Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userAutomated Instruction Stream Throughput Prediction for Intel and AMD Microarchitectures
65
0
0.0
(
0
)
Added by
Julian Hammer
Publication date
2018
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
An accurate prediction of scheduling and execution of instruction streams is a necessary prerequisite for predicting the in-core performance behavior of throughput-bound loop kernels on out-of-order processor architectures. Such predictions are an indispensable component of analytical performance models, such as the Roofline and the Execution-Cache-Memory (ECM) model, and allow a deep understanding of the performance-relevant interactions between hardware architecture and loop code. We present the Open Source Architecture Code Analyzer (OSACA), a static analysis tool for predicting the execution time of sequential loops comprising x86 instructions under the assumption of an infinite first-level cache and perfect out-of-order scheduling. We show the process of building a machine model from available documentation and semi-automatic benchmarking, and carry it out for the latest Intel Skylake and AMD Zen micro-architectures. To validate the constructed models, we apply them to several assembly kernels and compare runtime predictions with actual measurements. Finally we give an outlook on how the method may be generalized to new architectures.
rate research
Read More
Hardware platforms in high performance computing are constantly getting more complex to handle even when considering multicore CPUs alone. Numerous features and configuration options in the hardware and the software environment that are relevant for performance are not even known to most application users or developers. Microbenchmarks, i.e., simple codes that fathom a particular aspect of the hardware, can help to shed light on such issues, but only if they are well understood and if the results can be reconciled with known facts or performance models. The insight gained from microbenchmarks may then be applied to real applications for performance analysis or optimization. In this paper we investigate two modern Intel x86 server CPU architectures in depth: Broadwell EP and Cascade Lake SP. We highlight relevant hardware configuration settings that can have a decisive impact on code performance and show how to properly measure on-chip and off-chip data transfer bandwidths. The new victim L3 cache of Cascade Lake and its advanced replacement policy receive due attention. Finally we use DGEMM, sparse matrix-vector multiplication, and the HPCG benchmark to make a connection to relevant application scenarios.
Dynamic affinity scheduling has been an open problem for nearly three decades. The problem is to dynamically schedule multi-type tasks to multi-skilled servers such that the resulting queueing system is both stable in the capacity region (throughput optimality) and the mean delay of tasks is minimized at high loads near the boundary of the capacity region (heavy-traffic optimality). As for applications, data-intensive analytics like MapReduce, Hadoop, and Dryad fit into this setting, where the set of servers is heterogeneous for different task types, so the pair of task type and server determines the processing rate of the task. The load balancing algorithm used in such frameworks is an example of affinity scheduling which is desired to be both robust and delay optimal at high loads when hot-spots occur. Fluid model planning, the MaxWeight algorithm, and the generalized $cmu$-rule are among the first algorithms proposed for affinity scheduling that have theoretical guarantees on being optimal in different senses, which will be discussed in the related work section. All these algorithms are not practical for use in data center applications because of their non-realistic assumptions. The join-the-shortest-queue-MaxWeight (JSQ-MaxWeight), JSQ-Priority, and weighted-workload algorithms are examples of load balancing policies for systems with two and three levels of data locality with a rack structure. In this work, we propose the Generalized-Balanced-Pandas algorithm (GB-PANDAS) for a system with multiple levels of data locality and prove its throughput optimality. We prove this result under an arbitrary distribution for service times, whereas most previous theoretical work assumes geometric distribution for service times. The extensive simulation results show that the GB-PANDAS algorithm alleviates the mean delay and has a better performance than the JSQ-MaxWeight algorithm by twofold
With the advent of increasingly complex hardware in real-time embedded systems (processors with performance enhancing features such as pipelines, cache hierarchy, multiple cores), many processors now have a set-associative L2 cache. Thus, there is a need for considering cache hierarchies when validating the temporal behavior of real-time systems, in particular when estimating tasks worst-case execution times (WCETs). To the best of our knowledge, there is only one approach for WCET estimation for systems with cache hierarchies [Mueller, 1997], which turns out to be unsafe for set-associative caches. In this paper, we highlight the conditions under which the approach described in [Mueller, 1997] is unsafe. A safe static instruction cache analysis method is then presented. Contrary to [Mueller, 1997] our method supports set-associative and fully associative caches. The proposed method is experimented on medium-size and large programs. We show that the method is most of the time tight. We further show that in all cases WCET estimations are much tighter when considering the cache hierarchy than when considering only the L1 cache. An evaluation of the analysis time is conducted, demonstrating that analysing the cache hierarchy has a reasonable computation time.
Developing efficient GPU kernels can be difficult because of the complexity of GPU architectures and programming models. Existing performance tools only provide coarse-grained suggestions at the kernel level, if any. In this paper, we describe GPA, a performance advisor for NVIDIA GPUs that suggests potential code optimization opportunities at a hierarchy of levels, including individual lines, loops, and functions. To relieve users of the burden of interpreting performance counters and analyzing bottlenecks, GPA uses data flow analysis to approximately attribute measured instruction stalls to their root causes and uses information about a programs structure and the GPU to match inefficiency patterns with suggestions for optimization. To quantify each suggestions potential benefits, we developed PC sampling-based performance models to estimate its speedup. Our experiments with benchmarks and applications show that GPA provides an insightful report to guide performance optimization. Using GPA, we obtained speedups on a Volta V100 GPU ranging from 1.01$times$ to 3.53$times$, with a geometric mean of 1.22$times$.
This dissertation introduces measurement-based performance modeling and prediction techniques for dense linear algebra algorithms. As a core principle, these techniques avoid executions of such algorithms entirely, and instead predict their performance through runtime estimates for the underlying compute kernels. For a variety of operations, these predictions allow to quickly select the fastest algorithm configurations from available alternatives. We consider two scenarios that cover a wide range of computations: To predict the performance of blocked algorithms, we design algorithm-independent performance models for kernel operations that are generated automatically once per platform. For various matrix operations, instantaneous predictions based on such models both accurately identify the fastest algorithm, and select a near-optimal block size. For performance predictions of BLAS-based tensor contractions, we propose cache-aware micro-benchmarks that take advantage of the highly regular structure inherent to contraction algorithms. At merely a fraction of a contractions runtime, predictions based on such micro-benchmarks identify the fastest combination of tensor traversal and compute kernel.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
