mirror of
https://github.com/facebookresearch/faiss.git
synced 2025-06-03 08:19:58 +08:00
6d0bc58db6
Summary: IndexPQ and IndexIVFPQ implementations with AVX shuffle instructions. The training and computing of the codes does not change wrt. the original PQ versions but the code layout is "packed" so that it can be used efficiently by the SIMD computation kernels. The main changes are: - new IndexPQFastScan and IndexIVFPQFastScan objects - simdib.h for an abstraction above the AVX2 intrinsics - BlockInvertedLists for invlists that are 32-byte aligned and where codes are not sequential - pq4_fast_scan.h/.cpp: for packing codes and look-up tables + optmized distance comptuation kernels - simd_result_hander.h: SIMD version of result collection in heaps / reservoirs Misc changes: - added contrib.inspect_tools to access fields in C++ objects - moved .h and .cpp code for inverted lists to an invlists/ subdirectory, and made a .h/.cpp for InvertedListsIOHook - added a new inverted lists type with 32-byte aligned codes (for consumption by SIMD) - moved Windows-specific intrinsics to platfrom_macros.h Pull Request resolved: https://github.com/facebookresearch/faiss/pull/1542 Test Plan: ``` buck test mode/opt -j 4 //faiss/tests/:test_fast_scan_ivf //faiss/tests/:test_fast_scan buck test mode/opt //faiss/manifold/... ``` Reviewed By: wickedfoo Differential Revision: D25175439 Pulled By: mdouze fbshipit-source-id: ad1a40c0df8c10f4b364bdec7172e43d71b56c34
75 lines
2.0 KiB
Python
75 lines
2.0 KiB
Python
|
|
import time
|
|
import faiss
|
|
import numpy as np
|
|
|
|
def do_partition(n, qin, maxval=65536, seed=123, id_type='int64'):
|
|
print(
|
|
f"n={n} qin={qin} maxval={maxval} id_type={id_type} ",
|
|
end="\t", flush=True
|
|
)
|
|
|
|
# print("seed=", seed)
|
|
rs = np.random.RandomState(seed)
|
|
vals = rs.randint(maxval, size=n).astype('uint16')
|
|
ids = (rs.permutation(n) + 12345).astype(id_type)
|
|
|
|
sp = faiss.swig_ptr
|
|
|
|
tab_a = faiss.AlignedTableUint16()
|
|
faiss.copy_array_to_AlignedTable(vals, tab_a)
|
|
|
|
nrun = 2000
|
|
|
|
times = []
|
|
nerr = 0
|
|
stats = faiss.cvar.partition_stats
|
|
stats.reset()
|
|
for _run in range(nrun):
|
|
faiss.copy_array_to_AlignedTable(vals, tab_a)
|
|
t0 = time.time()
|
|
# print("tab a type", tab_a.get())
|
|
if type(qin) == int:
|
|
q = qin
|
|
faiss.CMax_uint16_partition_fuzzy(
|
|
tab_a.get(), sp(ids), n, q, q, None)
|
|
else:
|
|
q_min, q_max = qin
|
|
q = np.array([-1], dtype='uint64')
|
|
thresh2 = faiss.CMax_uint16_partition_fuzzy(
|
|
tab_a.get(), sp(ids), n,
|
|
q_min, q_max, sp(q)
|
|
)
|
|
q = q[0]
|
|
|
|
if not (q_min <= q <= q_max):
|
|
nerr += 1
|
|
|
|
t1 = time.time()
|
|
|
|
times.append(t1 - t0)
|
|
|
|
times = np.array(times[100:]) * 1000000
|
|
|
|
|
|
print(
|
|
f"times {times.mean():.3f} µs (± {times.std():.4f} µs) nerr={nerr} "
|
|
f"bissect {stats.bissect_cycles / 1e6:.3f} Mcy "
|
|
f"compress {stats.compress_cycles / 1e6:.3f} Mcy"
|
|
)
|
|
|
|
do_partition(200, (100, 100))
|
|
do_partition(200, (100, 150))
|
|
do_partition(2000, (1000, 1000))
|
|
do_partition(2000, (1000, 1500))
|
|
do_partition(20000, (10000, 10000))
|
|
do_partition(20000, (10000, 15000))
|
|
|
|
|
|
do_partition(200, (100, 100), id_type='int32')
|
|
do_partition(200, (100, 150), id_type='int32')
|
|
do_partition(2000, (1000, 1000), id_type='int32')
|
|
do_partition(2000, (1000, 1500), id_type='int32')
|
|
do_partition(20000, (10000, 10000), id_type='int32')
|
|
do_partition(20000, (10000, 15000), id_type='int32')
|