Build and Install

build pSZ/cuSZ

pSZ Backend overview: cuSZ (CUDA) and more

alias	status	backend	toolkit used	cmake	host copmiler	device compiler	cpp std
cuSZ	released	CUDA	CUDA toolkit 11.3+	3.17	nvcc *	gcc 8+	c++14
hipSZ	in dev.	HIP	HIP toolkit 5.x	3.21	hipcc *	hipcc	c++17
dpSZ	in dev.	oneAPI	oneAPI toolkit 2023.x	3.23	icpx *	icpx	c++17

* Device-code compiler is specified by the toolkit.

CUDA GPU architectures (SM version) and representative GPUs.

NVIDIA CUDA architectures and names and representative GPUs are listed below. More details can be found at CUDA GPUs.

SM id	arch.	grade/segment	GPU product example
60	Pascal	HPC/ML	P100
70	Volta	HPC/ML	V100
75	Turing	consumer/professional	RTX 20?0, Quadro RTX ?000
80	Ampere	HPC/ML	A100
86	Ampere	consumer/professional	RTX 30?0, RTX A?000
89 *	Ada	consumer/professional	RTX 40?0, RTX ?000
90 *	Hopper	HPC/ML	H100

* as of CUDA 11.8

CUDA toolkit compatibility matrix

The table gives a quick view of toolchain compatibility (i.e., GCC or LLVM only serves as host compiler)

• corresponding to 1dd0b8a
• more reference: CUDA compilers, and CUDA architectures & gencode.

CUDA		11.0	11.1	11.2	11.3	11.4	11.5	11.6	11.7	11.8	12.0	12.1	12.2
GCC	8.5	^	^	^	ok	ok	ok	ok	ok	ok
	9.3	^	^	^	ok	ok	ok	ok	ok	ok	ok	ok	ok
	10.2	^	^	^	ok	ok	ok	ok	ok	ok
	11.2	!	!	!	!	!	ok	ok	ok	ok
	11.4	!	!	!	!	!						ok	ok

 # only accessible using spack
 ! GCC version too high for CUDA SDK

AMD ROCm architectures and representative GPUs.

• AMD GPU arch. code can be found in LLVM Target.

arch. id	GPU	arch.
gfx906	MI50	Vega
gfx908	MI100	CDNA
gfx90a	MI210	CDNA2
gfx90a	MI250	CDNA2
gfx90a	MI250X	CDNA2

CUDA backend

• NVIDIA GPU with CUDA 11.3 onward
• cmake 3.17 onward
• C++14 enabled compiler, GCC 8 onward

# Example architectures (";" to separate multiple SM versions)
# Install to [/path/to/install/dir]

git clone https://github.com/szcompressor/cuSZ.git cusz-latest
cd cusz-latest && mkdir build && cd build

cmake .. \
    -DPSZ_BACKEND=cuda \
    -DPSZ_BUILD_EXAMPLES=on \
    -DCMAKE_CUDA_ARCHITECTURES="75;80;86" \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_COLOR_DIAGNOSTICS=on \
    -DCMAKE_INSTALL_PREFIX=[/path/to/install/dir]
make -j
make install
# `ctest` to perform testing

Build for development

If possible, use ninja for faster build process.

git clone https://github.com/szcompressor/cuSZ.git cusz-dev
cd cusz-dev && mkdir build && cd build
cmake .. \
    -DPSZ_BACKEND=cuda \
    -DPSZ_BUILD_EXAMPLES=on \
    -DCMAKE_CUDA_ARCHITECTURES="75;80;86" \
    -DCMAKE_BUILD_TYPE=Debug \
    -DCMAKE_COLOR_DIAGNOSTICS=on \
    -DPSZLOG_ENABLE_ALL=off \
    -GNinja
ninja
# `ctest` to perform testing

HIP-ROCm backend (dev mode)

Build for development

If possible, use ninja for faster build process.

git clone https://github.com/szcompressor/cuSZ.git cusz-dev
cd cusz-dev && mkdir build && cd build
cmake .. \
    -DPSZ_BACKEND=hip \
    -DPSZ_BUILD_EXAMPLES=on \
    -DCMAKE_BUILD_TYPE=Debug \
    -DCMAKE_C_COMPILER=hipcc \
    -DCMAKE_CXX_COMPILER=hipcc \
    -DCMAKE_HIP_ARCHITECTURES="gfx90a" \
    -GNinja
ninja
# `ctest` to perform testing

DPCPP backend (dev mode)

Build for development

If possible, use ninja for faster build process. Global -DCMAKE_CXX_STANDARD=17 ensures the correct inclusion of <dpct/dpct.hpp>.

git clone https://github.com/szcompressor/cuSZ.git cusz-dev
cd cusz-dev && mkdir build && cd build
source /opt/intel/oneapi/setvars.sh --force 
cmake .. \
    -DPSZ_BACKEND=1api\
    -DPSZ_BUILD_EXAMPLES=on \
    -DCMAKE_BUILD_TYPE=Debug \
    -DCMAKE_C_COMPILER=icx \
    -DCMAKE_CXX_COMPILER=icpx \
    -DCMAKE_CXX_STANDARD=17 \
    -DCMAKE_CXX_STANDARD_REQUIRED=ON \
    -GNinja
ninja
# `ctest` to perform testing