first commit set up scaffolding

Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>

Author: connortsui20

encodings/fastlanes/src/bitpacking/array/bitpack_decompress_vector.rs

@@ -0,0 +1,172 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright the Vortex contributors
+
+use vortex_buffer::BufferMut;
+use vortex_compute::filter::Filter;
+use vortex_dtype::match_each_integer_ptype;
+use vortex_mask::Mask;
+use vortex_vector::VectorMutOps;
+use vortex_vector::primitive::{PVector, PVectorMut, PrimitiveVector};
+
+use crate::BitPackedArray;
+use crate::bitpacking::array::BitPacked;
+
+/// Helper function to get the true count of a mask with a default if it doesn't exist.
+fn true_count_of_optional_mask(selection_mask: &Option<&Mask>, default: usize) -> usize {
+    selection_mask
+        .map(|mask| mask.true_count())
+        .unwrap_or(default)
+}
+
+/// Given a [`BitPackedArray`], unpacks all bitpacked values and creates a new [`PrimitiveVector`].
+///
+/// If a selection mask is passed in, the resultant vector will have the specified values filtered
+/// out.
+///
+/// Internally, this calls [`unpack_into_pvector`].
+pub fn unpack_into_vector(
+    array: &BitPackedArray,
+    selection_mask: Option<&Mask>,
+) -> PrimitiveVector {
+    match_each_integer_ptype!(array.ptype(), |T| {
+        unpack_into_pvector::<T>(array, selection_mask).into()
+    })
+}
+
+/// Given a [`BitPackedArray`] and a template type `T: BitPacked`, unpacks all bitpacked values and
+/// creates a new [`PVector<T>`].
+///
+/// If a selection mask is passed in, the resultant vector will have the specified values filtered
+/// out.
+///
+/// Internally, this calls [`write_unpacked_to_pvector`].
+pub fn unpack_into_pvector<T: BitPacked>(
+    array: &BitPackedArray,
+    selection_mask: Option<&Mask>,
+) -> PVector<T> {
+    let num_new_values = true_count_of_optional_mask(&selection_mask, array.len());
+    let mut pvector = PVectorMut::with_capacity(num_new_values);
+
+    if num_new_values == 0 {
+        return pvector.freeze();
+    }
+
+    write_unpacked_to_pvector(array, selection_mask, &mut pvector);
+
+    pvector.freeze()
+}
+
+/// Given a [`BitPackedArray`] and a template type `T: BitPacked`, unpacks all bitpacked values and
+/// writes them directly into an existing [`PVectorMut<T>`].
+///
+/// If a selection mask is passed in, the specified values will not be written.
+///
+/// Internally, this calls [`write_unpacked_to_buffer`].
+pub fn write_unpacked_to_pvector<T: BitPacked>(
+    array: &BitPackedArray,
+    selection_mask: Option<&Mask>,
+    vector: &mut PVectorMut<T>,
+) {
+    let num_new_values = true_count_of_optional_mask(&selection_mask, array.len());
+    if num_new_values == 0 {
+        return;
+    }
+
+    vector.reserve(num_new_values);
+
+    let validity_mask = match selection_mask {
+        Some(selection_mask) => array.validity_mask().filter(selection_mask),
+        None => array.validity_mask(),
+    };
+    debug_assert_eq!(validity_mask.len(), num_new_values);
+
+    // SAFETY: We add the same amount of elements to both the buffer and the validity mask.
+    let (buffer_mut, vector_validity_mut) = unsafe { vector.mut_parts() };
+
+    // We need to write the unpacked values to the buffer as well as update the validity mask.
+
+    // SAFETY: We reserved enough space in the `reserve` call above.
+    unsafe { write_unpacked_to_buffer(array, selection_mask, buffer_mut) };
+    vector_validity_mut.append_mask(&validity_mask);
+
+    debug_assert_eq!(buffer_mut.len(), vector_validity_mut.len());
+}
+
+/// Given a [`BitPackedArray`] and a template type `T: BitPacked`, unpacks all bitpacked values and
+/// writes them directly into an existing [`BufferMut<T>`].
+///
+/// If a selection mask is passed in, the specified values will not be written.
+///
+/// WARNING: this will completely ignore the validity mask of the [`BitPackedArray`]. This function
+/// should only be called from [`write_unpacked_to_pvector`].
+///
+/// # Safety
+///
+/// The caller must ensurre the buffer has enough capacity to write all of the unpacked values of
+/// the [`BitPackedArray`] into it.
+unsafe fn write_unpacked_to_buffer<T: BitPacked>(
+    array: &BitPackedArray,
+    selection_mask: Option<&Mask>,
+    buffer: &mut BufferMut<T>,
+) {
+    // TODO(connor): Where did these numbers come from? Add a public link after validating them.
+    let full_decompression_threshold = match size_of::<T>() {
+        1 => 0.03,
+        2 => 0.03,
+        4 => 0.075,
+        _ => 0.09,
+        // >8 bytes may have a higher threshold. These numbers are derived from a GCP c2-standard-4
+        // with a "Cascade Lake" CPU.
+    };
+
+    // If the selection mask is sparse, then we want to filter the bitpacked values while we unpack
+    // all of the values.
+    if let Some(selection) = selection_mask
+        && selection.density() < full_decompression_threshold
+    {
+        // SAFETY: The safety contract of `write_unpacked_to_buffer` ensures that we have enough
+        // capacity to write into the buffer.
+        unsafe { filter_while_unpacking_array(array, selection, buffer) };
+    }
+
+    // Otherwise, if the selection mask is dense, then we might as well unpack all of the values and
+    // then perform filtering.
+
+    // SAFETY: The safety contract of `write_unpacked_to_buffer` ensures that we have enough
+    // capacity to write into the buffer.
+    unsafe { unpack_array(array, buffer) };
+
+    if let Some(selection) = selection_mask {
+        todo!("Filter the `BufferMut` with the mask in place")
+    }
+}
+
+/// Unpacks the bitpacked values in the [`BitPackedArray`] directly into a mutable buffer.
+///
+/// WARNING: this will completely ignore the validity mask of the [`BitPackedArray`]. This function
+/// should only be called from [`write_unpacked_to_pvector`].
+///
+/// # Safety
+///
+/// The caller must ensurre the buffer has enough capacity to write all of the unpacked values of
+/// the [`BitPackedArray`] into it.
+unsafe fn unpack_array<T: BitPacked>(array: &BitPackedArray, buffer: &mut BufferMut<T>) {
+    todo!()
+}
+
+/// Unpacks the bitpacked array into the given buffer according to the given selection mask.
+///
+/// WARNING: this will completely ignore the validity mask of the [`BitPackedArray`]. This function
+/// should only be called from [`write_unpacked_to_pvector`].
+///
+/// # Safety
+///
+/// The caller must ensurre the buffer has enough capacity to write all of the unpacked values of
+/// the [`BitPackedArray`] into it.
+unsafe fn filter_while_unpacking_array<T: BitPacked>(
+    array: &BitPackedArray,
+    selection_mask: &Mask,
+    buffer: &mut BufferMut<T>,
+) {
+    todo!()
+}

encodings/fastlanes/src/bitpacking/array/mod.rs

@@ -13,6 +13,7 @@ use vortex_error::{VortexResult, vortex_bail, vortex_ensure};
 
 pub mod bitpack_compress;
 pub mod bitpack_decompress;
+pub mod bitpack_decompress_vector; // Name TBBS.
 pub mod unpack_iter;
 
 use crate::bitpack_compress::bitpack_encode;

encodings/fastlanes/src/bitpacking/mod.rs

@@ -2,7 +2,9 @@
 // SPDX-FileCopyrightText: Copyright the Vortex contributors
 
 mod array;
-pub use array::{BitPackedArray, bitpack_compress, bitpack_decompress, unpack_iter};
+pub use array::{
+    BitPackedArray, bitpack_compress, bitpack_decompress, bitpack_decompress_vector, unpack_iter,
+};
 
 mod vtable;
 pub use vtable::{BitPackedEncoding, BitPackedVTable};

vortex-vector/src/primitive/generic_mut.rs

@@ -78,13 +78,15 @@ impl<T> PVectorMut<T> {
         (self.elements, self.validity)
     }
 
-    /// Append n values to the vector.
-    pub fn append_values(&mut self, value: T, n: usize)
-    where
-        T: Copy,
-    {
-        self.elements.push_n(value, n);
-        self.validity.append_n(true, n);
+    /// Decomposes a mutable reference to the primitive vector into mutable references to it's
+    /// constituent parts (buffer and validity).
+    ///
+    /// # Safety
+    ///
+    /// The caller must ensure the [`BufferMut`] has the same length as the [`MaskMut`] once drop
+    /// the exclusive references.
+    pub unsafe fn mut_parts(&mut self) -> (&mut BufferMut<T>, &mut MaskMut) {
+        (&mut self.elements, &mut self.validity)
     }
 }
 

vortex-vector/src/primitive/generic_mut_impl.rs

@@ -62,6 +62,15 @@ impl<T: NativePType> PVectorMut<T> {
             self.validity.append_n(false, 1);
         }
     }
+
+    /// Append `n` of the same `value` to the vector.
+    pub fn append_values(&mut self, value: T, n: usize)
+    where
+        T: Copy,
+    {
+        self.elements.push_n(value, n);
+        self.validity.append_n(true, n);
+    }
 }
 
 impl<T: NativePType> AsRef<[T]> for PVectorMut<T> {