DYNAMICALLY PARALLELIZING SEQUENTIAL JAVA PROGRAMS AS INSTRUCTION-LEVEL PARALLELISM WITH A SINGLE THREAD OF CONTROL APPROACHES ITS PERFORMANCE LIMITS, DESIGNERS MUST FIND OTHER ARCHITECTURAL IMPROVEMENTS TO SPEED UP PROGRAM EXECUTION. THE JRPM SYSTEM TAKE

The quest to automatically parallelize general-purpose programs is a longstanding problem in the microarchitecture community. Solving this problem is critical to continued architectural improvements in microprocessor performance, as instructionlevel parallelism with a single thread of control approaches its performance limits. We’ve developed a solution that relies on four key technologies: chip multiprocessors, threadlevel speculation, dynamic compilation, and hardware-based profiling. We can combine these technologies and manage them inside a Java virtual machine (JVM). The resulting system, the Java runtime parallelizing machin