PatchedArray: basics and wiring by a10y · Pull Request #7002 · vortex-data/vortex

a10y · 2026-03-17T21:48:03Z

Summary

This is the first PR in a series addressing the PatchedArray RFC: vortex-data/rfcs#27

This PR adds a new PatchedArray array variant, which is slated to only be used with FastLanes array types. The design is largely documented in the RFC, but briefly

Wraps another child array
Holds a buffer of offsets and patch values. They are both sorted by chunk/lane so there is efficient random as well as bulk access for CUDA

We are able to pushdown the following at reduce time:

Filters pushes through to the child, and we trim the patch indices to the range of chunks covered by the filter
Slicing that reduces the chunk range
Compare always pushes through to the child
Take pushes through to the inner then constructs a hashmap of the patches

There will be follow ups to add the wiring into CUDA and to update how BitPacked and ALP arrays are written.

Testing

There are unit tests for all of the reducers and kernels

codspeed-hq · 2026-03-17T21:51:42Z

Merging this PR will degrade performance by 15.59%

⚡ 9 improved benchmarks
❌ 1 regressed benchmark
✅ 1096 untouched benchmarks
⏩ 1522 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
❌	Simulation	`chunked_bool_canonical_into[(1000, 10)]`	71.6 µs	84.8 µs	-15.59%
⚡	Simulation	`new_bp_prim_test_between[i16, 32768]`	146 µs	132.2 µs	+10.46%
⚡	Simulation	`new_bp_prim_test_between[i32, 16384]`	117.5 µs	103.2 µs	+13.8%
⚡	Simulation	`take_10k_contiguous`	331.2 µs	272.6 µs	+21.49%
⚡	Simulation	`new_bp_prim_test_between[i32, 32768]`	179.4 µs	150.6 µs	+19.14%
⚡	Simulation	`take_10k_random`	270.4 µs	212.2 µs	+27.44%
⚡	Simulation	`new_bp_prim_test_between[i64, 32768]`	242.3 µs	183.7 µs	+31.91%
⚡	Simulation	`new_bp_prim_test_between[i64, 16384]`	149 µs	119.7 µs	+24.51%
⚡	Simulation	`new_alp_prim_test_between[f64, 16384]`	154.7 µs	133.1 µs	+16.26%
⚡	Simulation	`old_alp_prim_test_between[f32, 32768]`	448.7 µs	402.9 µs	+11.38%

_{Comparing aduffy/patched-array (f3d9155) with develop (5ce08ed)}

1522 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

vortex-array/src/arrays/patched/array.rs

vortex-array/src/arrays/patched/compute/filter.rs

vortex-array/src/arrays/patched/array.rs

a10y · 2026-03-27T13:54:30Z

@claude review please

a10y · 2026-03-27T14:28:51Z

@claude i have addressed your main comments. please give another review.

a10y · 2026-03-27T15:14:10Z

@claude please review

a10y · 2026-03-27T15:50:06Z

@claude PR is ready for another review

vortex-buffer/src/buffer.rs

joseph-isaacs · 2026-03-30T10:41:58Z

vortex-test/compat-gen/src/fixtures/arrays/synthetic/encodings/patched.rs

shall we add a few more of these with different types? e.g. string?

Executing anything other than primitive would fail

vortex-array/src/arrays/patched/array.rs

joseph-isaacs · 2026-03-30T10:47:16Z

vortex-array/src/arrays/patched/array.rs

+    /// Number of lanes the patch indices and values have been split into. Each of the `n_chunks`
+    /// of 1024 values is split into `n_lanes` lanes horizontally, each lane having 1024 / n_lanes
+    /// values that might be patched.
+    pub(super) n_lanes: usize,


This I think we need to store incase we change the default ptype -> lane mapping in the future

I think this kind of change would require new arrays. The data layout would change

As long as you have the n_lanes saved you can always access the data without issue

vortex-array/src/arrays/patched/vtable/mod.rs

joseph-isaacs · 2026-03-30T10:52:00Z

vortex-array/src/arrays/patched/vtable/mod.rs

+}
+
+#[derive(Clone, prost::Message)]
+pub struct PatchedMetadata {


I wonder if you can pack this into a u128?

hm, is there a reason to do that over struct? can protobuf even encode u128?

I mean using less and smaller fields

I removed some fields and packed the rest into a single uint64 packed field

Sorry @a10y this was on me. I was thinking that we can have 4 fields that total u128 in size. I don't think we need to do this manual packing

reverted this the fields are now separated

vortex-array/src/arrays/patched/vtable/mod.rs

a10y · 2026-03-30T18:42:13Z

@claude review please

a10y · 2026-03-30T21:45:02Z

I'm pretty sure the CodSpeed results are just noise since the PatchedArray isn't wired into anything here

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

joseph-isaacs · 2026-03-31T16:26:33Z

Make sure when we wire this in we make it an unstable encoding

joseph-isaacs · 2026-03-31T16:27:04Z

vortex-file/src/strategy.rs

    session.register(Delta);
    session.register(FoR);
    session.register(FSST);
+    session.register(Patched);


Remove this

We should keep the old patches for a while and have this as an unstable_encoding

joseph-isaacs · 2026-03-31T16:30:23Z

You will also need to replace with children (with with_slots)

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

a10y · 2026-03-31T16:56:48Z

done

a10y added the changelog/feature A new feature label Mar 18, 2026

a10y marked this pull request as ready for review March 18, 2026 21:21

a10y requested a review from joseph-isaacs March 18, 2026 21:21

joseph-isaacs reviewed Mar 19, 2026

View reviewed changes

vortex-array/src/arrays/patched/array.rs Outdated Show resolved Hide resolved

joseph-isaacs reviewed Mar 19, 2026

View reviewed changes

vortex-array/src/arrays/patched/compute/filter.rs Show resolved Hide resolved

a10y force-pushed the aduffy/patched-array branch from c3d7434 to 6c1d7aa Compare March 19, 2026 13:24

joseph-isaacs reviewed Mar 19, 2026

View reviewed changes

vortex-array/src/arrays/patched/compute/filter.rs Outdated Show resolved Hide resolved

joseph-isaacs reviewed Mar 19, 2026

View reviewed changes

vortex-array/src/arrays/patched/array.rs Outdated Show resolved Hide resolved

a10y force-pushed the aduffy/patched-array branch from 860cc64 to 042394b Compare March 24, 2026 19:17

a10y mentioned this pull request Mar 24, 2026

[wip] PatchedArray wiring: Part 2 #7145

Draft

a10y force-pushed the aduffy/patched-array branch 4 times, most recently from df8dbf5 to 862d814 Compare March 26, 2026 19:54

This comment was marked as resolved.

Sign in to view

a10y force-pushed the aduffy/patched-array branch from daf258f to 578d83d Compare March 27, 2026 14:35

This comment was marked as resolved.

Sign in to view

a10y force-pushed the aduffy/patched-array branch from 8648244 to a9ce27f Compare March 27, 2026 19:33

joseph-isaacs reviewed Mar 30, 2026

View reviewed changes

This comment was marked as resolved.

Sign in to view

a10y added 20 commits March 31, 2026 12:18

use fixture

31cabc3

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

rename + doc comment

c596cce

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

add Patched as default codec

63ee5aa

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

save

8e82382

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

fixup

778e368

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

address comments about vtable

e5646ed

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

use child

f810e22

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

fix

ba3430a

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

more

de83227

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

fix

24133ba

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

more

220134e

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

fix

2d167c6

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

add with_children and build tests

4114b90

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

doc comment, capacity

ffadc1b

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

packed bitfield for PatchedMetadata

4369138

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

remove

ef628ce

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

clean

35aa0e8

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

validity check

54f9f06

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

clarify comments

2ce0cc5

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

bring back old PatchedMetadata

afe2d51

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

joseph-isaacs approved these changes Mar 31, 2026

View reviewed changes

joseph-isaacs requested changes Mar 31, 2026

View reviewed changes

update to use Slots

f3d9155

Signed-off-by: Andrew Duffy <andrew@a10y.dev>

a10y force-pushed the aduffy/patched-array branch from ad0eaa7 to f3d9155 Compare March 31, 2026 16:56

joseph-isaacs approved these changes Mar 31, 2026

View reviewed changes

a10y merged commit faf98f2 into develop Mar 31, 2026
107 of 110 checks passed

a10y deleted the aduffy/patched-array branch March 31, 2026 18:39

Conversation

a10y commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Testing

Uh oh!

codspeed-hq bot commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will degrade performance by 15.59%

Performance Changes

Footnotes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

a10y commented Mar 27, 2026

Uh oh!

This comment was marked as resolved.

a10y commented Mar 27, 2026

Uh oh!

This comment was marked as resolved.

a10y commented Mar 27, 2026

Uh oh!

This comment was marked as resolved.

a10y commented Mar 27, 2026

Uh oh!

This comment was marked as resolved.

This comment was marked as resolved.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

a10y commented Mar 30, 2026

Uh oh!

This comment was marked as resolved.

a10y commented Mar 30, 2026

Uh oh!

joseph-isaacs commented Mar 31, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joseph-isaacs commented Mar 31, 2026

Uh oh!

a10y commented Mar 31, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

a10y commented Mar 17, 2026 •

edited

Loading

codspeed-hq bot commented Mar 17, 2026 •

edited

Loading