Check out the new USENIX Web site. next up previous
Next: Related Work Up: Optimizations for the FastVM Previous: Floating-Point precision toggle

Overall performance improvement

Figure 13 shows the complete picture of the optimized code generator versus a non-optimized naive code generator.

Figure: Benchmark speedup of optimized version using register allocation, floating-point optimization, and method inlining compared to the non-optimized implementation of the FastVM on x86.

Benchmarks profiting most from the optimizations are compress and mpegaudio. Each of them improves by a different optimization technique, which demonstrates the necessity of multiple optimization passes. The optimization techniques are simple and inexpensive and can be widely applied.