Towards functional safety compliance of matrix–matrix multiplication for machine learning-based autonomous systems

Fernández, Javier; Pérez, Jon; Agirre, Irune; Allende, Imanol; Abella, Jaume; Cazorla, Francisco J.

doi:10.1016/j.sysarc.2021.102298

Cited by 8 publications

(10 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…return acc; 9: end function For the Fletcher, we employ the implementations provided by reference [14]. Finally, we resort to lookup tables for implementing the CRC.…”

Section: Diagnostic Techniquesmentioning

confidence: 99%

“…We launch a single matrix-matrix multiplication stream to the GPU to avoid the uncertainty in the order of execution of the streams associated with several applications running simultaneously [27]. We employ the same matrix dimensions and nomenclature as in reference [14], seeking a fair comparison. In reference [14], the authors also include a catalog of diagnostic techniques, but based on a sequential and an AVX-based implementation.…”

Section: Experimental Set-upmentioning

confidence: 99%

“…We employ the same matrix dimensions and nomenclature as in reference [14], seeking a fair comparison. In reference [14], the authors also include a catalog of diagnostic techniques, but based on a sequential and an AVX-based implementation. Table 2 shows the matrixmatrix multiplication dimensions, dividing them into two groups:…”

Section: Experimental Set-upmentioning

confidence: 99%

“…In DC experiments, we perform a bit-exhaustive fault-injection campaign in both A and B matrices (as in reference [14]). This evaluation requires as many matrix-matrix multiplication executions as bits positions have those matrices.…”

Section: Experimental Set-upmentioning

confidence: 99%

“…We analyze the performance impact and the diagnostic coverage (DC) of this catalog of mechanisms for multiple matrix dimensions in a NVIDIA Jetson Xavier NX GPU and compare our results against sequential and advanced vector extensions (AVX)-based implementations [14]; •…”

mentioning

confidence: 99%

See 4 more Smart Citations

On the Safe Deployment of Matrix Multiplication in Massively Parallel Safety-Related Systems

et al. 2022

Self Cite

View full text Add to dashboard Cite

Deep learning technology has enabled the development of increasingly complex safety-related autonomous systems using high-performance computers, such as GPU, which provide the required high computing performance for the execution of parallel computing algorithms, such as matrix–matrix multiplications (a central computing element of deep learning software libraries). However, the safety certification of parallel computing software algorithms and GPU-based safety-related systems is a challenge to be addressed. For example, achieving the required fault-tolerance and diagnostic coverage for random hardware errors. This paper contributes with a safe matrix–matrix multiplication software implementation for GPUs with random hardware error-detection capabilities (permanent, transient) that can be used with different architectural patterns for fault-tolerance, and which serves as a foundation for the implementation of safe deep learning libraries for GPUs. The proposed contribution is complementary and can be combined with other techniques, such as algorithm-based fault tolerance. In particular, (i) we provide the high-performance matrix multiplication CUTLASS library with a catalog of diagnostic mechanisms to detect random hardware errors down to the arithmetic operation level; and (ii) we measure the performance impact incurred by the adoption of these mechanisms and their achievable diagnostic coverage with a set of representative matrix dimensions. To that end, we implement these algebraic operations, targeting CUDA cores with single instructions and multiple-thread math instructions in an NVIDIA Xavier NX GPU.

show abstract

“…return acc; 9: end function For the Fletcher, we employ the implementations provided by reference [14]. Finally, we resort to lookup tables for implementing the CRC.…”

Section: Diagnostic Techniquesmentioning

confidence: 99%