[RISCV] Optimize undef Even vector in getWideningInterleave. (#88221)

We recently optimized the code when the Odd vector was undef to fix a
poison bug.

There are additional optimizations we can do if the even vector is
undef. With Zvbb, we can use a single vwsll. Without Zvbb, we can use a
vzext.vf2 and a vsll.

Author: topperc

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -4638,8 +4638,17 @@ static SDValue getWideningInterleave(SDValue EvenV, SDValue OddV,
                                            Subtarget.getXLenVT()));
     Interleaved = DAG.getNode(RISCVISD::VWSLL_VL, DL, WideContainerVT, OddV,
                               OffsetVec, Passthru, Mask, VL);
-    Interleaved = DAG.getNode(RISCVISD::VWADDU_W_VL, DL, WideContainerVT,
-                              Interleaved, EvenV, Passthru, Mask, VL);
+    if (!EvenV.isUndef())
+      Interleaved = DAG.getNode(RISCVISD::VWADDU_W_VL, DL, WideContainerVT,
+                                Interleaved, EvenV, Passthru, Mask, VL);
+  } else if (EvenV.isUndef()) {
+    Interleaved =
+        DAG.getNode(RISCVISD::VZEXT_VL, DL, WideContainerVT, OddV, Mask, VL);
+
+    SDValue OffsetVec =
+        DAG.getConstant(VecVT.getScalarSizeInBits(), DL, WideContainerVT);
+    Interleaved = DAG.getNode(RISCVISD::SHL_VL, DL, WideContainerVT,
+                              Interleaved, OffsetVec, Passthru, Mask, VL);
   } else {
     // FIXME: We should freeze the odd vector here. We already handled the case
     // of provably undef/poison above.

llvm/test/CodeGen/RISCV/rvv/vector-interleave.ll

@@ -674,6 +674,26 @@ define <vscale x 8 x i32> @vector_interleave_nxv8i32_nxv4i32_poison(<vscale x 4
   ret <vscale x 8 x i32> %res
 }
 
+define <vscale x 8 x i32> @vector_interleave_nxv8i32_nxv4i32_poison2(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: vector_interleave_nxv8i32_nxv4i32_poison2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e64, m4, ta, ma
+; CHECK-NEXT:    vzext.vf2 v12, v8
+; CHECK-NEXT:    li a0, 32
+; CHECK-NEXT:    vsll.vx v8, v12, a0
+; CHECK-NEXT:    ret
+;
+; ZVBB-LABEL: vector_interleave_nxv8i32_nxv4i32_poison2:
+; ZVBB:       # %bb.0:
+; ZVBB-NEXT:    li a0, 32
+; ZVBB-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; ZVBB-NEXT:    vwsll.vx v12, v8, a0
+; ZVBB-NEXT:    vmv4r.v v8, v12
+; ZVBB-NEXT:    ret
+  %res = call <vscale x 8 x i32> @llvm.experimental.vector.interleave2.nxv8i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a)
+  ret <vscale x 8 x i32> %res
+}
+
 declare <vscale x 64 x half> @llvm.experimental.vector.interleave2.nxv64f16(<vscale x 32 x half>, <vscale x 32 x half>)
 declare <vscale x 32 x float> @llvm.experimental.vector.interleave2.nxv32f32(<vscale x 16 x float>, <vscale x 16 x float>)
 declare <vscale x 16 x double> @llvm.experimental.vector.interleave2.nxv16f64(<vscale x 8 x double>, <vscale x 8 x double>)