Error when running on older Android device - FilteAttribute not found

[GregCornis]

Hi,

I've been using multik in an Android app without a problem for a while (amazing lib by the way!). However, it seems to crash on an older Android device, with a NoClassDefFoundError.
I have managed to fix the issue by replacing the multik-default dependency with multik-kotlin (the error pointed to openblas.JvmNativeEngine). However :

1. I am not sure what the difference is. Is there any performance impact ?
2. I am curious about the source of the error. Is it an android API which is missing, or something like that ?

java.util.ServiceConfigurationError: org.jetbrains.kotlinx.multik.api.Engine: Provider org.jetbrains.kotlinx.multik.openblas.JvmNativeEngine could not be instantiated: java.lang.NoClassDefFoundError: Failed resolution of: [Ljava/nio/file/attribute/FileAttribute;
at java.util.ServiceLoader.fail(ServiceLoader.java:225)
at java.util.ServiceLoader.-wrap1(ServiceLoader.java)
at java.util.ServiceLoader$LazyIterator.next(ServiceLoader.java:380)
at java.util.ServiceLoader$1.next(ServiceLoader.java:448)
at kotlin.collections.CollectionsKt___CollectionsKt.toCollection(_Collections.kt:1295)
at kotlin.collections.CollectionsKt___CollectionsKt.toMutableList(_Collections.kt:1328)
at kotlin.collections.CollectionsKt___CollectionsKt.toList(_Collections.kt:1319)
at org.jetbrains.kotlinx.multik.api.EngineKt.enginesProvider(Engine.kt:9)
at org.jetbrains.kotlinx.multik.api.Engine.<clinit>(Engine.kt:68)
at org.jetbrains.kotlinx.multik.api.Multik.getMath(Multik.kt:48)
at org.jetbrains.kotlinx.multik.ndarray.operations.IteratingNDArrayKt.sum(IteratingNDArray.kt:1242)

[devcrocod]

Hey, Native code is used In multik-openblas. Currently, only the arm8 architecture is supported, so it may not work on older devices. I plan to stabilize the native code for Android in the future.

[GregCornis]

Nice, thanks :) and regarding the performance, is there a significant difference between multik-kotlin and multik-openblas ?

[devcrocod]

multik-kotlin and multik-openblas implement the Math, LinearAlgebra, and Statistics interfaces. Everything else is implemented in multik-core. multik-openblas provides significant performance improvement when using linear algebra methods such as dot or solve on medium to large arrays. However, the overall impact may vary depending on your project.

[GregCornis]

Alright thanks 👍