[LegalizeTypes][VP] Add integer promotion support for vp.select

[victor-eds]

Summary

Promote select, vselect and vp.select in a similar way.

[victor-eds]

Call getSelect instead of getNode

[victor-eds]

Change indentation

[victor-eds]

Add comments for boolean arguments

[victor-eds]

Change commit message

[victor-eds]

Change commit message

[victor-eds]

Apply format changes

[frasercrmck]

LGTM other than nit. I'd prefer to leave official acceptance to someone outside our company, though.

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

1133

nit: maybe we should rename this function as it suggests it only handles (upper-case) ISD::SELECT rather than the family of selects. PromoteIntRes_Select? I dunno if that's a great name but SELECT is misleading.

[simoll]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

1133

+1 for PromoteIntRes_Select

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

242

Could check the opcode inside the function and have all 3 opcodes checked in the same place in the switch?

1133

+1 for PromoteIntRes_Select

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

1140

Use getNode(N->getOpcode() to match the widening patch and avoid a runtime check for of the condition type to pick the opcode.

[victor-eds]

Drop boolean argument

[victor-eds]

Done

[craig.topper]

LGTM with that one comment.

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

81

Restore the old formatting with the call and the break on the same line as the case.

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

#endif
   case ISD::CTTZ:        Res = PromoteIntRes_CTTZ(N); break;
   case ISD::EXTRACT_VECTOR_ELT:
                          Res = PromoteIntRes_EXTRACT_VECTOR_ELT(N); break;
   case ISD::LOAD:        Res = PromoteIntRes_LOAD(cast<LoadSDNode>(N)); break;
   case ISD::MLOAD:       Res = PromoteIntRes_MLOAD(cast<MaskedLoadSDNode>(N));
     break;
   case ISD::MGATHER:     Res = PromoteIntRes_MGATHER(cast<MaskedGatherSDNode>(N));
     break;
-  case ISD::SELECT:      Res = PromoteIntRes_SELECT(N); break;
-  case ISD::VSELECT:     Res = PromoteIntRes_VSELECT(N); break;
+  case ISD::SELECT:
+  case ISD::VSELECT:
+  case ISD::VP_SELECT:
+    Res = PromoteIntRes_Select(N);
+    break;
   case ISD::SELECT_CC:   Res = PromoteIntRes_SELECT_CC(N); break;
   case ISD::STRICT_FSETCC:
   case ISD::STRICT_FSETCCS:
   case ISD::SETCC:       Res = PromoteIntRes_SETCC(N); break;
   case ISD::SMIN:
   case ISD::SMAX:
     Res = PromoteIntRes_SExtIntBinOp(N, /*IsVP*/ false);
     break;
#endif
   case ISD::VECREDUCE_SMAX:
   case ISD::VECREDUCE_SMIN:
   case ISD::VECREDUCE_UMAX:
   case ISD::VECREDUCE_UMIN:
     Res = PromoteIntRes_VECREDUCE(N);
     break;
 
   case ISD::VP_REDUCE_ADD:
   case ISD::VP_REDUCE_MUL:
   case ISD::VP_REDUCE_AND:
   case ISD::VP_REDUCE_OR:
   case ISD::VP_REDUCE_XOR:
   case ISD::VP_REDUCE_SMAX:
   case ISD::VP_REDUCE_SMIN:
   case ISD::VP_REDUCE_UMAX:
   case ISD::VP_REDUCE_UMIN:
     Res = PromoteIntRes_VP_REDUCE(N);
SDValue DAGTypeLegalizer::PromoteIntRes_SADDSUBO(SDNode *N, unsigned ResNo) {
   Ofl = DAG.getSetCC(dl, N->getValueType(1), Ofl, Res, ISD::SETNE);
 
   // Use the calculated overflow everywhere.
   ReplaceValueWith(SDValue(N, 1), Ofl);
 
   return Res;
 }
 
-SDValue DAGTypeLegalizer::PromoteIntRes_SELECT(SDNode *N) {
-  SDValue LHS = GetPromotedInteger(N->getOperand(1));
-  SDValue RHS = GetPromotedInteger(N->getOperand(2));
-  return DAG.getSelect(SDLoc(N),
-                       LHS.getValueType(), N->getOperand(0), LHS, RHS);
-}
-
-SDValue DAGTypeLegalizer::PromoteIntRes_VSELECT(SDNode *N) {
+SDValue DAGTypeLegalizer::PromoteIntRes_Select(SDNode *N) {
   SDValue Mask = N->getOperand(0);
 
   SDValue LHS = GetPromotedInteger(N->getOperand(1));
   SDValue RHS = GetPromotedInteger(N->getOperand(2));
-  return DAG.getNode(ISD::VSELECT, SDLoc(N),
-                     LHS.getValueType(), Mask, LHS, RHS);
+
+  unsigned Opcode = N->getOpcode();
+  return Opcode == ISD::VP_SELECT
+             ? DAG.getNode(Opcode, SDLoc(N), LHS.getValueType(), Mask, LHS, RHS,
+                           N->getOperand(3))
+             : DAG.getNode(Opcode, SDLoc(N), LHS.getValueType(), Mask, LHS,
+                           RHS);
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_SELECT_CC(SDNode *N) {
   SDValue LHS = GetPromotedInteger(N->getOperand(2));
   SDValue RHS = GetPromotedInteger(N->getOperand(3));
   return DAG.getNode(ISD::SELECT_CC, SDLoc(N),
                      LHS.getValueType(), N->getOperand(0),
                      N->getOperand(1), LHS, RHS, N->getOperand(4));

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

private:
   SDValue PromoteIntRes_FP_TO_FP16(SDNode *N);
   SDValue PromoteIntRes_FREEZE(SDNode *N);
   SDValue PromoteIntRes_INT_EXTEND(SDNode *N);
   SDValue PromoteIntRes_LOAD(LoadSDNode *N);
   SDValue PromoteIntRes_MLOAD(MaskedLoadSDNode *N);
   SDValue PromoteIntRes_MGATHER(MaskedGatherSDNode *N);
   SDValue PromoteIntRes_Overflow(SDNode *N);
   SDValue PromoteIntRes_SADDSUBO(SDNode *N, unsigned ResNo);
-  SDValue PromoteIntRes_SELECT(SDNode *N);
-  SDValue PromoteIntRes_VSELECT(SDNode *N);
+  SDValue PromoteIntRes_Select(SDNode *N);
   SDValue PromoteIntRes_SELECT_CC(SDNode *N);
   SDValue PromoteIntRes_SETCC(SDNode *N);
   SDValue PromoteIntRes_SHL(SDNode *N, bool IsVP);
   SDValue PromoteIntRes_SimpleIntBinOp(SDNode *N, bool IsVP);
   SDValue PromoteIntRes_ZExtIntBinOp(SDNode *N, bool IsVP);
   SDValue PromoteIntRes_SExtIntBinOp(SDNode *N, bool IsVP);
   SDValue PromoteIntRes_UMINUMAX(SDNode *N);
   SDValue PromoteIntRes_SIGN_EXTEND_INREG(SDNode *N);

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vselect-vp.ll

 ; CHECK-NEXT:    vmandn.mm v9, v9, v0
 ; CHECK-NEXT:    vmand.mm v8, v8, v0
 ; CHECK-NEXT:    vmor.mm v0, v8, v9
 ; CHECK-NEXT:    ret
   %v = call <16 x i1> @llvm.vp.select.v16i1(<16 x i1> %a, <16 x i1> %b, <16 x i1> %c, i32 %evl)
   ret <16 x i1> %v
 }
 
+declare <8 x i7> @llvm.vp.select.v8i7(<8 x i1>, <8 x i7>, <8 x i7>, i32)
+
+define <8 x i7> @select_v8i7(<8 x i1> %a, <8 x i7> %b, <8 x i7> %c, i32 zeroext %evl) {
+; CHECK-LABEL: select_v8i7:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmerge.vvm v8, v9, v8, v0
+; CHECK-NEXT:    ret
+  %v = call <8 x i7> @llvm.vp.select.v8i7(<8 x i1> %a, <8 x i7> %b, <8 x i7> %c, i32 %evl)
+  ret <8 x i7> %v
+}
+
 declare <2 x i8> @llvm.vp.select.v2i8(<2 x i1>, <2 x i8>, <2 x i8>, i32)
 
 define <2 x i8> @select_v2i8(<2 x i1> %a, <2 x i8> %b, <2 x i8> %c, i32 zeroext %evl) {
 ; CHECK-LABEL: select_v2i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, ta, mu
 ; CHECK-NEXT:    vmerge.vvm v8, v9, v8, v0
 ; CHECK-NEXT:    ret

llvm/test/CodeGen/RISCV/rvv/vselect-vp.ll

 ; CHECK-NEXT:    vmandn.mm v9, v9, v0
 ; CHECK-NEXT:    vmand.mm v8, v8, v0
 ; CHECK-NEXT:    vmor.mm v0, v8, v9
 ; CHECK-NEXT:    ret
   %v = call <vscale x 64 x i1> @llvm.vp.select.nxv64i1(<vscale x 64 x i1> %a, <vscale x 64 x i1> %b, <vscale x 64 x i1> %c, i32 %evl)
   ret <vscale x 64 x i1> %v
 }
 
+declare <vscale x 8 x i7> @llvm.vp.select.nxv8i7(<vscale x 8 x i1>, <vscale x 8 x i7>, <vscale x 8 x i7>, i32)
+
+define <vscale x 8 x i7> @select_nxv8i7(<vscale x 8 x i1> %a, <vscale x 8 x i7> %b, <vscale x 8 x i7> %c, i32 zeroext %evl) {
+; CHECK-LABEL: select_nxv8i7:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmerge.vvm v8, v9, v8, v0
+; CHECK-NEXT:    ret
+  %v = call <vscale x 8 x i7> @llvm.vp.select.nxv8i7(<vscale x 8 x i1> %a, <vscale x 8 x i7> %b, <vscale x 8 x i7> %c, i32 %evl)
+  ret <vscale x 8 x i7> %v
+}
+
 declare <vscale x 1 x i8> @llvm.vp.select.nxv1i8(<vscale x 1 x i1>, <vscale x 1 x i8>, <vscale x 1 x i8>, i32)
 
 define <vscale x 1 x i8> @select_nxv1i8(<vscale x 1 x i1> %a, <vscale x 1 x i8> %b, <vscale x 1 x i8> %c, i32 zeroext %evl) {
 ; CHECK-LABEL: select_nxv1i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, ta, mu
 ; CHECK-NEXT:    vmerge.vvm v8, v9, v8, v0
 ; CHECK-NEXT:    ret