DICE: Enabling Efficient General-Purpose SIMT Execution with Statically Scheduled Coarse-Grained Reconfigurable Arrays

While general-purpose graphics processing units (GPGPUs) dominate massively parallel computing through the single-instruction, multiple-thread (SIMT) programming model, their underlying single-instruction, multiple-data (SIMD) execution incurs substantial energy overhead from frequent register file (RF) accesses and complex control logic. We present DICE, a novel architecture that addresses these inefficiencies by replacing the SIMD backend with minimaloverhead, statically scheduled coarse-grained reconfigurable arrays (CGRAs) while preserving the well-known SIMT programming model. Unlike SIMD units that execute warps of threads in lockstep, DICE dispatches active threads in a pipelined manner onto the CGRA fabric, where data flows directly between processing elements (PEs), reducing RF accesses for intermediate values. Moreover, DICE organizes threads into groups larger than warps, reducing overhead by eliminating the warp-level control logic and further amortizing remaining control overhead across hundreds of threads. To handle operations with runtime dynamism, such as variable-latency memory loads and data-dependent control flow, while preserving static scheduling, DICE compiles programs into p-graphs by partitioning dynamic dependence edges across separate CGRA configurations. DICE further introduces several key optimizations: double-buffered configuration memory to hide reconfiguration latency, compile-time $p$-graph unrolling to enhance resource utilization, and a temporal memory coalescing unit (TMCU) to merge memory requests from consecutive, pipelined threads. Evaluations on Rodinia benchmarks in Accelsim demonstrate that DICE reduces register file accesses by 68% on average. With equivalent computation and memory resources, DICE's CGRA Processors (CPs) achieve a geometric mean of $\text{1. 7 7 - 1. 9 0} \times$ dynamic energy efficiency and $\text{4 2. 0 \% - 4 5. 9 \%}$ average power reduction compared to the modeled NVIDIA Turing Streaming Multiprocessors (SMs), while the full DICE system achieves performance comparable to the modeled Turing GPU baselines. DICE demonstrates that spatial pipeline execution can deliver substantial energy savings without sacrificing performance.