Search in sources :

Example 26 with CudaTensor

use of com.simiacryptus.mindseye.lang.cudnn.CudaTensor in project MindsEye by SimiaCryptus.

the class SumReducerLayer method evalAndFree.

@Nullable
@Override
public Result evalAndFree(final Result... inObj) {
    if (!CudaSystem.isEnabled())
        return getCompatibilityLayer().evalAndFree(inObj);
    final Result input = inObj[0];
    final TensorList inputData = input.getData();
    @Nonnull final int[] inputSize = inputData.getDimensions();
    int length = inputData.length();
    CudaTensorList result = CudaSystem.run(gpu -> {
        CudaTensor inputTensor = gpu.getTensor(inputData, precision, MemoryType.Device, false);
        inputData.freeRef();
        CudaMemory inputMemory = inputTensor.getMemory(gpu);
        @Nonnull final CudaDevice.CudaTensorDescriptor outputDescriptor = gpu.newTensorDescriptor(precision, length, 1, 1, 1);
        long size = (long) precision.size * outputDescriptor.nStride * length;
        @Nonnull final CudaMemory outputMemory = gpu.allocate(size, MemoryType.Managed, true);
        CudaResource<cudnnReduceTensorDescriptor> reduceTensorDescriptor = gpu.cudnnCreateReduceTensorDescriptor(cudnnReduceTensorOp.CUDNN_REDUCE_TENSOR_ADD, precision.code, cudnnNanPropagation.CUDNN_NOT_PROPAGATE_NAN, cudnnReduceTensorIndices.CUDNN_REDUCE_TENSOR_NO_INDICES, cudnnIndicesType.CUDNN_32BIT_INDICES);
        @Nonnull final CudaMemory workspacePtr = gpu.allocate(inputMemory.size, MemoryType.Device, true);
        @Nonnull final CudaMemory indexPtr = gpu.allocate(12 * length, MemoryType.Device, false);
        // outputPtr.synchronize();
        gpu.cudnnReduceTensor(reduceTensorDescriptor.getPtr(), indexPtr.getPtr(), indexPtr.size, workspacePtr.getPtr(), workspacePtr.size, precision.getPointer(1.0), inputTensor.descriptor.getPtr(), inputMemory.getPtr(), precision.getPointer(0.0), outputDescriptor.getPtr(), outputMemory.getPtr());
        inputMemory.dirty();
        outputMemory.dirty();
        workspacePtr.dirty();
        Stream.of(inputTensor, inputMemory, reduceTensorDescriptor, workspacePtr, indexPtr).forEach(ReferenceCounting::freeRef);
        return CudaTensorList.wrap(CudaTensor.wrap(outputMemory, outputDescriptor, precision), length, new int[] { 1, 1, 1 }, precision);
    });
    return new Result(result, (DeltaSet<Layer> ctx, TensorList delta) -> {
        // Not supported by CuDNN?
        // CudaTensorList passback = CudaSystem.run(gpu -> {
        // CudaTensor deltaTensor = gpu.getTensor(delta, precision, MemoryType.Device, false);
        // CudaMemory deltaMemory = deltaTensor.getMemory(gpu);
        // 
        // @Nonnull final CudaDevice.CudaTensorDescriptor passbackDescriptor1 = gpu.newTensorDescriptor(
        // precision, length, inputSize[2], inputSize[1], inputSize[0]
        // );
        // @Nonnull final CudaMemory passbackPtr1 = gpu.allocate((long) precision.size * passbackDescriptor1.nStride * length, MemoryType.Device, false);
        // gpu.cudnnAddTensor(precision.getPointer(1.0), deltaTensor.descriptor.getPtr(), deltaMemory.getPtr(),
        // precision.getPointer(1.0), passbackDescriptor1.getPtr(), passbackPtr1.getPtr());
        // passbackPtr1.dirty();
        // 
        // Stream.of(deltaTensor, deltaMemory, passbackDescriptor1, passbackPtr1).forEach(ReferenceCounting::freeRef);
        // return CudaTensorList.wrap(CudaTensor.wrap(passbackPtr1, passbackDescriptor1, precision), length, inputSize, precision);
        // });
        TensorList passback = TensorArray.wrap(IntStream.range(0, length).mapToObj(i -> {
            Tensor tensor = delta.get(i);
            Tensor tensor1 = new Tensor(inputSize).setAll(tensor.get(0));
            tensor.freeRef();
            return tensor1;
        }).toArray(i -> new Tensor[i]));
        input.accumulate(ctx, passback);
    }) {

        @Override
        protected void _free() {
            super._free();
            input.freeRef();
        }
    };
}
Also used : IntStream(java.util.stream.IntStream) JsonObject(com.google.gson.JsonObject) Arrays(java.util.Arrays) CudaMemory(com.simiacryptus.mindseye.lang.cudnn.CudaMemory) jcuda.jcudnn.cudnnReduceTensorDescriptor(jcuda.jcudnn.cudnnReduceTensorDescriptor) Tensor(com.simiacryptus.mindseye.lang.Tensor) jcuda.jcudnn.cudnnReduceTensorOp(jcuda.jcudnn.cudnnReduceTensorOp) Result(com.simiacryptus.mindseye.lang.Result) DataSerializer(com.simiacryptus.mindseye.lang.DataSerializer) Precision(com.simiacryptus.mindseye.lang.cudnn.Precision) Map(java.util.Map) Layer(com.simiacryptus.mindseye.lang.Layer) ReferenceCounting(com.simiacryptus.mindseye.lang.ReferenceCounting) Nonnull(javax.annotation.Nonnull) Nullable(javax.annotation.Nullable) CudaResource(com.simiacryptus.mindseye.lang.cudnn.CudaResource) CudaDevice(com.simiacryptus.mindseye.lang.cudnn.CudaDevice) CudaTensor(com.simiacryptus.mindseye.lang.cudnn.CudaTensor) CudaTensorList(com.simiacryptus.mindseye.lang.cudnn.CudaTensorList) jcuda.jcudnn.cudnnIndicesType(jcuda.jcudnn.cudnnIndicesType) jcuda.jcudnn.cudnnNanPropagation(jcuda.jcudnn.cudnnNanPropagation) jcuda.jcudnn.cudnnReduceTensorIndices(jcuda.jcudnn.cudnnReduceTensorIndices) List(java.util.List) LayerBase(com.simiacryptus.mindseye.lang.LayerBase) Stream(java.util.stream.Stream) CudaSystem(com.simiacryptus.mindseye.lang.cudnn.CudaSystem) TensorList(com.simiacryptus.mindseye.lang.TensorList) MemoryType(com.simiacryptus.mindseye.lang.cudnn.MemoryType) TensorArray(com.simiacryptus.mindseye.lang.TensorArray) DeltaSet(com.simiacryptus.mindseye.lang.DeltaSet) CudaTensor(com.simiacryptus.mindseye.lang.cudnn.CudaTensor) Tensor(com.simiacryptus.mindseye.lang.Tensor) CudaTensor(com.simiacryptus.mindseye.lang.cudnn.CudaTensor) CudaDevice(com.simiacryptus.mindseye.lang.cudnn.CudaDevice) Nonnull(javax.annotation.Nonnull) jcuda.jcudnn.cudnnReduceTensorDescriptor(jcuda.jcudnn.cudnnReduceTensorDescriptor) CudaMemory(com.simiacryptus.mindseye.lang.cudnn.CudaMemory) DeltaSet(com.simiacryptus.mindseye.lang.DeltaSet) CudaTensorList(com.simiacryptus.mindseye.lang.cudnn.CudaTensorList) TensorList(com.simiacryptus.mindseye.lang.TensorList) Result(com.simiacryptus.mindseye.lang.Result) CudaTensorList(com.simiacryptus.mindseye.lang.cudnn.CudaTensorList) ReferenceCounting(com.simiacryptus.mindseye.lang.ReferenceCounting) Nullable(javax.annotation.Nullable)

Aggregations

CudaTensor (com.simiacryptus.mindseye.lang.cudnn.CudaTensor)26 Nullable (javax.annotation.Nullable)25 Nonnull (javax.annotation.Nonnull)24 TensorList (com.simiacryptus.mindseye.lang.TensorList)22 CudaTensorList (com.simiacryptus.mindseye.lang.cudnn.CudaTensorList)22 DeltaSet (com.simiacryptus.mindseye.lang.DeltaSet)21 Result (com.simiacryptus.mindseye.lang.Result)21 CudaDevice (com.simiacryptus.mindseye.lang.cudnn.CudaDevice)20 CudaMemory (com.simiacryptus.mindseye.lang.cudnn.CudaMemory)20 ReferenceCounting (com.simiacryptus.mindseye.lang.ReferenceCounting)16 JsonObject (com.google.gson.JsonObject)15 DataSerializer (com.simiacryptus.mindseye.lang.DataSerializer)15 Layer (com.simiacryptus.mindseye.lang.Layer)15 CudaSystem (com.simiacryptus.mindseye.lang.cudnn.CudaSystem)15 MemoryType (com.simiacryptus.mindseye.lang.cudnn.MemoryType)15 Precision (com.simiacryptus.mindseye.lang.cudnn.Precision)15 List (java.util.List)15 Map (java.util.Map)15 LayerBase (com.simiacryptus.mindseye.lang.LayerBase)14 Arrays (java.util.Arrays)14