[RISCV] Lower VECTOR_SHUFFLE to VSLIDEDOWN_VL.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

2905

I think you need to check that Mask[0] is >= 0.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

2905

I think you need to check that Mask[0] is >= 0.

I guess more precisely, it would need to be > 0 or it would definitely not be a slidedown.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

2901

I don’t think it’s enough to be contiguous. The first index must be 0. If it’s not you won’t extract the correct subvector at the end.

[HanKuanChen]

Apply Craig's comments.

1. Mask should not starts with -1.
2. Index of V1 EXTRACT_SUBVECTOR should be 0.

[craig.topper]

LGTM

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

2907

Checking Mask[0] < 0 would allow a faster exit for cases that obviously won't match slidedown.

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

SDValue Res =
       DAG.getNode(RISCVISD::VNSRL_VL, DL, IntContainerVT, Src, SplatShift,
                   DAG.getUNDEF(IntContainerVT), TrueMask, VL);
   // Cast back to FP if needed.
   Res = DAG.getBitcast(ContainerVT, Res);
 
   return convertFromScalableVector(VT, Res, DAG, Subtarget);
 }
 
+// Lower the following shuffle to vslidedown.
+// t49: v8i8 = extract_subvector t13, Constant:i64<0>
+// t109: v8i8 = extract_subvector t12, Constant:i64<8>
+// t108: v8i8 = vector_shuffle<1,2,3,4,5,6,7,8> t49, t106
+static SDValue lowerVECTOR_SHUFFLEAsVSlidedown(const SDLoc &DL, MVT VT,
+                                               SDValue V1, SDValue V2,
+                                               ArrayRef<int> Mask,
+                                               const RISCVSubtarget &Subtarget,
+                                               SelectionDAG &DAG) {
+  // Both input must be extracts.
+  if (V1.getOpcode() != ISD::EXTRACT_SUBVECTOR ||
+      V2.getOpcode() != ISD::EXTRACT_SUBVECTOR)
+    return SDValue();
+
+  // Extracting from the same source.
+  SDValue Src = V1.getOperand(0);
+  if (Src != V2.getOperand(0))
+    return SDValue();
+
+  // V1 and V2 are continuous.
+  if (V1.getConstantOperandVal(1) +
+          V1.getSimpleValueType().getVectorNumElements() !=
+      V2.getConstantOperandVal(1))
+    return SDValue();
+
+  // Mask is also continuous.
+  for (unsigned i = 1; i != Mask.size(); ++i)
+    if (Mask[i - 1] + 1 != Mask[i])
+      return SDValue();
+
+  MVT XLenVT = Subtarget.getXLenVT();
+  MVT SrcVT = Src.getSimpleValueType();
+  MVT ContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
+  auto [TrueMask, VL] = getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
+  SDValue Slidedown = DAG.getNode(
+      RISCVISD::VSLIDEDOWN_VL, DL, ContainerVT, DAG.getUNDEF(ContainerVT),
+      convertToScalableVector(ContainerVT, Src, DAG, Subtarget),
+      DAG.getConstant(Mask[0], DL, XLenVT), TrueMask, VL);
+  return DAG.getNode(
+      ISD::EXTRACT_SUBVECTOR, DL, VT,
+      convertFromScalableVector(SrcVT, Slidedown, DAG, Subtarget),
+      DAG.getConstant(0, DL, XLenVT));
+}
+
 static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG,
                                    const RISCVSubtarget &Subtarget) {
   SDValue V1 = Op.getOperand(0);
   SDValue V2 = Op.getOperand(1);
   SDLoc DL(Op);
   MVT XLenVT = Subtarget.getXLenVT();
   MVT VT = Op.getSimpleValueType();
   unsigned NumElts = VT.getVectorNumElements();
if (Lane >= 0) {
                                    V1, DAG.getConstant(Lane, DL, XLenVT),
                                    DAG.getUNDEF(ContainerVT), TrueMask, VL);
       return convertFromScalableVector(VT, Gather, DAG, Subtarget);
     }
   }
 
   ArrayRef<int> Mask = SVN->getMask();
 
+  if (SDValue V =
+          lowerVECTOR_SHUFFLEAsVSlidedown(DL, VT, V1, V2, Mask, Subtarget, DAG))
+    return V;
+
   // Lower rotations to a SLIDEDOWN and a SLIDEUP. One of the source vectors may
   // be undef which can be handled with a single SLIDEDOWN/UP.
   int LoSrc, HiSrc;
   int Rotation = isElementRotate(LoSrc, HiSrc, Mask);
   if (Rotation > 0) {
     SDValue LoV, HiV;
     if (LoSrc >= 0) {
       LoV = LoSrc == 0 ? V1 : V2;

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll

 ; RV64-NEXT:    ret
   %s = shufflevector <4 x double> %x, <4 x double> <double 2.0, double 2.0, double 2.0, double 2.0>, <4 x i32> <i32 0, i32 3, i32 6, i32 5>
   ret <4 x double> %s
 }
 
 define <4 x half> @shuffle_v8f16_to_vslidedown_1(<8 x half> %x) {
 ; CHECK-LABEL: shuffle_v8f16_to_vslidedown_1:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 3, e16, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 1
-; CHECK-NEXT:    vsetivli zero, 4, e16, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 4
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 3
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 1
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <8 x half> %x, <8 x half> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   ret <4 x half> %s
 }
 
 define <4 x half> @shuffle_v8f16_to_vslidedown_3(<8 x half> %x) {
 ; CHECK-LABEL: shuffle_v8f16_to_vslidedown_3:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 1, e16, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 3
-; CHECK-NEXT:    vsetivli zero, 4, e16, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 4
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 3
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <8 x half> %x, <8 x half> poison, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
   ret <4 x half> %s
 }
 
 define <2 x float> @shuffle_v4f32_to_vslidedown(<4 x float> %x) {
 ; CHECK-LABEL: shuffle_v4f32_to_vslidedown:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 1, e32, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 1
-; CHECK-NEXT:    vsetivli zero, 2, e32, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vsetivli zero, 2, e32, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 1
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <4 x float> %x, <4 x float> poison, <2 x i32> <i32 1, i32 2>
   ret <2 x float> %s
 }
 
 define <4 x half> @slidedown_v4f16(<4 x half> %x) {
 ; CHECK-LABEL: slidedown_v4f16:

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll

 ; RV64-NEXT:    ret
   %s = shufflevector <8 x i64> %x, <8 x i64> <i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5>, <8 x i32> <i32 0, i32 3, i32 10, i32 9, i32 4, i32 1, i32 7, i32 14>
   ret <8 x i64> %s
 }
 
 define <4 x i16> @shuffle_v8i16_to_vslidedown_1(<8 x i16> %x) {
 ; CHECK-LABEL: shuffle_v8i16_to_vslidedown_1:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 3, e16, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 1
-; CHECK-NEXT:    vsetivli zero, 4, e16, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 4
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 3
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 1
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <8 x i16> %x, <8 x i16> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   ret <4 x i16> %s
 }
 
 define <4 x i16> @shuffle_v8i16_to_vslidedown_3(<8 x i16> %x) {
 ; CHECK-LABEL: shuffle_v8i16_to_vslidedown_3:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 1, e16, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 3
-; CHECK-NEXT:    vsetivli zero, 4, e16, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 4
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 3
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <8 x i16> %x, <8 x i16> poison, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
   ret <4 x i16> %s
 }
 
 define <2 x i32> @shuffle_v4i32_to_vslidedown(<4 x i32> %x) {
 ; CHECK-LABEL: shuffle_v4i32_to_vslidedown:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vsetivli zero, 1, e32, mf2, ta, ma
-; CHECK-NEXT:    vslidedown.vi v9, v8, 1
-; CHECK-NEXT:    vsetivli zero, 2, e32, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    vsetivli zero, 2, e32, mf2, tu, ma
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
-; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 1
 ; CHECK-NEXT:    ret
 entry:
   %s = shufflevector <4 x i32> %x, <4 x i32> poison, <2 x i32> <i32 1, i32 2>
   ret <2 x i32> %s
 }
 
 define <4 x i8> @interleave_shuffles(<4 x i8> %x) {
 ; CHECK-LABEL: interleave_shuffles: