[NVPTX] Prefer prmt.b32 over bfi.b32 (#110766)

In [[NVPTX] Improve lowering of
v4i8](cbafb6f)
@Artem-B add the ability to lower ISD::BUILD_VECTOR with bfi PTX
instructions. @Artem-B did this because:
([source](#67866 (comment)))

> Under the hood byte extraction/insertion ends up as BFI/BFE
instructions, so we may as well do that in PTX, too.
https://godbolt.org/z/Tb3zWbj9b

However, the example that @Artem-B linked was targeting sm_52. On modern
architectures, ptxas uses prmt.b32.
[Example](https://godbolt.org/z/Ye4W1n84o).

Thus, remove uses of NVPTXISD::BFI in favor of NVPTXISD::PRMT.

Author: justinfargnoli

llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp

@@ -2332,20 +2332,23 @@ SDValue NVPTXTargetLowering::LowerBUILD_VECTOR(SDValue Op,
     // Lower non-const v4i8 vector as byte-wise constructed i32, which allows us
     // to optimize calculation of constant parts.
     if (VT == MVT::v4i8) {
-      SDValue C8 = DAG.getConstant(8, DL, MVT::i32);
-      SDValue E01 = DAG.getNode(
-          NVPTXISD::BFI, DL, MVT::i32,
-          DAG.getAnyExtOrTrunc(Op->getOperand(1), DL, MVT::i32),
-          DAG.getAnyExtOrTrunc(Op->getOperand(0), DL, MVT::i32), C8, C8);
-      SDValue E012 =
-          DAG.getNode(NVPTXISD::BFI, DL, MVT::i32,
-                      DAG.getAnyExtOrTrunc(Op->getOperand(2), DL, MVT::i32),
-                      E01, DAG.getConstant(16, DL, MVT::i32), C8);
-      SDValue E0123 =
-          DAG.getNode(NVPTXISD::BFI, DL, MVT::i32,
-                      DAG.getAnyExtOrTrunc(Op->getOperand(3), DL, MVT::i32),
-                      E012, DAG.getConstant(24, DL, MVT::i32), C8);
-      return DAG.getNode(ISD::BITCAST, DL, VT, E0123);
+      SDValue PRMT__10 = DAG.getNode(
+          NVPTXISD::PRMT, DL, MVT::v4i8,
+          {DAG.getAnyExtOrTrunc(Op->getOperand(0), DL, MVT::i32),
+           DAG.getAnyExtOrTrunc(Op->getOperand(1), DL, MVT::i32),
+           DAG.getConstant(0x3340, DL, MVT::i32),
+           DAG.getConstant(NVPTX::PTXPrmtMode::NONE, DL, MVT::i32)});
+      SDValue PRMT32__ = DAG.getNode(
+          NVPTXISD::PRMT, DL, MVT::v4i8,
+          {DAG.getAnyExtOrTrunc(Op->getOperand(2), DL, MVT::i32),
+           DAG.getAnyExtOrTrunc(Op->getOperand(3), DL, MVT::i32),
+           DAG.getConstant(0x4033, DL, MVT::i32),
+           DAG.getConstant(NVPTX::PTXPrmtMode::NONE, DL, MVT::i32)});
+      SDValue PRMT3210 = DAG.getNode(
+          NVPTXISD::PRMT, DL, MVT::v4i8,
+          {PRMT__10, PRMT32__, DAG.getConstant(0x5410, DL, MVT::i32),
+           DAG.getConstant(NVPTX::PTXPrmtMode::NONE, DL, MVT::i32)});
+      return DAG.getNode(ISD::BITCAST, DL, VT, PRMT3210);
     }
     return Op;
   }