Class GPUObject

  • public class GPUObject
    extends Object
    Handle to a matrix block on the GPU
    • Method Detail

      • getDensePointer

        public jcuda.Pointer getDensePointer()
        Pointer to dense matrix
        a pointer to the dense matrix
      • isDensePointerNull

        public boolean isDensePointerNull()
        Checks if the dense pointer is null
        if the state of dense pointer is null
      • clearDensePointer

        public void clearDensePointer()
        Removes the dense pointer and potential soft reference
      • setDensePointer

        public void setDensePointer​(jcuda.Pointer densePtr)
        Convenience method to directly set the dense matrix pointer on GPU
        densePtr - dense pointer
      • setDirty

        public void setDirty​(boolean flag)
      • clone

        public Object clone()
      • transpose

        public static jcuda.Pointer transpose​(GPUContext gCtx,
                                              jcuda.Pointer densePtr,
                                              int m,
                                              int n,
                                              int lda,
                                              int ldc)
        Transposes a dense matrix on the GPU by calling the cublasDgeam operation
        gCtx - a valid GPUContext
        densePtr - Pointer to dense matrix on the GPU
        m - rows in ouput matrix
        n - columns in output matrix
        lda - rows in input matrix
        ldc - columns in output matrix
        transposed matrix
      • columnMajorDenseToRowMajorSparse

        public static CSRPointer columnMajorDenseToRowMajorSparse​(GPUContext gCtx,
                                                                  jcuda.jcusparse.cusparseHandle cusparseHandle,
                                                                  jcuda.Pointer densePtr,
                                                                  int rows,
                                                                  int cols)
        Convenience method to convert a CSR matrix to a dense matrix on the GPU Since the allocated matrix is temporary, bookkeeping is not updated. Also note that the input dense matrix is expected to be in COLUMN MAJOR FORMAT Caller is responsible for deallocating memory on GPU.
        gCtx - a valid GPUContext
        cusparseHandle - handle to cusparse library
        densePtr - [in] dense matrix pointer on the GPU in row major
        rows - number of rows
        cols - number of columns
        CSR (compressed sparse row) pointer
      • getSparseMatrixCudaPointer

        public CSRPointer getSparseMatrixCudaPointer()
        Convenience method to directly examine the Sparse matrix on GPU
        CSR (compressed sparse row) pointer
      • setSparseMatrixCudaPointer

        public void setSparseMatrixCudaPointer​(CSRPointer sparseMatrixPtr)
        Convenience method to directly set the sparse matrix on GPU Needed for operations like cusparseDcsrgemm(cusparseHandle, int, int, int, int, int, cusparseMatDescr, int, Pointer, Pointer, Pointer, cusparseMatDescr, int, Pointer, Pointer, Pointer, cusparseMatDescr, Pointer, Pointer, Pointer)
        sparseMatrixPtr - CSR (compressed sparse row) pointer
      • denseToSparse

        public void denseToSparse()
        Converts this GPUObject from dense to sparse format.
      • denseRowMajorToColumnMajor

        public void denseRowMajorToColumnMajor()
        Convenience method. Converts Row Major Dense Matrix to Column Major Dense Matrix
      • denseColumnMajorToRowMajor

        public void denseColumnMajorToRowMajor()
        Convenience method. Converts Column Major Dense Matrix to Row Major Dense Matrix
      • sparseToDense

        public void sparseToDense()
        Convert sparse to dense (Performs transpose, use sparseToColumnMajorDense if the kernel can deal with column major format)
      • sparseToDense

        public void sparseToDense​(String instructionName)
        Convert sparse to dense (Performs transpose, use sparseToColumnMajorDense if the kernel can deal with column major format) Also records per instruction invokation of sparseToDense.
        instructionName - Name of the instruction for which statistics are recorded in GPUStatistics
      • sparseToColumnMajorDense

        public void sparseToColumnMajorDense()
        More efficient method to convert sparse to dense but returns dense in column major format
      • isSparse

        public boolean isSparse()
      • isAllocated

        public boolean isAllocated()
      • allocateSparseAndEmpty

        public void allocateSparseAndEmpty()
        Allocates a sparse and empty GPUObject This is the result of operations that are both non zero matrices.
      • allocateAndFillDense

        public void allocateAndFillDense​(double v)
        Allocates a dense matrix of size obtained from the attached matrix metadata and fills it up with a single value
        v - value to fill up the dense matrix
      • isSparseAndEmpty

        public boolean isSparseAndEmpty()
        If this GPUObject is sparse and empty Being allocated is a prerequisite to being sparse and empty.
        true if sparse and empty
      • getNnz

        public long getNnz​(String instName,
                           boolean recomputeDenseNNZ)
        Being allocated is a prerequisite for computing nnz. Note: if the matrix is in dense format, it explicitly re-computes the number of nonzeros.
        instName - instruction name
        recomputeDenseNNZ - recompute NNZ if dense
        the number of nonzeroes
      • acquireDeviceRead

        public boolean acquireDeviceRead​(String opcode)
      • acquireDeviceModifyDense

        public boolean acquireDeviceModifyDense()
      • acquireDeviceModifyDense

        public boolean acquireDeviceModifyDense​(boolean initialize)
      • acquireDeviceModifySparse

        public boolean acquireDeviceModifySparse()
      • acquireDeviceModifySparse

        public boolean acquireDeviceModifySparse​(boolean initialize)
      • acquireHostRead

        public boolean acquireHostRead​(String instName)
        if the data is allocated on the GPU and is dirty, it is copied back to the host memory
        instName - name of the instruction
        true if a copy to host happened, false otherwise
      • isLocked

        public boolean isLocked()
      • addReadLock

        public void addReadLock()
      • addWriteLock

        public void addWriteLock()
      • releaseReadLock

        public void releaseReadLock()
      • releaseWriteLock

        public void releaseWriteLock()
      • resetReadWriteLock

        public void resetReadWriteLock()
      • releaseInput

        public void releaseInput()
        Releases input allocated on GPU
      • releaseOutput

        public void releaseOutput()
        releases output allocated on GPU
      • getSizeOnDevice

        public long getSizeOnDevice()
      • getAllocatedSize

        public long getAllocatedSize()
      • toIntExact

        public static int toIntExact​(long l)
      • clearData

        public void clearData​(String opcode,
                              boolean eager)
                       throws DMLRuntimeException
        Clears the data associated with this GPUObject instance
        opcode - opcode of the instruction
        eager - whether to be done synchronously or asynchronously
        DMLRuntimeException - if error occurs
      • clearGPUObject

        public void clearGPUObject()
      • getJcudaSparseMatrixPtr

        public CSRPointer getJcudaSparseMatrixPtr()
        Pointer to sparse matrix
      • isDirty

        public boolean isDirty()
        Whether this block is dirty on the GPU
      • getDensePointerAddress

        public long getDensePointerAddress()
      • getPointerAddress

        public static long getPointerAddress​(jcuda.Pointer p)