apacheGH-35749: [C++] Handle run-end encoded filters in compute kerne…

…ls (apache#35750)

### Rationale for this change

Boolean arrays (bitmaps) used to represent filters in Arrow take 1 bit per boolean value. If the filter contains long runs, the filter can be run-end encoded and save even more memory.

Using POPCNT, a bitmap can be scanned efficiently for <64 runs of logical values, but a run-end encoded array gives the lengths of the run directly and go beyond word size per run.

These two observations make the case that, for the right dataset, REE filters can be more efficiently processed in compute kernels.

### What changes are included in this PR?

 - [x] `GetFilterOutputSize` can count number of emits from a REE filter
 - [x] `GetTakeIndices` can produce an array of logical indices from a REE filter
 - [x] `"array_filter"` can handle REE filters

### Are these changes tested?

Yes.

### Are there any user-facing changes?

Yes.
* Closes: apache#35749

Lead-authored-by: Felipe Oliveira Carvalho <[email protected]>
Co-authored-by: Antoine Pitrou <[email protected]>
Signed-off-by: Antoine Pitrou <[email protected]>

cpp/src/arrow/compute/api_vector.h

@@ -254,12 +254,18 @@ namespace internal {
 // These internal functions are implemented in kernels/vector_selection.cc
 
 /// \brief Return the number of selected indices in the boolean filter
+///
+/// \param filter a plain or run-end encoded boolean array with or without nulls
+/// \param null_selection how to handle nulls in the filter
 ARROW_EXPORT
 int64_t GetFilterOutputSize(const ArraySpan& filter,
                             FilterOptions::NullSelectionBehavior null_selection);
 
 /// \brief Compute uint64 selection indices for use with Take given a boolean
 /// filter
+///
+/// \param filter a plain or run-end encoded boolean array with or without nulls
+/// \param null_selection how to handle nulls in the filter
 ARROW_EXPORT
 Result<std::shared_ptr<ArrayData>> GetTakeIndices(
     const ArraySpan& filter, FilterOptions::NullSelectionBehavior null_selection,

cpp/src/arrow/compute/kernel.cc

@@ -338,6 +338,10 @@ std::shared_ptr<TypeMatcher> RunEndEncoded(
                                                 std::move(value_type_matcher));
 }
 
+std::shared_ptr<TypeMatcher> RunEndEncoded(Type::type value_type_id) {
+  return RunEndEncoded(SameTypeId(value_type_id));
+}
+
 std::shared_ptr<TypeMatcher> RunEndEncoded(
     std::shared_ptr<TypeMatcher> run_end_type_matcher,
     std::shared_ptr<TypeMatcher> value_type_matcher) {

cpp/src/arrow/compute/kernel.h

@@ -152,6 +152,12 @@ ARROW_EXPORT std::shared_ptr<TypeMatcher> RunEndInteger();
 ARROW_EXPORT std::shared_ptr<TypeMatcher> RunEndEncoded(
     std::shared_ptr<TypeMatcher> value_type_matcher);
 
+/// \brief Match run-end encoded types that use any valid run-end type and
+/// encode specific value types
+///
+/// @param[in] value_type_id a type id that the type of the values field should match
+ARROW_EXPORT std::shared_ptr<TypeMatcher> RunEndEncoded(Type::type value_type_id);
+
 /// \brief Match run-end encoded types that encode specific run-end and value types
 ///
 /// @param[in] run_end_type_matcher a matcher that is applied to the run_ends field

cpp/src/arrow/compute/kernels/ree_util_internal.h

@@ -83,6 +83,18 @@ class ReadWriteValue<ArrowType, in_has_validity_buffer, out_has_validity_buffer,
     return valid;
   }
 
+  /// Pre-conditions guaranteed by the callers:
+  /// - i and j are valid indices into the values buffer
+  /// - the values in i and j are valid
+  bool CompareValuesAt(int64_t i, int64_t j) const {
+    if constexpr (std::is_same_v<ArrowType, BooleanType>) {
+      return bit_util::GetBit(input_values_, i) == bit_util::GetBit(input_values_, j);
+    } else {
+      return (reinterpret_cast<const ValueRepr*>(input_values_))[i] ==
+             (reinterpret_cast<const ValueRepr*>(input_values_))[j];
+    }
+  }
+
   /// \brief Ensure padding is zeroed in validity bitmap.
   void ZeroValidityPadding(int64_t length) const {
     DCHECK(output_values_);
@@ -166,6 +178,11 @@ class ReadWriteValue<ArrowType, in_has_validity_buffer, out_has_validity_buffer,
     return valid;
   }
 
+  bool CompareValuesAt(int64_t i, int64_t j) const {
+    return 0 == memcmp(input_values_ + (i * byte_width_),
+                       input_values_ + (j * byte_width_), byte_width_);
+  }
+
   /// \brief Ensure padding is zeroed in validity bitmap.
   void ZeroValidityPadding(int64_t length) const {
     DCHECK(output_values_);
@@ -253,6 +270,14 @@ class ReadWriteValue<ArrowType, in_has_validity_buffer, out_has_validity_buffer,
     return valid;
   }
 
+  bool CompareValuesAt(int64_t i, int64_t j) const {
+    const offset_type len_i = input_offsets_[i + 1] - input_offsets_[i];
+    const offset_type len_j = input_offsets_[j + 1] - input_offsets_[j];
+    return len_i == len_j &&
+           memcmp(input_values_ + input_offsets_[i], input_values_ + input_offsets_[j],
+                  static_cast<size_t>(len_i));
+  }
+
   /// \brief Ensure padding is zeroed in validity bitmap.
   void ZeroValidityPadding(int64_t length) const {
     DCHECK(output_values_);

cpp/src/arrow/compute/kernels/vector_selection.cc

@@ -332,8 +332,8 @@ void RegisterVectorSelection(FunctionRegistry* registry) {
   VectorKernel filter_base;
   filter_base.init = FilterState::Init;
   RegisterSelectionFunction("array_filter", array_filter_doc, filter_base,
-                            /*selection_type=*/boolean(), filter_kernels,
-                            GetDefaultFilterOptions(), registry);
+                            std::move(filter_kernels), GetDefaultFilterOptions(),
+                            registry);
 
   DCHECK_OK(registry->AddFunction(MakeFilterMetaFunction()));
 
@@ -345,8 +345,7 @@ void RegisterVectorSelection(FunctionRegistry* registry) {
   take_base.init = TakeState::Init;
   take_base.can_execute_chunkwise = false;
   RegisterSelectionFunction("array_take", array_take_doc, take_base,
-                            /*selection_type=*/match::Integer(), take_kernels,
-                            GetDefaultTakeOptions(), registry);
+                            std::move(take_kernels), GetDefaultTakeOptions(), registry);
 
   DCHECK_OK(registry->AddFunction(MakeTakeMetaFunction()));