Specifications

Chapter 5 Peak Options for SPEC
®
-CPU Benchmark Programs 57
Compiler Usage Guidelines for AMD64 Platforms
32035 Rev. 3.22 November 2007
5.1.2 Peak Command-line Options
The table below specifies the best-known peak switches for various benchmarks in the SPECcpu2006
suite for the 64-bit PGI Release 7.1 compilers for Linux® on AMD Athlon™ 64 processor based
platforms and AMD Opteron™ processor-based platforms.
Table 10. Best-Known Peak Switches for the 64-Bit PGI Compilers for Linux
®
Application Area Benchmark Language Best Known Peak Switches
CINT2006
3
Programming
Language
400.perlbench ANSI C
pgcc -w -fast -O4 -Mfprelaxed -Msmartalloc=huge:448
-Mnounroll -Mpfi(pass 1) -Mpfo(pass 2)
-Mipa=inline(pass 2) -tp barcelona-64
-DSPEC_CPU_LP64 -DSPEC_CPU_LINUX_X64
Compression 401.bzip2 ANSI C
pgcc -w -fast -O4 -Msmartalloc=huge:448
-Mpfi(pass 1) -Mpfo(pass 2) -tp barcelona-64 -Mpfo
-DSPEC_CPU_LP64
GNU C compiler 403.gcc C
pgcc -w -fastsse -Mfprelaxed -Msmartalloc=huge:448
-Mipa=fast, inline -tp barcelona
Combinational
Optimization
429.mcf
ANSI C
1
pgcc -w -fastsse -Mipa=fast, inline:1
-Msmartalloc=huge:420 -tp barcelona
Artificial
Intelligence: Go
445.gobmk C
pgcc -w -fast -O4 -Msmartalloc=huge:448 -Mfprelaxed
-Mnovect -tp barcelona-64 -Mpfi(pass 1)
-Mpfo(pass 2) -Mipa=fast(pass 2)
-DSPEC_CPU_LP64
Search Gene
Sequence
456.hmmer C
pgcc -w -fast -Msmartalloc=huge:448 -Mfprelaxed
-Msafeptr -Mipa=const, ptr, arg -tp barcelona-64
-DSPEC_CPU_LP64
Artificial
Intelligence:
Chess
458.sjeng ANSI C
pgcc -w -fast -Msmartalloc=huge:448 -Mfprelaxed
-tp barcelona-64 -Mpfi(pass 1) -Mpfo(pass 2)
-Mipa=fast, inline:1, noarg(pass 2)
-DSPEC_CPU_LP64
Physics / Quantum
Computing
462.libquantum “C99”
pgcc -w -fast -Mfprelaxed -Msmartalloc=huge:448
-Munroll=m:4 -Mipa=fast, inline, noarg
-DSPEC_CPU_LP64 -DSPEC_CPU_LINUX
Video
compression
464.h264ref C Use base binaries and/or base results for peak.
Discrete Event
Simulation
471.omnetpp C++ Use base binaries and/or base results for peak.
Path-finding
Algorithms
473.astar C++ Use base binaries and/or base results for peak.
Notes:
1. Mathematical library (libm) required
2. Boost Library required
3. Smartheap libraries utilized. If the Smartheap libraries are not loaded, xalancbmk performs better with the
-Msmartalloc=huge:160 option.