Kamil Kedzierski, F. Cazorla, R. Gioiosa, A. Buyuktosunoglu, M. Valero
{"title":"Power and performance aware reconfigurable cache for CMPs","authors":"Kamil Kedzierski, F. Cazorla, R. Gioiosa, A. Buyuktosunoglu, M. Valero","doi":"10.1145/1882453.1882455","DOIUrl":"https://doi.org/10.1145/1882453.1882455","url":null,"abstract":"This paper investigates the problem of partitioning a shared cache among threads executing in a Chip Multi-Processor (CMP). We propose Reconfigurable Cache for CMPs (ReCaC), a low-overhead run-time mechanism that dynamically partitions the cache based on the phase behavior of threads. Unlike the previously proposed performance aware partitioning approaches, ReCaC targets to use a minimum number of ways to trade-off between power and performance. ReCaC dynamically reverts back to a performance centric cache partitioning scheme if the power savings are not achievable.\u0000 Our results show that in a 2-core architecture ReCaC saves on average 51.5% power in a L2 cache, which corresponds to 11.5% power savings of the processor chip and the memory. The overall processor energy efficiency is improved by up to 13.6%, achieving on average 8.5%. ReCaC proves to be scalable, saving on average 10.8% and 12.5% of the processor chip and memory power in 4-and 8-core architectures, respectively.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124351622","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An adaptive cache coherence protocol for chip multiprocessors","authors":"Abdullah Kayi, T. El-Ghazawi","doi":"10.1145/1882453.1882458","DOIUrl":"https://doi.org/10.1145/1882453.1882458","url":null,"abstract":"Multi-core architectures also referred to as Chip Multiprocessors (CMPs) have emerged as the dominant architecture for both desktop and high-performance systems. CMPs introduce many challenges that need to be addressed to achieve the best performance. One of the big challenges comes with the shared-memory model observed in such architectures which is the cache coherence overhead problem. Contemporary architectures employ write-invalidate based protocols which are known to generate coherence misses that yield to latency issues. On the other hand, write-update based protocols can solve the coherence misses problem but they tend to generate excessive network traffic which is especially not desirable for CMPs. Previous studies have shown that a single protocol approach is not sufficient for many sharing patterns. As a solution, this paper evaluates an adaptive protocol which targets write-update optimizations for producer-consumer sharing patterns. This work targets a minimalistic hardware extension approach to test the benefits of such adaptive protocols in a practical environment. Experimental study is conducted on a 16-core CMP by using a full-system simulator with selected scientific applications from SPLASH-2 and NAS parallel benchmark suites. Results show up to 40% improvement for coherence misses which corresponds to 15% application speedup.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"81 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130209790","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"LV*: a class of lazy versioning HTMs for low-cost integration of transactional memory systems","authors":"A. Negi, M. Waliullah, P. Stenström","doi":"10.1145/1882453.1882460","DOIUrl":"https://doi.org/10.1145/1882453.1882460","url":null,"abstract":"Transactional memory (TM) promises to unlock parallelism in software in a safer and easier way than lock-based approaches but the path to deployment is unclear for several reasons. First of all, since TM has not been deployed in any machine yet, experience of using it is limited. While software transactional memory implementations exist, they are too slow to provide useful experience. Existing hardware transactional memory implementations, on the other hand, can provide the efficiency required but they require a significant effort to integrate in cache coherence infrastructures or freeze critical policy parameters.\u0000 This paper proposes the LV* (lazy versioning and eager/lazy conflict resolution) class of hardware transactional memory protocols. This class of protocols has been implemented with ease of deployment in mind. LV* can be integrated with low additional complexity in standard snoopy-cache MESI-protocols and can be accommodated in a directory-based cache coherence infrastructure. Since the optimal conflict resolution policy (lazy or eager) depends on transactional characteristics of workloads, LV* supports a set of conflict resolution policies that range from LazEr -- a family of Lazy versioning Eager conflict resolution protocols -- to LL-MESI which provides lazy resolution. We show that LV* can be hosted in a MESI protocol through straightforward extensions and that the flexibility in the choice of conflict resolution strategy has a significant impact on performance.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126888911","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"SCMP architecture: an asymmetric multiprocessor system-on-chip for dynamic applications","authors":"N. Ventroux, R. David","doi":"10.1145/1882453.1882461","DOIUrl":"https://doi.org/10.1145/1882453.1882461","url":null,"abstract":"Future systems will have to support multiple and concurrent dynamic compute-intensive applications, while respecting real-time and energy consumption constraints. Within this framework, this paper presents an architecture, named SCMP. This asymmetric multiprocessor can support dynamic migration and preemption of tasks, thanks to a concurrent control of tasks, while offering a specific data sharing solution. Its tasks are controlled by a dedicated HW-RTOS that allows online scheduling of independent real-time and non-real-time tasks. By incorporating a connected component labeling algorithm into this platform, we have been able to measure its benefits for real-time and dynamic image processing.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"223 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114402661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Fitting FFT onto an energy efficient massively parallel architecture","authors":"I. Lorentz, M. Malita, Răzvan Andonie","doi":"10.1145/1882453.1882464","DOIUrl":"https://doi.org/10.1145/1882453.1882464","url":null,"abstract":"We present novel implementations of the Fast Fourier Transform on the massively parallel Connex Array#8482;(CA) circuit. The estimated performance is 19 GFlops (BenchFFT metric) of parallel computing 64 FFTs of size 1024, using 5 Watts. We compare the CA and NVIDIA's GTX 285 GPU performance. The CA is not a direct NVIDIA competitor, targeting a different application area. Considering its low power dissipation, the CA is a good solution for low cost mobile computing equipment, video processing, and multi-channel high-sampling audio processing.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124164103","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Accelerating data movement on future chip multi-processors","authors":"Junli Gu, Rakesh Kumar, S. Lumetta, Yihe Sun","doi":"10.1145/1882453.1882457","DOIUrl":"https://doi.org/10.1145/1882453.1882457","url":null,"abstract":"Moving data between cores on hardware coherent architectures suffers from memory latency and causes cache misses and coherence traffic, which are obstacles to achieving high performance. In this paper, we evaluate the potential for hardware optimization of message data transfer on chip multiprocessors with a combination of NAS parallel MPI benchmarks, Intel IMB MPI benchmarks, and a few microbenchmarks on a full-system simulator based on Simics and FeS2. We show that while passive hardware driven by cores can reduce cache traffic, it provides limited performance gains. We propose a data movement manager (DMM) that uses the on-chip coherence protocols to implement zero-copy message passing between separate address spaces and to remove synchronization and copy overheads from the processors. We also discuss methods for managing data placement in caches to reduce latency. We show that such a design shows substantial promise for both cache traffic reduction and performance improvements.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125847389","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Augusto J. Vega, Alejandro Rico, F. Cabarcas, Alex Ramírez, M. Valero
{"title":"Comparing last-level cache designs for CMP architectures","authors":"Augusto J. Vega, Alejandro Rico, F. Cabarcas, Alex Ramírez, M. Valero","doi":"10.1145/1882453.1882456","DOIUrl":"https://doi.org/10.1145/1882453.1882456","url":null,"abstract":"The emergence of hardware accelerators, such as graphics processing units (GPUs), has challenged the interaction between processing elements (PEs) and main memory. In architectures like the Cell/B.E. or GPUs, the PEs incorporate local memories which are fed with data transferred from memory using direct memory accesses (DMAs). We expect that chip multiprocessors (CMP) with DMA-managed local memories will become more popular in the near future due to the increasing interest in accelerators. In this work we show that, in that case, the way cache hierarchies are conceived should be revised.\u0000 Particularly for last-level caches, the norm today is to use latency-aware organizations. For instance, in dynamic nonuniform cache architectures (D-NUCA) data is migrated closer to the requester processor to optimize latency. However, in DMA-based scenarios, the memory system latency becomes irrelevant compared with the time consumed for moving the DMA data, so latency-aware designs are, a priori, inefficient.\u0000 In this work, we revisit the last-level cache designs in DMA-based CMP architectures with master-worker execution. Two scenarios are evaluated. First, we consider a set of private caches with data replication across them, where coherency of the copies is ensured through a hardware protocol. In this scenario, a PE has a nearby copy of the datum, improving cache access latency. Second, we consider a partitioned cache, where the allocation of a datum to a cache block is determined based on its physical address. In this scenario, there are no copies of data, and access to a datum has a variable latency. In contrast with traditional load/store-based architectures, we found that the partitioned last-level cache scheme outperforms the cache with data replication for DMA-based scenarios.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132917440","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Efficient parallel selective separable-kernel convolution on heterogeneous processors","authors":"A. El-Mahdy, H. El-Shishiny","doi":"10.1145/1882453.1882463","DOIUrl":"https://doi.org/10.1145/1882453.1882463","url":null,"abstract":"Separable kernel convolution is a fundamental operation for image processing and computer vision applications. Existing parallel implementations of convolution operation convolve a whole image with a separable kernel. While such algorithms are efficient, not all the computed convolutions are necessarily utilized by the underlying application. In this paper we introduce an integrated parallel selective separable-kernel convolution method with adaptive load balancing, for both homogeneous and heterogeneous multi-core processors. The method allows for only computing convolutions at selected points efficiently, making use of partial convolution results computed previously. The method also automatically load-balances the convolution computation among core/processors with different processing speeds. The load-balancing is a hybrid dynamic/static one that adapts load-balancing for sequences of video frames, allowing for fast convolution operation on video streams. Our method is studied on the computation of the Difference of Gaussian operation. Performance assessment results show a remarkable load-balancing reaching 80% of optimum load balance for both heterogeneous and homogeneous processor systems.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2010-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117305668","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An efficient in-place 3D transpose for multicore processors with software managed memory hierarchy","authors":"A. El-Moursy, A. El-Mahdy, H. El-Shishiny","doi":"10.1145/1463768.1463781","DOIUrl":"https://doi.org/10.1145/1463768.1463781","url":null,"abstract":"3D transpose is an important operation in many large scale scientific applications such as seismic and medical imaging. This paper proposes a novel algorithm for fast in-place 3D transpose operation. The algorithm exploits Single Instruction Multiple Data (SIMD) multicore architecture with software managed memory hierarchy. Such architectural features are present in the next generation processors, such as the Cell Broadband Engine (Cell BE) processor. The algorithm performs transposition at two levels of granularity: at coarse level, where logical transposition is done by merely transposing the address map at each access; and at a fine grain level, where physical transposition is done by actual element displacement/swapping. Such mix combines the benefits of allowing for fast on-chip bandwidth by providing for large transfer sizes, and at the same time allows for fine-grain SIMD operations. The transfer rate is further enhanced by allowing for batch transposing spatially joined data along a major axis. Results on the Cell BE processor show substantial utilisation of on-chip communication bandwidth, and negligible processing time.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124912022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Haitham Akkary, K. Jothi, Renjith Retnamma, Satyanarayana Nekkalapu, Doug Hall, S. Shahidzadeh
{"title":"On the potential of latency tolerant execution in speculative multithreading","authors":"Haitham Akkary, K. Jothi, Renjith Retnamma, Satyanarayana Nekkalapu, Doug Hall, S. Shahidzadeh","doi":"10.1145/1463768.1463772","DOIUrl":"https://doi.org/10.1145/1463768.1463772","url":null,"abstract":"High performance superscalar architectures used to exploit instruction level parallelism in single-thread applications have become too complex and too power hungry for the many-core processors era. We propose a new architecture that uses multiple latency-tolerant in-order cores to improve single-thread performance, without requiring complex out-of-order execution hardware or large, power hungry register files and instruction buffers. Using simple cores to provide improved single-thread performance for conventional difficult-to-parallelize applications allows designers to place many more of these cores on the same die. Consequently, emerging highly parallel applications can take full advantage of the many-core parallel hardware without sacrificing performance of inherently serial applications.\u0000 Our architecture splits single-thread program execution into disjoint control and data threads that execute concurrently on multiple latency-tolerant in-order cores. Hence we call this style of execution Disjoint Out-of-Order Execution (DOE). DOE is a novel implementation of Speculative Multithreading (SpMT). It uses latency tolerance to overcome performance issues of SpMT caused by load imbalance and inter-thread data communication delays.\u0000 Using control independence prediction hardware to spawn threads, we simulate the potential performance of DOE on a subset of Spec2000 integer benchmarks under various parallelism scenarios and for DOE configurations of 2, 4, 6 and 8 single-issue latency tolerant cores.","PeriodicalId":351385,"journal":{"name":"International Forum on Next-Generation Multicore/Manycore Technologies","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125489814","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}