[RISCV] Implement prefetch locality by NTLH

We add the MemOperand then backend will generate NTLH automatically.

```
__builtin_prefetch(ptr,  0 /* rw==read */, 0 /* locality */); => ntl.all + prefetch.r (ptr)
__builtin_prefetch(ptr,  0 /* rw==read */, 1 /* locality */); => ntl.pall + prefetch.r (ptr)
__builtin_prefetch(ptr,  0 /* rw==read */, 2 /* locality */); => ntl.p1 + prefetch.r (ptr)
__builtin_prefetch(ptr,  0 /* rw==read */, 3 /* locality */); => prefetch.r (ptr)
```

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D154691

Author: BeMg

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

@@ -2108,6 +2108,36 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
     ReplaceNode(Node, Load);
     return;
   }
+  case ISD::PREFETCH:
+    unsigned Locality = Node->getConstantOperandVal(3);
+    if (Locality > 2)
+      break;
+
+    if (auto *LoadStoreMem = dyn_cast<MemSDNode>(Node)) {
+      MachineMemOperand *MMO = LoadStoreMem->getMemOperand();
+      MMO->setFlags(MachineMemOperand::MONonTemporal);
+
+      int NontemporalLevel = 0;
+      switch (Locality) {
+      case 0:
+        NontemporalLevel = 3; // NTL.ALL
+        break;
+      case 1:
+        NontemporalLevel = 1; // NTL.PALL
+        break;
+      case 2:
+        NontemporalLevel = 0; // NTL.P1
+        break;
+      default:
+        llvm_unreachable("unexpected locality value.");
+      }
+
+      if (NontemporalLevel & 0b1)
+        MMO->setFlags(MONontemporalBit0);
+      if (NontemporalLevel & 0b10)
+        MMO->setFlags(MONontemporalBit1);
+    }
+    break;
   }
 
   // Select the default instruction.

llvm/test/CodeGen/RISCV/prefetch.ll

@@ -33,6 +33,7 @@ define void @test_prefetch_read_locality_0(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_read_locality_0:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.all
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.r 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 0, i32 1)
@@ -60,6 +61,7 @@ define void @test_prefetch_write_locality_0(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_write_locality_0:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.all
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.w 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 1, i32 0, i32 1)
@@ -87,6 +89,7 @@ define void @test_prefetch_instruction_locality_0(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_instruction_locality_0:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.all
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.i 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 0, i32 0)
@@ -114,6 +117,7 @@ define void @test_prefetch_read_locality_1(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_read_locality_1:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.pall
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.r 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 1, i32 1)
@@ -141,6 +145,7 @@ define void @test_prefetch_write_locality_1(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_write_locality_1:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.pall
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.w 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 1, i32 1, i32 1)
@@ -168,6 +173,7 @@ define void @test_prefetch_instruction_locality_1(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_instruction_locality_1:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.pall
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.i 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 1, i32 0)
@@ -195,6 +201,7 @@ define void @test_prefetch_read_locality_2(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_read_locality_2:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.p1
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.r 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 2, i32 1)
@@ -222,6 +229,7 @@ define void @test_prefetch_write_locality_2(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_write_locality_2:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.p1
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.w 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 1, i32 2, i32 1)
@@ -249,6 +257,7 @@ define void @test_prefetch_instruction_locality_2(ptr %a) nounwind {
 ;
 ; RV64ZICBOPZIHINTNTL-LABEL: test_prefetch_instruction_locality_2:
 ; RV64ZICBOPZIHINTNTL:       # %bb.0:
+; RV64ZICBOPZIHINTNTL-NEXT:    ntl.p1
 ; RV64ZICBOPZIHINTNTL-NEXT:    prefetch.i 0(a0)
 ; RV64ZICBOPZIHINTNTL-NEXT:    ret
   call void @llvm.prefetch(ptr %a, i32 0, i32 2, i32 0)