Fix ShardLevel Structure by AdrianGushin · Pull Request #765 · finch-tensor/Finch.jl

AdrianGushin · 2025-09-24T15:31:02Z

This pull request adds some fixes for the ShardLevel structure and adds associated test cases. It also updates the CPU struct to facilitate multiple distinct cores.

willow-ahrens

Almost there! This is looking great, only small changes requested.

willow-ahrens · 2025-09-24T15:33:29Z

Project.toml

 DataStructures = "0.18"
 Distributions = "0.25"
 HDF5 = "0.17"
+InteractiveUtils = "1.11.0"


InteractiveUtils shouldn't be a Finch dep, we need to remove before merging.

willow-ahrens · 2025-09-24T15:34:05Z

Project.toml

 NPZ = "15e1cf62-19b3-5cfa-8e77-841668bca605"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
-TensorMarket = "8b7d4fe7-0b45-4d0d-9dd8-5cc9b23b4b77"
+TensorMarket = "8b7d4fe7-0b45-4d0d-9dd8-5cc9b23b4b77"


Tensor market should be a test dep, but I don't think it's an extra right? do we need to move it to the test project.toml?

willow-ahrens · 2025-09-24T15:34:18Z

src/architecture.jl

@@ -1,8 +1,10 @@
+using InteractiveUtils


let's remove this, it was just for debugging

willow-ahrens · 2025-09-24T15:34:32Z

src/architecture.jl


 A datatype representing a device on which tasks can be executed.
 """
+


does this like break the documentation for abstract device?

willow-ahrens · 2025-09-24T15:35:16Z

src/architecture.jl

        end,
    )
-    VirtualCPU(value(n, Int))
+    VirtualCPU(value(n, Int), literal(id))


it's also okay to just use id without wrapping as literal, as long as you wrap it when needed. whatever's convenient

willow-ahrens · 2025-09-24T15:36:34Z

src/architecture.jl

 FinchNotation.finch_leaf(mem::VirtualCPULocalMemory) = virtual(mem)
-function virtualize(ctx, ex, ::Type{CPULocalMemory})
-    VirtualCPULocalMemory(virtualize(ctx, :($ex.device), CPU))
+function virtualize(ctx, ex, ::Type{CPULocalMemory{id}}) where {id}


instead of keying CPULocalMemory on id, let's put the whole CPU type in the type parameter so that further changes to CPU parameterization don't need to affect the localmem

then we can recursively virtualize the cpu

willow-ahrens · 2025-09-24T15:37:28Z

src/architecture.jl

-global_memory(device::CPU) = CPUSharedMemory(device)
+local_memory(device::CPU{id}) where {id} = CPULocalMemory{id}(device)
+shared_memory(device::CPU{id}) where {id} = CPUSharedMemory{id}(device)
+global_memory(device::CPU{id}) where {id} = CPUSharedMemory{id}(device)


same thing here, I think it makes more sense to key on the CPU than the ID. feel free to disagree here.

willow-ahrens · 2025-09-24T15:38:08Z

src/tensors/levels/shard_levels.jl

 function transfer(task::MemoryChannel, arr::MultiChannelBuffer)
    if task.device == arr.device
        temp = arr.data[task.t]
+        @assert isa(temp, Vector)


This is good for debugging, but I don't think this will always be the case, we might have different buffer types than vector

willow-ahrens · 2025-09-24T15:39:33Z

test/runtests.jl


    @testset "Finch" begin
        include("modules/checkoutput_testsetup.jl")
+        include("suites/constructors_tests.jl")


wow. It's crazy that this wasn't already included

willow-ahrens · 2025-09-24T15:40:15Z

test/suites/constructors_tests.jl

+        end
+
+        @test C[4,4] == 12
+    end


Good test! Let's also add a test in perhaps representation.jl which generates some reference output for a shard level kernel

willow-ahrens · 2025-09-24T15:41:06Z

We can merge this to main once we have some more tests.

willow-ahrens

Hi Adrian! This all looks good, the only requirement we need now is a test which calls check_output to compare the generated shardlevel code against a reference output. See elsewhere in the code where this function is used for examples. You'll need to follow instructions in the contributing guide to generate new 64-bit reference, and run the "fixbot" action to generate new 32-bit reference

willow-ahrens · 2025-10-18T00:00:25Z

I've added you to the repo as a collaborator, you can re-open the PR using finch-tensor as the remote and it will automatically run tests.

codecov · 2025-11-18T04:28:14Z

Codecov Report

❌ Patch coverage is 77.46479% with 16 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/architecture.jl	63.33%	11 Missing ⚠️
src/tensors/levels/shard_levels.jl	89.74%	4 Missing ⚠️
src/tensors/levels/sparse_list_levels.jl	0.00%	1 Missing ⚠️

Files with missing lines	Coverage Δ
src/lower.jl	`87.86% <ø> (ø)`
src/tensors/levels/sparse_dict_levels.jl	`88.41% <100.00%> (ø)`
src/tensors/levels/sparse_list_levels.jl	`92.41% <0.00%> (ø)`
src/tensors/levels/shard_levels.jl	`64.59% <89.74%> (ø)`
src/architecture.jl	`64.08% <63.33%> (ø)`

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

willow-ahrens

A summary of failing tests and next steps to merge this:

Windows tests are failing because we need to run fixbot (Contributing.md)
Ubuntu latest is failing for unrelated reasons (my bad, julia 1.12.3 is much stricter)
I'd like to see the CPU tagging and SparseDict, etc. fixes split out and merged to main first.

I'll leave 1 to you. I'll work on solving 2 asap. I won't be able to get to 3 until sometime tomorrow evening, feel free to do that earlier! I use "gitlens search & compare" vscode extension for stuff like that.

…o shard_dev

willow-ahrens · 2025-11-19T00:37:24Z

.github/workflows/FixBot.yml

      - uses: actions/checkout@v3
        with:
-          token: ${{ secrets.WILLOW_BOT_TOKEN }}
+          token: ${{ secrets.ADRIAN_BOT_TOKEN }}


You must have run this from your fork. If the willow_bot_token is not working, I should fix that rather than you using your own token on your fork, so that others can also trigger the action from the actions pane without generating their own tokens. I have verified you have write permissions, so in the future, if you make this a branch on the finch-tensor/Finch.jl repo, you should be able to manually run FixBot on this branch from the actions tab without changing this secret.

agushin101 added 30 commits August 21, 2025 17:46

initial attempt

358f453

remove extent from constructor

57f17c9

update creation of ext

670c8c1

fix typo

eeaa67c

modify eltype signature to account for scheduler

6833205

remove erroneous call to subcontext

a92044c

add other params to lambda

0faeb73

debugging

7b6da14

add 4 arguments to vpr lambda

5d706c5

fix error in getnumtasks method call

15f148a

unsafe removal of assert

44e9f57

more unsafe removals

26ca5df

add safety back

71cf3d8

debugging

fb1f84d

debugging

41c3edd

debug stmts

5b134cb

check in other dist

7c85760

check in other dist

222357c

try containing

ddeb820

debug stmts

92679f8

finalize results for thurs

cdd1adc

add debugging to shims

1df5b5c

relocate print

99d2cb8

bugfix

197bc74

printing arguments ONLY

7bab33e

debugging

dc0f56e

test if smaller or if resize

c3ed869

testing API

79346f7

delete error

5509898

checking inner

915bbca

willow-ahrens requested changes Sep 24, 2025

View reviewed changes

agushin101 added 3 commits September 24, 2025 12:01

update project.toml

52ffbd9

pr updates

f0afbbc

add representation test cases

92d4419

willow-ahrens requested changes Oct 17, 2025

View reviewed changes

agushin101 added 11 commits November 8, 2025 17:33

shard mux

ed06695

final commit before SL changes

d648ea2

parallel sparselist, update sparsedict with multiplexer

d7ac57c

fix shard doctests

1cf1880

repair test cases

a1fe56e

final fixes to tests

da2622d

fix project

4d6b1a1

repair projects for server compilation

007388f

generalize docs cases, repair parallel cases, delete concurrency checks

6f5472a

stylize passing tests

7752cdb

repair json version

3b38f28

willow-ahrens requested changes Nov 18, 2025

View reviewed changes

AdrianGushin and others added 8 commits November 18, 2025 17:41

Replace WILLOW_BOT_TOKEN with ADRIAN_BOT_TOKEN

d6fff38

Regenerate test output

9d78157

Regenerate test output

5c698cd

Merge branch 'shard_dev' of https://github.com/AdrianGushin/Finch int…

9a7130d

…o shard_dev

Regenerate test output

71611ea

Merge branch 'shard_dev' of https://github.com/AdrianGushin/Finch int…

effe159

…o shard_dev

Regenerate test output

47776f6

Merge branch 'shard_dev' of https://github.com/AdrianGushin/Finch int…

2dc9cc5

…o shard_dev

willow-ahrens reviewed Nov 19, 2025

View reviewed changes

AdrianGushin closed this Nov 19, 2025


		A datatype representing a device on which tasks can be executed.
		"""

Conversation

AdrianGushin commented Sep 24, 2025

Uh oh!

willow-ahrens left a comment

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!

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!

willow-ahrens commented Sep 24, 2025

Uh oh!

willow-ahrens left a comment

Choose a reason for hiding this comment

Uh oh!

willow-ahrens commented Oct 18, 2025

Uh oh!

codecov bot commented Nov 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

willow-ahrens left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov bot commented Nov 18, 2025 •

edited

Loading