Architectures for fine-grain parallelism