Stack size estimation on machine-independent intermediate code for OpenCL kernels

Abstract

Stack size is an important factor in the mapping decision when dealing with embedded heterogeneous architectures, where fast memory is a scarce resource. Trying to map a kernel onto a device with insufficient memory may lead to reduced performance or even failure to run the kernel. OpenCL kernels are often compiled just-intime, starting from the source code or an intermediate machineindependent representation. Precise stack size information, however, is only available in machine-dependent code. We provide a method for computing the stack size with sufficient accuracy on machine-independent code, given knowledge of the target ABI and register file architecture. This method can be applied to make mapping decisions early, thus avoiding to compile multiple times the code for each possible accelerator in a complex embedded heterogeneous system.

Automatic Optimizer Generation Method Based on Location and Context Information to Improve Mobile Services

“…Box 2 shows the optimization pattern of the SIL instruction defined according to the pattern specification grammar outlined in Box 1. The matching information of the optimization code for the parameter reduction, increase/decrease operation, and control instruction and the related conditions are provided in [8,9]. Box 3 presents part of the DFA optimization table generated by SOG.…”

“…It creates tokens, which are syntactically significant minimum units. In other words, it receives a pattern description as an input and generates a series of tokens [8,9]. The parser receives the token output from the scanner, checks for errors, and creates sentences according to the program syntax to generate an abstract syntax tree (AST).…”