[RISCV] Correct the VLOperand for masked vssrl/vssra intrinsics.

topperc · topperc · commit dbe9a602561d · 2023-12-17T17:42:08.000-08:00
Though I can't prove it matters for anything. The only use of
VLOperand I know of is for handling i64 splat operands to .vx
intrinsics on RV32. Shifts are special and always use XLen for .vx
so they are always legal.
diff --git a/llvm/include/llvm/IR/IntrinsicsRISCV.td b/llvm/include/llvm/IR/IntrinsicsRISCV.td
@@ -702,7 +702,7 @@ let TargetPrefix = "riscv" in {
                      LLVMMatchType<2>, LLVMMatchType<2>],
                     [ImmArg<ArgIndex<4>>,ImmArg<ArgIndex<6>>, IntrNoMem, IntrHasSideEffects]>,
                     RISCVVIntrinsic {
-    let VLOperand = 6;
+    let VLOperand = 5;
   }
   // For Saturating binary operations.
   // The destination vector type is NOT the same as first source vector.

Original file line number	Diff line number	Diff line change
`@@ -702,7 +702,7 @@ let TargetPrefix = "riscv" in {`
`702`	`702`	`LLVMMatchType<2>, LLVMMatchType<2>],`
`703`	`703`	`[ImmArg<ArgIndex<4>>,ImmArg<ArgIndex<6>>, IntrNoMem, IntrHasSideEffects]>,`
`704`	`704`	`RISCVVIntrinsic {`
`705`		`- let VLOperand = 6;`
	`705`	`+ let VLOperand = 5;`
`706`	`706`	`}`
`707`	`707`	`// For Saturating binary operations.`
`708`	`708`	`// The destination vector type is NOT the same as first source vector.`