GitLab

* Compile with -munsafe-fp-atomics to enable fast hardware f32 atomic
  add on global memory on pre-MI100 GPUs;
* Use fixed point charge spreading on other GPUs, otherwise float atomic
  add will be compiled as a slow CAS loop;
* Tune block sizes, use executeKernelFlat;
* Tune launch bounds of PME grid-related kernels: force the compiler to
  use all registers by limiting max waves per EU to 1.

Optimize findBlocksWithInteractions

* Replace volatile shared mem accesses with shuffles;
* Add NUM_TILES_IN_BATCH for processing block1 by multiple warps
  (for small systems);
* Cherry-pick missing changes from .cu;
* Tune MAX_BITS_FOR_PAIRS depending on device and the system size;
* Store single pairs immediately (if there are any), this allows not to
  store flags to shared memory and filter buffer and flagsBuffer after
  saving single pairs;
* Use fma explicitly and sign bit for better device code;
* Use CDNA's MFMA with singe/mixed precision;
* On CDNA the coarse grained stage processes warpSize blocks for
  one block1, the fine grained stage checks atoms of two block2 vs atoms
  of the same block1, singlePairs and interactingAtoms are also stored
  by warps, not half-warps;

Optimize findBlockBounds

* Use shuffles;
* Use executeKernelFlat;
* Process 2 tiles per warp 64 on CDNA;
* Use more uniformly distributed keys when sorting blocks;

Use compareInt2LargeSIMD when tile size < SIMD width

Fix exclusion tiles sorting on AMD CDNA (64 threads per wave)

    The nonbonded kernel uses USE_NEIGHBOR_LIST (useNeighborList)
    so host code also must check it instead of useCutoff.

    See also https://github.com/openmm/openmm/issues/3462

* All AMD GPUs support shuffle, double precision and 64-bit int atomics;
* Remove unused code: !ENABLE_SHUFFLE code paths in nonbonded.hip;
* Use intrinsics in single-precision;
* Use realToFixedPoint (faster float32-to-int64);
* Remove shared atomIndices, use shuffles;
* Check early if atoms are in the cutoff range, sometimes all lanes in
  a warp can skip computations, single pairs can also skip useless
  atomics with zero values;
* Remove volatile skipTiles access, use shuffles;
* Distribute work for warps in a strided order;
* Skip warps that may be still busy in the first loop;
* Unify conditions for excluded atoms with `includeInteraction`;
* Move multiprocessors to HipContext;
* Increase number of warps for computeNonbonded;
* Disable packed math for >=MI200 (it affects performance of some
  kernels like computeGKForces of amoebagk);
* Remove defaultOptimizationOptions and createModule's optimizationFlags
  as they are never used;
* Support -save-temps.

* Compile kernels with max block size of 256 threads:
  The default hipcc behavior since ROCm 4.2 is to compile kernels
  with 1024 threads unless __launch_bounds__ is specified. This
  significantly increases register pressure especially in heavy kernels
  (double precision, for example), requiring register spilling;
* Optimize computeRange by using multiple blocks for reduction;
* Use blocks of 1024 threads for computeBucketPositions - it is executed
  as a single work group so larger block size is faster;
* Sort up-to lenghtNextPow2 instead of blockDim.x (faster for short
  buckets);
* Optimize sortShortList2;
* Optimize sortBuckets with bit instructions;
* Decrease bucket size for non-uniform sorting: too many buckets may
  have sizes too large to sort in shared memory;
* Add more sizes in tests.

* Remove setting of link libraries, include and link dirs and compile
  flags for each target, instead let Cmake deal with them by linking the
  main library to hip::host hiprtc::hiprtc hip::hipfft;
* Fix: custom command without ADD_CUSTOM_TARGET and ADD_DEPENDENCIES is
  executed for both static and shared targets;
* Remove IF(APPLE) parts.

Port changes in CUDA backend to HIP

Fix a warning about arithmetic operations on void* in HipArray::uploadSubArray

Fix "Error Initializing context ROCm 5.3.0"

    https://github.com/StreamHPC/openmm-hip/issues/3


    hipDeviceSetCacheConfig returns hipErrorNotSupported on 5.3
Co-authored-by: Nick Curtis <nicholas.curtis@amd.com>

* ATMForce reorders inner contexts for better performance

* Fixed obsolete comments

* Minor optimization to validating exclusions

* Optimizations to findMoleculeGroups()

* Improved performance of CustomHbondForce on large systems

* Fixed CUDA compilation errors

* Made LangevinIntegrator identical to LangevinMiddleIntegrator

* Removed unused code

* VariableLangevinIntegrator uses LFMiddle

* Reference platform supports nested virtual sites

* Common platform supports nested virtual sites

* Fixed force distribution from nested virtual sites

* Fixed test failures

* Common implementation of BondedUtilities

* Common implementation of UpdateStateDataKernel

* Improved sorting of blocks when building neighbor list

* Improved block sorting for OpenCL

* Made sort keys more evenly distributed

* Fixed some incorrect comments

* Fixed some incorrect comments

* Fixed typo

Closes #2986

* Enable flush on Windows

- Implements perm 5 from https://github.com/openmm/openmm/issues/3937#issuecomment-1413872621

Co-Authored-By: Philip Turner <philipturner.AR@gmail.com>

* Add brackets for clarification
Co-authored-by: Philip Turner <philipturner.AR@gmail.com>

* Make this optimization only apply to AMD GPUs

* Switch to perm 1

- Flush before call to computeNonbonded since it works well on Windows and Linux

* Update OpenCLNonbondedUtilities.cpp

* Perm 4 is now significantly faster on Windows

* Use isAMD

* Fix indentation

* Fix missed variable

* Remove Mac check

* Remove isAMD out of Mac code

* Consistent (lack of) brackets style

---------
Co-authored-by: Philip Turner <philipturner.AR@gmail.com>

WIP - looking for a way to optimise performance of creating contexts by removing temporary arrays (and their associated mallocs/frees) (#4261)

* Suggesting a "haveSameParameters" function for CustomNonbondedForce which could be
used to avoid creating temporary copies of arrays when testing if particles are
the same.

Also updating "getParticleParameters" so that it re-uses the memory of the
passed vector argument, rather than deallocating and reallocating it
via a copy.

* Revert "Suggesting a "haveSameParameters" function for CustomNonbondedForce which could be"

This reverts commit e80ec2d2e9981abb90711636bf3a78d0c49e43fc.

* Moved to `thread_local static` as suggested to prevent new vector allocations on each function call.

Updated `getParameters` and `getBondParameters` to re-use the memory from the argument rather
than re-allocating via the copy.

* Forgot to reuse the memory for the groups...

* Reverted back the manual copies via memcpy as they aren't needed. Looking at the header
file and benchmarking shows that std::vector does the right thing.

* Confined `thread_local static` only to ForceInfo methods, and have also put declarations
for multiple variables back onto a single line

* Removed `thread_local static` from the constructor

* Moved constructor declarations back into the for loop

The nonbonded kernel uses USE_NEIGHBOR_LIST (useNeighborList)
so host code also must check it instead of useCutoff.

See also https://github.com/openmm/openmm/issues/3462

* Implemented CustomCPPForceImpl

* Documentation for CustomCPPForceImpl

* Attempt at fixing Windows compilation error

* Improved documentation

* remove std::binary_function from ConstraintOrderer functor

* whoops whitespace

* Amoeba minor cleanup

- Fix variable name in string
- Remove odd space between variable and period that is inconsistently styled

* Replaces random tabs with spaces in ATM Force

* Draft integration of the Alchemical Transfer Method (ATM) plugin

* Attempt to store and retrieve forces--does not compile

* Implement addForce()/getForce() methods

* Throw exception when specifying properties without a Platform (#4130)

* Fixed DOF calculation for NoseHooverIntegrator (#4128)

* Fix variance in documentation of VerletIntegrator (#4138)

* Python API for ATMForce

* Fixed compilation error

* Minor cleanup of formatting and documentation

* Files for ATMForce test cases

* More cleanup

* Removed variable groups

* Test ATMForce with two particles

* More tests for ATMForce plus fixes

* Added missing header

* Rework interface to pass displacements as vector of parameters

* Revert "Rework interface to pass displacements as vector of parameters"

This reverts commit 5e092031f31ded1137b677588f007add1c2d6f82.

* Test with nonbonded force

* Allow energy expression to be custom...

* Use large blocks to optimize building the neighbor list

* Large blocks optimization for OpenCL

* Fix test failures

* Select whether to use large blocks based on system size

* Always use nvrtc for compilation

* Install nvrtc on CI

* Workaround for compiler error

* Set empty values for deprecated properties

* save atomIndex order in MCBarostat kernel

* Update device indices

* Only setAtomIndex if there was a reorder

* Attempt at fixing errors with barostat

* Missing ContextSelector