Remove the single process, multi-gpu feature.

ARCHITECTURE.md

@@ -121,9 +121,7 @@ Array data:
 
 When using `GlobalArray` or `std::vector<T, hoomd::detail::managed_allocator<T>>`, call
 `cudaMemadvise` to set the appropriate memory hints for the array. Small parameter arrays should be
-set to `cudaMemAdviseSetReadMostly`. Larger arrays accessed in portions in single-process multi-GPU
-execution should be set to `cudaMemAdviseSetPreferredLocation` appropriately for the different
-portions of the array.
+set to `cudaMemAdviseSetReadMostly`.
 
 System data:
 

CHANGELOG.rst

@@ -61,7 +61,8 @@ Change Log
   (`#1931 <https://github.com/glotzerlab/hoomd-blue/pull/1931>`__).
 * ``Device.num_cpu_threads``.
   (`#1932 <https://github.com/glotzerlab/hoomd-blue/pull/1932>`__).
-
+* ``Device.gpu_ids`` and the single-process multi-GPU feature
+  (`#1936 <https://github.com/glotzerlab/hoomd-blue/pull/1936>`__).
 
 4.x
 ---

hoomd/CMakeLists.txt

@@ -136,7 +136,6 @@ set(_hoomd_headers
     GlobalArray.h
     GPUArray.h
     GPUFlags.h
-    GPUPartition.cuh
     GPUVector.h
     GSD.h
     GSDDequeWriter.h

hoomd/CellList.h

@@ -215,19 +215,6 @@ class PYBIND11_EXPORT CellList : public Compute
         m_params_changed = true;
         }
 
-    //! Request a multi-GPU cell list
-    virtual void setPerDevice(bool per_device)
-        {
-        // base class does nothing
-        }
-
-    //! Return true if we maintain a cell list per device
-    virtual bool getPerDevice() const
-        {
-        // base class doesn't support GPU
-        return false;
-        }
-
     // @}
     //! \name Get properties
     // @{
@@ -290,12 +277,6 @@ class PYBIND11_EXPORT CellList : public Compute
         return m_cell_size;
         }
 
-    //! Get the array of cell sizes (per device)
-    virtual const GlobalArray<unsigned int>& getCellSizeArrayPerDevice() const
-        {
-        throw std::runtime_error("Per-device cell size array not available in base class.\n");
-        }
-
     //! Get the adjacency list
     const GlobalArray<unsigned int>& getCellAdjArray() const
         {
@@ -330,13 +311,6 @@ class PYBIND11_EXPORT CellList : public Compute
         return m_idx;
         }
 
-    //! Get the cell list containing index (per device)
-    virtual const GlobalArray<unsigned int>& getIndexArrayPerDevice() const
-        {
-        // base class returns an empty array
-        throw std::runtime_error("Per-device cell index array not available in base class.\n");
-        }
-
     //! Compute the cell list given the current particle positions
     void compute(uint64_t timestep);