[MLIR] NFC. Drop trailing white space in GPU async ops print

[bondhugula]

Summary

NFC. Drop trailing end of line white space in GPU async ops' printer
whenever the list of async deps is empty.

[rriddle]

mlir/lib/Dialect/GPU/IR/GPUDialect.cpp

1056

It's not really clear to me why this is a custom directive. Feels like it could quite easily just be in the declarative format.

[bondhugula]

mlir/lib/Dialect/GPU/IR/GPUDialect.cpp

1056

The parse version of this requires a non-trivial check (looking at the number of results of parsed).

[mehdi_amini]

mlir/lib/Dialect/GPU/IR/GPUDialect.cpp

1056

Why though? To emit a nicer error?

[bondhugula]

mlir/lib/Dialect/GPU/IR/GPUDialect.cpp

1056

Not just for a nicer error, but for functionality AFAIU -- if the async keyword is used, a single result is expected. Can you see parseAsyncDependencies?

[bondhugula]

Any other comments here: @rriddle?

[rriddle]

LGTM (sorry, missed this)

mlir/lib/Dialect/GPU/IR/GPUDialect.cpp

if (succeeded(parser.parseOptionalKeyword("async"))) {
     if (parser.getNumResults() == 0)
       return parser.emitError(loc, "needs to be named when marked 'async'");
     asyncTokenType = parser.getBuilder().getType<AsyncTokenType>();
   }
   return parser.parseOperandList(asyncDependencies,
                                  OpAsmParser::Delimiter::OptionalSquare);
 }
 
+/// Prints optional async dependencies with its leading keyword.
+///   (`async`)? (`[` ssa-id-list `]`)?
+// Used by the tablegen assembly format for several async ops.
 static void printAsyncDependencies(OpAsmPrinter &printer, Operation *op,
                                    Type asyncTokenType,
                                    OperandRange asyncDependencies) {
   if (asyncTokenType)
-    printer << "async ";
+    printer << "async";
   if (asyncDependencies.empty())
     return;
-  printer << "[";
+  if (asyncTokenType)
+    printer << ' ';
+  printer << '[';
   llvm::interleaveComma(asyncDependencies, printer);
-  printer << "]";
+  printer << ']';
 }
 
 //===----------------------------------------------------------------------===//
 // GPU_SubgroupMmaLoadMatrixOp
 //===----------------------------------------------------------------------===//
 
 /// Return true if the last dimension of the MemRefType has unit stride. Also
 /// return true for memrefs with no strides.