pitrou commented on a change in pull request #10550:
URL: https://github.com/apache/arrow/pull/10550#discussion_r655195574
##########
File path: cpp/src/arrow/compute/kernels/hash_aggregate.cc
##########
@@ -442,6 +442,9 @@ struct GrouperImpl : Grouper {
};
struct GrouperFastImpl : Grouper {
+ static constexpr int kBitmapPaddingForSIMD = 64; // bits
+ static constexpr int kPaddingForSIMD = 32; // bytes
Review comment:
Unfortunately I have no idea why precisely though values. Perhaps
@michalursa can tell.
##########
File path: cpp/src/arrow/compute/kernels/hash_aggregate.cc
##########
@@ -442,6 +442,9 @@ struct GrouperImpl : Grouper {
};
struct GrouperFastImpl : Grouper {
+ static constexpr int kBitmapPaddingForSIMD = 64; // bits
+ static constexpr int kPaddingForSIMD = 32; // bytes
Review comment:
Unfortunately I have no idea why precisely those values. Perhaps
@michalursa can tell.
##########
File path: cpp/src/arrow/compute/kernels/scalar_cast_internal.cc
##########
@@ -255,7 +255,12 @@ static bool CanCastFromDictionary(Type::type type_id) {
void AddCommonCasts(Type::type out_type_id, OutputType out_ty, CastFunction*
func) {
// From null to this type
- DCHECK_OK(func->AddKernel(Type::NA, {null()}, out_ty, CastFromNull));
+ ScalarKernel kernel;
+ kernel.exec = CastFromNull;
+ kernel.signature = KernelSignature::Make({null()}, out_ty);
+ kernel.null_handling = NullHandling::COMPUTED_NO_PREALLOCATE;
+ kernel.mem_allocation = MemAllocation::NO_PREALLOCATE;
Review comment:
Not sure, because `MakeArrayOfNull` allocates a single
(sufficiently-sized) buffer of 0s for all array buffers. Preallocating in the
kernel execution machinery without any specific knowledge about the kernel
would allocate distinct buffers and consume more memory (though probably not
that much).
--
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:
us...@infra.apache.org
