luhenry commented on a change in pull request #30810:
URL: https://github.com/apache/spark/pull/30810#discussion_r545049557
##########
File path: mllib-local/src/main/scala/org/apache/spark/ml/linalg/BLAS.scala
##########
@@ -18,28 +18,46 @@
package org.apache.spark.ml.linalg
import com.github.fommil.netlib.{BLAS => NetlibBLAS, F2jBLAS}
-import com.github.fommil.netlib.BLAS.{getInstance => NativeBLAS}
+import scala.util.Try
+
+import org.apache.spark.util.Utils
/**
* BLAS routines for MLlib's vectors and matrices.
*/
private[spark] object BLAS extends Serializable {
- @transient private var _f2jBLAS: NetlibBLAS = _
+ @transient private var _javaBLAS: NetlibBLAS = _
@transient private var _nativeBLAS: NetlibBLAS = _
private val nativeL1Threshold: Int = 256
Review comment:
From the results of the benchmarks (see
https://github.com/apache/spark/pull/30810#issuecomment-747405702), the value
is likely to be a bit bigger when using vectorized implementation and we will
keep something similar for some operations (`dscal`, `dspr`, `dsyr`, `sdot`).
I'll look at why is the native implementation so much faster for simple
operations like `dscal` and `sdot` (especially when `ddot` is equivalent to
native), and what we can apply to the vectorized version.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
