[mlir][OpenMP] omp.single translation to LLVM IR

[shraiysh]

Summary

This patch adds translation from omp.single to LLVM IR.

Depends on D122288

[shraiysh]

Fixed spelling error.

[kiranchandramohan]

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

642

The code here is probably very similar to the master translation, is there any scope for codesharing?

655

Nit: Can you add a comment for the last field (which is nullptr)? Is. that required for copyprivate?

[shraiysh]

Addressed comments.

[shraiysh]

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

642

We can do that, but there are two things we have to think about:

1. Will it make it harder to understand and modify the code, because master and single are two unrelated constructs that have one function.
2. Master has no clauses, while single has clauses (the handling for allocate and nowait has not been added because it has not been handled by the IRBuilder yet, but they will be added) and will having a common function make it harder to handle both?

I suppose with careful handling, we could carve out a part of both these functions into one function, but intuitively what will that function will doing (what would be it's name)? I can give it a shot if you'd like to see how that would look like. Please let me know.

655

Sure, I will add it. I am not too sure what that is for. It is used here to set the value to i32 0.

[kiranchandramohan]

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

642

OK. Probably not worth it.

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

convertOmpSections(Operation &opInst, llvm::IRBuilderBase &builder,
 
   llvm::OpenMPIRBuilder::LocationDescription ompLoc(builder);
   builder.restoreIP(moduleTranslation.getOpenMPBuilder()->createSections(
       ompLoc, findAllocaInsertPoint(builder, moduleTranslation), sectionCBs,
       privCB, finiCB, false, sectionsOp.nowait()));
   return bodyGenStatus;
 }
 
+/// Converts an OpenMP single construct into LLVM IR using OpenMPIRBuilder.
+static LogicalResult
+convertOmpSingle(omp::SingleOp &singleOp, llvm::IRBuilderBase &builder,
+                 LLVM::ModuleTranslation &moduleTranslation) {
+  using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
+  llvm::OpenMPIRBuilder::LocationDescription ompLoc(builder);
+  LogicalResult bodyGenStatus = success();
+  auto bodyCB = [&](InsertPointTy allocaIP, InsertPointTy codegenIP,
+                    llvm::BasicBlock &continuationBB) {
+    convertOmpOpRegions(singleOp.region(), "omp.single.region",
+                        *codegenIP.getBlock(), continuationBB, builder,
+                        moduleTranslation, bodyGenStatus);
+  };
+  auto finiCB = [&](InsertPointTy codeGenIP) {};
+  builder.restoreIP(moduleTranslation.getOpenMPBuilder()->createSingle(
+      ompLoc, bodyCB, finiCB, /*DidIt=*/nullptr));
+  return bodyGenStatus;
+}
+
 /// Converts an OpenMP workshare loop into LLVM IR using OpenMPIRBuilder.
 static LogicalResult
 convertOmpWsLoop(Operation &opInst, llvm::IRBuilderBase &builder,
                  LLVM::ModuleTranslation &moduleTranslation) {
   auto loop = cast<omp::WsLoopOp>(opInst);
   // TODO: this should be in the op verifier instead.
   if (loop.lowerBound().empty())
     return failure();
return llvm::TypeSwitch<Operation *, LogicalResult>(op)
         return convertOmpAtomicUpdate(op, builder, moduleTranslation);
       })
       .Case([&](omp::AtomicCaptureOp op) {
         return convertOmpAtomicCapture(op, builder, moduleTranslation);
       })
       .Case([&](omp::SectionsOp) {
         return convertOmpSections(*op, builder, moduleTranslation);
       })
+      .Case([&](omp::SingleOp op) {
+        return convertOmpSingle(op, builder, moduleTranslation);
+      })
       .Case<omp::YieldOp, omp::TerminatorOp, omp::ReductionDeclareOp,
             omp::CriticalDeclareOp>([](auto op) {
         // `yield` and `terminator` can be just omitted. The block structure
         // was created in the region that handles their parent operation.
         // `reduction.declare` will be used by reductions and is not
         // converted directly, skip it.
         // `critical.declare` is only used to declare names of critical
         // sections which will be used by `critical` ops and hence can be

mlir/test/Target/LLVMIR/openmp-llvm.mlir

 llvm.func @repeated_successor(%arg0: i64, %arg1: i64, %arg2: i64, %arg3: i1) {
   omp.wsloop for (%arg4) : i64 = (%arg0) to (%arg1) step (%arg2)  {
     llvm.cond_br %arg3, ^bb1(%arg0 : i64), ^bb1(%arg1 : i64)
   ^bb1(%0: i64):  // 2 preds: ^bb0, ^bb0
     omp.yield
   }
   llvm.return
 }
+
+// -----
+
+// CHECK-LABEL: @single
+// CHECK-SAME: (i32 %[[x:.*]], i32 %[[y:.*]], i32* %[[zaddr:.*]])
+llvm.func @single(%x: i32, %y: i32, %zaddr: !llvm.ptr<i32>) {
+  // CHECK: call i32 @__kmpc_single
+  omp.single {
+    // CHECK: %[[z:.*]] = add i32 %[[x]], %[[y]]
+    %z = llvm.add %x, %y : i32
+    // CHECK: store i32 %[[z]], i32* %[[zaddr]]
+    llvm.store %z, %zaddr : !llvm.ptr<i32>
+    // CHECK: call void @__kmpc_end_single
+    omp.terminator
+  }
+  // CHECK: ret void
+  llvm.return
+}