Abstract The recent emergence of long-read sequencing technologies has enabled substantial improvements in accuracy and reduced computational costs. Nonetheless, pairwise sequence alignment remains a time-consuming step in common bioinformatics pipelines, becoming a bottleneck in de novo whole-genome assembly. Speeding up this step requires heuristics and the development of memory-frugal and efficient implementations. A promising candidate for all of the above is Myers’ algorithm. However, the state-of-the-art implementations face scalability challenges when dealing with longer reads and large datasets. To address these challenges, we propose SeqMatcher, a fast and memory-frugal genomics sequence aligner. By leveraging the long registers of AVX-512, SeqMatcher reduces the data movement and memory footprint. In a comprehensive performance evaluation, SeqMatcher achieves speedups of up to 12.32x for the unbanded version and 26.70x for the banded version compared to the non-vectorized implementation, along with energy footprint reductions of up to 2.59x. It also outperforms state-of-the-art implementations by factors of up to 29.21x, 17.56x, 13.47x, 9.12x, and 8.81x compared to Edlib, WFA2-lib, SeqAn, BSAlign, and QuickEd, while improving energy consumption with reductions of up to 6.78x.

Time series analysis is a key technique for extracting and predicting events in domains as diverse as epidemiology, genomics, neuroscience, environmental sciences, economics, etc. Matrix Profile, a state-of-the-art algorithm to perform time series analysis, finds out the most similar and dissimilar subsequences in a time series in deterministic time and it is exact. Matrix Profile has low arithmetic intensity and it operates on large amounts of time series data, which can be an issue in terms of memory requirements. On the other hand, Hardware Transactional Memory (HTM) is an alternative optimistic synchronization method that executes transactions speculatively in parallel while keeping track of memory accesses to detect and resolve conflicts. This work evaluates one of the best implementations of Matrix Profile exploring multiple multiprocessor variants and proposing new implementations that consider a variety of synchronization methods (HTM, locks, barriers), as well as algorithm organizations. We analyze these variants using real datasets, both short and large, in terms of speedup and memory requirements, the latter being a major issue when dealing with very large time series. The experimental evaluation shows that our proposals can achieve up to 100× speedup over the sequential algorithm for 128 threads, and up to 3× over the baseline, while keeping memory requirements low and even independent of the number of threads.

Abstract Time series analysis is an important research topic and a key step in monitoring and predicting events in many fields. Recently, the Matrix Profile method, and particularly two of its Euclidean-distance-based implementations—SCRIMP and SCAMP—have become the state-of-the-art approaches in this field. Those algorithms bring the possibility of obtaining exact motifs and discords from a time series, which can be used to infer events, predict outcomes, detect anomalies and more. While matrix profile is embarrassingly parallelizable, we find that auto-vectorization techniques fail to fully exploit the SIMD capabilities of modern CPU architectures. In this paper, we develop custom-vectorized SCRIMP and SCAMP implementations based on AVX2 and AVX-512 extensions, which we combine with multithreading techniques aimed at exploiting the potential of the underneath architectures. Our experimental evaluation, conducted using real data, shows a performance improvement of more than 4 $$\times$$ × with respect to the auto-vectorization.

Time series analysis is an important research topic and a key step in monitoring and predicting events in many fields. Recently, the Matrix Profile method, and particularly two of its Euclidean-distance-based implementations – SCRIMP and SCAMP – have become the state-of-the-art approaches in this field. Those algorithms bring the possibility of obtaining exact motifs and discords from a time series, which can be used to infer events, predict outcomes, detect anomalies and more. While matrix profile is embarrassingly parallelizable, we find that autovectorization techniques fail to fully exploit the SIMD capabilities of modern CPU architectures. In this paper, we develop custom-vectorized SCRIMP and SCAMP implementations based on AVX2 and AVX-512 extensions, which we combine with multithreading techniques aimed at exploiting the potential of the underneath architectures. Our experimental evaluation, conducted using real data, shows a performance improvement of more than 4× with respect to the autovectorization.

Transactional Memory (TM) is a synchronization model for parallel programming which provides optimistic concurrency control. Transactions can run in parallel and are only serialized in case of conflict. In this article we use hardware TM (HTM) to implement an optimisticspeculative barrier(SB) to replace the lock-based solution. SBs leverage HTM support to elide barriers speculatively. When a thread reaches an SB, a newSB transactionis started, keeping the updates private to the thread, and letting the HTM system detect potential conflicts. Once the last thread reaches the corresponding SB, the speculative threads can commit their changes. The main contributions of this work are: an API for SBs implemented with HTM extensions; a procedure to check the speculation state in between barriers to enable SBs with non-transactional codes; a HTM SB-aware conflict resolution enhancement where SB transactions stall on a conflict with a standard transaction; and a set of SB use guidelines derived from our experience on using SBs in a variety of applications. We evaluated our proposals in two different architectures with a full-system simulator and an IBM Power8 server. Results show an overall performance improvement of SBs over traditional barriers.

Time series analysis is a key technique for extracting and predicting events in domains as diverse as epidemiology, genomics, neuroscience, environmental sciences, economics, and more. Matrix profile, the state-of-the-art algorithm to perform time series analysis, computes the most similar subsequence for a given query subsequence within a sliced time series. Matrix profile has low arithmetic intensity, but it typically operates on large amounts of time series data. In current computing systems, this data needs to be moved between the off-chip memory units and the on-chip computation units for performing matrix profile. This causes a major performance bottleneck as data movement is extremely costly in terms of both execution time and energy. In this work, we present NATSA, the first Near-Data Processing accelerator for time series analysis. The key idea is to exploit modern 3D-stacked High Bandwidth Memory (HBM) to enable efficient and fast specialized matrix profile computation near memory, where time series data resides. NATSA provides three key benefits: 1) quickly computing the matrix profile for a wide range of applications by building specialized energy-efficient floating-point arithmetic processing units close to HBM, 2) improving the energy efficiency and execution time by reducing the need for data movement over slow and energy-hungry buses between the computation units and the memory units, and 3) analyzing time series data at scale by exploiting low-latency, high-bandwidth, and energy-efficient memory access provided by HBM. Our experimental evaluation shows that NATSA improves performance by up to 14.2× (9.9× on average) and reduces energy by up to 27.2 × (19.4 × on average), over the state-of-the-art multi-core implementation. NATSA also improves performance by 6.3 × and reduces energy by 10.2 × over a general-purpose NDP platform with 64 in-order cores.

Time series analysis is a key step in monitoring and predicting events over time in domains such as epidemiology, genomics, medicine, seismology, speech recognition, and economics. Matrix Profile has been recently proposed as a promising technique to perform time series analysis. For each subsequence, the matrix profile provides the most similar neighbour in the time series. This computation requires a huge amount of floating-point (FP) operations, which are a major contributor (approximately 50%) to the energy consumption in modern computing platforms. Transprecision Computing has recently emerged as a promising approach to improve energy efficiency and performance by tolerating some loss of precision in FP operations. In this work, we study how the matrix profile parallel algorithms benefit from transprecision computing using a recently proposed transprecision FPU. This FPU is intended to be integrated on embedded devices as part of RISC-V processors, FPGAs or ASICs to perform energy-efficient time series analysis. To this end, we propose an accuracy metric to compare the results with the double precision matrix profile. We use this metric to explore a wide range of exponent and mantissa combinations for a variety of datasets, as well as a mixed precision and a vectorized approach. Our analysis reveals that the energy consumption is reduced up to 3.3x compared with double precision approaches, while only slightly affecting the accuracy.

IBM and Intel now offer commercial systems with Transactional Memory (TM), a programming paradigm whose aim is to facilitate concurrent programming while maximizing parallelism. These TM systems are implemented in hardware and provide a software fallback path to overcome the hardware implementation limitations. They are known as best-effort hardware TM (BE-HTM) systems. The software fallback path must be provided by the user to ensure forward progress, which adds programming complexity to the TM paradigm. We propose a new type of irrevocability (a transactional mode that marks transactions as non-abortable) to deal with BE-HTM limitations in a more efficient manner, and to liberate the user from having to program a fallback path. It is based on the concept of lazy subscription used in the context of software fallback paths, where the fallback lock is checked at the end of the transaction instead of at the beginning. We propose a hardware lazy irrevocability mechanism that does not involve changes in the coherence protocol. It solves the unsafe execution problem of premature commits associated with lazy subscription fallbacks, and can be triggered by the user via an ISA extension, for the sake of versatility. It is compared with its software counterpart, which we propose as an enhanced lazy single global lock with escaped spinning at the end of the transaction. We also propose the lazy irrevocability with anticipation, a mechanism that cannot be implemented in software, which significantly improves the performance of codes with multiple cache evictions of transactional data. The evaluation of the proposals is carried out with the Simics/GEMS simulator along with the STAMP benchmark suite, and we obtain speedups from 14 to 28 percent over the fallback path approaches.

Signatures have been proposed in transactional memory systems to represent read and write sets and to decouple transaction conflict detection from private caches or to accelerate it. Generally, signatures are implemented as Bloom filters that allow unbounded read/write sets to be summarized in bounded space at the cost of false conflict detection. It is known that this behavior has great impact in parallel performance. In this work, a scalability study of state-of-the-art signature designs is presented, for different orthogonal transactional characteristics, including contention, length, concurrency and spatial locality. This study was accomplished using the Stanford EigenBench benchmark. This benchmark was modified to support spatial locality analysis using a Zipf address distribution. Experimental evaluation on a hardware transactional memory simulator shows the impact of those parameters in the behavior of state-of-the-art signatures.

Signatures have been proposed in Hardware Transactional Memory (HTM) to represent read and write sets of transactions and decouple transaction conflict detection from private caches. Generally, signatures are implemented as Bloom filters that allow unbounded read/write sets to be summarized in bounded hardware, at the cost of address aliasing that causes false conflict detection. Such conflicts rises exponentially as signature fills so they can lead a parallel program to perform worse than its sequential counterpart (we say that signature saturates). In this work, irrevocability is proposed to address the signature saturation problem. When a transaction is near to saturate its signature, the transaction enters an irrevocable state that prevents it from being aborted. Then, such a transaction keeps running while the others are either stalled or allowed to run concurrently. Two variants of irrevocability are analyzed in this paper. Experimental evaluation on an HTM simulator shows the benefits in performance and power consumption of the proposed irrevocability mechanisms.

Reductions are common operations in many real-world applications that may be responsible for a significant part of the computing time. Modern compilers implement parallel reductions by combining privatization, atomic operations and/or locks. In this paper we analyze how to address reductions in the transactional memory (TM) model, which is flourishing together with the modern shared-memory multicore-based parallel architectures. With this purpose, this paper studies which support needs to be added to a TM system to deal with reductions as a special case of conflicting memory accesses.

Transactional Memory (TM) is an alternative to conventional multithreaded programming to ease the writing of concurrent programs. In the context of unbounded TM, concurrent threads may use hardware signatures to record all the memory addresses issued inside a transaction to detect conflicts. Signatures are usually implemented as per-thread fixed hardware Bloom filters that summarize a very large amount of read and write memory addresses at the cost of false conflicts (detection of nonexisting conflicts). In this paper, to reduce the probability of false conflicts, a novel signature design that exploits spatial locality is proposed. The design is based on new hash function mappings, so that nearby located addresses share some bits inserted in the filters. This is favorable particularly for large transactions that usually exhibit some amount of spatial locality. Besides, its implementation does not require extra hardware. The proposed signature was experimentally evaluated using the GEMS simulator and all the codes of the STAMP benchmark suite. In most cases, the results show significant improvement, particularly in the codes that involve long-running, large-data transactions.

Transactional Memory (TM) systems must track memory accesses made by concurrent transactions in order to detect conflicts. Many TM implementations use signatures for this purpose, which summarize reads and writes in fixed-size bit registers at the cost of false positives (detection of nonexisting conflicts). Signatures are commonly implemented as two separate same-sized Bloom filters, one for reads and other for writes. In contrast, transactions frequently exhibit read and write sets of uneven cardinality. This mismatch between data sets and filter storage introduces inefficiencies in the use of signatures that have some impact on performance. This paper presents different signature designs as alternatives to the common scheme to deal with the asymmetry in transactional data sets in an effective way. Basically, we analyze two classes of new signatures, called multiset and reconfigurable asymmetric signatures. The first class uses only one Bloom filter to track both read and write sets, while the second class uses Bloom filters of configurable size for reads and writes. The main focus of this paper is a thorough study of these alternative signature designs, including a statistical analysis of false positives and an experimental evaluation, providing performance results and hardware area, time and energy requirements.

Transactional Memory (TM) systems must record the memory locations read and written (read and write sets) by concurrent transactions in order to detect conflicts. Some TM implementations use signatures for this purpose, which summarize read and write sets in bounded hardware at the cost of false positives (detection of non-existing conflicts).

Writing multithreaded programs is a fairly complex task that poses a major obstacle to exploit multicore processors. Transactional Memory (TM) emerges as an alternative to the conventional multithreaded programming to ease the writing of concurrent programs. Hardware Transactional Memory (HTM) implements most of the required mechanisms of TM at the core level, e.g. conflict detection. Signatures are designed to support the detection of conflicts amongst concurrent transactions, and are usually implemented as per-thread Bloom filters in HTM. Basically, signatures use fixed hardware to summarize an unbounded amount of read and write memory addresses at the cost of false conflicts (detection of non-existing conflicts). In this paper, a novel signature design that exploit locality is proposed to reduce the number of false conflicts. We show how that reduction translates into a performance improvement in the execution of concurrent transactions. Our signatures are based on address mappings of the hash functions that reduce the number of bits inserted in the filter for those addresses nearby located. This is specially favorable for large transactions, that usually exhibit some amount of spatial locality. Furthermore, the implementation do not require extra hardware. Our proposal was experimentally evaluated using the Wisconsin GEMS simulator and all codes from the STAMP benchmark suite. Results show a significant performance improvement in many cases, specially for those codes with long-running, large-data transactions.