AsmDB: Understanding and Mitigating Front-end Stalls in Warehouse-Scale Computers

Authors: Grant Ayers et al.
Venue: ISCA 2019

Previous works have highlighted a significant frontend bottleneck is warehouse-scale computers (WSC). A variety of solutions have been proposed, both on real hardware and architectural papers, to mitigate the issue. This paper performs deep analysis across 90%+ of Google's entire fleet to perform a fine-grain analysis of when, where, and how frontend bottlenecks occur. AsmDB comprises of post-processed last branch record (LBR) data to form control flow probabilities and precise information about what instructions triggered I-cache misses. The analysis shows several core reasons for I-cache misses: large jump distances (either via function call or indirect branch) and cold code being brought into the cache (either via cache blocks or prefetching).

The remainder of the paper transitions to being much more compiler focused. It describes a software prefetching algorithm which utilizes AsmDB information to inject prefetches at ideal points in code. Prefetches must be both timely, and accurate. The challenge is exacerbated in when performing software prefetching due to varying IPC, as well as fan-in and fan-out concerns. Fundamentally, software prefetching increases the code size by requiring another instruction, so even if a prefetch is useful, it naturally incurs some overhead. The authors describe in detail how the AsmDB information helps to determine exactly where to perform this prefetch, and the net results is about 0.5-1% overall performance gains. I absolutely loved the analysis component of this paper but was hoping for a bigger perf win, especially considering the complexity required for the prefetching implementation.

Full Text


Comments

Post a Comment

Popular posts from this blog

Fundamental Latency Trade-offs in Architecting DRAM Caches (Alloy Cache)

Authors: Moinuddin K. Qureshi and Gabriel H. Loh Venue:    MICRO 2012 The authors present a new architecture for DRAM cache which targets optimizing latency rather than hit rate. The authors show that for every optimization made to improve hit rate, one must also consider the added cache latency, thus resulting in a new "break even hit rate". The authors compare to three key baselines: an infeasibly large SRAM-tag array, the prior "LH-Cache", and a ideal latency-optimized cache. They find that "unoptimizations" improve latency, while minimally impacting hit-rate. Their final result is a cache which boosts performance 35% improvement over the baseline, and nearly 20% better than the previous LH-cache. The specific contributions are: Moving to a direct mapped cache provides significant improvement (by reducing latency) Combining tag-store and data-store into a single entity improves performance 21% Implementing a small per-core predictor enhances per...
Read more

ZCOMP: Reducing DNN Cross-Layer Memory Footprint Using Vector Extensions

Authors: Berkin Akin, Zeshan A. Christi, and Alaa R. Alameldeen Venue: MICRO, 2019 With accelerators dominating the deep learning space in architecture conferences, this paper stands out as it focuses on reducing DNN inference/training overhead while using a CPU. An obvious question is, with the wide-spread use of GPUs/TPUs for deep learning, why should we focus on optimizing CPUs for deep learning. In a recent paper published at HPCA 2019, Facebook claims that CPUs are preferred for applications where a tight integration is required between DNN and non-DNN tasks. Also, Intel's recent AVX512 has specialized support for DNNs in the form of new instructions called Vector Neural Network Instructions (VNNI). Broadly, optimizing DNNs can be viewed from two different perspectives; computation, and communication. This paper targets reducing communication overhead, more specifically, activation or feature-map communication overhead by compressing them. Compressing activations/weights...
Read more

AutoFDO: Automatic Feedback-Directed Optimization for Warehouse-Scale Applications

Authors: Dehao Chen, David Xinliang Li, Tipp Moseley Venue:    CGO 2016 This paper presents AutoFDO, a system used which profiles warehouse-scale applications, and applies feedback to the compilation for the next release. AutoFDO works by profiling and storing profiles in an aggregate database, annotating the profiles via an intermediate representation, and finally providing feedback. On average, this technique boosts performance by around 10% and works well even with stale releases. For profiles to be useful, they must be in an intermediate representation. This is built from binary-level profile, which uses LBR to map instruction frequencies. By using these frequencies, program counters, and the source, a source profile can be generated. The source profile can be used to build an annotated call-graph with edge frequencies, with some inaccuracy. These annotated call-graphs can then be used as feedback to the compiler. Because of the size of the applications and scale of d...
Read more