• -O3 -march=core-avx2 -mtune=corei7 -ffast-math for LLVM-3.3, -Ofast -march=core-avx2 -mtune=corei7 for the rest of compilers.
• The same set of options but -O4 instead of -O3 for LLVM-3.3, the same set of options plus -flto for the rest of compilers.

LLVM-3.3 does not support -Ofast. Therefore we use -O3 -ffast-math for it.

LLVM can not produce a correct code for 254.gap on any optimization level. Therefore -O0 was used instead of -O3 or -Ofast for all LLVM and GCC compilers for 254.gap.

LLVM-3.3 produces a wrong code for 256.bzip2 with -march=core-axv2. Therefore we use -march=corei7 instead for LLVM-3.3 and GCC-4.8.

The used machine is 3.4Ghz Intel Core i5-4670 (Haswell) with 16GB memory under Fedora Core20.

• Compilation Speed

  
  

  
  

• Code Size

  Here the code size (the corresponding text and data segment size) is.

  • SPECInt2000

    
    

• Generated Code Performace

  • SPECInt2000

    
    

    
    

Return to index page.