zhztheplayer commented on code in PR #9634:
URL: https://github.com/apache/incubator-gluten/pull/9634#discussion_r2109426519
##########
cpp/velox/compute/WholeStageResultIterator.cc:
##########
@@ -572,6 +576,12 @@ std::unordered_map<std::string, std::string>
WholeStageResultIterator::getQueryC
configs[velox::core::QueryConfig::kSparkLegacyStatisticalAggregate] =
std::to_string(veloxCfg_->get<bool>(kSparkLegacyStatisticalAggregate,
false));
+#ifdef GLUTEN_ENABLE_GPU
+ if (!enableCudf_) {
Review Comment:
@jinchengchenghh
Is there any reason for passing `enableCudf` by method calls down to here,
instead of reading a configuration option? I am a bit concerned about such many
APIs have to be changed for adding the parameter?
##########
gluten-substrait/src/main/scala/org/apache/gluten/execution/WholeStageTransformer.scala:
##########
@@ -198,6 +208,12 @@ trait TransformSupport extends ValidatablePlan {
// When true, it will not generate relNode, nor will it generate native
metrics.
def isNoop: Boolean = false
+
+ // If set the isCudf, the config will send to native backend,
+ // and enable the cudf plan or not in runtime stage
+ def setIsCudf: Unit = {
+ isCudf = true
+ }
Review Comment:
A flag like this will break Spark's plan immutability / equality? For
example, the subquery reuse rule will find a non-cudf node and use it to
replace a cudf node since they are actually equal.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
