RT Journal Article
T1 Micro-kernels for portable and efficient matrix multiplication in deep learning
A1 Alaejos, Guillermo
A1 Castelló, Adrián
A1 Martínez, Héctor
A1 Alonso-Jordá, Pedro
A1 Quintana-Ortí, Enrique S.
A1 Igual Peña, Francisco Daniel
AB We provide a practical demonstration that it is possible to systematically generate a variety of high-performance micro-kernels for the general matrix multiplication (gemm) via generic templates which can be easily customized to different processor architectures and micro-kernel dimensions. These generic templates employ vector intrinsics to exploit the SIMD (single instruction, multiple data) units in current general-purpose processors and, for the particular type of gemm problems encountered in deep learning, deliver a floating-point throughput rate on par with or even higher than that obtained with conventional, carefully tuned implementations of gemm in current linear algebra libraries (e.g., BLIS, AMD AOCL, ARMPL). Our work exposes the structure of the template-based micro-kernels for ARM Neon (128-bit SIMD), ARM SVE (variable-length SIMD) and Intel AVX512 (512-bit SIMD), showing considerable performance for an NVIDIA Carmel processor (ARM Neon), a Fujitsu A64FX processor (ARM SVE) and on an AMD EPYC 7282 processor (256-bit SIMD).
PB Springer
YR 2022
FD 2022-12-14
LK https://hdl.handle.net/20.500.14352/115347
UL https://hdl.handle.net/20.500.14352/115347
LA eng
NO Alaejos, G., Castelló, A., Martínez, H. et al. Micro-kernels for portable and efficient matrix multiplication in deep learning. J Supercomput 79, 8124–8147 (2023). https://doi.org/10.1007/s11227-022-05003-3
DS Docta Complutense
RD 18 abr 2025