Sun responds to compiler conundrum

We saw your article titled "Sun benchmark results pose big questions" in the Inquirer.com (sic) last Monday, November 25 (sic), and want to clarify some issues.

First of all, we were disappointed that the article's comments on our approach to compilers used for SPEC benchmarking appeared to be nothing more than speculation. We really wish that you had contacted us before publishing this material.

We would like to assure you and your readers that we performed the benchmarks using a pre-release version of what will be a publicly available compiler, the Sun Forte Developer 7 software. When systems integrating the UltraSPARC III Cu 1050 processor/Forte Developer 7 compiler combination ship to customers, we are confident that they will get the full measure of performance benefits in the real world as we communicated to SPEC.org.

Sun's Forte 7 SPARC compiler targets real world applications of interest to our customers. The UltraSPARC Cu 1050 Processor Sun Blade SPEC numbers we reported are the result of numerous improvements, in hardware as well as compiler optimizations. On the compiler side, extensive interprocedural analysis is followed by optimizations such as inlining, software pipelining, prefetching, and alias analysis to mention just a few. The result is efficient code.

Our policy is to leave backroom tricks and irreproduceable "benchmark-optimized" compiler technology to others. We stand by our results and look forward to delivering benefits derived from them to our customers.

In the meantime, we'd be happy to talk with you about the specifics of procedures and technologies we employ to keep the UltraSPARC processor at the forefront of workstation/server computing.

Ahmad Zandi
Technical Marketing Group Manager
Sun Microsystems
Processor Products Group