isDef());
+    return TLI->isCanonicalized(Dst.getReg(), MF);
+   }];
+}
 
 //===----------------------------------------------------------------------===//
 // MUBUF/SMEM Patterns

llvm/lib/Target/AMDGPU/SIInstructions.td

@@ -2946,30 +2946,12 @@ def : GCNPat<
 
 // If fcanonicalize's operand is implicitly canonicalized, we only need a copy.
 let AddedComplexity = 1000 in {
-def : GCNPat<
-  (is_canonicalized_1<fcanonicalize> f16:$src),
-  (COPY f16:$src)
->;
-
-def : GCNPat<
-  (is_canonicalized_1<fcanonicalize> v2f16:$src),
-  (COPY v2f16:$src)
->;
-
-def : GCNPat<
-  (is_canonicalized_1<fcanonicalize> f32:$src),
-  (COPY f32:$src)
->;
-
-def : GCNPat<
-  (is_canonicalized_1<fcanonicalize> v2f32:$src),
-  (COPY v2f32:$src)
->;
-
-def : GCNPat<
-  (is_canonicalized_1<fcanonicalize> f64:$src),
-  (COPY f64:$src)
->;
+foreach vt = [f16, v2f16, f32, v2f32, f64] in {
+  def : GCNPat<
+    (fcanonicalize (vt is_canonicalized:$src)),
+    (COPY vt:$src)
+  >;
+}
 }
 
 // Prefer selecting to max when legal, but using mul is always valid.