Order of Magnitude Instruction Level Parallelism

这个项目正在探索一种新的指令级并行方法。新的方法，不相交的急切执行，使用分支预测和急切执行来最大化在每个周期上可以有效执行的指令数量。这组分支路径形成一棵树，其中每个节点代表一个分支指令，每个边代表分支之间执行的代码。在不相交的紧急执行中，所有最可能的代码段都是推测性执行的候选代码段，而不仅仅是最可能路径上的那些代码段。换句话说，执行的指令子树可能不是简单的链。初步研究包括对基准的一些模拟，表明加速约30%是可能的。该项目包括更全面的模拟以确定好处，以及体系结构研究以确定并行性对机器周期时间的影响。

This project is exploring a new approach to instruction-level parallelism. The new approach, disjoint eager execution, uses branch prediction and eager execution to maximize the number of instructions that can be usefully executed on each cycle. The set of branch paths forms a tree, in which each node represents a branch instruction and each edge represents the code executed between branches. In disjoint eager execution, all of the most likely code segments are candidates for speculative execution, rather than only those code segments on the most likely path. In other words, a subtree of instructions is executed that may not be a simple chain. Preliminary research has included some simulations on benchmarks suggesting that speedup of about 30 is possible. This project includes more comprehensive simulations to determine the benefits, along with architectural studies to determine the effects of parallelism on cycle time of the machine.