The transformer’s attention mechanism has revolutionized AI and machine learning, with its efficient computation being crucial to its performance. However, calculating attention involves matrix operations interspersed with softmax rescaling, which inherently slows down computation and requires processing the entire input sequence. Building on online softmax computation, FlashAttention integrates softmax calculation with matrix arithmetic, enabling tiled computation independent of sequence length. While optimized for GPUs, FlashAttention’s simplicity makes it amenable to direct hardware acceleration. This work re-evaluates the core FlashAttention kernel, presenting FLASH-D a mathematically equivalent, yet simplified, formulation that achieves: (a) hiding softmax division within other nonlinear function evaluations; (b) inherently numerically stable computation of exponentials, eliminating the need for maximum value subtraction; and (c) a reduction in computational cost without introducing numerical approximations to the FlashAttention kernel. Importantly, the essential FlashAttention properties that facilitate efficient tiled implementation are fully preserved. Hardware implementation results at 28nm demonstrate that this proposed formulation achieves a 22.8% reduction in area and a 20.3% reduction in power, on average, compared to state-of-the-art parallel hardware architectures without any performance penalty.

Structured sparsity has been proposed as an efficient way to prune the complexity of Machine Learning (ML) applications and to simplify the handling of sparse data in hardware. Accelerating ML models, whether for training, or inference, heavily relies on matrix multiplications that can be efficiently executed on vector processors, or custom matrix engines. This work aims to integrate the simplicity of structured sparsity into vector execution to speed up the corresponding matrix multiplications. Initially, the implementation of structured-sparse matrix multiplication using the current RISC-V instruction set vector extension is comprehensively explored. Critical parameters that affect performance, such as the impact of data distribution across the scalar and vector register files, data locality, and the effectiveness of loop unrolling are analyzed both qualitatively and quantitatively. Furthermore, it is demonstrated that the addition of a single new instruction would reap even higher performance. The newly proposed instruction is calledvindexmac, i.e., vector index-multiply-accumulate. It allows for indirect reads from the vector register file and it reduces the number of instructions executed per matrix multiplication iteration, without introducing additional dependencies that would limit loop unrolling. The proposed new instruction was integrated in a decoupled RISC-V vector processor with negligible hardware cost. Experimental results demonstrate the runtime efficiency and the scalability offered by the introduced optimizations and the new instruction for the execution of state-of-the-art Convolutional Neural Networks. More particularly, the addition of a custom instruction improves runtime by 25% and 33%, when compared with highly-optimized vectorized kernels that use only the currently defined RISC-V instructions.

Multiterm floating-point (FP) addition appears in vector dot-product computations, matrix multiplications, and other forms of FP data aggregation. A critical step in multiterm floating-point addition is the alignment of fractions of the FP terms before adding them. Alignment is executed serially by identifying first the maximum of all exponents and then shifting the fraction of each term according to the difference of its exponent from the maximum one. Contrary to common practice, this work proposes a new online algorithm that splits the identification of the maximum exponent, the alignment shift for each fraction, and their addition to multiple fused incremental steps that can be computed in parallel. Each fused step is implemented by a new associative operator that allows the incremental alignment and addition for arbitrary number of operands. Experimental results show that employing the proposed align-and-add operators for the implementation of multiterm floating-point adders can improve delay or save significant area and power. The achieved area and power savings range between 3% and 23% and between 4% and 26%, respectively.

Structured sparsity has been proposed as an efficient way to prune the complexity of modern Machine Learning (ML) applications and to simplify the handling of sparse data in hardware. The acceleration of ML models - for both training and inference - relies primarily on equivalent matrix multiplications that can be executed efficiently on vector processors or custom matrix engines. The goal of this work is to incorporate the simplicity of structured sparsity into vector execution, thereby accelerating the corresponding matrix multiplications. Toward this objective, a new vector index-multiply-accumulate instruction is proposed, which enables the implementation of low-cost indirect reads from the vector register file. This reduces unnecessary memory traffic and increases data locality. The proposed new instruction was integrated in a decoupled RISC-V vector processor with negligible hardware cost. Extensive evaluation demonstrates significant speedups of$1.80\times-2.14\times$, as compared to state-of-the-art vectorized kernels, when executing layers of varying sparsity from state-of-the-art Convolutional Neural Networks (CNNs).