Joseph L. Greathouse

TOP-PIM: throughput-oriented programmable processing in memory

As computation becomes increasingly limited by data movement and energy consumption, exploiting locality throughout the memory hierarchy becomes critical to continued performance scaling. Moving computation closer to memory presents an opportunity to reduce both energy and data movement overheads. We explore the use of 3D die stacking to move memory-intensive computations closer to memory. This approach to processing in memory addresses some drawbacks of prior research on in-memory computing and is commercially viable in the foreseeable future. Because 3D stacking provides increased bandwidth, we study throughput-oriented computing using programmable GPU compute units across a broad range of benchmarks, including graph and HPC applications. We also introduce a methodology for rapid design space exploration by analytically predicting performance and energy of in-memory processors based on metrics obtained from execution on todays GPU hardware. Our results show that, on average, viable PIM configurations show moderate performance losses (27%) in return for significant energy efficiency improvements (76\% reduction in EDP) relative to a representative mainstream GPU at 22nm technology. At 16nm technology, on average, viable PIM configurations are performance competitive with a representative mainstream GPU (7% speedup) and provide even greater energy efficiency improvements (85\% reduction in EDP).

Efficient sparse matrix-vector multiplication on GPUs using the CSR storage format

The performance of sparse matrix vector multiplication (SpMV) is important to computational scientists. Compressed sparse row (CSR) is the most frequently used format to store sparse matrices. However, CSR-based SpMV on graphics processing units (GPUs) has poor performance due to irregular memory access patterns, load imbalance, and reduced parallelism. This has led researchers to propose new storage formats. Unfortunately, dynamically transforming CSR into these formats has significant runtime and storage overheads. We propose a novel algorithm, CSR-Adaptive, which keeps the CSR format intact and maps well to GPUs. Our implementation addresses the aforementioned challenges by (i) efficiently accessing DRAM by streaming data into the local scratchpad memory and (ii) dynamically assigning different numbers of rows to each parallel GPU compute unit. CSR-Adaptive achieves an average speedup of 14.7× over existing CSR-based algorithms and 2.3× over clSpMV cocktail, which uses an assortment of matrix formats.

GPGPU performance and power estimation using machine learning

Graphics Processing Units (GPUs) have numerous configuration and design options, including core frequency, number of parallel compute units (CUs), and available memory bandwidth. At many stages of the design process, it is important to estimate how application performance and power are impacted by these options. This paper describes a GPU performance and power estimation model that uses machine learning techniques on measurements from real GPU hardware. The model is trained on a collection of applications that are run at numerous different hardware configurations. From the measured performance and power data, the model learns how applications scale as the GPUs configuration is changed. Hardware performance counter values are then gathered when running a new application on a single GPU configuration. These dynamic counter values are fed into a neural network that predicts which scaling curve from the training data best represents this kernel. This scaling curve is then used to estimate the performance and power of the new application at different GPU configurations. Over an 8× range of the number of CUs, a 3.3× range of core frequencies, and a 2.9× range of memory bandwidth, our models performance and power estimates are accurate to within 15% and 10% of real hardware, respectively. This is comparable to the accuracy of cycle-level simulators. However, after an initial training phase, our model runs as fast as, or faster than the program running natively on real hardware.

PPEP: Online Performance, Power, and Energy Prediction Framework and DVFS Space Exploration

Performance, power, and energy (PPE) are critical aspects of modern computing. It is challenging to accurately predict, in real time, the effect of dynamic voltage and frequency scaling (DVFS) on PPE across a wide range of voltages and frequencies. This results in the use of reactive, iterative, and inefficient algorithms for dynamically finding good DVFS states. We propose PPEP, an online PPE prediction framework that proactively and rapidly searches the DVFS space. PPEP uses hardware events to implement both a cycles-per-instruction (CPI) model as well as a per-core power model in order to predict PPE across all DVFS states. We verify on modern AMD CPUs that the PPEP power model achieves an average error of 4.6% (2.8% standard deviation) on 152 benchmark combinations at 5 distinct voltage-frequency states. Predicting average chip power across different DVFS states achieves an average error of 4.2% with a 3.6% standard deviation. Further, we demonstrate the usage of PPEP by creating and evaluating a highly responsive power capping mechanism that can meet power targets in a single step. PPEP also provides insights for future development of DVFS technologies. For example, we find that it is important to carefully consider background workloads for DVFS policies and that enabling north bridge DVFS can offer up to 20% additional energy saving or a 1.4x performance improvement.

Adaptive GPU cache bypassing

Modern graphics processing units (GPUs) include hardware- controlled caches to reduce bandwidth requirements and energy consumption. However, current GPU cache hierarchies are inefficient for general purpose GPU (GPGPU) comput- ing. GPGPU workloads tend to include data structures that would not fit in any reasonably sized caches, leading to very low cache hit rates. This problem is exacerbated by the design of current GPUs, which share small caches be- tween many threads. Caching these streaming data struc- tures needlessly burns power while evicting data that may otherwise fit into the cache. We propose a GPU cache management technique to im- prove the efficiency of small GPU caches while further re- ducing their power consumption. It adaptively bypasses the GPU cache for blocks that are unlikely to be referenced again before being evicted. This technique saves energy by avoid- ing needless insertions and evictions while avoiding cache pollution, resulting in better performance. We show that, with a 16KB L1 data cache, dynamic bypassing achieves sim- ilar performance to a double-sized L1 cache while reducing energy consumption by 25% and power by 18%. The technique is especially interesting for programs that do not use programmer-managed scratchpad memories. We give a case study to demonstrate the inefficiency of current GPU caches compared to programmer-managed scratchpad memories and show the extent to which cache bypassing can make up for the potential performance loss where the effort to program scratchpad memories is impractical.

A new perspective on processing-in-memory architecture design

As computation becomes increasingly limited by data movement and energy consumption, exploiting locality throughout the memory hierarchy becomes critical for maintaining the performance scaling that many have come to expect from the computing industry. Moving computation closer to main memory presents an opportunity to reduce the overheads associated with data movement. We explore the potential of using 3D die stacking to move memory-intensive computations closer to memory. This approach to processing-in-memory addresses some drawbacks of prior research on in-memory computing and appears commercially viable in the foreseeable future. We show promising early results from this approach and identify areas that are in need of research to unlock its full potential.

Structural Agnostic SpMV: Adapting CSR-Adaptive for Irregular Matrices

Sparse matrix vector multiplication (SpMV) is an important linear algebra primitive. Recent research has focused on improving the performance of SpMV on GPUs when using compressed sparse row (CSR), the most frequently used matrix storage format on CPUs. Efficient CSR-based SpMV obviates the need for other GPU-specific storage formats, thereby saving runtime and storage overheads. However, existing CSR-based SpMV algorithms on GPUs perform poorly on irregular sparse matrices, limiting their usefulness. We propose a novel approach for SpMV on GPUs which works well for both regular and irregular matrices while keeping the CSR format intact. We start with CSR-Adaptive, which dynamically chooses between two SpMV algorithms depending on the length of each row. We then add a series of performance improvements, such as a more efficient reduction technique. Finally, we add a third algorithm which uses multiple parallel execution units when operating on irregular matrices with very long rows. Our implementation dynamically assigns the best algorithm to sets of rows in order to ensure that the GPU is efficiently utilized. We effectively double the performance of CSR-Adaptive, which had previously demonstrated better performance than algorithms that use other storage formats. In addition, our implementation is 36% faster than CSR5, the current state of the art for SpMV on GPUs.

Design and Analysis of an APU for Exascale Computing

The challenges to push computing to exaflop levels are difficult given desired targets for memory capacity, memory bandwidth, power efficiency, reliability, and cost. This paper presents a vision for an architecture that can be used to construct exascale systems. We describe a conceptual Exascale Node Architecture (ENA), which is the computational building block for an exascale supercomputer. The ENA consists of an Exascale Heterogeneous Processor (EHP) coupled with an advanced memory system. The EHP provides a high-performance accelerated processing unit (CPU+GPU), in-package high-bandwidth 3D memory, and aggressive use of die-stacking and chiplet technologies to meet the requirements for exascale computing in a balanced manner. We present initial experimental analysis to demonstrate the promise of our approach, and we discuss remaining open research challenges for the community.

A Taxonomy of GPGPU Performance Scaling

Graphics processing units (GPUs) range from small, embedded designs to large, high-powered discrete cards. While the performance of graphics workloads is generally understood, there has been little study of the performance of GPGPU applications across a variety of hardware configurations. This work presents performance scaling data gathered for 267 GPGPU kernels from 97 programs run on 891 hardware configurations of a modern GPU. We study the performance of these kernels across a 5× change in core frequency, 8.3× change in memory bandwidth, and 11× difference in compute units. We illustrate that many kernels scale in intuitive ways, such as those that scale directly with added computational capabilities or memory bandwidth. We also find a number of kernels that scale in non-obvious ways, such as losing performance when more processing units are added or plateauing as frequency and bandwidth are increased. In addition, we show that a number of current benchmark suites do not scale to modern GPU sizes, implying that either new benchmarks or new inputs are warranted.

Dynamic buffer overflow detection for GPGPUs

Buffer overflows are a common source of program crashes, data corruption, and security problems. In this work, we demonstrate that GPU-based workloads can also cause buffer overflows, a problem that was traditionally ignored because CPUs and GPUs had separate memory spaces. Modern GPUs share virtual, and sometimes physical, memory with CPUs, meaning that GPU-based buffer overflows are capable of producing the same program crashes, data corruption, and security problems as CPU-based overflows. While there are many tools to find buffer overflows in CPU-based applications, the shift towards GPU-enhanced programs has expanded the problem beyond their capabilities. This paper describes a tool that uses canaries to detect buffer overflows caused by GPGPU kernels. It wraps OpenCL™ API calls and alerts users to any kernel that writes outside of a memory buffer. We study a variety of optimizations, including using the GPU to perform the canary checks, which allow our tool to run at near application speeds. The resulting runtime overhead, which scales with the number of buffers used by the kernel, is 14% across 175 applications in 16 GPU benchmark suites. In these same suites, we found 13 buffer overflows in 7 benchmarks.