Examples with And org.apache.sysml.runtime.functionobjects.And used on opensource projects

Example 1 with And

use of org.apache.sysml.runtime.functionobjects.And in project incubator-systemml by apache.

the class LibMatrixCUDA method matrixScalarArithmetic.

/**
 * Entry point to perform elementwise matrix-scalar arithmetic operation specified by op
 *
 * @param ec                execution context
 * @param gCtx              a valid {@link GPUContext}
 * @param instName          the invoking instruction's name for record {@link Statistics}.
 * @param in                input matrix
 * @param outputName        output matrix name
 * @param isInputTransposed true if input transposed
 * @param op                scalar operator
 */
public static void matrixScalarArithmetic(ExecutionContext ec, GPUContext gCtx, String instName, MatrixObject in, String outputName, boolean isInputTransposed, ScalarOperator op) {
    if (ec.getGPUContext(0) != gCtx)
        throw new DMLRuntimeException("GPU : Invalid internal state, the GPUContext set with the ExecutionContext is not the same used to run this LibMatrixCUDA function");
    double constant = op.getConstant();
    if (LOG.isTraceEnabled()) {
        LOG.trace("GPU : matrixScalarArithmetic, scalar: " + constant + ", GPUContext=" + gCtx);
    }
    int outRLen = isInputTransposed ? (int) in.getNumColumns() : (int) in.getNumRows();
    int outCLen = isInputTransposed ? (int) in.getNumRows() : (int) in.getNumColumns();
    // if(!isCUDALibAvailable) {
    if (constant == 0) {
        if (op.fn instanceof Plus || (op.fn instanceof Minus && op instanceof RightScalarOperator) || op.fn instanceof Or) {
            deviceCopy(ec, gCtx, instName, in, outputName, isInputTransposed);
        } else if (op.fn instanceof Multiply || op.fn instanceof And) {
            setOutputToConstant(ec, gCtx, instName, 0.0, outputName, outRLen, outCLen);
        } else if (op.fn instanceof Power) {
            setOutputToConstant(ec, gCtx, instName, 1.0, outputName, outRLen, outCLen);
        } else // TODO:
        // x/0.0 is either +Infinity or -Infinity according to Java.
        // In the context of a matrix, different elements of the matrix
        // could have different values.
        // If the IEEE 754 standard defines otherwise, this logic needs
        // to be re-enabled and the Java computation logic for divide by zero
        // needs to be revisited
        // else if(op.fn instanceof Divide && isSparseAndEmpty(gCtx, in)) {
        // setOutputToConstant(ec, gCtx, instName, Double.NaN, outputName);
        // }
        // else if(op.fn instanceof Divide) {
        // //For division, IEEE 754 defines x/0.0 as INFINITY and 0.0/0.0 as NaN.
        // compareAndSet(ec, gCtx, instName, in, outputName, 0.0, 1e-6, Double.NaN, Double.POSITIVE_INFINITY, Double.POSITIVE_INFINITY);
        // }
        {
            // TODO: Potential to optimize
            matrixScalarOp(ec, gCtx, instName, in, outputName, isInputTransposed, op);
        }
    } else if (constant == 1.0 && op.fn instanceof Or) {
        setOutputToConstant(ec, gCtx, instName, 1.0, outputName, outRLen, outCLen);
    } else if (constant == 1.0 && (op.fn instanceof And || op.fn instanceof Power)) {
        deviceCopy(ec, gCtx, instName, in, outputName, isInputTransposed);
    } else {
        matrixScalarOp(ec, gCtx, instName, in, outputName, isInputTransposed, op);
    }
// }
// else {
// double alpha = 0;
// if(op.fn instanceof Multiply) {
// alpha = op.getConstant();
// }
// else if(op.fn instanceof Divide && op instanceof RightScalarOperator) {
// alpha = Math.pow(op.getConstant(), -1.0);
// }
// else {
// throw new DMLRuntimeException("Unsupported op");
// }
// TODO: Performance optimization: Call cublasDaxpy if(in.getNumRows() == 1 || in.getNumColumns() == 1)
// C = alpha* op( A ) + beta* op ( B )
// dgeam(ec, gCtx, instName, in, in, outputName, isInputTransposed, isInputTransposed, alpha, 0.0);
// }
}

Example 2 with And

use of org.apache.sysml.runtime.functionobjects.And in project systemml by apache.

the class LibMatrixCUDA method matrixScalarArithmetic.

/**
 * Entry point to perform elementwise matrix-scalar arithmetic operation specified by op
 *
 * @param ec                execution context
 * @param gCtx              a valid {@link GPUContext}
 * @param instName          the invoking instruction's name for record {@link Statistics}.
 * @param in                input matrix
 * @param outputName        output matrix name
 * @param isInputTransposed true if input transposed
 * @param op                scalar operator
 */
public static void matrixScalarArithmetic(ExecutionContext ec, GPUContext gCtx, String instName, MatrixObject in, String outputName, boolean isInputTransposed, ScalarOperator op) {
    if (ec.getGPUContext(0) != gCtx)
        throw new DMLRuntimeException("GPU : Invalid internal state, the GPUContext set with the ExecutionContext is not the same used to run this LibMatrixCUDA function");
    double constant = op.getConstant();
    if (LOG.isTraceEnabled()) {
        LOG.trace("GPU : matrixScalarArithmetic, scalar: " + constant + ", GPUContext=" + gCtx);
    }
    int outRLen = isInputTransposed ? (int) in.getNumColumns() : (int) in.getNumRows();
    int outCLen = isInputTransposed ? (int) in.getNumRows() : (int) in.getNumColumns();
    // if(!isCUDALibAvailable) {
    if (constant == 0) {
        if (op.fn instanceof Plus || (op.fn instanceof Minus && op instanceof RightScalarOperator) || op.fn instanceof Or) {
            deviceCopy(ec, gCtx, instName, in, outputName, isInputTransposed);
        } else if (op.fn instanceof Multiply || op.fn instanceof And) {
            setOutputToConstant(ec, gCtx, instName, 0.0, outputName, outRLen, outCLen);
        } else if (op.fn instanceof Power) {
            setOutputToConstant(ec, gCtx, instName, 1.0, outputName, outRLen, outCLen);
        } else // TODO:
        // x/0.0 is either +Infinity or -Infinity according to Java.
        // In the context of a matrix, different elements of the matrix
        // could have different values.
        // If the IEEE 754 standard defines otherwise, this logic needs
        // to be re-enabled and the Java computation logic for divide by zero
        // needs to be revisited
        // else if(op.fn instanceof Divide && isSparseAndEmpty(gCtx, in)) {
        // setOutputToConstant(ec, gCtx, instName, Double.NaN, outputName);
        // }
        // else if(op.fn instanceof Divide) {
        // //For division, IEEE 754 defines x/0.0 as INFINITY and 0.0/0.0 as NaN.
        // compareAndSet(ec, gCtx, instName, in, outputName, 0.0, 1e-6, Double.NaN, Double.POSITIVE_INFINITY, Double.POSITIVE_INFINITY);
        // }
        {
            // TODO: Potential to optimize
            matrixScalarOp(ec, gCtx, instName, in, outputName, isInputTransposed, op);
        }
    } else if (constant == 1.0 && op.fn instanceof Or) {
        setOutputToConstant(ec, gCtx, instName, 1.0, outputName, outRLen, outCLen);
    } else if (constant == 1.0 && (op.fn instanceof And || op.fn instanceof Power)) {
        deviceCopy(ec, gCtx, instName, in, outputName, isInputTransposed);
    } else {
        matrixScalarOp(ec, gCtx, instName, in, outputName, isInputTransposed, op);
    }
// }
// else {
// double alpha = 0;
// if(op.fn instanceof Multiply) {
// alpha = op.getConstant();
// }
// else if(op.fn instanceof Divide && op instanceof RightScalarOperator) {
// alpha = Math.pow(op.getConstant(), -1.0);
// }
// else {
// throw new DMLRuntimeException("Unsupported op");
// }
// TODO: Performance optimization: Call cublasDaxpy if(in.getNumRows() == 1 || in.getNumColumns() == 1)
// C = alpha* op( A ) + beta* op ( B )
// dgeam(ec, gCtx, instName, in, in, outputName, isInputTransposed, isInputTransposed, alpha, 0.0);
// }
}