org.apache.sysml.runtime.matrix.data

Class LibMatrixCUDA

java.lang.Object

org.apache.sysml.runtime.matrix.data.LibMatrixCUDA

Direct Known Subclasses:

LibMatrixCuDNN, LibMatrixCuDNNInputRowFetcher, LibMatrixCuMatMult

public class LibMatrixCUDA
extends Object

All CUDA kernels and library calls are redirected through this class

See Also:

GPUContext, GPUObject

Field Summary

Fields

Modifier and Type	Field and Description
static CudaSupportFunctions	cudaSupportFunctions
protected static int	CUDNN_DATA_TYPE
static String	customKernelSuffix
protected static long	maxNumElementsOfCuDNNTensor
static int	sizeOfDataType

Constructor Summary

Constructors

Constructor and Description
LibMatrixCUDA()

Method Summary

Modifier and Type	Method and Description
static void	abs(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "abs" operation on a matrix on the GPU
static void	acos(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "acos" operation on a matrix on the GPU
static void	asin(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "asin" operation on a matrix on the GPU
static void	atan(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "atan" operation on a matrix on the GPU
static void	axpy(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName, double constant) Performs daxpy operation
static void	biasAdd(GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, org.apache.sysml.runtime.controlprogram.caching.MatrixObject bias, org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock) Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input + matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_add(input, bias) built-in function
static void	biasMultiply(GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, org.apache.sysml.runtime.controlprogram.caching.MatrixObject bias, org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock) Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input * matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_add(input, bias) built-in function
static void	cbind(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName)
static void	ceil(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "ceil" operation on a matrix on the GPU
static void	channelSums(GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock, long C, long HW) Perform channel_sums operations: out = rowSums(matrix(colSums(A), rows=C, cols=HW))
static int	computeNNZ(GPUContext gCtx, jcuda.Pointer densePtr, int length) Utility to compute number of non-zeroes on the GPU
static void	cos(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "cos" operation on a matrix on the GPU
static void	cosh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "cosh" operation on a matrix on the GPU
protected static jcuda.Pointer	dataTypePointerTo(double value)
static void	denseTranspose(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, jcuda.Pointer A, jcuda.Pointer C, long numRowsA, long numColsA) Computes C = t(A)
static void	deviceCopy(String instName, jcuda.Pointer src, jcuda.Pointer dest, int rlen, int clen) Performs a deep copy of input device double pointer corresponding to matrix
static jcuda.Pointer	double2float(GPUContext gCtx, jcuda.Pointer A, jcuda.Pointer ret, int numElems)
static void	exp(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "exp" operation on a matrix on the GPU
static jcuda.Pointer	float2double(GPUContext gCtx, jcuda.Pointer A, jcuda.Pointer ret, int numElems)
static void	floor(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "floor" operation on a matrix on the GPU
protected static jcuda.jcublas.cublasHandle	getCublasHandle(GPUContext gCtx)
static JCudaKernels	getCudaKernels(GPUContext gCtx)
protected static jcuda.jcusparse.cusparseHandle	getCusparseHandle(GPUContext gCtx)
static org.apache.sysml.runtime.controlprogram.caching.MatrixObject	getDenseMatrixOutputForGPUInstruction(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, String instName, String name, long numRows, long numCols) Helper method to get the output block (allocated on the GPU) Also records performance information into Statistics
static jcuda.Pointer	getDensePointer(GPUContext gCtx, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, String instName) Convenience method to get jcudaDenseMatrixPtr.
static long	getNnz(GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject mo, boolean recomputeDenseNNZ) Note: if the matrix is in dense format, it explicitly re-computes the number of nonzeros.
protected static CSRPointer	getSparsePointer(GPUContext gCtx, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, String instName) Convenience method to get the sparse matrix pointer from a MatrixObject.
static boolean	isInSparseFormat(GPUContext gCtx, org.apache.sysml.runtime.controlprogram.caching.MatrixObject mo)
static void	log(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "log" operation on a matrix on the GPU
static void	matmultTSMM(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject left, String outputName, boolean isLeftTransposed) Performs tsmm, A %*% A' or A' %*% A, on GPU by exploiting cublasDsyrk(...)
static void	matrixMatrixArithmetic(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName, boolean isLeftTransposed, boolean isRightTransposed, org.apache.sysml.runtime.matrix.operators.BinaryOperator op) Performs elementwise arithmetic operation specified by op of two input matrices in1 and in2
static void	matrixMatrixRelational(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName, org.apache.sysml.runtime.matrix.operators.BinaryOperator op) Performs elementwise operation relational specified by op of two input matrices in1 and in2
static void	matrixScalarArithmetic(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in, String outputName, boolean isInputTransposed, org.apache.sysml.runtime.matrix.operators.ScalarOperator op) Entry point to perform elementwise matrix-scalar arithmetic operation specified by op
static void	matrixScalarOp(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in, String outputName, boolean isInputTransposed, org.apache.sysml.runtime.matrix.operators.ScalarOperator op) Utility to do matrix-scalar operation kernel
static void	matrixScalarRelational(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in, String outputName, org.apache.sysml.runtime.matrix.operators.ScalarOperator op) Entry point to perform elementwise matrix-scalar relational operation specified by op
static jcuda.Pointer	one() Convenience method to get a pointer to value '1.0' on device.
static void	rbind(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName)
static void	reluBackward(GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject input, org.apache.sysml.runtime.controlprogram.caching.MatrixObject dout, org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock) This method computes the backpropagation errors for previous layer of relu operation
static void	resetFloatingPointPrecision() Sets the internal state based on the DMLScript.DATA_TYPE
static void	round(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "round" operation on a matrix on the GPU
static void	sigmoid(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "sigmoid" operation on a matrix on the GPU
static void	sign(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "sign" operation on a matrix on the GPU
static void	sin(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "sin" operation on a matrix on the GPU
static void	sinh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "sinh" operation on a matrix on the GPU
protected static void	sliceDenseDense(GPUContext gCtx, String instName, jcuda.Pointer inPointer, jcuda.Pointer outPointer, int rl, int ru, int cl, int cu, int inClen) Perform slice operation on dense input and output it in dense format
static void	sliceOperations(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.util.IndexRange ixrange, String outputName) Method to perform rightIndex operation for a given lower and upper bounds in row and column dimensions.
protected static void	sliceSparseDense(GPUContext gCtx, String instName, CSRPointer inPointer, jcuda.Pointer outPointer, int rl, int ru, int cl, int cu, int inClen) Perform slice operation on sparse input and output it in dense format
static void	solve(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2, String outputName) Implements the "solve" function for systemml Ax = B (A is of size m*n, B is of size m*1, x is of size n*1)
static void	sqrt(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "sqrt" operation on a matrix on the GPU
static void	tan(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "tan" operation on a matrix on the GPU
static void	tanh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String outputName) Performs an "tanh" operation on a matrix on the GPU
static int	toInt(long num)
static void	transpose(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in, String outputName) Transposes the input matrix using cublasDgeam
static void	unaryAggregate(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec, GPUContext gCtx, String instName, org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1, String output, org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator op) Entry point to perform Unary aggregate operations on the GPU.
static jcuda.Pointer	zero() Convenience method to get a pointer to value '0.0f' on device.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

CUDNN_DATA_TYPE

protected static int CUDNN_DATA_TYPE

cudaSupportFunctions

public static CudaSupportFunctions cudaSupportFunctions

sizeOfDataType

public static int sizeOfDataType

customKernelSuffix

public static String customKernelSuffix

maxNumElementsOfCuDNNTensor

protected static long maxNumElementsOfCuDNNTensor

Constructor Detail

LibMatrixCUDA

public LibMatrixCUDA()

Method Detail

resetFloatingPointPrecision

public static void resetFloatingPointPrecision()

Sets the internal state based on the DMLScript.DATA_TYPE

isInSparseFormat

public static boolean isInSparseFormat(GPUContext gCtx,
                                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject mo)

getNnz

public static long getNnz(GPUContext gCtx,
                          String instName,
                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject mo,
                          boolean recomputeDenseNNZ)

Note: if the matrix is in dense format, it explicitly re-computes the number of nonzeros.

Parameters:

gCtx - a valid GPU context

instName - instruction name

mo - matrix object

recomputeDenseNNZ - recompute NNZ if dense

Returns:

number of non-zeroes

getCusparseHandle

protected static jcuda.jcusparse.cusparseHandle getCusparseHandle(GPUContext gCtx)

getCublasHandle

protected static jcuda.jcublas.cublasHandle getCublasHandle(GPUContext gCtx)

getCudaKernels

public static JCudaKernels getCudaKernels(GPUContext gCtx)
                                   throws DMLRuntimeException

Throws:

DMLRuntimeException

double2float

public static jcuda.Pointer double2float(GPUContext gCtx,
                                         jcuda.Pointer A,
                                         jcuda.Pointer ret,
                                         int numElems)

float2double

public static jcuda.Pointer float2double(GPUContext gCtx,
                                         jcuda.Pointer A,
                                         jcuda.Pointer ret,
                                         int numElems)

one

public static jcuda.Pointer one()

Convenience method to get a pointer to value '1.0' on device. Instead of allocating and deallocating it for every kernel invocation.

Returns:

jcuda pointer

zero

public static jcuda.Pointer zero()

Convenience method to get a pointer to value '0.0f' on device. Instead of allocating and deallocating it for every kernel invocation.

Returns:

jcuda pointer

getDensePointer

public static jcuda.Pointer getDensePointer(GPUContext gCtx,
                                            org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                                            String instName)
                                     throws DMLRuntimeException

Convenience method to get jcudaDenseMatrixPtr. This method explicitly converts sparse to dense format, so use it judiciously.

Parameters:

gCtx - a valid GPUContext

input - input matrix object

instName - the invoking instruction's name for record Statistics.

Returns:

jcuda pointer

Throws:

DMLRuntimeException

getSparsePointer

protected static CSRPointer getSparsePointer(GPUContext gCtx,
                                             org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                                             String instName)

Convenience method to get the sparse matrix pointer from a MatrixObject. Converts dense to sparse if necessary.

Parameters:

gCtx - a valid GPUContext

input - input matrix

instName - the invoking instruction's name for record Statistics.

Returns:

a sparse matrix pointer

dataTypePointerTo

protected static jcuda.Pointer dataTypePointerTo(double value)

reluBackward

public static void reluBackward(GPUContext gCtx,
                                String instName,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject dout,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock)

This method computes the backpropagation errors for previous layer of relu operation

Parameters:

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

input - input image

dout - next layer error propogation

outputBlock - output

channelSums

public static void channelSums(GPUContext gCtx,
                               String instName,
                               org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                               org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock,
                               long C,
                               long HW)

Perform channel_sums operations: out = rowSums(matrix(colSums(A), rows=C, cols=HW))

Parameters:

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

input - input image

outputBlock - output

C - number of channels

HW - height*width

biasMultiply

public static void biasMultiply(GPUContext gCtx,
                                String instName,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject bias,
                                org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock)

Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input * matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_add(input, bias) built-in function

Parameters:

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

input - input image

bias - bias

outputBlock - output

biasAdd

public static void biasAdd(GPUContext gCtx,
                           String instName,
                           org.apache.sysml.runtime.controlprogram.caching.MatrixObject input,
                           org.apache.sysml.runtime.controlprogram.caching.MatrixObject bias,
                           org.apache.sysml.runtime.controlprogram.caching.MatrixObject outputBlock)

Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input + matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_add(input, bias) built-in function

Parameters:

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

input - input image

bias - bias

outputBlock - output

matmultTSMM

public static void matmultTSMM(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                               GPUContext gCtx,
                               String instName,
                               org.apache.sysml.runtime.controlprogram.caching.MatrixObject left,
                               String outputName,
                               boolean isLeftTransposed)

Performs tsmm, A %*% A' or A' %*% A, on GPU by exploiting cublasDsyrk(...)

Memory Usage - If dense, input space - rows * cols, no intermediate memory, output - Max(rows*rows, cols*cols) If sparse, calls matmult

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

left - input matrix, as in a tsmm expression like A %*% A' or A' %*% A, we just need to check whether the left one is transposed or not, I named it 'left'

outputName - output matrix name

isLeftTransposed - if true, left transposed

unaryAggregate

public static void unaryAggregate(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                  GPUContext gCtx,
                                  String instName,
                                  org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                                  String output,
                                  org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator op)

Entry point to perform Unary aggregate operations on the GPU. The execution context object is used to allocate memory for the GPU.

Parameters:

ec - Instance of ExecutionContext, from which the output variable will be allocated

gCtx - a valid GPUContext

instName - name of the invoking instruction to recordStatistics.

in1 - input matrix

output - output matrix/scalar name

op - Instance of AggregateUnaryOperator which encapsulates the direction of reduction/aggregation and the reduction operation.

matrixScalarRelational

public static void matrixScalarRelational(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                          GPUContext gCtx,
                                          String instName,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in,
                                          String outputName,
                                          org.apache.sysml.runtime.matrix.operators.ScalarOperator op)

Entry point to perform elementwise matrix-scalar relational operation specified by op

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in - input matrix

outputName - output matrix name

op - scalar operator

matrixScalarArithmetic

public static void matrixScalarArithmetic(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                          GPUContext gCtx,
                                          String instName,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in,
                                          String outputName,
                                          boolean isInputTransposed,
                                          org.apache.sysml.runtime.matrix.operators.ScalarOperator op)

Entry point to perform elementwise matrix-scalar arithmetic operation specified by op

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in - input matrix

outputName - output matrix name

isInputTransposed - true if input transposed

op - scalar operator

matrixMatrixRelational

public static void matrixMatrixRelational(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                          GPUContext gCtx,
                                          String instName,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                                          String outputName,
                                          org.apache.sysml.runtime.matrix.operators.BinaryOperator op)

Performs elementwise operation relational specified by op of two input matrices in1 and in2

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix 1

in2 - input matrix 2

outputName - output matrix name

op - binary operator

matrixMatrixArithmetic

public static void matrixMatrixArithmetic(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                          GPUContext gCtx,
                                          String instName,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                                          org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                                          String outputName,
                                          boolean isLeftTransposed,
                                          boolean isRightTransposed,
                                          org.apache.sysml.runtime.matrix.operators.BinaryOperator op)

Performs elementwise arithmetic operation specified by op of two input matrices in1 and in2

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix 1

in2 - input matrix 2

outputName - output matrix name

isLeftTransposed - true if left-transposed

isRightTransposed - true if right-transposed

op - binary operator

matrixScalarOp

public static void matrixScalarOp(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                  GPUContext gCtx,
                                  String instName,
                                  org.apache.sysml.runtime.controlprogram.caching.MatrixObject in,
                                  String outputName,
                                  boolean isInputTransposed,
                                  org.apache.sysml.runtime.matrix.operators.ScalarOperator op)

Utility to do matrix-scalar operation kernel

Parameters:

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

ec - execution context

in - input matrix

outputName - output variable name

isInputTransposed - true if input is transposed

op - operator

deviceCopy

public static void deviceCopy(String instName,
                              jcuda.Pointer src,
                              jcuda.Pointer dest,
                              int rlen,
                              int clen)

Performs a deep copy of input device double pointer corresponding to matrix

Parameters:

instName - the invoking instruction's name for record Statistics.

src - source matrix

dest - destination matrix

rlen - number of rows

clen - number of columns

denseTranspose

public static void denseTranspose(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                  GPUContext gCtx,
                                  String instName,
                                  jcuda.Pointer A,
                                  jcuda.Pointer C,
                                  long numRowsA,
                                  long numColsA)
                           throws DMLRuntimeException

Computes C = t(A)

Parameters:

ec - execution context

gCtx - gpu context

instName - name of the instruction

A - pointer to the input matrix

C - pointer to the output matrix

numRowsA - number of rows of the input matrix

numColsA - number of columns of the output matrix

Throws:

DMLRuntimeException - if error

transpose

public static void transpose(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                             GPUContext gCtx,
                             String instName,
                             org.apache.sysml.runtime.controlprogram.caching.MatrixObject in,
                             String outputName)

Transposes the input matrix using cublasDgeam

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in - input matrix

outputName - output matrix name

toInt

public static int toInt(long num)

sliceOperations

public static void sliceOperations(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                   GPUContext gCtx,
                                   String instName,
                                   org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                                   org.apache.sysml.runtime.util.IndexRange ixrange,
                                   String outputName)

Method to perform rightIndex operation for a given lower and upper bounds in row and column dimensions.

Parameters:

ec - current execution context

gCtx - current gpu context

instName - name of the instruction for maintaining statistics

in1 - input matrix object

ixrange - index range (0-based)

outputName - output matrix object

sliceDenseDense

protected static void sliceDenseDense(GPUContext gCtx,
                                      String instName,
                                      jcuda.Pointer inPointer,
                                      jcuda.Pointer outPointer,
                                      int rl,
                                      int ru,
                                      int cl,
                                      int cu,
                                      int inClen)

Perform slice operation on dense input and output it in dense format

Parameters:

gCtx - gpu context

instName - instruction name

inPointer - dense input pointer

outPointer - dense output pointer (doesnot need to be zeroed out)

rl - row lower

ru - row upper

cl - column lower

cu - column upper

inClen - input number of columns

sliceSparseDense

protected static void sliceSparseDense(GPUContext gCtx,
                                       String instName,
                                       CSRPointer inPointer,
                                       jcuda.Pointer outPointer,
                                       int rl,
                                       int ru,
                                       int cl,
                                       int cu,
                                       int inClen)

Perform slice operation on sparse input and output it in dense format

Parameters:

gCtx - gpu context

instName - instruction name

inPointer - sparse CSR input pointer

outPointer - dense output pointer (expected to be zeroed out)

rl - row lower

ru - row upper

cl - column lower

cu - column upper

inClen - number of columns of input matrix

cbind

public static void cbind(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                         GPUContext gCtx,
                         String instName,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                         String outputName)

rbind

public static void rbind(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                         GPUContext gCtx,
                         String instName,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                         String outputName)

exp

public static void exp(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "exp" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

sqrt

public static void sqrt(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "sqrt" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

round

public static void round(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                         GPUContext gCtx,
                         String instName,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                         String outputName)

Performs an "round" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

abs

public static void abs(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "abs" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

log

public static void log(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "log" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

floor

public static void floor(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                         GPUContext gCtx,
                         String instName,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                         String outputName)

Performs an "floor" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

ceil

public static void ceil(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "ceil" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

sin

public static void sin(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "sin" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

cos

public static void cos(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "cos" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

tan

public static void tan(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                       GPUContext gCtx,
                       String instName,
                       org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                       String outputName)

Performs an "tan" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

sinh

public static void sinh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "sinh" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

cosh

public static void cosh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "cosh" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

tanh

public static void tanh(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "tanh" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

asin

public static void asin(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "asin" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

acos

public static void acos(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "acos" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

atan

public static void atan(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "atan" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

sign

public static void sign(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        String outputName)

Performs an "sign" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

sigmoid

public static void sigmoid(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                           GPUContext gCtx,
                           String instName,
                           org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                           String outputName)

Performs an "sigmoid" operation on a matrix on the GPU

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix

outputName - output matrix name

axpy

public static void axpy(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                        GPUContext gCtx,
                        String instName,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                        org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                        String outputName,
                        double constant)

Performs daxpy operation

Parameters:

ec - execution context

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix 1

in2 - input matrix 2

outputName - output matrix name

constant - pointer constant

solve

public static void solve(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                         GPUContext gCtx,
                         String instName,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in1,
                         org.apache.sysml.runtime.controlprogram.caching.MatrixObject in2,
                         String outputName)

Implements the "solve" function for systemml Ax = B (A is of size m*n, B is of size m*1, x is of size n*1)

Parameters:

ec - a valid ExecutionContext

gCtx - a valid GPUContext

instName - the invoking instruction's name for record Statistics.

in1 - input matrix A

in2 - input matrix B

outputName - name of the output matrix

getDenseMatrixOutputForGPUInstruction

public static org.apache.sysml.runtime.controlprogram.caching.MatrixObject getDenseMatrixOutputForGPUInstruction(org.apache.sysml.runtime.controlprogram.context.ExecutionContext ec,
                                                                                                                 String instName,
                                                                                                                 String name,
                                                                                                                 long numRows,
                                                                                                                 long numCols)

Helper method to get the output block (allocated on the GPU) Also records performance information into Statistics

Parameters:

ec - active ExecutionContext

instName - the invoking instruction's name for record Statistics.

name - name of input matrix (that the ExecutionContext is aware of)

numRows - number of rows of output matrix object

numCols - number of columns of output matrix object

Returns:

the matrix object

computeNNZ

public static int computeNNZ(GPUContext gCtx,
                             jcuda.Pointer densePtr,
                             int length)

Utility to compute number of non-zeroes on the GPU

Parameters:

gCtx - the associated GPUContext

densePtr - device pointer to the dense matrix

length - length of the dense pointer

Returns:

the number of non-zeroes