2008 GCC Developer Summit: Machine Learning, Compiler Enhancements, and Debugging Advances

需积分: 9 158 浏览量更新于2024-07-31 收藏 2.07MB PDF 举报

"GCC-2008 Proceedings是一份涵盖了2008年6月17日至19日在加拿大渥太华举行的GCC开发者峰会的重要论文集。这次会议聚焦于GCC（GNU Compiler Collection，GNU编译器套件）的最新进展和研究，展示了在机器学习驱动的编译器、工具链增强、调试优化、代码生成技术以及调试工具等方面的工作。 1. MILEPOST GCC: Machine Learning Based Research Compiler - Grigori Fursin等人探讨了如何利用机器学习改进GCC，通过将传统的grep和sed工具替换为更智能的解决方案，提升编译器的效率和性能。 2. Using GCC Instead of Grep and Sed - 这篇论文介绍了如何利用GCC内置的功能来替代繁琐的文本处理工具，简化开发者的日常维护工作。 3. Middle-End Array Expressions - Taras Glek和David Mandelin的工作关注于优化GCC中的数组操作，提升代码的执行效率。 4. Adding Coding Rule Checking Capabilities - Richard Guenther的论文涉及增强GCC工具链，以便在编译过程中检查和纠正编码规范，保证代码质量。 5. Improving GCC’s Debug Information Precision - 通过精确的调试信息，Marpons、Polo和Mariño团队致力于提高GCC对程序内部结构的理解，帮助开发者更好地定位和修复问题。 6. Coarrays in GNU Fortran - Michael Matz提出了在GNU Fortran中实现并行编程特性——Coarrays，这对于多核处理器时代具有重要意义。 7. A Plan to Fix Local Variable Debug Information - Toon Moene的贡献在于提出了解决GCC本地变量调试信息问题的策略，增强了代码可读性和调试体验。 8. Streamization Infrastructure Improvements - Alexandre Oliva的工作着重于GCC基础设施的优化，特别是在数据流处理方面的技术进步。 9. Feedback-Directed Optimizations with Estimated Edge Profiles - Pop等人通过硬件事件采样获取边缘-profile信息，实现了在GCC中进行反馈导向的优化。 10. Superoptimizer Analysis of Multiway Branch Code Generation - V. Ramasamy等人对多路分支代码生成进行了深度分析，寻求性能的极致优化。 11. Non-stop Multi-threaded Debugging in GDB - Roger A. Sayle讨论了在多线程环境下，如何改进GDB（GNU Debugger）以提供无缝的调试体验。 12. A New ELF Linker - N. Sidwell等人介绍了新一代的ELF（Executable and Linkable Format）链接器，对链接过程进行了革新。 13. Incremental Compilation for GCC - Ian Lance Taylor和Tom Tromey的研究关注于增量编译，减少编译时间，提高开发效率。 14. Conference Overview - 最后，会议还概述了其他议题，如编译器的持续改进和未来趋势等。这些论文代表了GCC及相关工具在2008年的前沿研究和技术创新，为未来的软件开发提供了宝贵的参考和实践指南。"

16 • MILEPOST GCC: machine learning based research compiler

1.1

1.2

1.3

1.4

1.5

bitcount

susan_c

susan_e

susan_s

jpeg_c

jpeg_d

dijkstra

patricia

blowfish_d

blowfish_e

rijndael_d

rijndael_e

sha

adpcm_c

adpcm_d

CRC32

gsm

qsort1

stringsearch1

speedups

2.24 2.51 2.51 1.97

AMD IA32 IA64

(a)

0.8

0.9

1.1

1.2

1.3

1.4

bitcount

susan_c

susan_e

susan_s

jpeg_c

jpeg_d

dijkstra

patricia

blowfish_d

blowfish_e

rijndael_d

rijndael_e

sha

adpcm_c

adpcm_d

CRC32

gsm

qsort1

stringsearch1

AVERAGE

speedups

Iterative compilation Predicted optimizations using MILEPOST GCC and ML

(b)

Figure 5: Experimental results when using iterative compilation with random search strategy (500 iterations; 50%

probability to select each ﬂags; AMD,IA32,IA64) and when predicting best optimization passes based on program

features and ML model (ARC)

Firstly, features for each benchmark were extracted

from programs using the new pass within MILEPOST

GCC, and these features then sent to the Global Op-

timization Database within CCC Framework. An ML

model for each benchmark was built, using the execu-

tion time gathered from 500 separate runs using differ-

ent random sequences of passes, and a ﬁxed data set.

Each run was repeated 5 times so speedups were not

caused by cache priming, etc.). After each run, the ex-

perimental results including execution time, compila-

tion time, code size, and program features are sent to

the database, where they are stored for future reference.

Figure 5a shows that considerable speedups can be al-

ready obtained after iterative compilation on all plat-

forms. However, this is a time-consuming process and

different speedups across different platforms motivates

the use of machine learning to automatically build spe-

cialized compilers and predict the best optimization

ﬂags or sequences of passes for different architectures.

5.2 Evaluating Model Performance

Once a model has been built for each of our bench-

marks, we can evaluate the results by introducing a

new program to the system, and measuring how well

the prediction provided by our model performs. In this

work we did not use a separate testing benchmark suite

due to time constraints, so instead leave-one-out-cross-

validation was used. Using this method, all training data

relating to the benchmark being tested is excluded from

the training process, and the models rebuilt. When a sec-

ond benchmark is tested, the training data pertaining to

the ﬁrst benchmark is returned to the training set, that of

the second benchmark excluded, and so on. In this way

we ensure that each benchmark is tested as a new pro-

gram entering the system for the ﬁrst time—of course,

in real-world usage, this process is unnecessary.

2008 GCC Developers’ Summit • 17

When a new program is compiled, features are ﬁrst

generated using MILEPOST GCC. These features are

then sent to our ML model within the CCC Framework

(implemented as a MATLAB server), which processes

them and returns a predicted sequence of passes which

should either improve execution time, reduce code size,

or both. We then evaluate the prediction by compil-

ing the program with the suggested sequence of passes,

measure the execution time, and compare with the orig-

inal time for the default -O3 optimization level. It

is important to note that only one compilation occurs

at evaluation—there is no search involved. Figure 5b

shows these results for the ARC725D. It demonstrates

that except a few pathological cases where predicted

ﬂags degraded performance and which analysis we leave

for future work, using CCC Framework, MILEPOST

GCC and Machine Learning Models we can improve

original ARC GCC by around 11%.

This suggests that our techniques and tools can be efﬁ-

cient to build future iterative adaptive specialized com-

pilers.

6 Conclusions and Future Work

In this paper we have shown that MILEPOST GCC has

signiﬁcant potential in the automatic tuning of GCC op-

timization. We plan to use these techniques and tools

to further investigate the automatic selection of opti-

mal orders of optimization passes and ﬁne-grain tun-

ing of transformation parameters. The overall frame-

work will also allow the analysis of interactions be-

tween optimizations and investigation of the inﬂuence

of program inputs and run-time state on program opti-

mizations. Future work will also include ﬁne-grain run-

time adaptation for multiple program inputs on hetero-

geneous multi-core architectures.

7 Acknowledgments

The authors would like to thank Sebastian Pop for the

help with modifying and debugging GCC and John W.

Lockhart for the help with editing this paper. This re-

search is supported by the European Project MILEPOST

(Machine Learning for Embedded Programs Optimi-

sation) [4] and by the HiPEAC European Network of

Excellence (High-Performance Embedded Architecture

and Compilation) [5].

References

[1] ACOVEA: Using Natural Selection to Investigate

Software Complexities. http://www.

coyotegulch.com/products/acovea.

[2] Continuous Collective Compilation Framework.

http://cccpf.sourceforge.net.

[3] ESTO: Expert System for Tuning Optimizations.

http:

//www.haifa.ibm.com/projects/

systems/cot/esto/index.html.

[4] EU Milepost project (MachIne Learning for

Embedded PrOgramS opTimization).

http://www.milepost.eu.

[5] European Network of Excellence on

High-Performance Embedded Architecture and

Compilation (HiPEAC).

http://www.hipeac.net.

[6] MiDataSets SourceForge development site.

http://midatasets.sourceforge.net.

[7] Plugin project.

http://libplugin.sourceforge.net.

[8] QLogic PathScale EKOPath Compilers.

http://www.pathscale.com.

[9] F. Agakov, E. Bonilla, J.Cavazos, B.Franke,

G. Fursin, M. O’Boyle, J. Thomson,

M. Toussaint, and C. Williams. Using machine

learning to focus iterative optimization. In

Proceedings of the International Symposium on

Code Generation and Optimization (CGO), 2006.

[10] F. Bodin, T. Kisuki, P. Knijnenburg, M. O’Boyle,

and E. Rohou. Iterative compilation in a

non-linear optimisation space. In Proceedings of

the Workshop on Proﬁle and Feedback Directed

Compilation, 1998.

[11] E. V. Bonilla, C. K. I. Williams, F. V. Agakov,

J. Cavazos, J. Thomson, and M. F. P. O’Boyle.

Predictive search distributions. In W. W. Cohen

and A. Moore, editors, Proceedings of the 23rd

International Conference on Machine learning,

pages 121–128, New York, NY, USA, 2006.

ACM.

18 • MILEPOST GCC: machine learning based research compiler

[12] S. Callanan, D. J. Dean, and E. Zadok. Extending

gcc with modular gimple optimizations. In

Proceedings of the GCC Developers’

Summit’2007, 2007.

[13] J. Cavazos, G. Fursin, F. Agakov, E. Bonilla,

M. O’Boyle, and O. Temam. Rapidly selecting

good compiler optimizations using performance

counters. In Proceedings of the 5th Annual

International Symposium on Code Generation

and Optimization (CGO), March 2007.

[14] K. Cooper, A. Grosul, T. Harvey, S. Reeves,

D. Subramanian, L. Torczon, and T. Waterman.

ACME: adaptive compilation made efﬁcient. In

Proceedings of the Conference on Languages,

Compilers, and Tools for Embedded Systems

(LCTES), 2005.

[15] K. Cooper, P. Schielke, and D. Subramanian.

Optimizing for reduced code space using genetic

algorithms. In Proceedings of the Conference on

Languages, Compilers, and Tools for Embedded

Systems (LCTES), pages 1–9, 1999.

[16] B. Elliston. Studying optimisation sequences in

the gnu compiler collection. Technical Report

ZITE8199, School of Information Technology

and Electical Engineering, Australian Defence

Force Academy, 2005.

[17] B. Franke, M. O’Boyle, J. Thomson, and

G. Fursin. Probabilistic source-level optimisation

of embedded programs. In Proceedings of the

Conference on Languages, Compilers, and Tools

for Embedded Systems (LCTES), 2005.

[18] G. Fursin. Iterative Compilation and Performance

Prediction for Numerical Applications. PhD

thesis, University of Edinburgh, United Kingdom,

2004.

[19] G. Fursin, J. Cavazos, M. O’Boyle, and

O. Temam. Midatasets: Creating the conditions

for a more realistic evaluation of iterative

optimization. In Proceedings of the International

Conference on High Performance Embedded

Architectures & Compilers (HiPEAC 2007),

January 2007.

[20] G. Fursin and A. Cohen. Building a practical

iterative interactive compiler. In 1st Workshop on

Statistical and Machine Learning Approaches

Applied to Architectures and Compilation

(SMART’07), colocated with HiPEAC 2007

conference, January 2007.

[21] G. Fursin, A. Cohen, M. O’Boyle, and O. Temam.

A practical method for quickly evaluating

program optimizations. In Proceedings of the

International Conference on High Performance

Embedded Architectures & Compilers (HiPEAC

2005), pages 29–46, November 2005.

[22] M. R. Guthaus, J. S. Ringenberg, D. Ernst, T. M.

Austin, T. Mudge, and R. B. Brown. Mibench: A

free, commercially representative embedded

benchmark suite. In IEEE 4th Annual Workshop

on Workload Characterization, Austin, TX,

December 2001.

[23] K. Heydemann and F. Bodin. Iterative

compilation for two antagonistic criteria:

Application to code size and performance. In

Proceedings of the 4th Workshop on

Optimizations for DSP and Embedded Systems,

colocated with CGO, 2006.

[24] K. Hoste and L. Eeckhout. Cole: Compiler

optimization level exploration. In Proceedings of

the International Symposium on Code Generation

and Optimization (CGO), 2008.

[25] P. Kulkarni, W. Zhao, H. Moon, K. Cho,

D. Whalley, J. Davidson, M. Bailey, Y. Paek, and

K. Gallivan. Finding effective optimization phase

sequences. In Proc. Languages, Compilers, and

Tools for Embedded Systems (LCTES), pages

12–23, 2003.

[26] P. Larrañaga and J. A. Lozano. Estimation of

Distribution Algorithms: A New Tool for

Evolutionary Computation. Kluwer Academic

Publishers, Norwell, MA, USA, 2001.

[27] F. Matteo and S. Johnson. FFTW: An adaptive

software architecture for the FFT. In Proceedings

of the IEEE International Conference on

Acoustics, Speech, and Signal Processing,

volume 3, pages 1381–1384, Seattle, WA, May

1998.

[28] A. Monsifrot, F. Bodin, and R. Quiniou. A

machine learning approach to automatic

production of compiler heuristics. In Proceedings

of the International Conference on Artiﬁcial

剩余141页未读，继续阅读

lichunqiang11

粉丝: 0
资源: 11

2008 GCC Developer Summit: Machine Learning, Compiler Enhancemen...

最新资源

2008 GCC Developer Summit: Machine Learning, Compiler Enhancemen...

自相关与互相关的C++实现

离线安装GCC与GCC-C++

gcc-4.8.0 gcc-4.9.0 gcc-5.1.0 gcc-6.1.0 gcc-7.1.0 gcc-8.1.0 gcc-9.1.0.7z

gcc-4.1.0.tar.gz_gcc for linux_gcc-2.95.3_gcc-4.1.0.tar.gz _linu

riscv-gcc-riscv-gcc-9.2.0-pulp.zip

gcc-g77-3.4.6-3.1.x86_64.zip_SD_g77_gcc-3.4.6.rpm_gcc-g77

gcc-toolset-10-gcc-c++-10.2.1-8.2.el8.aarch64.rpm

gcc-toolset-11-gcc-c++-11.2.1-1.1.el8.aarch64.rpm

gcc-toolset-9-gcc-gfortran-9.2.1-2.3.el8.aarch64.rpm

gcc-toolset-9-gcc-c++-9.2.1-2.3.el8.aarch64.rpm

最新资源