Engineering topics
Corporaal, Henk
Publications and source records attributed to Corporaal, Henk.
Dissecting Tensor Cores via Microbenchmarks: Latency, Throughput and Numeric Behaviors
Tensor Cores have been an important unit to accelerate Fused Matrix Multiplication Accumulation (MMA) in all NVIDIA GPUs since Volta Architecture. To program Tensor Cores, users have to use either legacy wmma APIs or current mma APIs. Legacy wmma APIs are more easy-to-use but can only exploit limited features and power of Tensor Cores. Specifically, wmma APIs support fewer operand shapes and can not leverage the new sparse matrix multiplication feature of the newest Ampere Tensor Cores. However, the performance of current programming interface has not been well explored. Furthermore, the computation numeric behaviors of low-precision floating points (TF32, BF16, and FP16) supported by the newest Ampere Tensor Cores are also mysterious. In this paper, we explore the throughput and latency of current programming APIs. Further, we intuitively study the numeric behaviors of Tensor Cores MMA and profile the intermediate operations including multiplication, addition of inner product, and accumulation. All codes used in this work can be found in https://github.com/sunlex0717/DissectingTensorCores.
Locality-Aware CTA Clustering For Modern GPUs
In this paper, we proposed a novel clustering technique for tapping into the performance potential of a largely ignored type of locality: inter-CTA locality. We first demonstrated the capability of the existing GPU hardware to exploit such locality, both spatially and temporally, on L1 or L1/Tex unified cache. To verify the potential of this locality, we quantified its existence in a broad spectrum of applications and discussed its sources of origin. Based on these insights, we proposed the concept of CTA-Clustering and its associated software techniques. Finally, We evaluated these techniques on all modern generations of NVIDIA GPU architectures. The experimental results showed that our proposed clustering techniques could significantly improve on-chip cache performance.
SFU-Driven Transparent Approximation Acceleration on GPUs
Approximate computing, the technique that sacrifices certain amount of accuracy in exchange for substantial performance boost or power reduction, is one of the most promising solutions to enable power control and performance scaling towards exascale. Although most existing approximation designs target the emerging data-intensive applications that are comparatively more error-tolerable, there is still high demand for the acceleration of traditional scientific applications (e.g., weather and nuclear simulation), which often comprise intensive transcendental function calls and are very sensitive to accuracy loss. To address this challenge, we focus on a very important but often ignored approximation unit on GPUs.
X: A Comprehensive Analytic Model for Parallel Machines
To continuously comply with Moore’s Law, modern parallel machines become increasingly complex. Effectively tuning application performance for these machines therefore becomes a daunting task. Moreover, identifying performance bottlenecks at application and architecture level, as well as evaluating various optimization strategies, are becoming extremely difficult when the entanglement of numerous correlated factors is being presented. To tackle these challenges, we present a visual analytical model named “X”. It is intuitive and sufficiently flexible to track all the typical features of a parallel machine.
Critical Points Based Register-Concurrency Autotuning for GPUs
The unprecedented prevalence of GPGPU is largely attributed to its abundant on-chip register resources, which allow massively concurrent threads and extremely fast context switch. However, due to internal memory capacity constraints, there is a tradeoff between the per-thread register usage and the overall concurrency. This becomes a design problem in terms of performance tuning, since the performance “sweet spot” which can be significantly affected by these two factors is generally unknown beforehand. In this paper, we propose an effective autotuning solution to quickly and efficiently select the optimal number of registers perthread for delivering the best GPU performance. Experiments on three generations of GPUs (Nvidia Fermi, Kepler and Maxwell) demonstrate that our simple strategy can achieve an average of 10% performance improvement while a max of 50% over the original version without modifying the user program. Additionally, to reduce local cache misses due to register spilling and further improve performance, we explore three optimization schemes (i.e. bypass L1 for global memory access, enlarge local L1 cache and spill into shared memory) and discuss their impact on performance on a Kepler GPU.