[
https://issues.apache.org/jira/browse/MAHOUT-2003?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16088757#comment-16088757
]
Andrew Palumbo commented on MAHOUT-2003:
----------------------------------------
It seems that {{CUSPARSE}} {{dgemm}} requires sorted CSC format.
Testing with unsorted crows (transposed into columns) without sorting results
in:
{code}
CUDATestSuite:
log4j:WARN No appenders could be found for logger
(org.apache.mahout.math.backend.RootSolverFactory$).
log4j:WARN Please initialize the log4j system properly.
log4j:WARN See http://logging.apache.org/log4j/1.2/faq.html#noconfig for more
info.
Mahout JVM Sparse multiplication time: 1898 ms.
- sparse mmul at geometry of 1000 x 1000 %*% 1000 x 1000 density = .2. 5 runs
*** FAILED ***
jcuda.CudaException: cudaErrorIllegalAddress
at jcuda.runtime.JCuda.checkResult(JCuda.java:460)
at jcuda.runtime.JCuda.cudaMemcpy(JCuda.java:4592)
at org.apache.mahout.cuda.package$.prod(package.scala:311)
at
org.apache.mahout.cuda.CUDATestSuite$.getAverageTimeSparse(CUDATestSuite.scala:68)
at
org.apache.mahout.cuda.CUDATestSuite$$anonfun$1.apply$mcV$sp(CUDATestSuite.scala:14)
at
org.apache.mahout.cuda.CUDATestSuite$$anonfun$1.apply(CUDATestSuite.scala:14)
at
org.apache.mahout.cuda.CUDATestSuite$$anonfun$1.apply(CUDATestSuite.scala:14)
at
org.scalatest.Transformer$$anonfun$apply$1.apply$mcV$sp(Transformer.scala:22)
at org.scalatest.OutcomeOf$class.outcomeOf(OutcomeOf.scala:85)
at org.scalatest.OutcomeOf$.outcomeOf(OutcomeOf.scala:104)
...
- sparse mmul at geometry of 1000 x 1000 %*% 1000 x 1000 density = .02. 5 runs
*** FAILED **
{code}
> Check to see if CUSPARSE library requires sorted CSC format
> --------------------------------------------------------------
>
> Key: MAHOUT-2003
> URL: https://issues.apache.org/jira/browse/MAHOUT-2003
> Project: Mahout
> Issue Type: Test
> Affects Versions: 0.13.1
> Reporter: Andrew Palumbo
> Assignee: Andrew Palumbo
> Fix For: 0.13.2
>
>
> Test whether CUSPARSE requires sorted CSC(R) if not, it should be an easy
> win- to shave some time off of the conversion to compressed format
> without any sorting.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
