Lost in Abstraction: Pitfalls of Analyzing GPUs at the Intermediate Language Level

Gutierrez, Anthony; Beckmann, Bradford M.; Dutu, Alexandru; Gross, Joseph; LeBeane, Michael; Kalamatianos, John; Kayıran, Onur; Poremba, Matthew; Potter, Brandon; Puthoor, Sooraj; Sinclair, Matthew; Wyse, Mark; Yin, Jieming; Zhang, Xianwei; Jain, Akshay; Rogers, Timothy G.

doi:10.1109/hpca.2018.00058

Cited by 49 publications

(20 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prior work has shown how to use AMD's ROCm ecosystem to simulate HCC and HIP applications in gem5 with high fidelity compared to an AMD APU [5] [6]. In this work, we build from the existing ROCm support to simulate MI applications that use the MIOpen library [36].…”

Section: A the Gem5 Simulatormentioning

confidence: 99%

“…Although these tools could be used to perform similar studies, GPGPU-Sim focuses on discrete GPUs, not tightly coupled ones like gem5 models (GPGPU-Sim's recent updates could be integrated into gem5-gpu [34] to allow such a study). Moreover, recent work has shown that simulating at a higher level, like GPGPU-Sim, loses important architectural details and may lead to incorrect conclusions [5]. Thus, when combined with the importance of being able to recompile the libraries to run APU-compliant code, we believe gem5 and MIOpen represent the best combination between simulator and MI library.…”

Section: Pc-based L2 Bypassingmentioning

confidence: 99%

“…This motivates a coherence strategy which prioritizes memory throughput and scalability, sometimes at the cost of cache reuse. In an effort to better understand the trade-offs of different caching strategies for MI workloads, we evaluate the performance of these applications with multiple levels of GPU caching enabled using the publicly available AMD gem5 APU simulator [5].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Optimizing GPU Cache Policies for MI Workloads

Alsop¹,

Zhang²,

Yeh

et al. 2019

2019 IEEE International Symposium on Workload Characterization (IISWC)

Self Cite

View full text Add to dashboard Cite

In recent years, machine intelligence (MI) applications have emerged as a major driver for the computing industry. Optimizing these workloads is important but complicated. As memory demands grow and data movement overheads increasingly limit performance, determining the best GPU caching policy to use for a diverse range of MI workloads represents one important challenge. To study this, we evaluate 17 MI applications and characterize their behaviors using a range of GPU caching strategies. In our evaluations, we find that the choice of caching policy in GPU caches involves multiple performance trade-offs and interactions, and there is no one-size-fits-all GPU caching policy for MI workloads. Based on detailed simulation results, we motivate and evaluate a set of cache optimizations that consistently match the performance of the best static GPU caching policies.

show abstract

Section: A the Gem5 Simulatormentioning

confidence: 99%

Section: Pc-based L2 Bypassingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Optimizing GPU Cache Policies for MI Workloads

Alsop¹,

Zhang²,

Yeh

et al. 2019

2019 IEEE International Symposium on Workload Characterization (IISWC)

Self Cite

View full text Add to dashboard Cite

show abstract

“…While Central Processing Unit (CPU) simulation techniques have reached maturity, GPU simulation often suffers from the following problems: (a) instruction sets are not accurately modeled, but approximated by an artificial, low-level intermediate representation [7], [8], (b) GPU simulators do not model existing commercial GPUs, but only simplified GPU architectures [9], (c) instead of using vendor provided driver stacks and compilers, GPU simulators often rely on simplified system software, which may behave entirely differently to original tools [10], [11], and (d) GPUs are treated as standalone devices, not modeling any CPU-GPU transactions [12]. This has led researchers using GPU simulation to rely on tools providing questionable accuracy [13]. For example, many existing GPU simulators including gem5-GPU [14], GpuTejas [15], Multi2Sim [10], GPGPU-Sim [16], and Multi2Sim-Kepler [11] claim cycle-accuracy.…”

Section: Introductionmentioning

confidence: 99%

“…GPU arithmetic cycles in the selected kernel differ by 47% (6.0 to 6.1). It is more than likely that simplified or non-vendor supplied tool chains used by other GPU simulators introduce even greater error, as also highlighted in [13].…”

Section: Introductionmentioning

confidence: 99%

Full-System Simulation of Mobile CPU/GPU Platforms

Kaszyk

Wagstaff

Spink

et al. 2019

2019 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)

View full text Add to dashboard Cite

Graphics Processing Units (GPUs) critically rely on a complex system software stack comprising kernel-and userspace drivers and Just-in-time (JIT) compilers. Yet, existing GPU simulators typically abstract away details of the software stack and GPU instruction set. Partly, this is because GPU vendors rarely release sufficient information about their latest GPU products. However, this is also due to the lack of an integrated CPU/GPU simulation framework, which is complete and powerful enough to drive the complex GPU software environment. This has led to a situation where research on GPU architectures and compilers is largely based on outdated or greatly simplified architectures and software stacks, undermining the validity of the generated results. In this paper we develop a full-system system simulation environment for a mobile platform, which enables users to run a complete and unmodified software stack for a state-of-the-art mobile Arm CPU and Mali-G71 GPU powered device. We validate our simulator against a hardware implementation and Arm's stand-alone GPU simulator, achieving 100% architectural accuracy across all available toolchains. We demonstrate the capability of our GPU simulation framework by optimizing an advanced Computer Vision application using simulated statistics unavailable with other simulation approaches or physical GPU implementations. We demonstrate that performance optimizations for desktop GPUs trigger bottlenecks on mobile GPUs, and show the importance of efficient memory use.

show abstract