gfx936-dtk26.04.patch

diff --git a/examples/benchmarks/benchmark.py b/examples/benchmarks/benchmark.py
index f3eb95a88..f9fea45c2 100644
--- a/examples/benchmarks/benchmark.py
+++ b/examples/benchmarks/benchmark.py
@@ -354,6 +354,8 @@ def runOneTest(testName, options):
             initialSteps = 250
     if options.disable_pme_stream:
         properties['DisablePmeStream'] = 'true'
+    if options.platform == 'HIP' and options.fft_backend is not None and 'FFTBackend' in platform.getPropertyNames():
+        properties['FFTBackend'] = options.fft_backend
     if options.opencl_platform is not None and 'OpenCLPlatformIndex' in platform.getPropertyNames():
         properties['OpenCLPlatformIndex'] = options.opencl_platform
     if (options.precision is not None) and ('Precision' in platform.getPropertyNames()):
@@ -481,6 +483,7 @@ parser.add_argument('--polarization', default='mutual', dest='polarization', cho
 parser.add_argument('--mutual-epsilon', default=1e-5, dest='epsilon', type=float, help='mutual induced epsilon for AMOEBA [default: 1e-5]')
 parser.add_argument('--bond-constraints', default='hbonds', dest='bond_constraints', help=f'hbonds: constrain bonds to hydrogen, use 1.5*amu H mass; allbonds: constrain all bonds, use 4*amu H mass, and use larger timestep. This option is ignored for AMOEBA: {BOND_CONSTRAINTS} [default: hbonds]')
 parser.add_argument('--disable-pme-stream', default=False, action='store_true', dest='disable_pme_stream', help='disable use of a separate GPU stream for PME')
+parser.add_argument('--fft-backend', default=None, dest='fft_backend', help='FFT backend for HIP platform (for example: vkfft or hipfft)')
 parser.add_argument('--device', default=None, dest='device', help='device index for CUDA, HIP, or OpenCL')
 parser.add_argument('--opencl-platform', default=None, dest='opencl_platform', help='platform index for OpenCL')
 parser.add_argument('--precision', default='single', dest='precision', help=f'precision modes for CUDA, HIP, or OpenCL: {PRECISIONS} [default: single]')
diff --git a/platforms/common/include/openmm/common/CommonCalcNonbondedForce.h b/platforms/common/include/openmm/common/CommonCalcNonbondedForce.h
index 91416216e..e5fd1f6ec 100644
--- a/platforms/common/include/openmm/common/CommonCalcNonbondedForce.h
+++ b/platforms/common/include/openmm/common/CommonCalcNonbondedForce.h
@@ -45,7 +45,7 @@ namespace OpenMM {
 class CommonCalcNonbondedForceKernel : public CalcNonbondedForceKernel {
 public:
     CommonCalcNonbondedForceKernel(std::string name, const Platform& platform, ComputeContext& cc, const System& system) : CalcNonbondedForceKernel(name, platform),
-            hasInitializedKernel(false), cc(cc), pmeio(NULL), stepsToSort(0) {
+            hasInitializedKernel(false), cc(cc), pmeio(NULL), stepsToSort(0), dispersionStepsToSort(0) {
     }
     ~CommonCalcNonbondedForceKernel();
     /**
@@ -140,6 +140,7 @@ private:
     ComputeArray pmeDispersionBsplineModuliY;
     ComputeArray pmeDispersionBsplineModuliZ;
     ComputeArray pmeAtomGridIndex;
+    ComputeArray pmeDispersionAtomGridIndex;
     ComputeArray pmeEnergyBuffer;
     ComputeArray chargeBuffer;
     ComputeSort sort;
@@ -167,6 +168,7 @@ private:
     int gridSizeX, gridSizeY, gridSizeZ;
     int dispersionGridSizeX, dispersionGridSizeY, dispersionGridSizeZ;
     int stepsToSort;
+    int dispersionStepsToSort;
     bool usePmeQueue, deviceIsCpu, useFixedPointChargeSpreading, useCpuPme;
     bool hasCoulomb, hasLJ, doLJPME, usePosqCharges, recomputeParams, hasOffsets;
     NonbondedMethod nonbondedMethod;
diff --git a/platforms/common/src/CommonCalcNonbondedForce.cpp b/platforms/common/src/CommonCalcNonbondedForce.cpp
index bab0679b7..27277ebfc 100644
--- a/platforms/common/src/CommonCalcNonbondedForce.cpp
+++ b/platforms/common/src/CommonCalcNonbondedForce.cpp
@@ -457,6 +457,8 @@ void CommonCalcNonbondedForceKernel::commonInitialize(const System& system, cons
                     pmeDispersionBsplineModuliZ.initialize(cc, dispersionGridSizeZ, elementSize, "pmeDispersionBsplineModuliZ");
                 }
                 pmeAtomGridIndex.initialize<mm_int2>(cc, numParticles, "pmeAtomGridIndex");
+                if (doLJPME)
+                    pmeDispersionAtomGridIndex.initialize<mm_int2>(cc, numParticles, "pmeDispersionAtomGridIndex");
                 int energyElementSize = (cc.getUseDoublePrecision() || cc.getUseMixedPrecision() ? sizeof(double) : sizeof(float));
                 pmeEnergyBuffer.initialize(cc, cc.getNumThreadBlocks()*ComputeContext::ThreadBlockSize, energyElementSize, "pmeEnergyBuffer");
                 cc.clearBuffer(pmeEnergyBuffer);
@@ -836,7 +838,7 @@ double CommonCalcNonbondedForceKernel::execute(ContextImpl& context, bool includ
                 pmeDispersionEvalEnergyKernel = program->createKernel("gridEvaluateEnergy");
                 pmeDispersionInterpolateForceKernel = program->createKernel("gridInterpolateForce");
                 pmeDispersionGridIndexKernel->addArg(cc.getPosq());
-                pmeDispersionGridIndexKernel->addArg(pmeAtomGridIndex);
+                pmeDispersionGridIndexKernel->addArg(pmeDispersionAtomGridIndex);
                 for (int i = 0; i < 8; i++)
                     pmeDispersionGridIndexKernel->addArg();
                 pmeDispersionSpreadChargeKernel->addArg(cc.getPosq());
@@ -846,7 +848,7 @@ double CommonCalcNonbondedForceKernel::execute(ContextImpl& context, bool includ
                     pmeDispersionSpreadChargeKernel->addArg(pmeGrid1);
                 for (int i = 0; i < 8; i++)
                     pmeDispersionSpreadChargeKernel->addArg();
-                pmeDispersionSpreadChargeKernel->addArg(pmeAtomGridIndex);
+                pmeDispersionSpreadChargeKernel->addArg(pmeDispersionAtomGridIndex);
                 pmeDispersionSpreadChargeKernel->addArg(sigmaEpsilon);
                 pmeDispersionConvolutionKernel->addArg(pmeGrid2);
                 pmeDispersionConvolutionKernel->addArg(pmeDispersionBsplineModuliX);
@@ -869,7 +871,7 @@ double CommonCalcNonbondedForceKernel::execute(ContextImpl& context, bool includ
                 pmeDispersionInterpolateForceKernel->addArg(pmeGrid1);
                 for (int i = 0; i < 8; i++)
                     pmeDispersionInterpolateForceKernel->addArg();
-                pmeDispersionInterpolateForceKernel->addArg(pmeAtomGridIndex);
+                pmeDispersionInterpolateForceKernel->addArg(pmeDispersionAtomGridIndex);
                 pmeDispersionInterpolateForceKernel->addArg(sigmaEpsilon);
                 if (useFixedPointChargeSpreading) {
                     pmeDispersionFinishSpreadChargeKernel = program->createKernel("finishSpreadCharge");
@@ -962,7 +964,7 @@ double CommonCalcNonbondedForceKernel::execute(ContextImpl& context, bool includ
         // Execute the reciprocal space kernels.
 
         if (hasCoulomb) {
-            if (stepsToSort <= 0 || doLJPME || cc.getNumAtoms() > 15000) {
+            if (stepsToSort <= 0) {
                 setPeriodicBoxArgs(cc, pmeGridIndexKernel, 2);
                 if (cc.getUseDoublePrecision()) {
                     pmeGridIndexKernel->setArg(7, recipBoxVectors[0]);
@@ -1033,19 +1035,24 @@ double CommonCalcNonbondedForceKernel::execute(ContextImpl& context, bool includ
         }
 
         if (doLJPME && hasLJ) {
-            setPeriodicBoxArgs(cc, pmeDispersionGridIndexKernel, 2);
-            if (cc.getUseDoublePrecision()) {
-                pmeDispersionGridIndexKernel->setArg(7, recipBoxVectors[0]);
-                pmeDispersionGridIndexKernel->setArg(8, recipBoxVectors[1]);
-                pmeDispersionGridIndexKernel->setArg(9, recipBoxVectors[2]);
-            }
-            else {
-                pmeDispersionGridIndexKernel->setArg(7, recipBoxVectorsFloat[0]);
-                pmeDispersionGridIndexKernel->setArg(8, recipBoxVectorsFloat[1]);
-                pmeDispersionGridIndexKernel->setArg(9, recipBoxVectorsFloat[2]);
+            if (dispersionStepsToSort <= 0) {
+                setPeriodicBoxArgs(cc, pmeDispersionGridIndexKernel, 2);
+                if (cc.getUseDoublePrecision()) {
+                    pmeDispersionGridIndexKernel->setArg(7, recipBoxVectors[0]);
+                    pmeDispersionGridIndexKernel->setArg(8, recipBoxVectors[1]);
+                    pmeDispersionGridIndexKernel->setArg(9, recipBoxVectors[2]);
+                }
+                else {
+                    pmeDispersionGridIndexKernel->setArg(7, recipBoxVectorsFloat[0]);
+                    pmeDispersionGridIndexKernel->setArg(8, recipBoxVectorsFloat[1]);
+                    pmeDispersionGridIndexKernel->setArg(9, recipBoxVectorsFloat[2]);
+                }
+                pmeDispersionGridIndexKernel->execute(cc.getNumAtoms());
+                sort->sort(pmeDispersionAtomGridIndex);
+                dispersionStepsToSort = 3;
             }
-            pmeDispersionGridIndexKernel->execute(cc.getNumAtoms());
-            sort->sort(pmeAtomGridIndex);
+            else
+                dispersionStepsToSort--;
             if (useFixedPointChargeSpreading)
                 cc.clearBuffer(pmeGrid2);
             else
diff --git a/platforms/hip/CMakeLists.txt b/platforms/hip/CMakeLists.txt
index 7c8dfca8f..2ca6f35d7 100644
--- a/platforms/hip/CMakeLists.txt
+++ b/platforms/hip/CMakeLists.txt
@@ -12,7 +12,23 @@
 #   libOpenMMHIP_static.a
 #----------------------------------------------------
 
-FIND_PACKAGE(HIPRTC CONFIG)
+IF(NOT TARGET hiprtc::hiprtc)
+    add_library(hiprtc::hiprtc SHARED IMPORTED)
+    set_target_properties(hiprtc::hiprtc PROPERTIES
+        IMPORTED_LOCATION "/opt/dtk/hip/lib/libhiprtc.so"
+        INTERFACE_INCLUDE_DIRECTORIES "/opt/dtk/hip/include"
+    )
+ENDIF()
+
+FIND_PACKAGE(HIPFFT CONFIG QUIET)
+
+IF(NOT TARGET hiprtc::hiprtc)
+    add_library(hiprtc::hiprtc SHARED IMPORTED)
+    set_target_properties(hiprtc::hiprtc PROPERTIES
+        IMPORTED_LOCATION "/opt/dtk/hip/lib/libhiprtc.so"
+        INTERFACE_INCLUDE_DIRECTORIES "/opt/dtk/hip/include"
+    )
+ENDIF()
 
 SET(OPENMM_BUILD_HIP_TESTS TRUE CACHE BOOL "Whether to build HIP test cases")
 IF(BUILD_TESTING AND OPENMM_BUILD_HIP_TESTS)
@@ -103,6 +119,10 @@ IF(OPENMM_BUILD_SHARED_LIB)
     TARGET_LINK_LIBRARIES(${SHARED_TARGET} PUBLIC ${OPENMM_LIBRARY_NAME} hip::host hiprtc::hiprtc)
     SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_COMMON_BUILDING_SHARED_LIBRARY")
     SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_LINK_FLAGS}")
+    IF(HIPFFT_FOUND)
+        TARGET_LINK_LIBRARIES(${SHARED_TARGET} PUBLIC hip::hipfft)
+        TARGET_COMPILE_OPTIONS(${SHARED_TARGET} PUBLIC "-DOPENMM_HIP_WITH_HIPFFT")
+    ENDIF()
 
     INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${SHARED_TARGET})
 ENDIF(OPENMM_BUILD_SHARED_LIB)
@@ -116,6 +136,10 @@ IF(OPENMM_BUILD_STATIC_LIB)
     TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${OPENMM_LIBRARY_NAME} hip::host hiprtc::hiprtc)
     SET_TARGET_PROPERTIES(${STATIC_TARGET} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_COMMON_BUILDING_STATIC_LIBRARY")
     SET_TARGET_PROPERTIES(${STATIC_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_LINK_FLAGS}")
+    IF(HIPFFT_FOUND)
+        TARGET_LINK_LIBRARIES(${STATIC_TARGET} PUBLIC hip::hipfft)
+        TARGET_COMPILE_OPTIONS(${STATIC_TARGET} PUBLIC "-DOPENMM_HIP_WITH_HIPFFT")
+    ENDIF()
 
     INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${STATIC_TARGET})
 ENDIF(OPENMM_BUILD_STATIC_LIB)
diff --git a/platforms/hip/include/HipHipFFT3D.h b/platforms/hip/include/HipHipFFT3D.h
new file mode 100644
index 000000000..af509e4fc
--- /dev/null
+++ b/platforms/hip/include/HipHipFFT3D.h
@@ -0,0 +1,60 @@
+#ifndef __OPENMM_HIPHIPFFT3D_H__
+#define __OPENMM_HIPHIPFFT3D_H__
+
+/* -------------------------------------------------------------------------- *
+ *                                   OpenMM                                   *
+ * -------------------------------------------------------------------------- *
+ * This is part of the OpenMM molecular simulation toolkit.                   *
+ * See https://openmm.org/development.                                        *
+ *                                                                            *
+ * Portions copyright (c) 2009-2026 Stanford University and the Authors.      *
+ * Contributors:                                                              *
+ *                                                                            *
+ * This program is free software: you can redistribute it and/or modify       *
+ * it under the terms of the GNU Lesser General Public License as published   *
+ * by the Free Software Foundation, either version 3 of the License, or       *
+ * (at your option) any later version.                                        *
+ *                                                                            *
+ * This program is distributed in the hope that it will be useful,            *
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of             *
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the              *
+ * GNU Lesser General Public License for more details.                        *
+ *                                                                            *
+ * You should have received a copy of the GNU Lesser General Public License   *
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.      *
+ * -------------------------------------------------------------------------- */
+
+#include "openmm/common/windowsExportCommon.h"
+#include "openmm/common/FFT3D.h"
+#include "openmm/common/ArrayInterface.h"
+
+#ifdef OPENMM_HIP_WITH_HIPFFT
+#if __has_include(<hipfft/hipfft.h>)
+#include <hipfft/hipfft.h>
+#else
+#include <hipfft.h>
+#endif
+#endif
+
+namespace OpenMM {
+
+class HipContext;
+
+class OPENMM_EXPORT_COMMON HipHipFFT3D : public FFT3DImpl {
+public:
+    HipHipFFT3D(HipContext& context, int xsize, int ysize, int zsize, bool realToComplex);
+    ~HipHipFFT3D();
+    void execFFT(ArrayInterface& in, ArrayInterface& out, bool forward=true) override;
+private:
+    HipContext& context;
+    bool realToComplex;
+    bool doublePrecision;
+#ifdef OPENMM_HIP_WITH_HIPFFT
+    hipfftHandle forwardPlan;
+    hipfftHandle backwardPlan;
+#endif
+};
+
+} // namespace OpenMM
+
+#endif // __OPENMM_HIPHIPFFT3D_H__
diff --git a/platforms/hip/include/HipPlatform.h b/platforms/hip/include/HipPlatform.h
index a0665d45c..43eba16af 100644
--- a/platforms/hip/include/HipPlatform.h
+++ b/platforms/hip/include/HipPlatform.h
@@ -111,13 +111,21 @@ public:
         static const std::string key = "DeterministicForces";
         return key;
     }
+    /**
+     * This is the name of the parameter for selecting which FFT backend to use.
+     */
+    static const std::string& HipFFTBackend() {
+        static const std::string key = "FFTBackend";
+        return key;
+    }
 };
 
 class OPENMM_EXPORT_COMMON HipPlatform::PlatformData {
 public:
     PlatformData(ContextImpl* context, const System& system, const std::string& deviceIndexProperty, const std::string& blockingProperty, const std::string& precisionProperty,
             const std::string& cpuPmeProperty, const std::string& tempProperty,
-            const std::string& pmeStreamProperty, const std::string& deterministicForcesProperty, int numThreads, ContextImpl* originalContext);
+            const std::string& pmeStreamProperty, const std::string& deterministicForcesProperty, const std::string& fftBackendProperty,
+            int numThreads, ContextImpl* originalContext);
     ~PlatformData();
     void initializeContexts(const System& system);
     void syncContexts();
diff --git a/platforms/hip/src/HipContext.cpp b/platforms/hip/src/HipContext.cpp
index 0050a8c46..dc4290b3c 100644
--- a/platforms/hip/src/HipContext.cpp
+++ b/platforms/hip/src/HipContext.cpp
@@ -31,6 +31,7 @@
 #include "HipBondedUtilities.h"
 #include "HipEvent.h"
 #include "HipFFT3D.h"
+#include "HipHipFFT3D.h"
 #include "HipIntegrationUtilities.h"
 #include "HipKernels.h"
 #include "HipKernelSources.h"
@@ -80,6 +81,12 @@
 using namespace OpenMM;
 using namespace std;
 
+static string normalizeFftBackendName(const string& value) {
+    string result = value;
+    transform(result.begin(), result.end(), result.begin(), ::tolower);
+    return result;
+}
+
 const int HipContext::ThreadBlockSize = 64;
 const int HipContext::TileSize = 32;
 bool HipContext::hasInitializedHip = false;
@@ -712,7 +719,12 @@ ComputeSort HipContext::createSort(ComputeSortImpl::SortTrait* trait, unsigned i
 }
 
 FFT3D HipContext::createFFT(int xsize, int ysize, int zsize, bool realToComplex) {
-    return FFT3D(new HipFFT3D(*this, xsize, ysize, zsize, realToComplex));
+    string backend = normalizeFftBackendName(platformData.propertyValues[HipPlatform::HipFFTBackend()]);
+    if (backend.empty() || backend == "vkfft")
+        return FFT3D(new HipFFT3D(*this, xsize, ysize, zsize, realToComplex));
+    if (backend == "hipfft")
+        return FFT3D(new HipHipFFT3D(*this, xsize, ysize, zsize, realToComplex));
+    throw OpenMMException("Illegal value for FFTBackend: "+backend);
 }
 
 int HipContext::findLegalFFTDimension(int minimum) {
diff --git a/platforms/hip/src/HipHipFFT3D.cpp b/platforms/hip/src/HipHipFFT3D.cpp
new file mode 100644
index 000000000..a0a408c00
--- /dev/null
+++ b/platforms/hip/src/HipHipFFT3D.cpp
@@ -0,0 +1,100 @@
+/* -------------------------------------------------------------------------- *
+ *                                   OpenMM                                   *
+ * -------------------------------------------------------------------------- *
+ * This is part of the OpenMM molecular simulation toolkit.                   *
+ * See https://openmm.org/development.                                        *
+ *                                                                            *
+ * Portions copyright (c) 2009-2026 Stanford University and the Authors.      *
+ * Contributors:                                                              *
+ *                                                                            *
+ * This program is free software: you can redistribute it and/or modify       *
+ * it under the terms of the GNU Lesser General Public License as published   *
+ * by the Free Software Foundation, either version 3 of the License, or       *
+ * (at your option) any later version.                                        *
+ *                                                                            *
+ * This program is distributed in the hope that it will be useful,            *
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of             *
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the              *
+ * GNU Lesser General Public License for more details.                        *
+ *                                                                            *
+ * You should have received a copy of the GNU Lesser General Public License   *
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.      *
+ * -------------------------------------------------------------------------- */
+
+#include "HipHipFFT3D.h"
+#include "HipContext.h"
+#include "openmm/OpenMMException.h"
+
+using namespace OpenMM;
+using namespace std;
+
+HipHipFFT3D::HipHipFFT3D(HipContext& context, int xsize, int ysize, int zsize, bool realToComplex) :
+        context(context), realToComplex(realToComplex), doublePrecision(context.getUseDoublePrecision())
+#ifdef OPENMM_HIP_WITH_HIPFFT
+        , forwardPlan(0), backwardPlan(0)
+#endif
+{
+#ifndef OPENMM_HIP_WITH_HIPFFT
+    throw OpenMMException("hipFFT backend requested but OpenMM was built without HIPFFT support");
+#else
+    hipfftType forwardType, backwardType;
+    if (realToComplex) {
+        forwardType = doublePrecision ? HIPFFT_D2Z : HIPFFT_R2C;
+        backwardType = doublePrecision ? HIPFFT_Z2D : HIPFFT_C2R;
+    }
+    else {
+        forwardType = doublePrecision ? HIPFFT_Z2Z : HIPFFT_C2C;
+        backwardType = forwardType;
+    }
+    hipfftResult result = hipfftPlan3d(&forwardPlan, xsize, ysize, zsize, forwardType);
+    if (result != HIPFFT_SUCCESS)
+        throw OpenMMException("Error initializing hipFFT forward plan: "+context.intToString(result));
+    result = hipfftPlan3d(&backwardPlan, xsize, ysize, zsize, backwardType);
+    if (result != HIPFFT_SUCCESS) {
+        hipfftDestroy(forwardPlan);
+        throw OpenMMException("Error initializing hipFFT backward plan: "+context.intToString(result));
+    }
+#endif
+}
+
+HipHipFFT3D::~HipHipFFT3D() {
+#ifdef OPENMM_HIP_WITH_HIPFFT
+    if (forwardPlan != 0)
+        hipfftDestroy(forwardPlan);
+    if (backwardPlan != 0)
+        hipfftDestroy(backwardPlan);
+#endif
+}
+
+void HipHipFFT3D::execFFT(ArrayInterface& in, ArrayInterface& out, bool forward) {
+#ifndef OPENMM_HIP_WITH_HIPFFT
+    throw OpenMMException("hipFFT backend requested but OpenMM was built without HIPFFT support");
+#else
+    hipfftHandle plan = forward ? forwardPlan : backwardPlan;
+    hipfftResult result = hipfftSetStream(plan, context.getCurrentStream());
+    if (result != HIPFFT_SUCCESS)
+        throw OpenMMException("Error setting hipFFT stream: "+context.intToString(result));
+    if (realToComplex) {
+        if (forward) {
+            if (doublePrecision)
+                result = hipfftExecD2Z(plan, (double*) context.unwrap(in).getDevicePointer(), (double2*) context.unwrap(out).getDevicePointer());
+            else
+                result = hipfftExecR2C(plan, (float*) context.unwrap(in).getDevicePointer(), (float2*) context.unwrap(out).getDevicePointer());
+        }
+        else {
+            if (doublePrecision)
+                result = hipfftExecZ2D(plan, (double2*) context.unwrap(in).getDevicePointer(), (double*) context.unwrap(out).getDevicePointer());
+            else
+                result = hipfftExecC2R(plan, (float2*) context.unwrap(in).getDevicePointer(), (float*) context.unwrap(out).getDevicePointer());
+        }
+    }
+    else {
+        if (doublePrecision)
+            result = hipfftExecZ2Z(plan, (double2*) context.unwrap(in).getDevicePointer(), (double2*) context.unwrap(out).getDevicePointer(), forward ? HIPFFT_FORWARD : HIPFFT_BACKWARD);
+        else
+            result = hipfftExecC2C(plan, (float2*) context.unwrap(in).getDevicePointer(), (float2*) context.unwrap(out).getDevicePointer(), forward ? HIPFFT_FORWARD : HIPFFT_BACKWARD);
+    }
+    if (result != HIPFFT_SUCCESS)
+        throw OpenMMException("Error executing hipFFT: "+context.intToString(result));
+#endif
+}
diff --git a/platforms/hip/src/HipNonbondedUtilities.cpp b/platforms/hip/src/HipNonbondedUtilities.cpp
index 01cc7f826..60548054d 100644
--- a/platforms/hip/src/HipNonbondedUtilities.cpp
+++ b/platforms/hip/src/HipNonbondedUtilities.cpp
@@ -65,16 +65,15 @@ HipNonbondedUtilities::HipNonbondedUtilities(HipContext& context) : context(cont
     string errorMessage = "Error initializing nonbonded utilities";
     CHECK_RESULT(hipEventCreateWithFlags(&downloadCountEvent, context.getEventFlags()));
     CHECK_RESULT(hipHostMalloc((void**) &pinnedCountBuffer, 2*sizeof(unsigned int), context.getHostMallocFlags()));
-    numForceThreadBlocks = 5*4*context.getMultiprocessors();
-    forceThreadBlockSize = 64;
-    findInteractingBlocksThreadBlockSize = context.getSIMDWidth();
+    numForceThreadBlocks = 16*4*context.getMultiprocessors();
+    forceThreadBlockSize = 256;
+    findInteractingBlocksThreadBlockSize = 128;
 
     // When building the neighbor list, we can optionally use large blocks (32 * warpSize atoms) to
     // accelerate the process.  This makes building the neighbor list faster, but it prevents
     // us from sorting atom blocks by size, which leads to a slightly less efficient neighbor
     // list.  We guess based on system size which will be faster.
-
-    useLargeBlocks = (context.getNumAtoms() > 90000);
+    useLargeBlocks = false;
     setKernelSource(HipKernelSources::nonbonded);
 }
 
@@ -400,7 +399,7 @@ double HipNonbondedUtilities::getMaxCutoffDistance() {
 }
 
 double HipNonbondedUtilities::padCutoff(double cutoff) {
-    double padding = (usePadding ? 0.08*cutoff : 0.0);
+    double padding = (usePadding ? 0.12*cutoff : 0.0);
     return cutoff+padding;
 }
 
diff --git a/platforms/hip/src/HipPlatform.cpp b/platforms/hip/src/HipPlatform.cpp
index e6e68e2dd..3c3936da6 100644
--- a/platforms/hip/src/HipPlatform.cpp
+++ b/platforms/hip/src/HipPlatform.cpp
@@ -69,6 +69,7 @@ HipPlatform::HipPlatform() {
     deprecatedPropertyReplacements["HipTempDirectory"] = HipTempDirectory();
     deprecatedPropertyReplacements["HipDisablePmeStream"] = HipDisablePmeStream();
     deprecatedPropertyReplacements["HipDeterministicForces"] = HipDeterministicForces();
+    deprecatedPropertyReplacements["HipFFTBackend"] = HipFFTBackend();
     HipKernelFactory* factory = new HipKernelFactory();
     registerKernelFactory(CalcForcesAndEnergyKernel::Name(), factory);
     registerKernelFactory(UpdateStateDataKernel::Name(), factory);
@@ -122,6 +123,7 @@ HipPlatform::HipPlatform() {
     platformProperties.push_back(HipTempDirectory());
     platformProperties.push_back(HipDisablePmeStream());
     platformProperties.push_back(HipDeterministicForces());
+    platformProperties.push_back(HipFFTBackend());
     setPropertyDefaultValue(HipDeviceIndex(), "");
     setPropertyDefaultValue(HipDeviceName(), "");
     setPropertyDefaultValue(HipUseBlockingSync(), "true");
@@ -129,6 +131,7 @@ HipPlatform::HipPlatform() {
     setPropertyDefaultValue(HipUseCpuPme(), "false");
     setPropertyDefaultValue(HipDisablePmeStream(), "false");
     setPropertyDefaultValue(HipDeterministicForces(), "false");
+    setPropertyDefaultValue(HipFFTBackend(), "vkfft");
 #ifdef _MSC_VER
     setPropertyDefaultValue(HipTempDirectory(), string(getenv("TEMP")));
 #else
@@ -213,11 +216,14 @@ void HipPlatform::contextCreated(ContextImpl& context, const map<string, string>
             getPropertyDefaultValue(HipDisablePmeStream()) : properties.find(HipDisablePmeStream())->second);
     string deterministicForcesValue = (properties.find(HipDeterministicForces()) == properties.end() ?
             getPropertyDefaultValue(HipDeterministicForces()) : properties.find(HipDeterministicForces())->second);
+    string fftBackendValue = (properties.find(HipFFTBackend()) == properties.end() ?
+            getPropertyDefaultValue(HipFFTBackend()) : properties.find(HipFFTBackend())->second);
     transform(blockingPropValue.begin(), blockingPropValue.end(), blockingPropValue.begin(), ::tolower);
     transform(precisionPropValue.begin(), precisionPropValue.end(), precisionPropValue.begin(), ::tolower);
     transform(cpuPmePropValue.begin(), cpuPmePropValue.end(), cpuPmePropValue.begin(), ::tolower);
     transform(pmeStreamPropValue.begin(), pmeStreamPropValue.end(), pmeStreamPropValue.begin(), ::tolower);
     transform(deterministicForcesValue.begin(), deterministicForcesValue.end(), deterministicForcesValue.begin(), ::tolower);
+    transform(fftBackendValue.begin(), fftBackendValue.end(), fftBackendValue.begin(), ::tolower);
     vector<string> pmeKernelName;
     pmeKernelName.push_back(CalcPmeReciprocalForceKernel::Name());
     if (!supportsKernels(pmeKernelName))
@@ -227,7 +233,7 @@ void HipPlatform::contextCreated(ContextImpl& context, const map<string, string>
     if (threadsEnv != NULL)
         stringstream(threadsEnv) >> threads;
     context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, tempPropValue,
-            pmeStreamPropValue, deterministicForcesValue, threads, NULL));
+            pmeStreamPropValue, deterministicForcesValue, fftBackendValue, threads, NULL));
 }
 
 void HipPlatform::linkedContextCreated(ContextImpl& context, ContextImpl& originalContext) const {
@@ -239,9 +245,10 @@ void HipPlatform::linkedContextCreated(ContextImpl& context, ContextImpl& origin
     string tempPropValue = platform.getPropertyValue(originalContext.getOwner(), HipTempDirectory());
     string pmeStreamPropValue = platform.getPropertyValue(originalContext.getOwner(), HipDisablePmeStream());
     string deterministicForcesValue = platform.getPropertyValue(originalContext.getOwner(), HipDeterministicForces());
+    string fftBackendValue = platform.getPropertyValue(originalContext.getOwner(), HipFFTBackend());
     int threads = reinterpret_cast<PlatformData*>(originalContext.getPlatformData())->threads.getNumThreads();
     context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, tempPropValue,
-            pmeStreamPropValue, deterministicForcesValue, threads, &originalContext));
+            pmeStreamPropValue, deterministicForcesValue, fftBackendValue, threads, &originalContext));
 }
 
 void HipPlatform::contextDestroyed(ContextImpl& context) const {
@@ -250,8 +257,8 @@ void HipPlatform::contextDestroyed(ContextImpl& context) const {
 }
 
 HipPlatform::PlatformData::PlatformData(ContextImpl* context, const System& system, const string& deviceIndexProperty, const string& blockingProperty, const string& precisionProperty,
-            const string& cpuPmeProperty, const string& tempProperty, const string& pmeStreamProperty,
-            const string& deterministicForcesProperty, int numThreads, ContextImpl* originalContext) :
+              const string& cpuPmeProperty, const string& tempProperty, const string& pmeStreamProperty,
+              const string& deterministicForcesProperty, const string& fftBackendProperty, int numThreads, ContextImpl* originalContext) :
                 context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0), hasInitializedContexts(false),
                 threads(numThreads) {
     bool blocking = (blockingProperty == "true");
@@ -306,6 +313,7 @@ HipPlatform::PlatformData::PlatformData(ContextImpl* context, const System& syst
     propertyValues[HipPlatform::HipTempDirectory()] = tempProperty;
     propertyValues[HipPlatform::HipDisablePmeStream()] = disablePmeStream ? "true" : "false";
     propertyValues[HipPlatform::HipDeterministicForces()] = deterministicForces ? "true" : "false";
+    propertyValues[HipPlatform::HipFFTBackend()] = fftBackendProperty;
     contextEnergy.resize(contexts.size());
 
     // Determine whether peer-to-peer copying is supported, and enable it if so.
diff --git a/platforms/hip/tests/TestHipFFT3D.cpp b/platforms/hip/tests/TestHipFFT3D.cpp
index 170ad1fe0..8aff3112e 100644
--- a/platforms/hip/tests/TestHipFFT3D.cpp
+++ b/platforms/hip/tests/TestHipFFT3D.cpp
@@ -58,7 +58,8 @@ void testTransform(bool realToComplex, int xsize, int ysize, int zsize, double e
     system.addParticle(0.0);
     HipPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("HipPrecision"), "false",
             platform.getPropertyDefaultValue(HipPlatform::HipTempDirectory()),
-            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false",
+            platform.getPropertyDefaultValue(HipPlatform::HipFFTBackend()), 1, NULL);
     HipContext& context = *platformData.contexts[0];
     context.initialize();
     context.setAsCurrent();
diff --git a/platforms/hip/tests/TestHipRandom.cpp b/platforms/hip/tests/TestHipRandom.cpp
index 94212702c..54f869b35 100644
--- a/platforms/hip/tests/TestHipRandom.cpp
+++ b/platforms/hip/tests/TestHipRandom.cpp
@@ -55,7 +55,8 @@ void testGaussian() {
         system.addParticle(1.0);
     HipPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("HipPrecision"), "false",
             platform.getPropertyDefaultValue(HipPlatform::HipTempDirectory()),
-            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false",
+            platform.getPropertyDefaultValue(HipPlatform::HipFFTBackend()), 1, NULL);
     HipContext& context = *platformData.contexts[0];
     context.initialize();
     context.setAsCurrent();
diff --git a/platforms/hip/tests/TestHipSort.cpp b/platforms/hip/tests/TestHipSort.cpp
index fe8b1c56a..03e7ee2c5 100644
--- a/platforms/hip/tests/TestHipSort.cpp
+++ b/platforms/hip/tests/TestHipSort.cpp
@@ -65,7 +65,8 @@ void verifySorting(vector<float> array, bool uniform) {
     system.addParticle(0.0);
     HipPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("HipPrecision"), "false",
             platform.getPropertyDefaultValue(HipPlatform::HipTempDirectory()),
-            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(HipPlatform::HipDisablePmeStream()), "false",
+            platform.getPropertyDefaultValue(HipPlatform::HipFFTBackend()), 1, NULL);
     HipContext& context = *platformData.contexts[0];
     context.initialize();
     context.setAsCurrent();