On Identifying Segments of Traces for Dynamic Compilation

Abstract

Typical computing systems based on general purpose processors (GPPs) are extended with coarse-grained reconfigurable arrays (CGRAs) to provide higher performance and/or energy savings. In order for applications to take advantage of these computing systems, efficient dynamic mapping techniques are required. Those dynamic mapping techniques will be responsible for automatically moving computations originally running in the GPP to the CGRA. The concept of dynamic compilation, widespread in the context of JIT compilation to GPPs, is receiving more attention by the reconfigurable computing community. This paper presents our approach to dynamically map computations to CGRAs coupled to a GPP. Specifically, we present the identification of large sequences of instructions, MegaBlocks, being executed in a GPP. These MegaBlocks are then mapped to the target CGRA. We evaluate the potential of the MegaBlocks over Basic Blocks and SuperBlocks to increase the IPC when targeting a CGRA and considering the execution of a number of representative benchmarks.

Transparent Trace-Based Binary Acceleration for Reconfigurable HW/SW Systems

“…One particular research topic has been the use of reconfigurable fabrics (e.g., FPGAs) to accelerate execution of general purpose applications in embedded scenarios [12], [13]. Some approaches include the Warp Processor [14], [15], the AMBER architecture [16], [17], the Configurable Compute Array (CCA) [18], [19], the DIM Reconfigurable System [20], [21], and the Megablock approach [4], [22].…”

“…Bispo and Cardoso [4], [22] propose the mapping of repetitive trace-based patterns of instructions, named Megablocks, in the context of dynamically moving instructions executing on a CPU to reconfigurable arrays. The objective of the Megablock is to address several difficulties related to mapping loops of sequential code to hardware, such as determination of the number of loop iterations, and the handling of jumps and branches in sequential code.…”

Techniques for Dynamically Mapping Computations to Coprocessors

“…The Megablock [6] was proposed in the context of dynamically moving large instruction sequences executing in a GPP to a coprocessor. The Megablock [6], represents repetitive patterns of instructions in the execution trace of a program.…”

“…The Megablock [6], represents repetitive patterns of instructions in the execution trace of a program. Each Megablock represents a single, common path of a loop, which can cross several Basic Blocks.…”

“…We have proposed techniques for efficient detection of Megablocks [6], and developed a hardware module which can detect the kind of patterns needed for identifying Megablocks.…”