Merge branch 'CUDA' into MAHOUT-1974

apache · May 9, 2017 · dce10d7 · dce10d7
2 parents aa8fdcf + e073884
commit dce10d7
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 2 deletions.
diff --git a/cuda/pom.xml b/cuda/pom.xml
@@ -192,7 +192,6 @@
       <artifactId>jcusparse</artifactId>
       <version>${jcuda.jcudaVersion}</version>
     </dependency>
-
   </dependencies>
 
 

diff --git a/cuda/src/main/scala/org/apache/mahout/cuda/Context.scala b/cuda/src/main/scala/org/apache/mahout/cuda/Context.scala
@@ -28,6 +28,7 @@ import jcublas._
 import JCublas._
 
 final class Context {
+
   // Enable exceptions for all CUDA libraries
   JCuda.setExceptionsEnabled(true)
   JCusparse.setExceptionsEnabled(true)
@@ -42,6 +43,5 @@ final class Context {
   //TODO: is this needed somehow- via the cusparse library?
   // cusparseCreate(denseHandle)
 
-
 }
 
diff --git a/cuda/src/main/scala/org/apache/mahout/cuda/package.scala b/cuda/src/main/scala/org/apache/mahout/cuda/package.scala
@@ -31,12 +31,16 @@ import scala.collection.JavaConversions._
 import jcuda.runtime.JCuda._
 import jcuda.runtime.cudaMemcpyKind._
 
+
 import jcuda._
 import jcuda.jcublas._
 
 
 
 
+import jcuda.jcusparse.JCusparse._
+
+
 package object cuda {
 
   private implicit val log = getLog(GPUMMul.getClass)
@@ -124,6 +128,7 @@ package object cuda {
 //    }
 //  }
 
+
   /**
     *
     * @param mxSrc
@@ -231,6 +236,7 @@ package object cuda {
     (jumpers, colIdcs, els)
   }
 
+
   /**
     * Dense %*% Dense
     * @param a
@@ -285,10 +291,12 @@ package object cuda {
     // step 1: compute nnz count
     var nnzC = new Array[Int](1)
     nnzC(0) = 0
+
     cusparseXcsrgemmNnz(ctx.sparseHandle, a.trans, b.trans, m, n, k,
       a.descr, a.nonz, a.row_ptr, a.col_ind,
       b.descr, b.nonz, b.row_ptr, b.col_ind,
       c.descr, c.row_ptr, jcuda.Pointer.to(nnzC))
+
     c.nonz = nnzC(0)
     if (c.nonz == 0) {
       var baseC = new Array[Int](1)
@@ -300,6 +308,7 @@ package object cuda {
     // step 2: allocate and compute matrix product
     cudaMalloc(c.col_ind, jcuda.Sizeof.INT * c.nonz);
     cudaMalloc(c.vals, jcuda.Sizeof.DOUBLE * c.nonz);
+
     cusparseDcsrgemm(ctx.sparseHandle, a.trans, b.trans, m, n, k,
       a.descr, a.nonz,
       a.vals, a.row_ptr, a.col_ind,