Skip to main content

vortex_array/arrays/primitive/vtable/
mod.rs

1// SPDX-License-Identifier: Apache-2.0
2// SPDX-FileCopyrightText: Copyright the Vortex contributors
3
4use kernel::PARENT_KERNELS;
5use vortex_error::VortexResult;
6use vortex_error::vortex_bail;
7use vortex_error::vortex_ensure;
8use vortex_error::vortex_panic;
9
10use crate::ArrayRef;
11use crate::ExecutionCtx;
12use crate::ExecutionResult;
13use crate::array::Array;
14use crate::array::ArrayView;
15use crate::array::VTable;
16use crate::arrays::primitive::PrimitiveData;
17use crate::buffer::BufferHandle;
18use crate::dtype::DType;
19use crate::dtype::PType;
20use crate::serde::ArrayChildren;
21use crate::validity::Validity;
22mod kernel;
23mod operations;
24mod validity;
25
26use std::hash::Hasher;
27
28use vortex_buffer::Alignment;
29use vortex_session::VortexSession;
30
31use crate::Precision;
32use crate::array::ArrayId;
33use crate::arrays::primitive::array::SLOT_NAMES;
34use crate::arrays::primitive::compute::rules::RULES;
35use crate::hash::ArrayEq;
36use crate::hash::ArrayHash;
37
38/// A [`Primitive`]-encoded Vortex array.
39pub type PrimitiveArray = Array<Primitive>;
40
41impl ArrayHash for PrimitiveData {
42    fn array_hash<H: Hasher>(&self, state: &mut H, precision: Precision) {
43        self.buffer.array_hash(state, precision);
44    }
45}
46
47impl ArrayEq for PrimitiveData {
48    fn array_eq(&self, other: &Self, precision: Precision) -> bool {
49        self.buffer.array_eq(&other.buffer, precision)
50    }
51}
52
53impl VTable for Primitive {
54    type ArrayData = PrimitiveData;
55
56    type OperationsVTable = Self;
57    type ValidityVTable = Self;
58
59    fn id(&self) -> ArrayId {
60        Self::ID
61    }
62
63    fn nbuffers(_array: ArrayView<'_, Self>) -> usize {
64        1
65    }
66
67    fn buffer(array: ArrayView<'_, Self>, idx: usize) -> BufferHandle {
68        match idx {
69            0 => array.buffer_handle().clone(),
70            _ => vortex_panic!("PrimitiveArray buffer index {idx} out of bounds"),
71        }
72    }
73
74    fn buffer_name(_array: ArrayView<'_, Self>, idx: usize) -> Option<String> {
75        match idx {
76            0 => Some("values".to_string()),
77            _ => None,
78        }
79    }
80
81    fn serialize(
82        _array: ArrayView<'_, Self>,
83        _session: &VortexSession,
84    ) -> VortexResult<Option<Vec<u8>>> {
85        Ok(Some(vec![]))
86    }
87
88    fn validate(
89        &self,
90        data: &PrimitiveData,
91        dtype: &DType,
92        len: usize,
93        slots: &[Option<ArrayRef>],
94    ) -> VortexResult<()> {
95        let DType::Primitive(_, nullability) = dtype else {
96            vortex_bail!("Expected primitive dtype, got {dtype:?}");
97        };
98        vortex_ensure!(
99            data.len() == len,
100            "PrimitiveArray length {} does not match outer length {}",
101            data.len(),
102            len
103        );
104        let validity = crate::array::child_to_validity(&slots[0], *nullability);
105        if let Some(validity_len) = validity.maybe_len() {
106            vortex_ensure!(
107                validity_len == len,
108                "PrimitiveArray validity len {} does not match outer length {}",
109                validity_len,
110                len
111            );
112        }
113
114        Ok(())
115    }
116
117    fn deserialize(
118        &self,
119        dtype: &DType,
120        len: usize,
121        metadata: &[u8],
122
123        buffers: &[BufferHandle],
124        children: &dyn ArrayChildren,
125        _session: &VortexSession,
126    ) -> VortexResult<crate::array::ArrayParts<Self>> {
127        if !metadata.is_empty() {
128            vortex_bail!(
129                "PrimitiveArray expects empty metadata, got {} bytes",
130                metadata.len()
131            );
132        }
133        if buffers.len() != 1 {
134            vortex_bail!("Expected 1 buffer, got {}", buffers.len());
135        }
136        let buffer = buffers[0].clone();
137
138        let validity = if children.is_empty() {
139            Validity::from(dtype.nullability())
140        } else if children.len() == 1 {
141            let validity = children.get(0, &Validity::DTYPE, len)?;
142            Validity::Array(validity)
143        } else {
144            vortex_bail!("Expected 0 or 1 child, got {}", children.len());
145        };
146
147        let ptype = PType::try_from(dtype)?;
148
149        vortex_ensure!(
150            buffer.is_aligned_to(Alignment::new(ptype.byte_width())),
151            "Misaligned buffer cannot be used to build PrimitiveArray of {ptype}"
152        );
153
154        if buffer.len() != ptype.byte_width() * len {
155            vortex_bail!(
156                "Buffer length {} does not match expected length {} for {}, {}",
157                buffer.len(),
158                ptype.byte_width() * len,
159                ptype.byte_width(),
160                len,
161            );
162        }
163
164        vortex_ensure!(
165            buffer.is_aligned_to(Alignment::new(ptype.byte_width())),
166            "PrimitiveArray::build: Buffer (align={}) must be aligned to {}",
167            buffer.alignment(),
168            ptype.byte_width()
169        );
170
171        // SAFETY: checked ahead of time
172        let slots = PrimitiveData::make_slots(&validity, len);
173        let data = unsafe { PrimitiveData::new_unchecked_from_handle(buffer, ptype, validity) };
174        Ok(crate::array::ArrayParts::new(self.clone(), dtype.clone(), len, data).with_slots(slots))
175    }
176
177    fn slot_name(_array: ArrayView<'_, Self>, idx: usize) -> String {
178        SLOT_NAMES[idx].to_string()
179    }
180
181    fn execute(array: Array<Self>, _ctx: &mut ExecutionCtx) -> VortexResult<ExecutionResult> {
182        Ok(ExecutionResult::done(array))
183    }
184
185    fn reduce_parent(
186        array: ArrayView<'_, Self>,
187        parent: &ArrayRef,
188        child_idx: usize,
189    ) -> VortexResult<Option<ArrayRef>> {
190        RULES.evaluate(array, parent, child_idx)
191    }
192
193    fn execute_parent(
194        array: ArrayView<'_, Self>,
195        parent: &ArrayRef,
196        child_idx: usize,
197        ctx: &mut ExecutionCtx,
198    ) -> VortexResult<Option<ArrayRef>> {
199        PARENT_KERNELS.execute(array, parent, child_idx, ctx)
200    }
201}
202
203#[derive(Clone, Debug)]
204pub struct Primitive;
205
206impl Primitive {
207    pub const ID: ArrayId = ArrayId::new_ref("vortex.primitive");
208}
209
210#[cfg(test)]
211mod tests {
212    use vortex_buffer::ByteBufferMut;
213    use vortex_buffer::buffer;
214    use vortex_session::registry::ReadContext;
215
216    use crate::ArrayContext;
217    use crate::IntoArray;
218    use crate::LEGACY_SESSION;
219    use crate::arrays::PrimitiveArray;
220    use crate::assert_arrays_eq;
221    use crate::serde::SerializeOptions;
222    use crate::serde::SerializedArray;
223    use crate::validity::Validity;
224
225    #[test]
226    fn test_nullable_primitive_serde_roundtrip() {
227        let array = PrimitiveArray::new(
228            buffer![1i32, 2, 3, 4],
229            Validity::from_iter([true, false, true, false]),
230        );
231        let dtype = array.dtype().clone();
232        let len = array.len();
233
234        let ctx = ArrayContext::empty();
235        let serialized = array
236            .clone()
237            .into_array()
238            .serialize(&ctx, &LEGACY_SESSION, &SerializeOptions::default())
239            .unwrap();
240
241        let mut concat = ByteBufferMut::empty();
242        for buf in serialized {
243            concat.extend_from_slice(buf.as_ref());
244        }
245        let parts = SerializedArray::try_from(concat.freeze()).unwrap();
246        let decoded = parts
247            .decode(
248                &dtype,
249                len,
250                &ReadContext::new(ctx.to_ids()),
251                &LEGACY_SESSION,
252            )
253            .unwrap();
254
255        assert_arrays_eq!(decoded, array);
256    }
257}