AMD Opteron processors code named Abu Dhabi and AMD Opteron processors code named Seoul

Compiler Options Quick Reference Guide

Open64
Latest release: 4.5.1, December 2011 http://developer.amd.com/open64 Architecture
Generate instructions specific to Abu Dhabi -march=bdver1

GCC
Latest release: 4.7, March 2012 http://gcc.gnu.org Architecture
Generate instructions specific -march=bdver2 to Abu Dhabi

Generate instructions for the -march=auto local machine

Optimization Levels
Disable all optimizations (default) Local optimizations Global optimizations Additional aggressive optimizations Maximize performance -O0 -O1 -O2 -O3 -Ofast

Optimization Levels
Disable all optimizations Local optimizations Global optimizations -O0 -O1 -O2

Additional aggressive optimi- -O3 zations Maximize performance -Ofast

Additional Optimizations
Schedule instructions AVX tuning Enable unrolling Disable prefetch instructions for loops Inline string operations Link to ACML 5.1.0 -fschedule-insns -fscheduleinsns2 -fsched-pressure -mprefer-avx128 -funroll-all-loops -fno-prefetch-loop-arrays -minline-all-stringops -L/opt/acml5.1.0/ gfortran64_fma4/lib -lacml -fopenmp -fprofile-generate, -fprofile-use -fno-tree-pre -ftree-vectorize

Additional Optimizations
Autoparallelization Feedback directed optimization Huge pages Interprocedural Analysis and Optimizations Link to ACML 5.1.0 Loop nest optimizations: vectorization, fission, fusion -apo -fb-create -fb-opt -HP -ipa -L/opt/acml5.1.0/ open64_64_fma4/lib -lacml -LNO:simd=n -LNO:fission=n -LNO:fusion=n Multicore scalability OpenMP Prefetch (disabled by default) -mso -mp -LNO:prefetch -LNO:prefetch_ahead Turn off partial redundancy elimination Vectorization

OpenMP Profile guided optimization

Floating point accuracy

Enable generation of code that follows IEEE arithmetic Enable faster, less precise math operations -mieee-fp

Floating point accuracy

Floating point accuracy -fp-accuracy

For more information, visit http://developer.amd.com

-ffast-math

Advanced Micro Devices One AMD Place P.O. Box 3453 Sunnyvale, CA 94088-3453

2012 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, AMD Opteron, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Linux is a registered trademark of Linus Torvalds. Other names are for informational purposes only and may be trademarks of their respective owners.

AMD Opteron processors code named Abu Dhabi and AMD Opteron processors code named Seoul

Compiler Options Quick Reference Guide

ICC
Latest release: 12U10, April 2012 http://software.intel.com Architecture
Generate instructions that run on Abu Dhabi -msse3 (avoid mAVX)

PGI
Latest release: 12.3, March 2012 http://www.pgroup.com Architecture
Generate instructions that run on Abu Dhabi -tp bulldozer

Optimization Levels
Disable all optimizations Speed optimization without code growth Enable optimization including vectorization Aggressive optimization Maximize performance -O0 -O1

Optimization Levels
Disable all optimizations Local optimization Global optimization -O2 Aggressive global optimization -O3 -fast Hoist guarded invariant floating point expressions Maximize performance -unroll-aggressive -no-prec-div -O4 -O0 -O1 -O2 -O3

Additional Optimizations
Aggressive unrolling Disable improved precision floating divides Enable vectorization Inter procedural Optimization Link to ACML to 5.1.0

-fast

Additional Optimizations
Huge pages Autoparallelize loops -Msmartalloc=huge -Mconcur -Mvect -Mipa=fast,inline -L/opt/acml5.1.0/ pgi64_fma4/lib -lacml -mp -Mvect=prefetch -Mpfi, -Mpfo -Munroll

-vec -ipo

Enable vectorization Interprocedural Optimization Link to ACML to 5.1.0

L/opt/acml5.1.0/ ifort64_fma4/lib lacml -openmp -opt-prefetch -prof-gen -prof-use

OpenMP Prefetch optimization Profile generated optimization

OpenMP Prefetch instructions Profile guided optimization Unroll loops

Use optimized header defini- use-intel-optimized-headers tions

Floating point accuracy

Generate relaxed precision code -Mfprelaxed Perform floating point operations in conformance with IEEE standard -Kieee

Floating point accuracy

Floating point accuracy -fp-model

For more information, visit http://developer.amd.com

Advanced Micro Devices One AMD Place P.O. Box 3453 Sunnyvale, CA 94088-3453

2012 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, AMD Opteron, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Linux is a registered trademark of Linus Torvalds. Other names are for informational purposes only and may be trademarks of their respective owners.