Add 'make test' target to build and run correctness tests

[v1kko]

Add a top-level test target that builds the library and the test programs against it, then runs the GPU-vs-CPU correctness tests for the 4th-order, GRAPE5 (2nd-order) and 6th-order integrators. The target dispatches to the CUDA or OpenCL test Makefile based on the selected BACKEND, and sets LD_LIBRARY_PATH so the freshly built shared libraries are found at runtime. A build-tests target is also provided to compile the tests without running them, and clean now also cleans the tests directory.

Fix two pre-existing bugs in tests/Makefile_ocl that prevented the OpenCL tests from building/running: link against libsapporo2 (not the old libsapporo name) and symlink kernels from src/OpenCL.

[HannoSpreeuw]

Some progress report on this.

Commit c1292cb completes both make all BACKEND=CUDA and make test BACKEND=CUDA.
This is on node021 on DAS6 after module load cuda12.6/toolkit/12.6.

make all BACKEND=OpenCL also completes without error after module load opencl-nvidia/12.3.

make test BACKEND=OpenCL fails with

Cuda driver error <CUDA_ERROR_INVALID_IMAGE> in file 'src/cudadev.h' in line 763.
test_gravity_block_ocl: src/cudadev.h:763: void dev::kernel::load_source(const char*, const char*, const char*): Assertion `false' failed.
/bin/sh: line 1: 1803410 Aborted                 (core dumped)

when running test_gravity_block_ocl, i.e. after building the eight executables in the tests folder.

Commit 2bcbd49 does not fix that error, so I reverted it (locally).

[HannoSpreeuw]

On a different machine, with NVIDIA GPU and CUDA 13.1, after branching off master and git cherry-pick c1292cb, the error from make test BACKEND=OpenCL is slightly different:

=== Running test_gravity_block_ocl ===
 n = 1024
./test_gravity_block_ocl is using: n: 1024	Devices: 0	IntegrationOrder: 1	IntegrationPrecision: 1	File: OpenCL/kernels4th.cl
Integration order used: 1 (0=GRAPE5, 1=4th, 2=6th, 3=8th)
Integration precision used: 1 (0=FLOAT, 1 = DOUBLESINGLE, 2=DOUBLE)
Getting list of OpenCL devices ...
oclSafeCall() Runtime API error in file <src/ocldev.h>, line 251 : Unknown OpenCL error
.
test_gravity_block_ocl: src/ocldev.h:179: void dev::__oclsafeCall(cl_int, const char*, int): Assertion `false' failed.
make: *** [Makefile:241: test] Error 1

[HannoSpreeuw]

My AI tool insists that this error occurs because OpenCL packages have not been installed completely and/or my environment is not setup to point to the OpenCL libraries.

Indeed, I could bypass that error through conda install pocl, next make test BACKEND=OpenCL will yield a (slightly) different error

=== Running test_gravity_block_ocl ===
 n = 1024
./test_gravity_block_ocl is using: n: 1024	Devices: 0	IntegrationOrder: 1	IntegrationPrecision: 1	File: OpenCL/kernels4th.cl
Integration order used: 1 (0=GRAPE5, 1=4th, 2=6th, 3=8th)
Integration precision used: 1 (0=FLOAT, 1 = DOUBLESINGLE, 2=DOUBLE)
Getting list of OpenCL devices ...
 0: Portable Computing Language
Using platform 0 
oclSafeCall() Runtime API error in file <src/ocldev.h>, line 263 : Device not found.
.
test_gravity_block_ocl: src/ocldev.h:179: void dev::__oclsafeCall(cl_int, const char*, int): Assertion `false' failed.
make: *** [Makefile:241: test] Error 1

[v1kko]

Right, that one was fixed by claude by temporarily patching the code and setting the Opencl target to CPU, it is hardcoded somewhere to be GPU, and pocl is a cpu-library...

[HannoSpreeuw]

I also added pocl through my OS package manager.
Fresh start: make clean.
make all BACKEND=OpenCL; make test BACKEND=OpenCL now yields another slightly different error:

=== Running test_gravity_block_ocl ===
 n = 1024
./test_gravity_block_ocl is using: n: 1024	Devices: 0	IntegrationOrder: 1	IntegrationPrecision: 1	File: OpenCL/kernels4th.cl
Integration order used: 1 (0=GRAPE5, 1=4th, 2=6th, 3=8th)
Integration precision used: 1 (0=FLOAT, 1 = DOUBLESINGLE, 2=DOUBLE)
Getting list of OpenCL devices ...
 0: Portable Computing Language
Using platform 0 
No GPU devices found on platform, trying CPU devices...
Found 1 suitable devices: 
 0: pthread-AMD Ryzen 9 7900X 12-Core Processor	Vendor: AuthenticAMD
Number of cpus available: 24
Number of gpus available: 1
integrationOrder : 1
Getting list of OpenCL devices ...
 0: Portable Computing Language
Using platform 0 
No GPU devices found on platform, trying CPU devices...
Found 1 suitable devices: 
 0: pthread-AMD Ryzen 9 7900X 12-Core Processor	Vendor: AuthenticAMD
Using device: 0
Device has: 24 	 multiprocessors 
Using  2 blocks per multi-processor for a total of : 48
Loading file:  OpenCL/kernels4th.cl 
Opening kernel file: OpenCL/kernels4th.cl
Compilation of the source file failed: OpenCL/kernels4th.cl 
Compiler output: 
 error: unknown target CPU 'generic'
 
test_gravity_block_ocl: src/ocldev.h:660: void dev::kernel::print_compiler_output(): Assertion `false' failed.
make: *** [Makefile:241: test] Error 1

My AI tool (GH Copilot) suggests that I should install an OpenCL runtime for my hardware (AMD CPU, NVIDIA GPU).

[HannoSpreeuw]

With a fresh Python 3.14 Conda env and everything fresh: make clean; make all BACKEND=OpenCL; make test BACKEND=OpenCL:

=== Running test_gravity_block_ocl ===
 n = 1024
./test_gravity_block_ocl is using: n: 1024	Devices: 0	IntegrationOrder: 1	IntegrationPrecision: 1	File: OpenCL/kernels4th.cl
Integration order used: 1 (0=GRAPE5, 1=4th, 2=6th, 3=8th)
Integration precision used: 1 (0=FLOAT, 1 = DOUBLESINGLE, 2=DOUBLE)
Getting list of OpenCL devices ...
 0: AMD Accelerated Parallel Processing
 1: NVIDIA CUDA
Using platform 0 
Found 1 suitable devices: 
 0: gfx1036	Vendor: Advanced Micro Devices, Inc.
Number of cpus available: 24
Number of gpus available: 1
integrationOrder : 1
Getting list of OpenCL devices ...
 0: AMD Accelerated Parallel Processing
 1: NVIDIA CUDA
Using platform 0 
Found 1 suitable devices: 
 0: gfx1036	Vendor: Advanced Micro Devices, Inc.
Using device: 0
Device has: 1 	 multiprocessors 
Using  2 blocks per multi-processor for a total of : 2
Loading file:  OpenCL/kernels4th.cl 
Opening kernel file: OpenCL/kernels4th.cl
Loading file:  OpenCL/kernels4th.cl 
Opening kernel file: OpenCL/kernels4th.cl
Loading file:  OpenCL/kernels4th.cl 
Opening kernel file: OpenCL/kernels4th.cl
Loading file:  OpenCL/kernels4th.cl 
Opening kernel file: OpenCL/kernels4th.cl
Kernel files found .. building compute kernels! 
Creating kernel dev_copy_particles 
Maximum work group size: 256 Optimal work group multiple: 32 
Creating kernel dev_predictor 
Maximum work group size: 256 Optimal work group multiple: 32 
Creating kernel dev_evaluate_gravity_fourth_DS 
Maximum work group size: 256 Optimal work group multiple: 32 
Creating kernel dev_reset_buffers 
Maximum work group size: 256 Optimal work group multiple: 32 
oclSafeCall() Runtime API error in file <src/ocldev.h>, line 881 : Invalid work group size
. Kernel name: dev_evaluate_gravity_fourth_DS
test_gravity_block_ocl: src/ocldev.h:187: void dev::__oclsafeCallKernel(cl_int, const char*, const char*, int): Assertion `false' failed.
make: *** [Makefile:241: test] Fout 1

Invalid work group size is something that has to do with the code, not with with my env.

And it is using platform 0 (CPU) instead of my GPU.

[HannoSpreeuw]

And it is using platform 0 (CPU) instead of my GPU.

Got that fixed to use my GPU and it runs at 100% utilisation.

Unfortunately, it is still running, way longer than with the CUDA backend (2s or so).