Towards Cross-Platform Performance Portability of DNN Models using SYCL

Goli, Mehdi; Narasimhan, Kumudha; Reyes, Ruyman; Tracy, Ben; Soutar, Daniel; Georgiev, Svetlozar; Fomenko, Evarist; Chereshnev, Eugene

doi:10.1109/p3hpc51967.2020.00008

Cited by 16 publications

(5 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, the overhead of the SYCL runtime, and especially kernel dispatch, affects the overall performance our library. This observation is in direct agreement with previous work [8]; although the runtime overheads are significant for small problem sizes, larger problems are impacted less as the gap between runtime and kernel execution increases.…”

Section: Discussionsupporting

confidence: 92%

“…2b shows that optimal SYCL-FFT run-times (chosen as the smallest of the 1000 test runs) do not differ largely from cuFFT or rocFFT. Similar behavior has been observed in previous studies [8], where the overhead of SYCL runtime interacting with CUDA was constant 30 𝜇s. Sequence Length A similar trend in launch overheads is observed for the Intel Iris P580 integrated GPU (iGPU) and ARM Neoverse CPU, as seen in Fig.…”

Section: Computational Performancesupporting

confidence: 88%

See 1 more Smart Citation

Benchmarking a Proof-of-Concept Performance Portable SYCL-based Fast Fourier Transformation Library

Pascuzzi,

Goli

2022

Preprint

Self Cite

View full text Add to dashboard Cite

In this paper, we present an early version of a SYCL-based FFT library, capable of running on all major vendor hardware, including CPUs and GPUs from AMD, ARM, Intel and NVIDIA. Although preliminary, the aim of this work is to seed further developments for a rich set of features for calculating FFTs. It has the advantage over existing portable FFT libraries in that it is single-source, and therefore removes the complexities that arise due to abundant use of pre-process macros and auto-generated kernels to target different architectures. We exercise two SYCL-enabled compilers, Codeplay ComputeCpp and Intel's open-source LLVM project, to evaluate performance portability of our SYCL-based FFT on various heterogeneous architectures. The current limitations of our library is it supports single-dimension FFTs up to 2 11 in length and base-2 input sequences. We compare our results with highly optimized vendor specific FFT libraries and provide a detailed analysis to demonstrate a fair level of performance, as well as potential sources of performance bottlenecks. CCS Concepts: • Computing methodologies → Distributed programming languages; • Mathematics of computing → Mathematical software performance.

show abstract

Section: Discussionsupporting

confidence: 92%

Section: Computational Performancesupporting

confidence: 88%

Benchmarking a Proof-of-Concept Performance Portable SYCL-based Fast Fourier Transformation Library

Pascuzzi,

Goli

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Interoperability is enabled via the aforementioned low-level APIs by facilitating the SYCL runtime interaction with native objects for the supported backends [13,24].…”

Section: Sycl Overviewmentioning

confidence: 99%

Achieving near native runtime performance and cross-platform performance portability for random number generation through SYCL interoperability

Pascuzzi,

Goli

2021

Preprint

Self Cite

View full text Add to dashboard Cite

High-performance computing (HPC) is a major driver accelerating scientific research and discovery, from quantum simulations to medical therapeutics. The growing number of new HPC systems coming online are being furnished with various hardware components, engineered by competing industry entities, each having their own architectures and platforms to be supported. While the increasing availability of these resources is in many cases pivotal to successful science, even the largest collaborations lack the computational expertise required for maximal exploitation of current hardware capabilities. The need to maintain multiple platform-specific codebases further complicates matters, potentially adding a constraint on the number of machines that can be utilized. Fortunately, numerous programming models are under development that aim to facilitate software solutions for heterogeneous computing. In particular is SYCL, an open standard, C++-based single-source programming paradigm. Among SYCL's features is interoperability, a mechanism through which applications and third-party libraries coordinate sharing data and execute collaboratively. In this paper, we leverage the SYCL programming model to demonstrate cross-platform performance portability across heterogeneous resources. We detail our NVIDIA and AMD random number generator extensions to the oneMKL open-source interfaces library. Performance portability is measured relative to platformspecific baseline applications executed on four major hardware platforms using two different compilers supporting SYCL. The utility of our extensions are exemplified in a real-world setting via a high-energy physics simulation application. We show the performance of implementations that capitalize on SYCL interoperability are at par with native implementations, attesting to the cross-platform performance portability of a SYCL-based approach to scientific codes.

show abstract

“…Moreover, oneAPI is becoming very popular. It has shown promising results in various computer fields, such as machine learning (Goli et al, 2020) or decision-making (Constantinescu et al, 2020). One of the keys to their growing impact on the heterogeneous field is the existence of optimized libraries that can be used together with oneAPI.…”

Section: Introductionmentioning

confidence: 99%

Performance portability in a real world application: PHAST applied to Caffe

Martínez

Peccerillo

Bartolini

et al. 2022

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

This work covers the PHAST Library’s employment, a hardware-agnostic programming library, to a real-world application like the Caffe framework. The original implementation of Caffe consists of two different versions of the source code: one to run on CPU platforms and another one to run on the GPU side. With PHAST, we aim to develop a single-source code implementation capable of running efficiently on CPU and GPU. In this paper, we start by carrying out a basic Caffe implementation performance analysis using PHAST. Then, we detail possible performance upgrades. We find that the overall performance is dominated by few ‘heavy’ layers. In refining the inefficient parts of this version, we find two different approaches: improvements to the Caffe source code and improvements to the PHAST Library itself, which ultimately translates into improved performance in the PHAST version of Caffe. We demonstrate that our PHAST implementation achieves performance portability on CPUs and GPUs. With a single source, the PHAST version of Caffe provides the same or even better performance than the original version of Caffe built from two different codebases. For the MNIST database, the PHAST implementation takes an equivalent amount of time as native code in CPU and GPU. Furthermore, PHAST achieves a speedup of 51% and a 49% with the CIFAR-10 database against native code in CPU and GPU, respectively. These results provide a new horizon for software development in the upcoming heterogeneous computing era.

show abstract

Towards Cross-Platform Performance Portability of DNN Models using SYCL

Cited by 16 publications

References 5 publications

Benchmarking a Proof-of-Concept Performance Portable SYCL-based Fast Fourier Transformation Library

Benchmarking a Proof-of-Concept Performance Portable SYCL-based Fast Fourier Transformation Library

Achieving near native runtime performance and cross-platform performance portability for random number generation through SYCL interoperability

Performance portability in a real world application: PHAST applied to Caffe

Contact Info

Product

Resources

About