edponce commented on a change in pull request #11793:
URL: https://github.com/apache/arrow/pull/11793#discussion_r760207224
##########
File path: cpp/src/arrow/compute/kernels/scalar_compare.cc
##########
@@ -439,6 +472,329 @@ struct ScalarMinMax {
}
};
+template <typename Type, typename Op>
+struct BinaryScalarMinMax {
+ using ArrayType = typename TypeTraits<Type>::ArrayType;
+ using BuilderType = typename TypeTraits<Type>::BuilderType;
+ using offset_type = typename Type::offset_type;
+
+ static Status Exec(KernelContext* ctx, const ExecBatch& batch, Datum* out) {
+ const ElementWiseAggregateOptions& options = MinMaxState::Get(ctx);
+ if (std::all_of(batch.values.begin(), batch.values.end(),
+ [](const Datum& d) { return d.is_scalar(); })) {
+ return ExecOnlyScalar(ctx, options, batch, out);
+ }
+ return ExecContainingArrays(ctx, options, batch, out);
+ }
+
+ static Status ExecOnlyScalar(KernelContext* ctx,
+ const ElementWiseAggregateOptions& options,
+ const ExecBatch& batch, Datum* out) {
+ if (batch.values.empty()) {
+ return Status::OK();
+ }
+ auto output = checked_cast<BaseBinaryScalar*>(out->scalar().get());
+ if (!options.skip_nulls) {
+ // any nulls in the input will produce a null output
+ for (const auto& value : batch.values) {
+ if (!value.scalar()->is_valid) {
+ output->is_valid = false;
+ return Status::OK();
+ }
+ }
+ }
+ const auto& first_scalar = *batch.values.front().scalar();
+ string_view result = UnboxScalar<Type>::Unbox(first_scalar);
+ bool valid = first_scalar.is_valid;
+ for (size_t i = 1; i < batch.values.size(); i++) {
+ const auto& scalar = *batch[i].scalar();
+ if (!scalar.is_valid) {
+ DCHECK(options.skip_nulls);
+ continue;
+ } else {
+ string_view value = UnboxScalar<Type>::Unbox(scalar);
+ result = !valid ? value : Op::Call(result, value);
+ valid = true;
+ }
+ }
+ if (valid) {
+ ARROW_ASSIGN_OR_RAISE(output->value, ctx->Allocate(result.size()));
+ std::copy(result.begin(), result.end(), output->value->mutable_data());
+ output->is_valid = true;
+ } else {
+ output->is_valid = false;
+ }
+ return Status::OK();
+ }
+
+ static Status ExecContainingArrays(KernelContext* ctx,
+ const ElementWiseAggregateOptions&
options,
+ const ExecBatch& batch, Datum* out) {
+ // Presize data to avoid reallocations
+ int64_t final_size = 0;
+ for (int64_t i = 0; i < batch.length; i++) {
+ auto size = CalculateRowSize(options, batch, i);
+ if (size > 0) final_size += size;
+ }
+ BuilderType builder(ctx->memory_pool());
+ RETURN_NOT_OK(builder.Reserve(batch.length));
+ RETURN_NOT_OK(builder.ReserveData(final_size));
+
+ std::vector<util::optional<string_view>> valid_cols(batch.values.size());
+ for (size_t row = 0; row < static_cast<size_t>(batch.length); row++) {
+ size_t num_valid = 0;
+ for (size_t col = 0; col < batch.values.size(); col++) {
+ if (batch[col].is_scalar()) {
+ const auto& scalar = *batch[col].scalar();
+ if (scalar.is_valid) {
+ valid_cols[col] = UnboxScalar<Type>::Unbox(scalar);
+ num_valid++;
+ } else {
+ valid_cols[col] = util::nullopt;
+ }
+ } else {
+ const ArrayData& array = *batch[col].array();
+ if (!array.MayHaveNulls() ||
+ bit_util::GetBit(array.buffers[0]->data(), array.offset + row)) {
+ const offset_type* offsets = array.GetValues<offset_type>(1);
+ const uint8_t* data = array.GetValues<uint8_t>(2,
/*absolute_offset=*/0);
+ const int64_t length = offsets[row + 1] - offsets[row];
+ valid_cols[col] =
+ string_view(reinterpret_cast<const char*>(data +
offsets[row]), length);
+ num_valid++;
+ } else {
+ valid_cols[col] = util::nullopt;
+ }
+ }
+ }
+
+ if (num_valid == 0 || (num_valid < batch.values.size() &&
!options.skip_nulls)) {
+ // We had some nulls
+ builder.UnsafeAppendNull();
+ continue;
+ }
+ util::optional<string_view> result = valid_cols.front();
+ for (size_t col = 1; col < batch.values.size(); ++col) {
+ util::optional<string_view> value = valid_cols[col];
+ if (!value) {
+ DCHECK(options.skip_nulls);
+ continue;
+ }
+ result = !result ? *value : Op::Call(*result, *value);
+ }
+ if (result) {
+ builder.UnsafeAppend(*result);
+ } else {
+ builder.UnsafeAppendNull();
+ }
+ }
+
+ std::shared_ptr<Array> string_array;
+ RETURN_NOT_OK(builder.Finish(&string_array));
+ *out = *string_array->data();
+ out->mutable_array()->type = batch[0].type();
+ DCHECK_EQ(batch.length, out->array()->length);
+ DCHECK_GE(final_size,
+ checked_cast<const
ArrayType&>(*string_array).total_values_length());
+ return Status::OK();
+ }
+
+ // Compute the length of the output for the given position, or -1 if it
would be null.
+ static int64_t CalculateRowSize(const ElementWiseAggregateOptions& options,
+ const ExecBatch& batch, const int64_t index)
{
+ const auto num_args = batch.values.size();
+ int64_t final_size = 0;
+ for (size_t i = 0; i < num_args; i++) {
+ int64_t element_size = 0;
+ bool valid = true;
+ if (batch[i].is_scalar()) {
+ const auto& scalar = *batch[i].scalar();
+ valid = scalar.is_valid;
+ element_size =
static_cast<int64_t>(UnboxScalar<Type>::Unbox(scalar).size());
+ } else {
+ const ArrayData& array = *batch[i].array();
+ valid = !array.MayHaveNulls() ||
+ bit_util::GetBit(array.buffers[0]->data(), array.offset +
index);
+ const offset_type* offsets = array.GetValues<offset_type>(1);
+ element_size = offsets[index + 1] - offsets[index];
+ }
+ if (!valid) {
+ if (options.skip_nulls) {
+ continue;
+ }
+ return -1;
+ }
+ // Conservative estimation of the element size.
+ final_size = std::max(final_size, element_size);
+ }
+ return final_size;
+ }
+};
+
+template <typename Op>
+struct FixedSizeBinaryScalarMinMax {
Review comment:
Ok, thanks for the clarification. It is fine as is.
--
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]
