Discovering and Exploiting Program Phases

Discovering and Exploiting Program Phases Timothy Sherwood, Erez Perelman, Greg Hamerly, Suleyman Sair, Brad Calder CSE 231 Presentation by Justin Ma 400 Million Instructions Non-Existent Processor New Processor New Compiler Spec2000 Benchmark Simulator 400 Million Instructions • Suppose you have a time budget… • Less than half second of execution time • What would you simulate? – Beginning? – Middle? – End? 400 Million Instructions Programs exhibit diverse modes of behavior gzip gcc 400 Million Instructions • Suppose you have a time budget… • Less than half second of execution time • What would you simulate? – – – – Beginning? Middle? End? Samples of different modes of behavior Program Phases • Observation: programs exhibit various modes of periodic behavior • These modes are program phases • Challenge: Extract these automatically Phase Basics • Intervals – slices in times • Phases – intervals with similar behavior IPC Time (Instruction Count) Phase Basics • Intervals – slices in times • Phases – intervals with similar behavior IPC Time (Instruction Count) Defining “Similar Behavior” • Metric for comparing intervals? – Cache misses? – IPC? – Branch misprediction rates? • Problem: Performance alone is too architecture dependent Defining “Similar Behavior” • Code path traversal – Directly affects time-varying behavior – Execute same code, same performance – Architecture independent • Metrics for code path traversal – Frequency of branches – Frequency of function calls – Frequency of basic block calls Basic Block Vector Time t 0 B1 0 B2 0 B3 0 B4 B1 B2 B3 B4 Basic Block Vector Time t 1 B1 1 B2 0 B3 1 B4 B1 B2 B3 B4 Basic Block Vector Time t 2 B1 1 B2 1 B3 2 B4 B1 B2 B3 B4 Basic Block Vector Time t 2 B1 1 B2 1 B3 2 B4 B1 B2 B3 Time t + 1 0 B1 0 B2 0 B3 0 B4 B4 Basic Block Vector Time t 2 B1 1 B2 1 B3 2 B4 B1 B2 B3 Time t + 1 1 B1 1 B2 0 B3 1 B4 B4 Basic Block Vector Time t 2 B1 1 B2 1 B3 2 B4 B1 B2 B3 Time t + 1 2 B1 2 B2 0 B3 2 B4 B4 Manhattan Distance = |1 – 2| + |1 – 0| = 2 Euclidian Distance = sqrt((1 – 2)2 + (1 – 0)2) = sqrt(2) Basic Block Similarity Matrix • gzip Basic Block Similarity Matrix • gcc BBV similarity between intervals reflects performance similarity Automatic Phase Classification • Classify intervals into phases – We do not know which BBVs correspond to particular phases a priori • k-means clustering – Iterative clustering algorithm – Dimension Reduction • Random Linear Projection – Try different k values • Use BIC to choose best Automatic Phase Classification Automatic Phase Classification Clustering accurately distinguishes phases automatically SimPoint • Simulate large programs on a budget • Perform detailed simulation on representative code snippets – Choose centroid interval from each phase (10 million instructions) • Extrapolate large program performance – Weighted by frequency of phase SimPoint • Simulate 400 million instructions total Accurate estimate despite instruction budget Why SimPoint Succeeds • Program behavior varies over time • SimPoint intelligently chooses which intervals to simulate • Regularity within program phases allows accurate extrapolation Online Classification • Detect phases as program is running • Applications – Thread scheduling – Power management – Predicting future phases • Challenges – One pass of input – Limited storage Online Classification Online Classification High variance in metrics across full trace Low variance shows online classification succeeds in finding phases Conclusions • Phases are a vital abstraction – Performance varies greatly w/in program – Attributable to different modes of behavior – Offline: k-means clustering – Online • Can discover phases automatically • Code path characterization – Strong correlation with actual performance – SimPoint exploits this with great success Outline • Introduction (motivate) • Basics (definitions, BBV, BBMatrix) • Offline Phase Classification – SimPoints • Online Phase Classification • Conclusions Limitations of Clustering Bayesian Information Criterion • Fit to Gaussians Self-Modifying Code Self-modifying code 85o Program Phases Learning Phases Learning Phases