TALU:用于RISC-V的可重构 tiled-ALU 阵列
TALU: A Reconfigurable Tiled-ALU Grid for RISC-V
AI总结:
针对 Lane 型 RISC-V 向量处理器归约开销高、深度依赖表达式链无法融合的问题,提出可重构 tiled-ALU 阵列架构 TALU,其在基准测试中性能显著优于 RISC-V2 和 Vicuna,面积仅为 RISC-V2 的一半。
摘要:
基于 Lane 的 RISC-V 向量处理器能很好地利用数据并行性,但在两种常见模式下会产生重复开销:归约操作需要单独的树遍历过程,深度依赖表达式链无法融合,会作为串行向量指令发射。我们提出 TALU,这是一款 5 级 RISC-V 核心,其执行阶段是 N×N 的寄存器化 ALU 单元阵列,而非单个 ALU。该阵列是可重构的前馈空间结构:单元仅从非后向邻居处路由操作数,新增的对角转发机制可让归约操作在空间上合并。一个极简的自定义扩展用于配置每个单元的操作、路由及单元内常量,并将操作数流式传输通过阵列。在单个结构上,TALU 可运行普通标量代码、单深度路径表达式及多 Lane 向量内核。在 RISC-V2 向量处理器基准测试套件中,当 Lane 数量相同时,TALU 的速度是 RISC-V2 的 1.8-5.5 倍,是 Vicuna 的 4.5-15.5 倍;其归约成本(点积÷加法)为 1.05 倍,而向量核心的该值分别为 1.96 倍和 1.26 倍。经时序重定时的实现,在 45nm 工艺下可达到 1.03 GHz 的工作频率,面积为 0.30 mm²,仅为 RISC-V2 面积的一半。
英文摘要:
Lane-based RISC-V vector processors exploit data parallelism well but pay a recurring cost for two common patterns: reductions, which require a separate tree pass, and deep dependent expression chains, which cannot be fused and are issued as serial vector instructions. We present TALU, a 5-stage RISC-V core whose execute stage is an NxN grid of registered ALU tiles instead of a single ALU. The grid is a reconfigurable, feed-forward spatial fabric: tiles route operands only from non-backward neighbours, and added diagonal forwarding lets a reduction collapse spatially. A minimal custom extension configures the per-tile operation, routing, and per-tile constants, and streams operands through the array. On a single fabric, TALU runs ordinary scalar code, single deep-path expressions, and multi-lane vector kernels. On the benchmark suite of the RISC-V2 vector processor, TALU is 1.8-5.5x faster than RISC-V2 and 4.5-15.5x faster than Vicuna at equal lane count, with a reduction cost (dot÷add) of 1.05x versus 1.96x and 1.26x for the vector cores. A retimed implementation places and routes at 1.03 GHz in 45 nm in 0.30 mm2, half the area of RISC-V2.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。