Download presentation
Presentation is loading. Please wait.
Published byPatricia Turner Modified over 8 years ago
1
JILP RESULTS 1
2
JILP Experimental Framework Goal Simplicity of a trace based simulator Flexibility to model special predictors ( e.g., using data values) Trace driven with pipeline timing information Tracing Methodology: Detailed timing simulator with perfect branch predictor 50M uops per trace Traces include pipeline behavior/timing, instruction address, uop type etc. 2
3
JILP Experiments Workloads 40 workloads selected from a large pool of applications 5 classes: CLIENT 16, INT 6, MM 7, SERVER 5, WS 6 Metrics Arithmetic average of MPPKI ( Misprediction Penalty per Kilo Instructions) 3 No secret workloads
4
JILP Conditional predictor results #1 A. Seznec, A 64 Kbytes ISL-TAGE branch predictor, MPPKI 568 #2 Y. Ishii, K. Kuroyanagi, T. Sawada, M. Inaba, K. Hiraki, Revisiting Local History for Improving Fused Two-Level Branch Predictor, MPPKI 581 #3 D. Jimenez, OH-SNAP: Optimized Hybrid Scaled Neural Analog Predictor, MPPKI 598 #4 Y. Hu, D. Koppelman and L. Peng, Penalty-Sensitive L-TAGE Predictor, MPPKI 608 #5 G. Shi and M. Lipasti, Perceptron Branch Prediction with Separated Taken/Not-Taken Weight Tables, MPPKI 677 4
5
JILP Indirect predictor results #1 A. Seznec, A 64-Kbytes ITTAGE indirect branch predictor, MPPKI 34.1 #2 Y. Ishii, T. Sawada, K. Kuroyanagi, M. Inaba, K. Hiraki, Bimode Cascading: Adaptive Rehashing for ITTAGE Indirect Branch Predictor, MPPKI 37.0 #3 N. Bhansali, C. Panirwala, H. Zhou, Exploring Correlation for Indirect Branch Prediction, MPPKI 51.6 #4 Daniel A. Jimenez, SNIP: Scaled Neural Indirect Predictor, MPPKI 52.9 5
6
JILP Discussions From industry perspective It is important! Perf gain from better branch predictors Branch prediction papers Total perf gain from uarch changes uarch papers What do we need from researchers? What’s next? 6
7
JILP Detailed Results 7 Conditional Track 01_shi02_jimenez03_seznec04_hu05_ishii CLIENT01195.6329134.769694.2842115.9801122.0738 CLIENT022246.37042029.13392180.85442213.6722082.369 CLIENT03142.251497.234359.525581.773884.3104 CLIENT04521.9194483.7377450.4976454.617448.4412 CLIENT05441.8335359.2672230.9359303.9691358.7924 CLIENT0692.46276.159934.450861.321157.6427 CLIENT07555.418461.9348274.5406323.4801380.713 CLIENT08586.2423446.227792.8724108.0867162.3047 CLIENT09208.0368166.5278153.311165.8465165.7341 CLIENT10300.3148214.1727161.4855191.2055205.3421 CLIENT11175.246188.048283.089399.365276.5518 CLIENT12482.0413348.066268.6224301.2785339.7142 CLIENT13175.2024107.411666.471391.3964101.5649 CLIENT14238.0711183.485115.7987168.6111175.4837 CLIENT15318.7824274.3057190.0626236.2642262.8464 CLIENT16208.8891177.529122.1048150.7502171.7198 INT011109.14651044.9755977.02871011.1441019.732 INT021312.93491283.69981318.65591358.9651278.442 INT035.92414.73464.21335.01715.108 INT0484.855316.014513.845914.69315.3078 INT05306.668264.7533256.906295.6024260.1517 INT06311.4544269.2155263.2841297.5455268.2491 MM01553.1553420.0735365.5334440.022399.1041 MM02550.0649424.4888357.273427.1802387.6036 MM03496.1162414.9755399.1166451.5351409.5461 MM04421.1688367.3337363.4216398.1258358.9154 MM051813.9131791.24261719.49641747.5071567.24 MM0619.787513.05665.45818.83928.9297 MM074583.23374474.32394630.55294874.4594396.604 SERVER01411.1282390.5577388.7235406.2656393.3057 SERVER02396.0853258.2285196.0167219.92260.3895 SERVER03359.8728244.1589167.9817190.0911224.7323 SERVER04338.5081255.3887229.8285240.2901249.9335 SERVER05347.7651259.1003236.5313247.4486254.0217 WS01557.3669443.504335.4763388.4488357.2099 WS02528.4282446.5272347.1347419.3035440.7037 WS033994.01213834.95913948.16884080.8153834.593 WS041368.41311133.02281478.67641548.431475.929 WS05173.42693.708561.711675.628480.3839 WS06143.8643114.759380.6703107.6931114.1167 AVG676.900165597.7703475568.11532608.0647581.3964 Indirect Track 06_bhansali07_seznec08_ishii09_jimenez CLIENT0112.51257.93378.741317.1601 CLIENT020.81620.61040.610.6155 CLIENT0310.06935.8118.107221.2596 CLIENT0423.679219.193422.085729.8753 CLIENT05184.018786.523897.2576185.1857 CLIENT063.32732.4952.67475.5105 CLIENT0717.075211.104313.022126.734 CLIENT0823.012116.140319.243822.95 CLIENT0921.597416.008718.028620.3676 CLIENT10119.12651.724860.5654117.9056 CLIENT1125.74499.49117.388617.9866 CLIENT12109.386547.340650.6063123.9733 CLIENT1316.08859.852612.0826.9447 CLIENT1431.022420.316420.558529.6097 CLIENT1548.005931.063731.858349.9706 CLIENT1628.690719.597319.878833.8436 INT010.08570.04150.0451 INT020.99890.67130.7490.7462 INT033.45632.17591.67341.7106 INT040.32870.29040.31740.2939 INT05268.2163187.9945205.5175286.9901 INT06248.4637179.3377194.1275264.2514 MM0132.20625.278925.090627.0786 MM0226.757620.716621.934127.1227 MM0374.226853.436261.268974.8308 MM0446.294430.34733.32332.7072 MM050000 MM061.19590.19830.25150.6047 MM070000 SERVER01323.493276.1373293.9561274.5287 SERVER0264.670643.845646.088573.6275 SERVER0367.273943.357548.5589.9561 SERVER0427.668519.461620.264543.8979 SERVER0532.518223.895424.811746.8056 WS0184.329344.337650.008548.0442 WS0263.82441.147741.957866.0846 WS030000 WS043.40632.44792.54012.6119 WS0513.9638.93899.082116.8948 WS066.88684.83414.84365.7889 AVG51.610917534.10247536.97769552.8628475
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.