[llvm] [NVPTX] Add Bulk Copy Prefetch Intrinsics (PR #123226)

[Artem Belevich via llvm-commits]

================
@@ -3105,6 +3105,30 @@ void NVPTXDAGToDAGISel::SelectCpAsyncBulkG2S(SDNode *N) {
   ReplaceNode(N, CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops));
 }
 
+void NVPTXDAGToDAGISel::SelectCpAsyncBulkPrefetchL2(SDNode *N) {
+  // We have {Chain, Intrinsic-ID} followed by the actual intrisic args:
+  // src, size, cache_hint, cache_hint_flag
+  // NumOperands = {Chain, IID} + {Actual intrinsic args}
+  //             = {2}          + {4}
+  size_t NumOps = N->getNumOperands();
+  bool IsCacheHint = N->getConstantOperandVal(NumOps - 1) == 1;
+  size_t NumArgs = IsCacheHint ? 3 : 2; // src, size, cache_hint
+
+  SDLoc DL(N);
+  SmallVector<SDValue, 8> Ops(N->ops().slice(2, NumArgs));
+  Ops.push_back(N->getOperand(0)); // Chain operand
+  //if (IsCacheHint) {
+  //  Ops.push_back(N->getOperand(2));
+  //}
+  
+  unsigned Opcode;
+  if (IsCacheHint)
+    Opcode = NVPTX::CP_ASYNC_BULK_PREFETCH_CH;
+  else
+    Opcode = NVPTX::CP_ASYNC_BULK_PREFETCH;
----------------
Artem-B wrote:

Indeed:
```
unsigned Opcode = IsCacheHint 
  ?  NVPTX::CP_ASYNC_BULK_PREFETCH_CH
  :  NVPTX::CP_ASYNC_BULK_PREFETCH;
```

https://github.com/llvm/llvm-project/pull/123226