k-Wave

This version of the k-Wave Toolbox has been modified by bobqianic. The modifications include adding support for the latest hardware and fixing bugs. These modifications are also distributed under the terms of the GNU Lesser General Public License as specified above.

Important

If you are an Imperial HPC user, please see this

The following guidance is for non-HPC users

Install Guide (Linux)

1. Install libraries

Requires at least 8 GiB of RAM (Support Ubuntu, Debian, CentOS)

sudo curl -L https://raw.githubusercontent.com/bobqianic/k-Wave/main/install_lib.sh | bash

The default install location is /usr/local. If you want to install to a custom location, you can use the following command:

sudo curl -L https://raw.githubusercontent.com/bobqianic/k-Wave/main/install_lib.sh HDF5_DIR ZLIB_DIR SZIP_DIR | bash

2. Install k-Wave

Download the latest source code from this repo.

wget https://github.com/bobqianic/k-Wave/archive/refs/heads/main.zip

Unzip the souce code.

unzip ./main.zip

Change current directory to kspaceFirstOrder-CUDA

cd ./k-Wave-main/kspaceFirstOrder-CUDA

Modify the CUDA_DIR, HDF5_DIR, ZLIB_DIR, and SZIP_DIR in the Makefile to point to your library installation locations.

CUDA_DIR = "/usr/local/cuda-12.1"
HDF5_DIR = "/usr/local/hdf5"
ZLIB_DIR = "/usr/local/zlib"
SZIP_DIR = "/usr/local/szip"

Run make to compile k-Wave.

make -j $(nproc)

Install Guide (Windows)

1. Install libraries & IDE

1. Download Visual Studio 2022 (https://visualstudio.microsoft.com/vs/) and install the Desktop development with C++ workload.

2. Download and unzip VCPKG source code (https://github.com/microsoft/vcpkg). Run the bootstrap-vcpkg.bat in the folder to compile vcpkg and generate the executable file.

3. Run ./vcpkg.exe install --triplet x64-windows hdf5 zlib szip in PowerShell to install libraries.

2. Install k-Wave

1. Modify kspaceFirstOrder-CUDA.vcxproj. Replace "$(VCTargetsPath)\BuildCustomizations\CUDA 12.4.props" and "$(VCTargetsPath)\BuildCustomizations\CUDA 12.4.targets" to your CUDA Toolkit version.

For example, if your CUDA Toolkit version is 12.6.X:

"$(VCTargetsPath)\BuildCustomizations\CUDA 12.4.props" -> "$(VCTargetsPath)\BuildCustomizations\CUDA 12.6.props"
"$(VCTargetsPath)\BuildCustomizations\CUDA 12.4.targets" -> "$(VCTargetsPath)\BuildCustomizations\CUDA 12.6.targets"

2. Add new CUDA architecture for you GPU to <CodeGeneration> section if its a newer GPU (B100, B200, RTX5090, etc.)

<CodeGeneration>compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;compute_89,sm_89;compute_90,sm_90;</CodeGen

3. If you encounter error LNK1811: cannot open input file 'libszip.lib'/'libhdf5.lib'..., please remove them from Properties->Linker->Input->Additional Dependencies.

C++ vs MATLAB performance