Kernels are nests of loops that process multidimensional arrays. Hamsa is a tool to extract kernels from programs and infer their asymptotic complexity. Figure 1 illustrates this:
For linear loop bounds, inferring the complexity is trivial. However, for non-linear bounds this task is more challenging. We intend to use the approach proposed by Gulwani et al. to implement it.
First of all, you need to inform the directory where LLVM is built. You can do this by changing the value of LLVM_BUILD_DIR at tool/setup.sh.
After configuring the build dir, you can setup the tool by running this command:
./setup.sh
After you've built the tool, you can run it as a Clang plugin. However the tool/run.sh script makes it easier to run. You can use it like this:
./run.sh inputFile outputFile outputFormat
You can also watch our tutorial video about how to setup the tool and run it inside a Docker container.