Performance improvements observed in Open XL C/C++ 2.2 for z/OS
With the newly released Open XL C/C++ 2.2 for z/OS, IBM reaffirms the commitment on combining the benefits and innovation from open-source Clang and LLVM technology framework with the leading-edge compiler technology and modern z/Architecture from IBM to deliver a best of breed C/C++ compiler.
This release supports the latest IBM z17 hardware. Open XL C/C++ 2.2 shows substantial performance improvement over z/OS XL C/C++ from the latest tests on IBM z17 ME1:
- Recompiling 64-bit compute intensive applications with IBM Open XL C/C++ 2.2 for z/OS reduces execution CPU time by 18% on average compared to compiling the applications with z/OS V2R4 XL C/C++ compiler on IBM z17 ME1 (Disclaimer: The performance improvements are based on internal IBM laboratory measurements on IBM z17 ME1. Results were obtained by taking a geomean average of reduction in execution CPU time across a set of compute intensive benchmarks that were compiled on z/OS Unix System Services with XPLINK linkage in 64- bit addressing mode. The -march=z17 and -O3 compiler options were used with the IBM Open XL C/C++ 2.2 for z/OS. The -qarch=13 and -O3 compiler options were used with the z/OS V2R4 XL C/C++. All benchmarks were executed on an IBM z/OS 3.2 LPAR with 1 CP and 32GB Central Storage on IBM z17 ME1. Performance results for specific applications will vary, depending on the source code, the compiler options specified, and other factors.)
- Recompiling 31-bit XPLINK compute intensive applications with IBM Open XL C/C++ 2.2 for z/OS reduces execution CPU time by 19% on average compared to compiling the applications with z/OS V2R4 XL C/C++ compiler on IBM z17 ME1 (Disclaimer: The performance improvements are based on internal IBM laboratory measurements on IBM z17 ME1. Results were obtained by taking a geomean average of reduction in execution CPU time across a set of compute intensive benchmarks that were compiled on z/OS Unix System Services with XPLINK linkage in 31-bit addressing mode. The -march=z17 and -O3 compiler options were used with the IBM Open XL C/C++ 2.2 for z/OS. The -qarch=13 and -O3 compiler options were used with the z/OS V2R4 XL C/C++. All benchmarks were executed on an IBM z/OS 3.2 LPAR with 1 CP and 32GB Central Storage on IBM z17 ME1. Performance results for specific applications will vary, depending on the source code, the compiler options specified, and other factors.)
- Recompiling 31-bit non-XPLINK compute intensive applications with IBM Open XL C/C++ 2.2 for z/OS reduces execution CPU time by 10% on average compared to compiling the applications with z/OS V2R4 XL C/C++ compiler on IBM z17 ME1 （Disclaimer: The performance improvements are based on internal IBM laboratory measurements on IBM z17 ME1. Results were obtained by taking a geomean average of reduction in execution CPU time across a set of compute intensive benchmarks that were compiled on z/OS Unix System Services with Standard C linkage in 31-bit addressing mode. The -march=z17 and -O3 compiler options were used with the IBM Open XL C/C++ 2.2 for z/OS. The -qarch=13 and -O3 compiler options were used with the z/OS V2R4 XL C/C++. All benchmarks were executed on an IBM z/OS 3.2 LPAR with 1 CP and 32GB Central Storage on IBM z17 ME1. Performance results for specific applications will vary, depending on the source code, the compiler options specified, and other factors.）
Attached is a chart format with the above performance claims:
