This is a framework for compiling programs with instrumentation such that they can call user-specified code at fixed intervals during the program's run time.
a. To build Compiler Interrupt transformation pass, and its helper library, run: cd src; make; cd - (The generated libraries will be exported to ./lib path) b. For creating experiment related setup: cd src; ./setup.sh; cd -
a. Build the example: cd example; make clean; make; cd - b. Run the program as it is: cd example; ./gcc_demo; cd - c. Run the program with interrupt_handler(), defined in example/demo.c, called every 1000 instructions: cd example; ./ci_llvm_demo; cd - Expect to see a set of "Compiler interrupt called" print from interrupt_handler() in demo.c.
a. An analysis & transformation pass (CompilerInterrupt.so) to be run with LLVM's opt at the time of compilation b. An intermediate library (libci.so) to be linked with the application - provides the user with API to register interrupt handlers, enable or disable them
The sources for both are in src/ directory. libci.so is a simple library built from ci_lib.* files and the api names are self-explanatory. The Compiler Interrupt pass is built from CompilerInterrupt.cpp, which is complex, and the code starts at runOnModule() function. This file has explanatory comments, but they are probably not at a level (yet) that can be easily deciphered by someone unacquainted with this.
a. MTCP experiments (path: benchmarks/mtcp) b. Shenango experiments (path: benchmarks/shenango, benchmarks/cpuminer-multi/) c. Server delegation experiments (path: benchmarks/server_delegation) d. Accuracy & Overhead experiments (path: benchmarks/accuracy_and_overhead)
For assistance in understanding/using CI, or reporting bugs, mail to: nbasu4@uic.edu