[LoongArch] Remove wrong vector shuffle lowering for lasx. (llvm#140688)

PR llvm#137918 introduces a wrong
lowering for v4f64/v4i64 to generate xvshuf4i.d instruction.
This PR reverts the wrong part of lasx.

Author: tangaac

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

@@ -1366,10 +1366,8 @@ static SDValue lowerVECTOR_SHUFFLE_VSHUF4I(const SDLoc &DL, ArrayRef<int> Mask,
                                            SelectionDAG &DAG) {
 
   unsigned SubVecSize = 4;
-  if (VT == MVT::v2f64 || VT == MVT::v2i64 || VT == MVT::v4f64 ||
-      VT == MVT::v4i64) {
+  if (VT == MVT::v2f64 || VT == MVT::v2i64)
     SubVecSize = 2;
-  }
 
   int SubMask[4] = {-1, -1, -1, -1};
   for (unsigned i = 0; i < SubVecSize; ++i) {
@@ -1407,9 +1405,8 @@ static SDValue lowerVECTOR_SHUFFLE_VSHUF4I(const SDLoc &DL, ArrayRef<int> Mask,
     Imm |= M & 0x3;
   }
 
-  // Return vshuf4i.d and xvshuf4i.d
-  if (VT == MVT::v2f64 || VT == MVT::v2i64 || VT == MVT::v4f64 ||
-      VT == MVT::v4i64)
+  // Return vshuf4i.d
+  if (VT == MVT::v2f64 || VT == MVT::v2i64)
     return DAG.getNode(LoongArchISD::VSHUF4I, DL, VT, V1, V2,
                        DAG.getConstant(Imm, DL, MVT::i64));
 
@@ -1797,6 +1794,10 @@ static SDValue lowerVECTOR_SHUFFLE_XVREPLVEI(const SDLoc &DL,
 static SDValue lowerVECTOR_SHUFFLE_XVSHUF4I(const SDLoc &DL, ArrayRef<int> Mask,
                                             MVT VT, SDValue V1, SDValue V2,
                                             SelectionDAG &DAG) {
+  // When the size is less than or equal to 4, lower cost instructions may be
+  // used.
+  if (Mask.size() <= 4)
+    return SDValue();
   return lowerVECTOR_SHUFFLE_VSHUF4I(DL, Mask, VT, V1, V2, DAG);
 }
 
@@ -2178,9 +2179,6 @@ static SDValue lower256BitShuffle(const SDLoc &DL, ArrayRef<int> Mask, MVT VT,
     return Result;
   if ((Result = lowerVECTOR_SHUFFLE_XVPICKOD(DL, NewMask, VT, V1, V2, DAG)))
     return Result;
-  if ((VT.SimpleTy == MVT::v4i64 || VT.SimpleTy == MVT::v4f64) &&
-      (Result = lowerVECTOR_SHUFFLE_XVSHUF4I(DL, NewMask, VT, V1, V2, DAG)))
-    return Result;
   if ((Result =
            lowerVECTOR_SHUFFLEAsShift(DL, NewMask, VT, V1, V2, DAG, Zeroable)))
     return Result;

llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td

@@ -23,7 +23,7 @@ def SDT_LoongArchV2R : SDTypeProfile<1, 2, [SDTCisVec<0>,
                                      SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>]>;
 def SDT_LoongArchV1RUimm: SDTypeProfile<1, 2, [SDTCisVec<0>,
                                         SDTCisSameAs<0,1>, SDTCisVT<2, i64>]>;
-def SDT_LoongArchVShuf4i_D
+def SDT_LoongArchV2RUimm
     : SDTypeProfile<1, 3,
                     [SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>,
                      SDTCisVT<3, i64>]>;
@@ -57,8 +57,7 @@ def loongarch_vilvl: SDNode<"LoongArchISD::VILVL", SDT_LoongArchV2R>;
 def loongarch_vilvh: SDNode<"LoongArchISD::VILVH", SDT_LoongArchV2R>;
 
 def loongarch_vshuf4i: SDNode<"LoongArchISD::VSHUF4I", SDT_LoongArchV1RUimm>;
-def loongarch_vshuf4i_d
-    : SDNode<"LoongArchISD::VSHUF4I", SDT_LoongArchVShuf4i_D>;
+def loongarch_vshuf4i_d : SDNode<"LoongArchISD::VSHUF4I", SDT_LoongArchV2RUimm>;
 def loongarch_vreplvei: SDNode<"LoongArchISD::VREPLVEI", SDT_LoongArchV1RUimm>;
 def loongarch_vreplgr2vr: SDNode<"LoongArchISD::VREPLGR2VR", SDT_LoongArchVreplgr2vr>;
 

llvm/test/CodeGen/LoongArch/lasx/ir-instruction/shuffle-as-xvshuf4i.ll

@@ -40,24 +40,4 @@ define <8 x float> @shufflevector_xvshuf4i_v8f32(<8 x float> %a, <8 x float> %b)
 ; CHECK-NEXT:    ret
     %c = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
     ret <8 x float> %c
-}
-
-;; xvshuf4i.d
-define <4 x i64> @shufflevector_xvshuf4i_v4d64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: shufflevector_xvshuf4i_v4d64:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    xvshuf4i.d $xr0, $xr1, 9
-; CHECK-NEXT:    ret
-    %c = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 1, i32 2, i32 5, i32 6>
-    ret <4 x i64> %c
-}
-
-;; xvshuf4i.d
-define <4 x double> @shufflevector_xvshuf4i_v4f64(<4 x double> %a, <4 x double> %b) {
-; CHECK-LABEL: shufflevector_xvshuf4i_v4f64:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    xvshuf4i.d $xr0, $xr1, 9
-; CHECK-NEXT:    ret
-    %c = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 1, i32 2, i32 5, i32 6>
-    ret <4 x double> %c
-}
+}