Sharif Digital Repository

In this project, a method is emoloyed to implement a Viterbi decoder on GPGPU. This method is based on combining all steps of the algorithm. This combination has some challenges that are related to differences between different steps of the algorithm. So in this project, some solutions are found to handle these challenges and a high-throughput Viterbi decoder is acheived  

As technology shrinks, the static power consumption is getting worse. Moreover, considering high usage of General-Purpose Graphics Processing Units (GPGPU), reducing the static power of GPGPUs is becoming an important issue. Execution units in GPGPUs are one of the most power hungry units that play an essential role in total power consumption of GPGPUs. On the other hand power gating is a promising method to reduce static power consumption. In this project, we propose a novel method to implement power-gating method for execution units with the negligible performance and power overheads. We utilize out of order execution in intra warp to keep the power-gated resources in off state more than... 

In recent years, GPUs are becoming ideal candidates for processing a variety of high performance applications. By relying on thousands of concurrent threads in applications and the computational power of large numbers of computing units, GPGPUs have provided high efficiency and throughput. To achieve the potential computational power of GPGPUs in broader types of applications, we need to apply some modifications. By understanding the features and properties of applications, we can execute them in a more proper way on GPUs. Therefore, considering applications’ behavior, we define 5 different categories for them. Every category has special definitions, and we change the configuration of GPU... 

The objective of OCR systems is to retrieve machine-encoded text from a raster image. Despite the abundance of powerful OCR algorithms for English, there are not many for Farsi. Our proposed algorithm is comprised of pre-processing, line detection, sub-word detection and segmentation, feature extraction and classification. Furthermore, hardware implementation and acceleration of this system on a GPGPU is presented. This algorithm was tested on 5 fonts including Titr, Lotus,Yekan, Koodak and Nazanin and an average accuracy above 90% was achieved  

The importance of shared memory (scratchpad memory) in GPGPU programming, the memory size limits of GPGPUs and the influence of shared memory on overall performance of the GPGPU has led to its performance optimization. Moreover, the trend of new GPGPUs design shows that the ratio of shared memory to processing elements is going smaller. As a result, the limited capacity of shared memory becomes a bottleneck for a GPU to host a high number of thread blocks, limiting the otherwise available thread-level parallelism (TLP). In this thesis we introduced a reconfigurable and adaptive shared memory architecture for GPGPUs based on resource sharing which can be exploited for throughput improvement... 

The number of transistors on a single chip is growing exponentially which results in a huge increase in consumed power and temperature. Parallel processing is a solution which concentrates on increasing the number of cores instead of improving single thread performance. Graphics Processing Units (GPUs) are parallel accelerators which are categorized as manycore systems. However, recent research shows that their consumed power and energy are increasing. In this research, we aim to propose methods to make GPGPUs energy effiecient. In this regard, we evaluated the detailed consummed power of GPGPUs. Our results show that memory sub-system is a critical bottelneck in terms of performance and... 

Access to global memory is one of the bottlenecks in performance and energy in GPUs. Graphical processors uses multi-thread in streaming multiprocessors to reduce memory access latency. However, due to the high number of concurrent memory requests, the memory bandwidth of low level memorties and the interconnection network are quickly saturated. Recent research suggests that adjacent thread blocks share a significant amount of data blocks. If the adjacent thread blocks are assigned to specific streaming multiprocessor, shared data block can be rused by these thread blocks. However the thread block scheduler assigns adjacent thread blocks to different streaming multiprocessors that increase... 

Graphics Processing Units (GPUs) employ a scratch-pad memory (a.k.a., shared memory) in each streaming multiprocessor to accelerate data sharing among the threads in a thread block and provide a software-managed cache for the programmers.However, we observe that about 60% of GPU workloads of several well-known benchmark suites do not use shared memory. Morever, among those workloads that use shared memory, about 42% of shared memory is not utilized, on average. On the other hand, we observe that many general purpose GPU applications suffer from the low hit rate and limited bandwidth of L1 data cache.We aim to use shared memory space and its corrsponding bandwidth for improving L1 data cache,... 

GPUs employ large register files to reduce the performance and energy overhead of memory accesses through improving the thread-level parallelism, and reducing the number of data movements from the off-chip memory. Recently, latency-tolerant register file (LTRF) is proposed to enable high-capacity register files with low power and area cost. LTRF is a two-level register file in which the first level is a small fast register cache, and the second level is a large slow main register file. LTRF uses a near-perfect register prefetching mechanism that warp registers are prefetched from the main register file to the register file cache before scheduling the warp, and hiding the register prefetching... 

Given a social network modeled as a weighted graph G, the influence maximization problem seeks k vertices to become initially influenced, to maximize the expected number of influenced nodes under a particular diffusion model. The influence maximization problem has been proven to be NP-hard, and most proposed solutions to the problem are approximate greedy algorithms, which can guarantee a tunable approximation ratio for their results with respect to the optimal solution. The state-of-the-art algorithms are based on Reverse Influence Sampling (RIS) technique, which can offer both computational efficiency and non-trivial (1-1/e-ϵ)-approximation ratio guarantee for any ϵ>0. RIS-based... 

Due to the high processing capacity of GPGPUs and their requirement to a large and high speed shared memory between thread processors clusters, exploiting Spin-Transfer Torque (STT) RAM as a replacement with SRAM can result in significant reduction in power consumption and linear enhancement of memory capacity in GPGPUs. In the GPGPU (as a many-core) with ability of parallel thread executing, advantages of STT-RAM technology, such as low read latency and high density, could be so effective. However, the usage of STT-RAM will be grantee applications run time reduction and growth threads throughput, when write operations manages and schedules to have least overhead on read operations. The...