Support casting between day-time interval and string #5155
Conversation
Signed-off-by: Chong Gao <res_life@163.com>
sql-plugin/src/main/330+/scala/com/nvidia/spark/rapids/shims/GpuIntervalUtils.scala
Outdated
Show resolved
Hide resolved
| val numRows = micros.getRowCount | ||
| val from = DT.fieldToString(startField).toUpperCase | ||
| val to = DT.fieldToString(endField).toUpperCase | ||
| val prefixStr = "INTERVAL '" |
There was a problem hiding this comment.
We can has this prefixStr as a const member of GpuIntervalUtils, instead of creating one each time this method is called.
| // prefix part: INTERVAL ' | ||
| parts += getConstStringVector(prefixStr, numRows) | ||
|
|
||
| // sign part: - or empty |
There was a problem hiding this comment.
We may merge the sign vector and the prefix vector into one vector, to reduce the vectors to be concatenated.
withResource(getConstStringVector("INTERVAL '-", numRows)) { neg =>
withResource(getConstStringVector("INTERVAL '", numRows)) { empty =>
| /** | ||
| * return (micros / div).toString | ||
| */ | ||
| private def devResult(micros: ColumnVector, div: Long): ColumnVector = { |
There was a problem hiding this comment.
devResult -> divResult ?
| /** | ||
| * return (micros / div).toString with padding zero | ||
| */ | ||
| private def devResultWithPadding(micros: ColumnVector, div: Long): ColumnVector = { |
There was a problem hiding this comment.
devResultWithPadding -> divResultWithPadding
res-life
changed the title
|
Depending on #5020 to merge. |
| // `restHolder` only hold one rest Cv; | ||
| // use `resetRest` to close the old one and set a new one | ||
| // make a copy of micros | ||
|
|
There was a problem hiding this comment.
| // `restHolder` only hold one rest Cv; | |
| // use `resetRest` to close the old one and set a new one | |
| // make a copy of micros |
| withResource(getConstStringVector(prefixStr + "-", numRows)) { neg => | ||
| withResource(getConstStringVector(prefixStr, numRows)) { empty => | ||
| less.ifElse(neg, empty) | ||
| } | ||
| } |
There was a problem hiding this comment.
ifElse supports operating on scalar values directly which is more efficient, as we don't need to manifest a column of identical scalars.
| withResource(getConstStringVector(prefixStr + "-", numRows)) { neg => | |
| withResource(getConstStringVector(prefixStr, numRows)) { empty => | |
| less.ifElse(neg, empty) | |
| } | |
| } | |
| withResource(Scalar.fromString(prefixStr + "-")) { negPrefix => | |
| withResource(Scalar.fromString(prefixStr)) { prefix => | |
| less.ifElse(negPrefix, prefix) | |
| } | |
| } |
|
build |
| @@ -120,4 +124,5 @@ object GpuTypeShims { | |||
| */ | |||
| def additionalCommonOperatorSupportedTypes: TypeSig = TypeSig.none | |||
|
|
|||
| def isDayTimeType(t: DataType): Boolean = false | |||
There was a problem hiding this comment.
The name here is problematic on Spark 3.2.x which does support DayTimeIntervalType and could return false when called with a DayTimeIntervalType instance. Maybe this should be named something like isSupportedDayTimeType?
There was a problem hiding this comment.
done, already have isSupportedDayTimeType.
tests/src/test/330/scala/com/nvidia/spark/rapids/GpuIntervalUtilsTest.scala
Outdated
Show resolved
Hide resolved
tests/src/test/330/scala/com/nvidia/spark/rapids/GpuIntervalUtilsTest.scala
Outdated
Show resolved
Hide resolved
|
build |
Contributes #4811
Signed-off-by: Chong Gao res_life@163.com