Extending OpenMP* with Vector Constructs for Modern Multicore SIMD Architectures

Klemm, Michael; Durán, Alejandro; Tian, Xin; Saito, Hideki; Caballero, Diego; Martorell, Xavier

doi:10.1007/978-3-642-30961-8_5

Cited by 23 publications

(17 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…List 1 is the original serial code. vector float c = vec vsx ld(0, ba+i); 10 vector float out = c1 * a + c2 * b + c3 * c ; 11 vec vsx st(out, 0, gray+i);…”

Section: Motivationmentioning

confidence: 99%

“…For example, in order to multiply a vector with a scalar, the programmer has to construct a vector from the scalar first (Line 2,3,4 in Listing 2 and Listing 3). To access and operate on SIMD data, the programmer has to explicitly call the load and store intrinsics (Line 7,8,9,12 in Listing 2 and Line 7, 8,9,11 in Listing 3) and other intrinsics for basic arithmetic operations over SIMD vectors (Line 10,11 in Listing 2). All of these combined result in a SIMD code that is much more complex and more difficult to maintain than the original scalar code.…”

Section: Motivationmentioning

confidence: 99%

“…This programming model includes OpenACC [18], OpenCL [10], OpenMP [5,11,16], ISPC [19], and CUDA. Under the SPMD-on-SIMD model, programmers specify the parallelism and the compiler takes care of SIMD code generation to achieve portability and a reasonable performance.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Simple, portable and fast SIMD intrinsic programming

Wang

Tănase

et al. 2014

Proceedings of the 2014 Workshop on Programming Models for SIMD/Vector Processing

View full text Add to dashboard Cite

Using SIMD (Single Instruction Multiple Data) is a costeffective way to explore data parallelism on modern processors. Most processor vendors today provide SIMD engines, such as Altivec/VSX for POWER, SSE/AVX for Intel processors, and NEON for ARM. While high-level SIMD programming models are rapidly evolving, for many SIMD developers, the most effective way to get the performance out of SIMD is still by programming directly via vendorprovided SIMD intrinsics. However, intrinsics programming is both tedious and error-prone, and worst of all, introduces non-portable codes.This paper presents the Generic SIMD Library (https://github.com/genericsimd/generic simd/), an open-source, portable C++ interface that provides an abstraction of short vectors and overloads most C/C++ operators for short vectors. The library provides several mappings from platform-specific intrinsics to the generic SIMD intrinsic interface so that codes developed based on the library are portable across different SIMD platforms.We have evaluated the library with several applications from the multimedia, data analytics and math domains. Compared with platform-specific intrinsics codes, using Generic SIMD Library results in less line-of-code, a 22% reduction on average, and achieves similar performance as platform-specific intrinsics versions.

show abstract

“…List 1 is the original serial code. vector float c = vec vsx ld(0, ba+i); 10 vector float out = c1 * a + c2 * b + c3 * c ; 11 vec vsx st(out, 0, gray+i);…”

Section: Motivationmentioning

confidence: 99%

Section: Motivationmentioning

confidence: 99%

See 1 more Smart Citation

Simple, portable and fast SIMD intrinsic programming

Wang

Tănase

et al. 2014

Proceedings of the 2014 Workshop on Programming Models for SIMD/Vector Processing

View full text Add to dashboard Cite

show abstract

“…This directive instructs the compiler to vectorize that loop, relaxing constraints that could prevent vectorization (pointer aliasing, data dependencies, cost models, etc.) [18,28]. Figure 1b shows the code resulting from the outer loop vectorization [27].…”

Section: Overviewmentioning

confidence: 99%

“…In an attempt to reduce these issues and ease SIMD exploitation, programming languages and programming models [18,11,25,23] are introducing SIMD support. In particular, the OpenMP* parallel programming model added SIMD extensions to its release 4.0 [28].…”

Section: Introductionmentioning

confidence: 99%

Optimizing Overlapped Memory Accesses in User-directed Vectorization

Caballero

Royuela

Ferrer

et al. 2015

Proceedings of the 29th ACM on International Conference on Supercomputing

Self Cite

View full text Add to dashboard Cite

Current processors incorporate wide and powerful vector units whose optimal exploitation is crucial to reach peak performance. However, present autovectorizing compilers fall short of that goal. Exploiting some vector instructions requires aggressive approaches that are not affordable in production compilers. Thus, advanced programmers pursuing the best performance from their applications are compelled to manually vectorize them using low-level SIMD intrinsics.We propose a user-directed code optimization that targets overlapped vector loads, i.e., vector loads that read scalar elements redundantly from memory. Instead, our optimization loads these elements once and combines them using advanced register-to-register vector instructions. This code is potentially more efficient and it uses advanced vector instructions that compilers do not widely exploit automatically. We also extend the OpenMP* SIMD directives with a new clause called overlap that allows users to easily enable and tune this optimization on demand. We implement our proposal for the Intel R Xeon Phi TM coprocessor.Our evaluation shows up to 29% speed-up over five highlyoptimized stencil kernels and workloads from real-world applications. Results also demonstrate how important user hints are to maximize performance.

show abstract

OpenMP in VASP: Threading and SIMD

Wende

Marsman

Kim

et al. 2018

Int J of Quantum Chemistry

View full text Add to dashboard Cite

The Vienna Ab initio Simulation Package (VASP) is a widely used electronic structure code that originally exploits process‐level parallelism through the Message Passing Interface (MPI) for work distribution within and across nodes. Architectural changes of modern parallel processors urge programmers to address thread‐ and data‐level parallelism as well to benefit most from the available compute resources within a node. We describe for VASP how to approach for an MPI + OpenMP parallelization including data‐level parallelism through OpenMP SIMD constructs together with a generic high‐level vector coding scheme. We can demonstrate an improved scalability of VASP and more than 20% gain over the MPI‐only version as well as a 2× increased performance of collective operations using the multiple‐endpoint MPI feature. The high‐level vector coding scheme applied to VASP's general gradient approximation routine gives up 9× performance gain on AVX512 platforms with the Intel compiler.

show abstract

Extending OpenMP* with Vector Constructs for Modern Multicore SIMD Architectures

Cited by 23 publications

References 13 publications

Simple, portable and fast SIMD intrinsic programming

Simple, portable and fast SIMD intrinsic programming

Optimizing Overlapped Memory Accesses in User-directed Vectorization

OpenMP in VASP: Threading and SIMD

Contact Info

Product

Resources

About