GitLab

It may contain a garbage value, and if it is large then
updateNeighborListSize does not force reorder atoms after 25 steps
in extremal cases.

* Use 768 instead of 384 threads in generic kernels

* Use 1536 instead of 1024 threads in force kernels.

- Appear to be copy and pasted from getPositions and were not removed

* Flushing optimization

* Remove unnecessary checks

* Profiling of OpenCL kernels

* Minor improvements to profiling

* Use VkFFT for OpenCL

* Updated comments for OpenCLFFT3D

This initialization is needed to prevent segfault during object destruction in certain circumstances (e.g., when a `Force` is changed before calling `Context.reinitialize()` and this change causes the corresponding `ForceImpl` to throw an exception).

* Support kernel files with extensions of any length (like .hip)

* Do not allow to replace symbols in single-line comments

* Add OPENMM_BUILD_COMMON CMake option

It allows to build and install common platform files even if
CUDA or OpenCL platforms are not built.
This is required for HIP platform (openmm-hip) if ROCm OpenCL
packages are not installed.

* Add an option for Python wrapper to install into user packages

OPENMM_PYTHON_USER_INSTALL is OFF be default.

* Support FFT backends in Amoeba plugin

The HIP platform supports FFT backends, this commit moves
findLegalFFTDimension to ComputeContext, so platforms can have their own
implementations.

* Compatibility for common platform w/ new HIP platform

* Do not use volatile with private and local AtomData parameters on HIP

The generated code is not optimal, for example, the compiler generates
flat_load instructions instead of ds_read.

* Tune launch bounds for PME grid-related kernels and add WA for RDNA

Force the compiler to use all registers for gridSpreadCharge and
gridInterpolateForce by limiting max waves per EU to 1 on CDNA GPUs,
RDNA GPUs work better without it.

* Optimize atom data structs in GBSA and Amoeba on HIP

Manually rearrange fields, add paddings and force alignments to
have faster accesses to shared memory: ds_read and ds_write may
work slower if addresses are not aligned by 16 bytes.
Co-authored-by: Anton Gorenko <anton@streamhpc.com>
Co-authored-by: Nick Curtis <nicholas.curtis@amd.com>

* Use PocketFFT instead of FFTW

* Minor cleanup

* Use PocketFFT instead of fftpack for reference platform

* Remove FFTW as a dependency

* Converted a test case to use PocketFFT

* Fixed an incorrect comment

* add fix to Prevent Windows from defining macros that interfere with other code

* add fix to the tippy top of the file

Without this fix, we see cases in which not all work-items in a thread group end up hitting the same number of barriers, which leads to a hang in OpenCL GPU execution.

* Added FAQ links to error messages

* Added missing Windows export

* Vectorize NonbondedForce with no cutoff

* Vectorize CustomNonbondedForce with no cutoff

* Memory efficient dense neighbor list

* Fixed errors

* Began vectorizing CustomNonbondedForce

* Refactored CpuCustomNonbondedForce to support multiple vector sizes

* AVX implementation of CpuCustomNonbondedForce

* Fixed compilation errors

It allows to use a faster float-to-int64 in the HIP platform.

* Minor optimizations to computing single pairs

* Adjusted MAX_BITS_FOR_PAIRS on Ampere