Examples with ConvolutionParameters org.apache.sysml.runtime.matrix.data.ConvolutionParameters used on opensource projects

Example 1 with ConvolutionParameters

use of org.apache.sysml.runtime.matrix.data.ConvolutionParameters in project incubator-systemml by apache.

the class ConvolutionOp method inferOutputCharacteristics.

@Override
protected long[] inferOutputCharacteristics(MemoTable memo) {
    // [numRows, numCols, NNZ] 
    long[] ret = new long[3];
    if (op == ConvOp.BIAS_ADD || op == ConvOp.BIAS_MULTIPLY) {
        MatrixCharacteristics[] mc = memo.getAllInputStats(getInput());
        ret[0] = mc[0].rowsKnown() ? mc[0].getRows() : -1;
        ret[1] = mc[0].colsKnown() ? mc[0].getCols() : -1;
        ret[2] = -1;
        return (ret[0] > 0 && ret[1] > 0) ? ret : null;
    }
    ConvolutionParameters params;
    try {
        params = parseInput();
    } catch (DMLRuntimeException e) {
        throw new RuntimeException(e);
    }
    switch(op) {
        case MAX_POOLING:
            {
                // input
                long N = getInput().get(0)._dim1;
                ret[0] = N;
                ret[1] = getExtractedVal(params.C, params.P, params.Q);
                ret[2] = -1;
                break;
            }
        case DIRECT_CONV2D:
            {
                // input, filter
                long N = getInput().get(0)._dim1;
                ret[0] = N;
                ret[1] = getExtractedVal(params.K, params.P, params.Q);
                ret[2] = -1;
                break;
            }
        case DIRECT_CONV2D_BACKWARD_FILTER:
            {
                // input, dout	
                ret[0] = params.K;
                ret[1] = getExtractedVal(params.C, params.R, params.S);
                ret[2] = -1;
                break;
            }
        case MAX_POOLING_BACKWARD:
            {
                // input, dout
                ret[0] = getInput().get(0)._dim1;
                ret[1] = getInput().get(0)._dim2;
                ret[2] = -1;
                break;
            }
        case DIRECT_CONV2D_BACKWARD_DATA:
            {
                // filter, dout
                long N = getInput().get(1)._dim1;
                ret[0] = N;
                ret[1] = getExtractedVal(params.C, params.H, params.W);
                ret[2] = -1;
                break;
            }
        default:
            throw new RuntimeException("Unsupported op:" + op.name());
    }
    if (LOG.isDebugEnabled() && (ret[0] <= 0 || ret[1] <= 0)) {
        LOG.debug("Unknown dimensions for ConvolutionOp in inferOutputCharacteristics:" + op.name() + " " + ret[0] + " " + ret[1] + " img_dim=[" + params.N + " " + params.C + " " + params.H + " " + params.W + "]" + " filter_dim=[" + params.K + " " + params.C + " " + params.H + " " + params.W + "]" + " output_feature_map=[" + params.P + " " + params.Q + "] stride=[" + params.stride_h + " " + params.stride_w + "]" + " pad=[" + params.pad_h + " " + params.pad_w + "]");
    }
    //safe return (create entry only if at least dims known)
    return (ret[0] > 0 && ret[1] > 0) ? ret : null;
}

Example 2 with ConvolutionParameters

use of org.apache.sysml.runtime.matrix.data.ConvolutionParameters in project incubator-systemml by apache.

the class ConvolutionSPInstruction method processInstruction.

@Override
public void processInstruction(ExecutionContext ec) throws DMLRuntimeException {
    SparkExecutionContext sec = (SparkExecutionContext) ec;
    if (instOpcode.equalsIgnoreCase("conv2d") || instOpcode.equalsIgnoreCase("conv2d_bias_add") || instOpcode.equalsIgnoreCase("maxpooling") || instOpcode.equalsIgnoreCase("relu_maxpooling")) {
        String rddVar = input1.getName();
        int numRowsPerBlock = 1;
        JavaPairRDD<MatrixIndexes, MatrixBlock> inputRDD = reblockAsRectangularMatrices(sec, rddVar, numRowsPerBlock);
        MatrixCharacteristics mcRdd = sec.getMatrixCharacteristics(rddVar);
        // ------------------------------------
        // TODO: Handle large filters > 2G
        Broadcast<MatrixBlock> filterBroadcast = null;
        Broadcast<MatrixBlock> biasBroadcast = null;
        if (instOpcode.equalsIgnoreCase("conv2d")) {
            filterBroadcast = getBroadcast(sec, _in2.getName());
        } else if (instOpcode.equalsIgnoreCase("conv2d_bias_add")) {
            filterBroadcast = getBroadcast(sec, _in3.getName());
            biasBroadcast = getBroadcast(sec, _in2.getName());
        }
        // ------------------------------------
        int pad_h = getScalarInput(ec, _padding, 0);
        int pad_w = getScalarInput(ec, _padding, 1);
        int stride_h = getScalarInput(ec, _stride, 0);
        int stride_w = getScalarInput(ec, _stride, 1);
        // int N = getScalarInput(ec, _input_shape, 0);
        int C = getScalarInput(ec, _input_shape, 1);
        int H = getScalarInput(ec, _input_shape, 2);
        int W = getScalarInput(ec, _input_shape, 3);
        int K = getScalarInput(ec, _filter_shape, 0);
        int R = getScalarInput(ec, _filter_shape, 2);
        int S = getScalarInput(ec, _filter_shape, 3);
        int P = (int) ConvolutionUtils.getP(H, R, stride_h, pad_h);
        int Q = (int) ConvolutionUtils.getQ(W, S, stride_w, pad_w);
        ConvolutionParameters params = new ConvolutionParameters(numRowsPerBlock, C, H, W, K, R, S, stride_h, stride_w, pad_h, pad_w, 1);
        boolean enableNativeBLAS = NativeHelper.isNativeLibraryLoaded();
        JavaPairRDD<MatrixIndexes, MatrixBlock> out = inputRDD.mapPartitionsToPair(new RDDConv2dMapMMFunction(filterBroadcast, params, instOpcode, biasBroadcast, mcRdd.getRows(), enableNativeBLAS), true);
        //put output RDD handle into symbol table
        sec.setRDDHandleForVariable(output.getName(), out);
        sec.addLineageRDD(output.getName(), rddVar);
        // TODO: Handle nnz
        long nnz = -1;
        long numCols = ((long) K) * ((long) P) * ((long) Q);
        if (instOpcode.equalsIgnoreCase("maxpooling") || instOpcode.equalsIgnoreCase("relu_maxpooling")) {
            numCols = ((long) C) * ((long) P) * ((long) Q);
        }
        if (numCols > Integer.MAX_VALUE) {
            throw new DMLRuntimeException("The current operator doesnot support large outputs.");
        }
        sec.setMetaData(output.getName(), new MatrixFormatMetaData(new MatrixCharacteristics(mcRdd.getRows(), numCols, numRowsPerBlock, (int) numCols, nnz), OutputInfo.BinaryBlockOutputInfo, InputInfo.BinaryBlockInputInfo));
    } else {
        throw new DMLRuntimeException("Not implemented: " + instOpcode);
    }
}

Example 3 with ConvolutionParameters

use of org.apache.sysml.runtime.matrix.data.ConvolutionParameters in project incubator-systemml by apache.

the class ConvolutionCPInstruction method processInstruction.

@Override
public void processInstruction(ExecutionContext ec) throws DMLRuntimeException {
    if (instOpcode.equalsIgnoreCase("bias_add")) {
        processBiasAddInstruction(ec);
        return;
    } else if (instOpcode.equalsIgnoreCase("bias_multiply")) {
        processBiasMultiplyInstruction(ec);
        return;
    } else if (instOpcode.equalsIgnoreCase("relu_backward")) {
        processReluBackwardInstruction(ec);
        return;
    }
    // acquire inputs
    MatrixBlock outputBlock = null;
    MatrixBlock matBlock = ec.getMatrixInput(input1.getName());
    int pad_h = getScalarInput(ec, _padding, 0);
    int pad_w = getScalarInput(ec, _padding, 1);
    int stride_h = getScalarInput(ec, _stride, 0);
    int stride_w = getScalarInput(ec, _stride, 1);
    int N = getScalarInput(ec, _input_shape, 0);
    int C = getScalarInput(ec, _input_shape, 1);
    int H = getScalarInput(ec, _input_shape, 2);
    int W = getScalarInput(ec, _input_shape, 3);
    int K = getScalarInput(ec, _filter_shape, 0);
    int R = getScalarInput(ec, _filter_shape, 2);
    int S = getScalarInput(ec, _filter_shape, 3);
    int P = (int) ConvolutionUtils.getP(H, R, stride_h, pad_h);
    int Q = (int) ConvolutionUtils.getQ(W, S, stride_w, pad_w);
    ConvolutionParameters params = new ConvolutionParameters(N, C, H, W, K, R, S, stride_h, stride_w, pad_h, pad_w, _numThreads);
    params.enableNative = NativeHelper.isNativeLibraryLoaded();
    if (instOpcode.equalsIgnoreCase("maxpooling") || instOpcode.equalsIgnoreCase("relu_maxpooling")) {
        if (matBlock.isEmpty()) {
            outputBlock = new MatrixBlock(N, C * P * Q, true);
        } else {
            outputBlock = getDenseOutputBlock(N, C * P * Q);
            if (instOpcode.equalsIgnoreCase("maxpooling"))
                Arrays.fill(outputBlock.getDenseBlock(), -Double.MAX_VALUE);
            LibMatrixDNN.maxpooling(matBlock, outputBlock, params);
        }
    } else if (instOpcode.equalsIgnoreCase("maxpooling_backward") || instOpcode.equalsIgnoreCase("relu_maxpooling_backward")) {
        MatrixBlock dout = ec.getMatrixInput(_in2.getName());
        if (matBlock.isEmpty() || dout.isEmpty()) {
            outputBlock = new MatrixBlock(N, C * H * W, true);
        } else {
            outputBlock = getDenseOutputBlock(N, C * H * W);
            if (instOpcode.equalsIgnoreCase("maxpooling_backward"))
                LibMatrixDNN.maxpoolingBackward(matBlock, dout, outputBlock, params, false);
            else
                LibMatrixDNN.maxpoolingBackward(matBlock, dout, outputBlock, params, true);
        }
        ec.releaseMatrixInput(_in2.getName());
    } else if (instOpcode.equalsIgnoreCase("conv2d")) {
        MatrixBlock filter = ec.getMatrixInput(_in2.getName());
        if (filter.isEmpty() || matBlock.isEmpty()) {
            outputBlock = new MatrixBlock(N, K * P * Q, true);
        } else {
            outputBlock = getDenseOutputBlock(N, K * P * Q);
            if (params.enableNative && !isFilterSparse(filter) && !matBlock.isInSparseFormat())
                LibMatrixNative.conv2d(matBlock, filter, outputBlock, params);
            else
                LibMatrixDNN.conv2d(matBlock, filter, outputBlock, params);
        }
        ec.releaseMatrixInput(_in2.getName());
    } else if (instOpcode.equalsIgnoreCase("conv2d_bias_add")) {
        MatrixBlock filter = ec.getMatrixInput(_in3.getName());
        MatrixBlock bias = ec.getMatrixInput(_in2.getName());
        if (bias.getNumRows() != params.K || bias.getNumColumns() != 1) {
            throw new DMLRuntimeException("Incorrect shape of bias matrix: [" + bias.getNumRows() + " " + bias.getNumColumns() + "]. " + "Expected: [" + params.K + ", 1]");
        }
        boolean isOutputConvEmpty = filter.isEmpty() || matBlock.isEmpty();
        if (isOutputConvEmpty && bias.isEmpty()) {
            // bias_add(empty mb, empty mb) = empty mb
            outputBlock = new MatrixBlock(N, K * P * Q, true);
        } else if (isOutputConvEmpty && !bias.isEmpty()) {
            // Add bias to empty output block
            // bias_add(empty mb, bias)
            outputBlock = getDenseOutputBlock(N, K * P * Q);
            for (int n = 0; n < params.N; n++) ConvolutionUtils.fillBias(bias, outputBlock.getDenseBlock(), n, n + 1, params.N, params.K, params.P * params.Q);
        } else {
            outputBlock = getDenseOutputBlock(N, K * P * Q);
            if (!bias.isEmpty()) {
                // Handle situation where both input and filter are non empty, but bias is empty
                params.bias = bias;
            }
            if (params.enableNative && !isFilterSparse(filter) && !matBlock.isInSparseFormat())
                LibMatrixNative.conv2d(matBlock, filter, outputBlock, params);
            else
                LibMatrixDNN.conv2d(matBlock, filter, outputBlock, params);
        }
        ec.releaseMatrixInput(_in3.getName());
        ec.releaseMatrixInput(_in2.getName());
    } else if (instOpcode.equalsIgnoreCase("conv2d_backward_filter")) {
        MatrixBlock dout = ec.getMatrixInput(_in2.getName());
        if (dout.isEmpty() || matBlock.isEmpty()) {
            outputBlock = new MatrixBlock(K, C * R * S, true);
        } else {
            outputBlock = getDenseOutputBlock(K, C * R * S);
            if (params.enableNative && !matBlock.isInSparseFormat() && !dout.isInSparseFormat())
                LibMatrixNative.conv2dBackwardFilter(matBlock, dout, outputBlock, params);
            else
                LibMatrixDNN.conv2dBackwardFilter(matBlock, dout, outputBlock, params);
        }
        ec.releaseMatrixInput(_in2.getName());
    } else if (instOpcode.equalsIgnoreCase("conv2d_backward_data")) {
        MatrixBlock dout = ec.getMatrixInput(_in2.getName());
        if (dout.isEmpty() || matBlock.isEmpty()) {
            outputBlock = new MatrixBlock(N, C * H * W, true);
        } else {
            outputBlock = getDenseOutputBlock(N, C * H * W);
            if (params.enableNative && !isFilterSparse(matBlock) && !dout.isInSparseFormat())
                LibMatrixNative.conv2dBackwardData(matBlock, dout, outputBlock, params);
            else
                LibMatrixDNN.conv2dBackwardData(matBlock, dout, outputBlock, params);
        }
        ec.releaseMatrixInput(_in2.getName());
    } else {
        throw new DMLRuntimeException("Unsupported op code " + instOpcode);
    }
    // release inputs/outputs
    ec.releaseMatrixInput(input1.getName());
    ec.setMatrixOutput(getOutputVariableName(), outputBlock);
}

Example 4 with ConvolutionParameters

use of org.apache.sysml.runtime.matrix.data.ConvolutionParameters in project incubator-systemml by apache.

the class ConvolutionOp method refreshSizeInformation.

@Override
public void refreshSizeInformation() {
    if (op == ConvOp.BIAS_ADD || op == ConvOp.BIAS_MULTIPLY) {
        Hop input1 = getInput().get(0);
        setDim1(input1.getDim1());
        setDim2(input1.getDim2());
        return;
    }
    ConvolutionParameters params;
    try {
        params = parseInput();
    } catch (DMLRuntimeException e) {
        throw new RuntimeException(e);
    }
    switch(op) {
        case MAX_POOLING:
            {
                // input
                long N = getInput().get(0)._dim1;
                _dim1 = N;
                _dim2 = getExtractedVal(params.C, params.P, params.Q);
                // cannot infer stats
                _nnz = -1;
                break;
            }
        case MAX_POOLING_BACKWARD:
            {
                // input, dout
                _dim1 = getInput().get(0)._dim1;
                _dim2 = getInput().get(0)._dim2;
                _nnz = -1;
                break;
            }
        case DIRECT_CONV2D:
            {
                // input, filter
                long N = getInput().get(0)._dim1;
                _dim1 = N;
                _dim2 = getExtractedVal(params.K, params.P, params.Q);
                // cannot infer stats
                _nnz = -1;
                break;
            }
        case DIRECT_CONV2D_BACKWARD_DATA:
            {
                // filter, dout
                long N = getInput().get(1)._dim1;
                _dim1 = N;
                _dim2 = getExtractedVal(params.C, params.H, params.W);
                // cannot infer stats
                _nnz = -1;
                break;
            }
        case DIRECT_CONV2D_BACKWARD_FILTER:
            {
                // input, dout	
                _dim1 = params.K;
                _dim2 = getExtractedVal(params.C, params.R, params.S);
                // cannot infer stats
                _nnz = -1;
                break;
            }
        default:
            throw new RuntimeException("The sizes are not refreshed for " + op.name());
    }
    if (LOG.isDebugEnabled() && (_dim1 <= 0 || _dim2 <= 0)) {
        LOG.debug("Unknown dimensions for ConvolutionOp in refreshSizeInformation:" + op.name() + " " + _dim1 + " " + _dim2 + " img_dim=[" + params.N + " " + params.C + " " + params.H + " " + params.W + "]" + " filter_dim=[" + params.K + " " + params.C + " " + params.H + " " + params.W + "]" + " output_feature_map=[" + params.P + " " + params.Q + "] stride=[" + params.stride_h + " " + params.stride_w + "]" + " pad=[" + params.pad_h + " " + params.pad_w + "]");
    }
}