[mlir] remove-dead-values pass throws error when module has a name

[mrakitaTT]

When the input mlir module has a name, like this for example:

module @test_opt attributes {} {
  func.func public @test_opt(%arg0: tensor<128x10xf32>) -> tensor<128x10xf32> {
    %0 = tensor.empty() : tensor<64x96xbf16>
    // CHECK-NOT: %[[C:.*]] = tensor.empty[[C:.*]]
    return %arg0 : tensor<128x10xf32>
  }
}

then remove-dead-values pass throws this error:

error: cannot optimize an IR with non-function symbol ops, non-call symbol user ops or branch ops

After removing module name (@test_opt) everything works. Seems to be a bug.

[CoTinker]

llvm-project/mlir/lib/Transforms/RemoveDeadValues.cpp

Lines 575 to 587 in ffcff2f

	// The removal of non-live values is performed iff there are no branch ops,
	// all symbol ops present in the IR are function-like, and all symbol user ops
	// present in the IR are call-like.
	WalkResult acceptableIR = module->walk([&](Operation *op) {
	if (isa<BranchOpInterface>(op) ||
	(isa<SymbolOpInterface>(op) && !isa<FunctionOpInterface>(op)) ||
	(isa<SymbolUserOpInterface>(op) && !isa<CallOpInterface>(op))) {
	op->emitError() << "cannot optimize an IR with non-function symbol ops, "
	"non-call symbol user ops or branch ops\n";
	return WalkResult::interrupt();
	}
	return WalkResult::advance();
	});

It's just unsupported.

[joker-eph]

It's not clear to me that the "unsupported" aspect is intentional or just an accident of the implementation that didn't consider that the walk would apply to the module itself and that it could also have a name?

That is can we just add:

 WalkResult acceptableIR = module->walk([&](Operation *op) { 
    if (op == module) return  WalkResult::advance();