Abstract: Dynamic optimization relies on runtime profile information to improvethe performance of program execution. Traditional profiling techniquesincur significant overhead and are not suitable for dynamicoptimization. In this paper, a new profiling technique is proposed, thatincorporates the strength of both software and hardware to achievenear-zero overhead profiling. The compiler passes profiling requests asa few bits of information in branch instructions to the hardware, andthe processor executes profiling operations asynchronously in availablefree slots or on dedicated hardware. The compiler instrumentation ofthis technique is implemented using an Itanium research compiler.The result shows that the accurate block profiling incurs very littleoverhead to the user program in terms of the program scheduling cycles.For example, the average overhead is 0.6% for the SPECint95 benchmarks.The hardware support required for the new profiling is practical.The technique is extended to collect edge profiles for continuous phasetransition detection. It is believed that the hardware-softwarecollaborative scheme will enable many profile-driven dynamicoptimizations for EPIC processors such as the Itanium processors.