#![allow(clippy::expect_used)]
use std::alloc::{GlobalAlloc, Layout, System};
use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
use zeppelin_embed::ingest::{DocId, DocumentVersion, IngestDocument, Revision};
use zeppelin_embed_text::runtime::EmbeddingBatch;
struct CountingAllocator;
static COUNTING: AtomicBool = AtomicBool::new(false);
static ALLOCATIONS: AtomicUsize = AtomicUsize::new(0);
unsafe impl GlobalAlloc for CountingAllocator {
unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
if COUNTING.load(Ordering::Relaxed) {
ALLOCATIONS.fetch_add(1, Ordering::Relaxed);
}
unsafe { System.alloc(layout) }
}
unsafe fn dealloc(&self, pointer: *mut u8, layout: Layout) {
unsafe { System.dealloc(pointer, layout) }
}
}
#[global_allocator]
static ALLOCATOR: CountingAllocator = CountingAllocator;
#[test]
fn the_embed_to_writer_hand_off_allocates_zero_bytes_per_document() {
let batch = EmbeddingBatch::new(vec![1.0; 128 * 768], 128, 768).expect("batch");
ALLOCATIONS.store(0, Ordering::Relaxed);
COUNTING.store(true, Ordering::Relaxed);
let values = batch.into_values();
COUNTING.store(false, Ordering::Relaxed);
assert_eq!(values.len(), 128 * 768);
let core_vectors = values
.chunks_exact(768)
.map(<[f32]>::to_vec)
.collect::<Vec<_>>();
let mut documents = Vec::with_capacity(core_vectors.len());
COUNTING.store(true, Ordering::Relaxed);
for (row, vector) in core_vectors.into_iter().enumerate() {
documents.push(IngestDocument::new(
DocumentVersion::new(DocId::new(row as u128 + 1), Revision::new(1)),
vector,
));
}
COUNTING.store(false, Ordering::Relaxed);
assert_eq!(documents.len(), 128);
assert_eq!(ALLOCATIONS.load(Ordering::Relaxed), 0);
}