Heterogeneous architectures became a compelling choice for edge processors executing complex DNN workloads, as they provide an ideal blend of openness, customization, energy-efficient heterogeneity, and scalable performance. Compiler optimization for DNNs on heterogeneous System-on-Chip (SoC) architectures however, must navigate complex hardware-software co-design, data movement minimization, aggressive parallelism exploitation, and advanced static/dynamic code transformations to deliver high performance and energy efficiency.This paper presents a novel compiler ecosystem for highly heterogeneous SoCs with multiple back-end targets, spanning from typical CPUs, to programmable RISC-V clusters and up to dedicated and reconfigurable accelerators. It puts together static analysis, optimization, and scheduling infrastructure to overcome the limitations of current state-of-the-art tools for heterogeneous edge AI processors. Our compilation pipeline introduces several innovative features: (1) an automatic end-to-end flow for RISC-V-based platforms, (2) efficient data layout remapping (reducing memory footprint by 35% on average) and recognition of complex ternary reductions for auto-vectorization, (3) code layout adaptation for hardware simplification, (4) a novel MLIR-based RISC-V backend supporting optimized matrix-multiplication micro-kernels that reach 90% of peak performance, (5) periodic scheduling capabilities for layer-fused CNNs, and (6) automated mapping and scheduling onto heterogeneous CGRA templates for advanced parallel kernel execution, delivering 33% higher energy efficiency than the scalar implementation and up to 3.6× higher performance. These advances enable hardware-aware compilation that reduces manual optimization effort, lowers energy consumption through memory and computation optimization, and minimizes memory footprint and data transfers.

The evolution of quantization and mixed-precision techniques has unlocked new possibilities for enhancing the speed and energy efficiency of Neural Networks (NNs). Several recent studies indicate that adapting precision levels across different parameters can maintain accuracy comparable to full-precision models while significantly reducing computational demands. However, existing embedded microprocessors lack sufficient architectural support for efficiently executing mixed-precision NNs, both in terms of ISA extensions and hardware design. This limitation results in inefficiencies such as excessive data packing/unpacking and underutilized arithmetic units, leading to performance bottlenecks. In this work, to address these challenges, we propose novel ISA extensions and the micro-architecture implementation specifically designed to optimize mixed-precision execution, enabling energy-efficient deep learning inference on RISC-V architectures. We introduceMaRVIn, a cross-layer hardware-software co-design framework that enhances power efficiency and performance through a combination of hardware improvements, mixed-precision quantization, ISA-level optimizations, and cycle-accurate emulation. At the hardware level, we enhance the ALU with configurable mixed-precision arithmetic (2-, 4-, and 8-bit) for weights and/or activations. To further improve execution efficiency, we employ multi-pumping to reduce execution latency and implement soft SIMD for efficient 2-bit operations. We also extend ISA to support these mixed-precision operations. At the software level, we integrate a pruning-aware fine-tuning method to optimize model compression. Additionally, we introduce a greedy-based design space exploration (DSE) approach to efficiently search for Pareto-optimal mixed-quantized models. Finally, we incorporate voltage scaling to boost the power efficiency of our system. Our extensive experimental evaluation over widely used DNNs and datasets, such as CIFAR10 and ImageNet, demonstrates that our framework can achieve, on average, 17.6× speedup for less than 1% accuracy loss and outperforms the ISA-agnostic state-of-the-art RISC-V cores, delivering up to 1.8 TOPs/W.

Recent advancements in quantization and mixed-precision approaches offers substantial opportunities to improve the speed and energy efficiency of Neural Networks (NN). Research has shown that individual parameters with varying low precision, can attain accuracies comparable to full-precision counterparts. However, modern embedded microprocessors provide very limited support for mixed-precision NNs regarding both Instruction Set Architecture (ISA) extensions and their hardware design for efficient execution of mixed-precision operations, i.e., introducing several performance bottlenecks due to numerous instructions for data packing and unpacking, arithmetic unit under-utilizations etc. In this work, we bring together, for the first time, ISA extensions tailored to mixed-precision hardware optimizations, targeting energy-efficient DNN inference on leading RISC-V CPU architectures. We introduce a hardware-software co-design framework that supports cooperative hardware design, mixed-precision quantization, ISA extensions, and cycle-accurate emulations. At the hardware level, we expand the ALU unit in our micro-architecture for configurable mixed-precision arithmetic operations and implement multi-pumping to reduce execution latency, with soft SIMD optimization for 2-bit operations. At the ISA level, we encode three distinct MAC instructions extending the RISC-V ISA, each for different mixed-precision modes, and expose them to the compiler. Our extensive experimental evaluation over widely used DNNs and datasets, such as CIFAR10 and ImageNet, demonstrates that our framework can achieve, on average, 15× energy reduction for less than 1% accuracy loss and outperforms the ISA-agnostic state-of-the-art RISC-V cores.