Remove unnecessary async group creates and awaits.

[mehdi_amini]

I'm surprised there is no test where this is reflected, can you add one?

[ezhulenev]

+1 to check that group is not created if dispatch is sync

[bakhtiyarneyman]

Fix the test.

mlir/lib/Dialect/Async/Transforms/AsyncParallelFor.cpp

static void doAsyncDispatch(ImplicitLocOpBuilder &b, PatternRewriter &rewriter,
   // Add one more level of indirection to dispatch parallel compute functions
   // using async operations and recursive work splitting.
   FuncOp asyncDispatchFunction =
       createAsyncDispatchFunction(parallelComputeFunction, rewriter);
 
   Value c0 = b.create<ConstantIndexOp>(0);
   Value c1 = b.create<ConstantIndexOp>(1);
 
-  // Create an async.group to wait on all async tokens from the concurrent
-  // execution of multiple parallel compute function. First block will be
-  // executed synchronously in the caller thread.
-  Value groupSize = b.create<SubIOp>(blockCount, c1);
-  Value group = b.create<CreateGroupOp>(GroupType::get(ctx), groupSize);
-
   // Appends operands shared by async dispatch and parallel compute functions to
   // the given operands vector.
   auto appendBlockComputeOperands = [&](SmallVector<Value> &operands) {
     operands.append(tripCounts);
     operands.append(op.lowerBound().begin(), op.lowerBound().end());
     operands.append(op.upperBound().begin(), op.upperBound().end());
     operands.append(op.step().begin(), op.step().end());
     operands.append(parallelComputeFunction.captures);
auto syncDispatch = [&](OpBuilder &nestedBuilder, Location loc) {
 
     nb.create<CallOp>(parallelComputeFunction.func.sym_name(),
                       parallelComputeFunction.func.getCallableResults(),
                       operands);
     nb.create<scf::YieldOp>();
   };
 
   auto asyncDispatch = [&](OpBuilder &nestedBuilder, Location loc) {
+    // Create an async.group to wait on all async tokens from the concurrent
+    // execution of multiple parallel compute function. First block will be
+    // executed synchronously in the caller thread.
+    Value groupSize = b.create<SubIOp>(blockCount, c1);
+    Value group = b.create<CreateGroupOp>(GroupType::get(ctx), groupSize);
+
     ImplicitLocOpBuilder nb(loc, nestedBuilder);
 
     // Launch async dispatch function for [0, blockCount) range.
     SmallVector<Value> operands = {group, c0, blockCount, blockSize};
     appendBlockComputeOperands(operands);
 
     nb.create<CallOp>(asyncDispatchFunction.sym_name(),
                       asyncDispatchFunction.getCallableResults(), operands);
+
+    // Wait for the completion of all parallel compute operations.
+    b.create<AwaitAllOp>(group);
+
     nb.create<scf::YieldOp>();
   };
 
   // Dispatch either single block compute function, or launch async dispatch.
   b.create<scf::IfOp>(TypeRange(), isSingleBlock, syncDispatch, asyncDispatch);
-
-  // Wait for the completion of all parallel compute operations.
-  b.create<AwaitAllOp>(group);
 }
 
 // Dispatch parallel compute functions by submitting all async compute tasks
 // from a simple for loop in the caller thread.
 static void
 doSequentialDispatch(ImplicitLocOpBuilder &b, PatternRewriter &rewriter,
                      ParallelComputeFunction &parallelComputeFunction,
                      scf::ParallelOp op, Value blockSize, Value blockCount,

mlir/test/Dialect/Async/async-parallel-for-async-dispatch.mlir

 // RUN: mlir-opt %s -split-input-file -async-parallel-for=async-dispatch=true  \
 // RUN: | FileCheck %s --dump-input=always
 
 // CHECK-LABEL: @loop_1d(
 // CHECK-SAME:    %[[LB:.*]]: index, %[[UB:.*]]: index, %[[STEP:.*]]: index
 func @loop_1d(%arg0: index, %arg1: index, %arg2: index, %arg3: memref<?xf32>) {
   // CHECK:      %[[C0:.*]] = constant 0 : index
 
   // CHECK:      %[[RANGE:.*]] = subi %[[UB]], %[[LB]]
   // CHECK:      %[[TRIP_CNT:.*]] = ceildivi_signed %[[RANGE]], %[[STEP]]
   // CHECK:      %[[IS_NOOP:.*]] = cmpi eq, %[[TRIP_CNT]], %[[C0]] : index
 
   // CHECK:      scf.if %[[IS_NOOP]] {
   // CHECK-NEXT: } else {
-  // CHECK:        %[[GROUP:.*]] = async.create_group
     // CHECK:        scf.if {{.*}} {
   // CHECK:          call @parallel_compute_fn(%[[C0]]
   // CHECK:        } else {
+  // CHECK:          %[[GROUP:.*]] = async.create_group
   // CHECK:          call @async_dispatch_fn
-  // CHECK:        }
-  // CHECK:        async.await_all %[[GROUP]]
-  // CHECK:      }
+  // CHECK:          async.await_all %[[GROUP]]
+  // CHECK:        }
+  // CHECK:      }
   scf.parallel (%i) = (%arg0) to (%arg1) step (%arg2) {
     %one = constant 1.0 : f32
     memref.store %one, %arg3[%i] : memref<?xf32>
   }
   return
 }
 
 // CHECK-LABEL: func private @parallel_compute_fn