The Cambrian Explosion of Mixed-Precision Matrix Multiplication for Quantized Deep Learning Inference

[EN] Recent advances in deep learning (DL) have promoted to a shift from traditional 64-bit floating point (FP64) arithmetic for scientific computing toward reduced-precision formats-such as FP16, BF16, or even 8-bit integers-combined with mixed-precision arithmetic. This transition enhances computa...

Descripción completa

Detalles Bibliográficos
Autores: Martínez, Héctor, Igual, Francisco D., Castelló, Adrián|||0000-0002-8576-8451, Quintana-Ortí, Enrique S.|||0000-0002-5454-165X
Tipo de recurso: artículo
Fecha de publicación:2026
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:riunet.upv.es:10251/232328
Acceso en línea:https://riunet.upv.es/handle/10251/232328
Access Level:acceso abierto
Palabra clave:Matrix multiplication
Mixed-precision arithmetic
SIMD
Matrix engines
Deep learning inference
Descripción
Sumario:[EN] Recent advances in deep learning (DL) have promoted to a shift from traditional 64-bit floating point (FP64) arithmetic for scientific computing toward reduced-precision formats-such as FP16, BF16, or even 8-bit integers-combined with mixed-precision arithmetic. This transition enhances computational throughput, reduces memory and bandwidth usage, and improves energy efficiency, offering significant advantages for resource-constrained edge devices. To support this shift, hardware architectures have evolved accordingly, now including adapted ISAs (Instruction Set Architectures) that expose mixed-precision vector units and matrix engines tailored for DL workloads. At the heart of many DL and scientific computing tasks is the general matrix-matrix multiplication (GEMM), a fundamental kernel historically optimized using fused multiply-add (FMA) vector instructions on SIMD (single instruction, multiple data) units. However, as hardware moves toward mixed-precision dot (or inner)-product-centric operations optimized for quantized inference, these legacy approaches are being phased out. In response to this, our paper revisits the conventional, high-performance implementation of GEMM and describes strategies for adapting it to mixed integer precision (MIP) arithmetic across modern ISAs, including x86_64, Arm, and RISC-V. Concretely, we illustrate novel micro-kernel designs and data layouts that better exploit today's specialized hardware and demonstrate significant performance gains from MIP arithmetic over floatingpoint implementations across three representative CPUs. These contributions highlight a new era of GEMM optimization-driven by the demands of DL inference on heterogeneous architectures, marking what we term as the "Cambrian period" for matrix multiplication.