[llvm] [RISCV] Combine vmv.s.x to vmv.v.i if VL is known > 0 (PR #67155)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

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

<details>
<summary>Changes</summary>

We currently combine vmv.s.x with an undef passthru to vmv.v.i. Assuming a VL
toggle is cheaper than a li, we can also do this if the passthru isn't undef by
setting VL=1, provided that we know VL > 0 to begin with. (vmv.s.x has only two
behaviours with respect to VL: VL = 0, and VL > 0)


---
Full diff: https://github.com/llvm/llvm-project/pull/67155.diff


2 Files Affected:

- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+10-2) 
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll (+6-8) 


``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 8b745b2afaf956b..d14ca45821888bb 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -14480,8 +14480,16 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
     // no purpose.
     if (ConstantSDNode *Const = dyn_cast<ConstantSDNode>(Scalar);
         Const && !Const->isZero() && isInt<5>(Const->getSExtValue()) &&
-        VT.bitsLE(getLMUL1VT(VT)) && Passthru.isUndef())
-      return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Scalar, VL);
+        VT.bitsLE(getLMUL1VT(VT))) {
+      if (Passthru.isUndef())
+        return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Scalar, VL);
+
+      // If we know VL isn't zero, then we can use VL=1.
+      if (isa<ConstantSDNode>(VL) &&
+          cast<ConstantSDNode>(VL)->getZExtValue() > 0)
+        return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Scalar,
+                           DAG.getVectorIdxConstant(1, DL));
+    }
 
     break;
   }
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
index 78cafbb84bb926e..c16df830073f110 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
@@ -361,10 +361,9 @@ define <8 x i8> @splat_ve4_ins_i0ve2(<8 x i8> %v) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.v.i v10, 4
-; CHECK-NEXT:    li a0, 2
-; CHECK-NEXT:    vsetvli zero, zero, e8, mf2, tu, ma
-; CHECK-NEXT:    vmv.s.x v10, a0
-; CHECK-NEXT:    vsetvli zero, zero, e8, mf2, ta, ma
+; CHECK-NEXT:    vsetivli zero, 1, e8, mf2, tu, ma
+; CHECK-NEXT:    vmv.v.i v10, 2
+; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vrgather.vv v9, v8, v10
 ; CHECK-NEXT:    vmv1r.v v8, v9
 ; CHECK-NEXT:    ret
@@ -407,10 +406,9 @@ define <8 x i8> @splat_ve2_we0_ins_i0ve4(<8 x i8> %v, <8 x i8> %w) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.v.i v11, 2
-; CHECK-NEXT:    li a0, 4
-; CHECK-NEXT:    vsetvli zero, zero, e8, mf2, tu, ma
-; CHECK-NEXT:    vmv.s.x v11, a0
-; CHECK-NEXT:    vsetvli zero, zero, e8, mf2, ta, mu
+; CHECK-NEXT:    vsetivli zero, 1, e8, mf2, tu, ma
+; CHECK-NEXT:    vmv.v.i v11, 4
+; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, mu
 ; CHECK-NEXT:    li a0, 66
 ; CHECK-NEXT:    vmv.s.x v0, a0
 ; CHECK-NEXT:    vrgather.vv v10, v8, v11

``````````

</details>


https://github.com/llvm/llvm-project/pull/67155