[AArch64][GlobalISel] Add selection support for fpr bank source variants of G_SITOFP and G_UITOFP.

[aemerson]

Summary

In order to import patterns for these, we need to define new ops that can map to the AArch64ISD::[SU]ITOF nodes. We then transform fpr->fpr variants of the generic opcodes to these custom opcodes in preisel-lowering. We have to do it here and not the PostLegalizer combiner because this has to run after regbankselect.

[paquette]

LGTM

llvm/lib/Target/AArch64/AArch64InstrGISel.td

 }
 
 // Represents a vector G_LSHR with an immediate.
 def G_VLSHR : AArch64GenericInstruction {
   let OutOperandList = (outs type0:$dst);
   let InOperandList = (ins type0:$src1, untyped_imm_0:$imm);
 }
 
+// Represents an integer to FP conversion on the FPR bank.
+def G_SITOF : AArch64GenericInstruction {
+  let OutOperandList = (outs type0:$dst);
+  let InOperandList = (ins type0:$src);
+}
+def G_UITOF : AArch64GenericInstruction {
+  let OutOperandList = (outs type0:$dst);
+  let InOperandList = (ins type0:$src);
+}
+
 def : GINodeEquiv<G_REV16, AArch64rev16>;
 def : GINodeEquiv<G_REV32, AArch64rev32>;
 def : GINodeEquiv<G_REV64, AArch64rev64>;
 def : GINodeEquiv<G_UZP1, AArch64uzp1>;
 def : GINodeEquiv<G_UZP2, AArch64uzp2>;
 def : GINodeEquiv<G_ZIP1, AArch64zip1>;
 def : GINodeEquiv<G_ZIP2, AArch64zip2>;
 def : GINodeEquiv<G_DUP, AArch64dup>;
 def : GINodeEquiv<G_DUPLANE8, AArch64duplane8>;
 def : GINodeEquiv<G_DUPLANE16, AArch64duplane16>;
 def : GINodeEquiv<G_DUPLANE32, AArch64duplane32>;
 def : GINodeEquiv<G_DUPLANE64, AArch64duplane64>;
 def : GINodeEquiv<G_TRN1, AArch64trn1>;
 def : GINodeEquiv<G_TRN2, AArch64trn2>;
 def : GINodeEquiv<G_EXT, AArch64ext>;
 def : GINodeEquiv<G_VASHR, AArch64vashr>;
 def : GINodeEquiv<G_VLSHR, AArch64vlshr>;
+def : GINodeEquiv<G_SITOF, AArch64sitof>;
+def : GINodeEquiv<G_UITOF, AArch64uitof>;
 
 def : GINodeEquiv<G_EXTRACT_VECTOR_ELT, vector_extract>;
 
 // These are patterns that we only use for GlobalISel via the importer.
 def : Pat<(f32 (fadd (vector_extract (v2f32 FPR64:$Rn), (i64 0)),
                      (vector_extract (v2f32 FPR64:$Rn), (i64 1)))),
            (f32 (FADDPv2i32p (v2f32 FPR64:$Rn)))>;

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

case AArch64::G_DUP: {
     MachineIRBuilder MIB(I);
     auto NewSrc = MIB.buildCopy(LLT::scalar(64), I.getOperand(1).getReg());
     MRI.setType(I.getOperand(0).getReg(),
                 DstTy.changeElementType(LLT::scalar(64)));
     MRI.setRegBank(NewSrc.getReg(0), RBI.getRegBank(AArch64::GPRRegBankID));
     I.getOperand(1).setReg(NewSrc.getReg(0));
     return true;
   }
+  case TargetOpcode::G_UITOFP:
+  case TargetOpcode::G_SITOFP: {
+    // If both source and destination regbanks are FPR, then convert the opcode
+    // to G_SITOF so that the importer can select it to an fpr variant.
+    // Otherwise, it ends up matching an fpr/gpr variant and adding a cross-bank
+    // copy.
+    Register SrcReg = I.getOperand(1).getReg();
+    if (MRI.getType(SrcReg).isVector())
+      return false;
+    if (RBI.getRegBank(SrcReg, MRI, TRI)->getID() == AArch64::FPRRegBankID) {
+      if (I.getOpcode() == TargetOpcode::G_SITOFP)
+        I.setDesc(TII.get(AArch64::G_SITOF));
+      else
+        I.setDesc(TII.get(AArch64::G_UITOF));
+      return true;
+    }
+    return false;
+  }
   default:
     return false;
   }
 }
 
 /// This lowering tries to look for G_PTR_ADD instructions and then converts
 /// them to a standard G_ADD with a COPY on the source.
 ///

llvm/test/CodeGen/AArch64/GlobalISel/select-fp-casts.mir

bb.0:
     ; CHECK: [[FCVTLv2i32:%[0-9]+]]:fpr128 = FCVTLv2i32 [[COPY]]
     ; CHECK: $q0 = COPY [[FCVTLv2i32]]
     %0(<2 x s32>) = COPY $d0
     %1(<2 x s64>) = G_FPEXT %0
     $q0 = COPY %1(<2 x s64>)
 ...
 
 ---
-name:            sitofp_s32_s32_fpr
+name:            sitofp_s32_s32_fpr_gpr
 legalized:       true
 regBankSelected: true
 
 registers:
   - { id: 0, class: gpr }
   - { id: 1, class: fpr }
 
 body:             |
   bb.0:
     liveins: $w0
 
-    ; CHECK-LABEL: name: sitofp_s32_s32_fpr
+    ; CHECK-LABEL: name: sitofp_s32_s32_fpr_gpr
     ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
     ; CHECK: [[SCVTFUWSri:%[0-9]+]]:fpr32 = SCVTFUWSri [[COPY]]
     ; CHECK: $s0 = COPY [[SCVTFUWSri]]
     %0(s32) = COPY $w0
     %1(s32) = G_SITOFP %0
     $s0 = COPY %1(s32)
 ...
 
 ---
+name:            sitofp_s32_s32_fpr_fpr
+legalized:       true
+regBankSelected: true
+
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+
+body:             |
+  bb.0:
+    liveins: $s0
+
+    ; CHECK-LABEL: name: sitofp_s32_s32_fpr_fpr
+    ; CHECK: [[COPY:%[0-9]+]]:fpr32 = COPY $s0
+    ; CHECK: [[SCVTFv1i32:%[0-9]+]]:fpr32 = SCVTFv1i32 [[COPY]]
+    ; CHECK: $s0 = COPY [[SCVTFv1i32]]
+    %0(s32) = COPY $s0
+    %1(s32) = G_SITOFP %0
+    $s0 = COPY %1(s32)
+...
+
+---
+name:            uitofp_s32_s32_fpr_fpr
+legalized:       true
+regBankSelected: true
+
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+
+body:             |
+  bb.0:
+    liveins: $s0
+
+    ; CHECK-LABEL: name: uitofp_s32_s32_fpr_fpr
+    ; CHECK: [[COPY:%[0-9]+]]:fpr32 = COPY $s0
+    ; CHECK: [[UCVTFv1i32:%[0-9]+]]:fpr32 = UCVTFv1i32 [[COPY]]
+    ; CHECK: $s0 = COPY [[UCVTFv1i32]]
+    %0(s32) = COPY $s0
+    %1(s32) = G_UITOFP %0
+    $s0 = COPY %1(s32)
+...
+
+---
 name:            sitofp_s32_s64_fpr
 legalized:       true
 regBankSelected: true
 
 registers:
   - { id: 0, class: gpr }
   - { id: 1, class: fpr }