Nakul Jindal created SYSTEMML-1138:
--------------------------------------

             Summary: Exception thrown when a GPU sparse-sparse matrix multiply 
is performed
                 Key: SYSTEMML-1138
                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1138
             Project: SystemML
          Issue Type: Bug
            Reporter: Nakul Jindal


For a simple program like so:

A = rand(rows=50000, cols=100000, sparsity=0.0003)
B = rand(rows=100000, cols=100, sparsity=0.7)
C = A %*% B
print(toString(C))


This is the exception:
Caused by: jcuda.CudaException: cudaErrorIllegalAddress
        at jcuda.runtime.JCuda.checkResult(JCuda.java:460)
        at jcuda.runtime.JCuda.cudaDeviceSynchronize(JCuda.java:7361)
        at 
org.apache.sysml.runtime.instructions.gpu.context.JCudaObject.columnMajorDenseToRowMajorSparse(JCudaObject.java:1130)
        at 
org.apache.sysml.runtime.matrix.data.LibMatrixCUDA.sparseDenseMatmult(LibMatrixCUDA.java:668)
        at 
org.apache.sysml.runtime.matrix.data.LibMatrixCUDA.eitherSparseMatmult(LibMatrixCUDA.java:573)
        at 
org.apache.sysml.runtime.matrix.data.LibMatrixCUDA.matmult(LibMatrixCUDA.java:538)
        at 
org.apache.sysml.runtime.instructions.gpu.AggregateBinaryGPUInstruction.processInstruction(AggregateBinaryGPUInstruction.java:98)
        at 
org.apache.sysml.runtime.controlprogram.ProgramBlock.executeSingleInstruction(ProgramBlock.java:290)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to