Batched Kernels with Memory Allocations (BKMA)

This code implements optimal parallel loops for BKMA-like algorithm. Two use case implementations are available in the code, a 1D semi-lagrangian advection operator and a 1D Convolution operator.

1D Convolution operator

Implement a 1D convolution operator in-place using BKMA strategies.

Lagrangian Advection

Implements a 1D advection operator inside a multidimensionnal space. It implements a semi-Lagrangian scheme using the SYCL 2020 progamming models.

To reproduce the benchmark, follow the benchmark README.md instructions.

SYCL Implementations

The algorithm is implemented in various ways using different SYCL constructs. It requires local memory allocation via the local accessor. The implementations are in the src/core directory.

• BasicRange (out of place), no hierarchical parallelism involved
• NDRange (in-place), work-groups and work-items, direct mapping of the problem dimensions
• AdaptiveWg (in-place or out-of-place), optimized work-group sizes, streaming, optimal local memory usage

Build the project:

You can use the compile.sh script to compile for various hardware and sycl-implementations. For multi-device compilation flows, build the project manually. Use the ./compile.sh --help to see the options.

Example usage:

#generate the advection executable and advection.ini file
./compile.sh --hw cpu --sycl dpcpp 

#create build_dpcpp_a100 folder with benchmarks
./compile.sh --hw a100 --sycl dpcpp --benchmark_DIR=/path/to/google/benchmark/build 

#create build_acpp_mi300 folder with tests and execute tests
./compile.sh --hw mi300 --sycl acpp --build-tests --run-tests 

Manually build the project

Flags varies on the SYCL implementation you are using.

• For DPC++, add the correct flags via the -DDPCPP_FSYCL_TARGETS cmake variable.
• For acpp, export the ACPP_TARGETS environment variable before compiling

Run the executable

1. Set the runtime parameters in build/src/<conv1d|advection>.ini
2. Run the executable build/src/advection/<conv1d|advection>

Credits

The advection operator in this code is largely inspired by the vlp4D code.