[RISCV] Use SelectionDAG::getVScale in lowerVPReverseExperimental. NFCI

[lukel97]

Use a slightly more idiomatic way of getting vscale. getVScale
performs additional constant folding, but I presume computeKnownBits
also catches these cases too.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Luke Lau (lukel97)

Changes

Use a slightly more idiomatic way of getting vscale. getVScale
performs additional constant folding, but I presume computeKnownBits
also catches these cases too.

---

Full diff: https://github.com/llvm/llvm-project/pull/81694.diff

1 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+2-2)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 73492c28815b1d..4caadef694d725 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -11106,8 +11106,8 @@ RISCVTargetLowering::lowerVPReverseExperimental(SDValue Op,
       // Slide off any elements from past EVL that were reversed into the low
       // elements.
       unsigned MinElts = GatherVT.getVectorMinNumElements();
-      SDValue VLMax = DAG.getNode(ISD::VSCALE, DL, XLenVT,
-                                  DAG.getConstant(MinElts, DL, XLenVT));
+      SDValue VLMax =
+          DAG.getVScale(DL, XLenVT, APInt(XLenVT.getSizeInBits(), MinElts));
       SDValue Diff = DAG.getNode(ISD::SUB, DL, XLenVT, VLMax, EVL);
 
       Result = getVSlidedown(DAG, Subtarget, DL, GatherVT,

[topperc]

LGTM