Example #1
0
        /// <summary>
        /// Constructs a new OpenCL accelerator.
        /// </summary>
        /// <param name="context">The ILGPU context.</param>
        /// <param name="acceleratorId">The accelerator id.</param>
        public CLAccelerator(Context context, CLAcceleratorId acceleratorId)
            : base(context, AcceleratorType.OpenCL)
        {
            if (acceleratorId == null)
            {
                throw new ArgumentNullException(nameof(acceleratorId));
            }

            PlatformId = acceleratorId.PlatformId;
            DeviceId   = acceleratorId.DeviceId;

            PlatformName = CLAPI.GetPlatformInfo(
                PlatformId,
                CLPlatformInfoType.CL_PLATFORM_NAME);

            VendorName = CLAPI.GetPlatformInfo(
                PlatformId,
                CLPlatformInfoType.CL_PLATFORM_VENDOR);

            // Create new context
            CLException.ThrowIfFailed(
                CLAPI.CreateContext(DeviceId, out contextPtr));

            // Resolve device info
            Name = CLAPI.GetDeviceInfo(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_NAME);

            MemorySize = CLAPI.GetDeviceInfo <long>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_GLOBAL_MEM_SIZE);

            DeviceType = (CLDeviceType)CLAPI.GetDeviceInfo <long>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_TYPE);

            // Max grid size
            int workItemDimensions = IntrinsicMath.Max(CLAPI.GetDeviceInfo <int>(
                                                           DeviceId,
                                                           CLDeviceInfoType.CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS), 3);
            var workItemSizes = new IntPtr[workItemDimensions];

            CLAPI.GetDeviceInfo(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_MAX_WORK_ITEM_SIZES,
                workItemSizes);
            MaxGridSize = new Index3(
                workItemSizes[0].ToInt32(),
                workItemSizes[1].ToInt32(),
                workItemSizes[2].ToInt32());

            // Resolve max threads per group
            MaxNumThreadsPerGroup = CLAPI.GetDeviceInfo <IntPtr>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_MAX_WORK_GROUP_SIZE).ToInt32();

            // Resolve max shared memory per block
            MaxSharedMemoryPerGroup = (int)IntrinsicMath.Min(
                CLAPI.GetDeviceInfo <long>(
                    DeviceId,
                    CLDeviceInfoType.CL_DEVICE_LOCAL_MEM_SIZE),
                int.MaxValue);

            // Resolve total constant memory
            MaxConstantMemory = (int)CLAPI.GetDeviceInfo <long>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_MAX_PARAMETER_SIZE);

            // Resolve clock rate
            ClockRate = CLAPI.GetDeviceInfo <int>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_MAX_CLOCK_FREQUENCY);

            // Resolve number of multiprocessors
            NumMultiprocessors = CLAPI.GetDeviceInfo <int>(
                DeviceId,
                CLDeviceInfoType.CL_DEVICE_MAX_COMPUTE_UNITS);

            // Result max number of threads per multiprocessor
            MaxNumThreadsPerMultiprocessor = MaxNumThreadsPerGroup;

            InitVendorFeatures();
            InitSubGroupSupport(acceleratorId);

            Bind();
            DefaultStream = CreateStreamInternal();
            base.Backend  = new CLBackend(Context, Backends.Backend.OSPlatform, Vendor);
        }