Remzi Yang
Remzi Yang
Blocked by https://github.com/rapidsai/cudf/issues/11530
After removing the `hasNan` config from`GpuArrayMin`(#6317), we can close this issue without fixing it.
If we want to add JNIs for them, should we put them in cudf or spark-rapids-jni?
Hi @revans2 @abellina @ttnghia I want to replace all `Nan`s with `null`s in the `inputProjection` stage: ```scala override lazy val inputProjection: Seq[Expression] = Seq( child, GpuOr(GpuIsNan(child), GpuIsNull(child)), GpuIsNan(child), // Replace...
> I think you want to use [GpuIf](https://github.com/NVIDIA/spark-rapids/blob/8d6f6f5cc115e44d3b6afcdf85f4207f35912a69/sql-plugin/src/main/scala/com/nvidia/spark/rapids/conditionalExpressions.scala#L180). It does the If/Else that you want. There does appear to be an optimized [replace_nans](https://github.com/rapidsai/cudf/blob/11d40a01064c9b4846e6db7d7b299350d0ba5074/cpp/include/cudf/replace.hpp#L138) operation in cudf that you could write...
I guess the bug is that we add `PsNote` on `FloatType` and `DoubleType`, but `ArrayMin` only supports ArrayType. We should add the PsNote for `ArrayType(FloatType)` and `ArrayType(DoubleType)` I guess. But...