Use Intel OpenCL for CI (#4366)

* Use Intel OpenCL for CI

* Set environment variables

* Try to get CI to run

* Debugging

* Debugging

* Fixes for Intel OpenCL

.github/workflows/CI.yml

@@ -70,7 +70,7 @@ jobs:
               -DCUDA_TOOLKIT_INCLUDE=/usr/local/cuda-11.2/include \
               -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-11.2/ \
 
-          - name: Linux AMD OpenCL
+          - name: Linux Intel OpenCL
             os: ubuntu-latest
             env: ubuntu-latest
             python-version: "3.13"
@@ -88,8 +88,6 @@ jobs:
               -DOPENMM_BUILD_PYTHON_WRAPPERS=OFF \
               -DOPENMM_BUILD_C_AND_FORTRAN_WRAPPERS=OFF \
               -DOPENMM_BUILD_EXAMPLES=OFF \
-              -DOPENCL_INCLUDE_DIR=${GITHUB_WORKSPACE}/AMDAPPSDK/include \
-              -DOPENCL_LIBRARY=${GITHUB_WORKSPACE}/AMDAPPSDK/lib/x86_64/libOpenCL.so \
 
           - name: Linux HIP
             os: ubuntu-latest
@@ -213,7 +211,7 @@ jobs:
 
       - name: "Install OpenCL on Ubuntu (if needed)"
         if: matrix.OPENCL == true && startsWith(matrix.os, 'ubuntu')
-        run: source devtools/ci/gh-actions/scripts/install_amd_opencl.sh
+        run: source devtools/ci/gh-actions/scripts/install_intel_opencl.sh
 
       - name: "Install conda-forge compilers on Ubuntu or M1 (if needed)"
         if: matrix.compilers == 'conda-forge'
@@ -278,6 +276,9 @@ jobs:
         run: |
           set -x
           cd build
+          if [[ ${{ matrix.OPENCL }} == true ]]; then
+            source /opt/intel/oneapi/setvars.sh
+          fi
           python ../devtools/run-ctest.py --parallel 2 --timeout 600 --job-duration 900 --attempts 3
 
           if [[ ${{ matrix.os }} == ubuntu-* ]]; then SHLIB=so; else SHLIB=dylib; fi

devtools/ci/gh-actions/scripts/install_intel_opencl.sh

+# This script installs Intel's OpenCL for CPUs.
+
+set -euxo pipefail
+
+wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
+echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
+sudo apt update
+sudo apt install intel-basekit intel-hpckit intel-oneapi-runtime-opencl opencl-headers ocl-icd-opencl-dev -y

platforms/common/src/CommonKernels.cpp

@@ -4465,7 +4465,7 @@ double CommonCalcCustomHbondForceKernel::execute(ContextImpl& context, bool incl
     setPeriodicBoxArgs(cc, forceKernel, 6);
     int numDonorBlocks = (numDonors+31)/32;
     int numAcceptorBlocks = (numAcceptors+31)/32;
-    forceKernel->execute(numDonorBlocks*numAcceptorBlocks*32, cc.getIsCPU() ? 32 : 128);
+    forceKernel->execute(numDonorBlocks*numAcceptorBlocks*32, cc.getSIMDWidth() < 32 ? 32 : 128);
     return 0.0;
 }
 

platforms/opencl/src/OpenCLContext.cpp

@@ -125,7 +125,7 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
             string platformVendor = platforms[j].getInfo<CL_PLATFORM_VENDOR>();
             vector<cl::Device> devices;
             try {
-                platforms[j].getDevices(CL_DEVICE_TYPE_ALL, &devices);
+                platforms[j].getDevices(CL_DEVICE_TYPE_GPU | CL_DEVICE_TYPE_CPU, &devices);
             }
             catch (...) {
                 // There are no devices available for this platform.
@@ -197,7 +197,7 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
             cout << "WARNING: Using an unsupported OpenCL implementation.  Results may be incorrect." << endl;
 
         vector<cl::Device> devices;
-        platforms[bestPlatform].getDevices(CL_DEVICE_TYPE_ALL, &devices);
+        platforms[bestPlatform].getDevices(CL_DEVICE_TYPE_GPU | CL_DEVICE_TYPE_CPU, &devices);
         string platformVendor = platforms[bestPlatform].getInfo<CL_PLATFORM_VENDOR>();
         device = devices[bestDevice];
 

platforms/opencl/src/OpenCLFFT3D.cpp

@@ -56,6 +56,12 @@ OpenCLFFT3D::OpenCLFFT3D(OpenCLContext& context, int xsize, int ysize, int zsize
     config.inputBufferStride[0] = zsize;
     config.inputBufferStride[1] = ysize*zsize;
     config.inputBufferStride[2] = xsize*ysize*zsize;
+    cl::Platform platform(context.getDevice().getInfo<CL_DEVICE_PLATFORM>());
+    string platformVendor = platform.getInfo<CL_PLATFORM_VENDOR>();
+    if (platformVendor.size() >= 5 && platformVendor.substr(0, 5) == "Intel") {
+        // Intel's OpenCL uses low accuracy trig functions, so tell VkFFT to use lookup tables instead.
+        config.useLUT = 1;
+    }
     VkFFTResult result = initializeVkFFT(&app, config);
     if (result != VKFFT_SUCCESS)
         throw OpenMMException("Error initializing VkFFT: "+context.intToString(result));

platforms/opencl/tests/TestOpenCLNonbondedForce.cpp

@@ -54,9 +54,9 @@ bool canRunHugeTest() {
     platforms[platformIndex].getDevices(CL_DEVICE_TYPE_ALL, &devices);
     long long memory = devices[deviceIndex].getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>();
 
-    // Only run the huge test if the device has at least 4 GB of memory.
+    // Only run the huge test if the device has at least 8 GB of memory.
 
-    return (memory >= 4*(long long)(1<<30));
+    return (memory >= 8*(long long)(1<<30));
 }
 
 void runPlatformTests() {