This repository is a collection of applications ported for HammerBlade. These benchmarks are meant to be optimized for execution on HB using HammerBlade's CUDA-style runtime.
This repository is meant to be cloned into bsg_replicant which is meant to be cloned into bsg_bladerunner.
For initial setup do the following:
- Clone bsg_bladerunner and follow the setup instructions.
- From bsg_bladerunner,
cd bsg_replicant/examples; git clone [email protected]:bespoke-silicon-group/hb_hammerbench
Each application is meant to be standalone and will have a README explaining how to run it. All applications live in the apps/ directory.
Currently, go into each benchmark and follow the instructions to run it until you have run them all.
If you want an easy way to setup a parameterizable testbench, a helper script exists in mk/make_testbench.mk
.
Here's how to use it:
cd apps/
make -f ../mk/make_testbench.mk TESTBENCH="benchmark-name" PARAMETERS="first-parameter [second-parameter etc...]"
This will generate a template testbench from which multiple program executions can be run. You will need to do the following:
cd apps/[benchmark-name]
- Edit
template.mk
to use the test parameters (these are included in the generatedparameters.mk
) - Edit main.c and kernel.cpp to implement your benchmark.
2022-07-26: 11 apps present, none working.