Wma/shard levels #697

willow-ahrens · 2025-02-06T00:35:03Z

Add ShardLevels and rework parallelism in Finch

github-actions

Remaining comments which cannot be posted as a review comment to avoid GitHub Rate Limit

JuliaFormatter

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 142 in 2289266

    
           VirtualShardLevel(device_2, lvl_2, sym, ptr, task, val, typeof(level_fill_value(Lvl)), Device, Lvl, Ptr, Task, Val)

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 147 in 2289266

    
           virtual_level_resize!(ctx, lvl::VirtualShardLevel, dims...) = (lvl.lvl = virtual_level_resize!(ctx, lvl.lvl, dims...); lvl)

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 154 in 2289266

push_preamble!(ctx, quote

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Lines 157 to 158 in 2289266

    
               end) 
        
           push_epilogue!(ctx, quote

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 160 in 2289266

end)

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 165 in 2289266

push_preamble!(ctx,

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 167 in 2289266

    
           lvl_2 = virtualize(ctx_2, :($(lvl.ex).val[$(ctx_2(get_task_num(get_task(ctx_2))))]), lvl.Lvl) #TODO should this virtualize the eltype of Val?

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 169 in 2289266

end

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Lines 191 to 195 in 2289266

    
           push_preamble!(ctx, quote 
        
               Finch.resize_if_smaller!($(lvl.task), $(ctx(pos_stop))) 
        
               Finch.resize_if_smaller!($(lvl.ptr), $(ctx(pos_stop))) 
        
               Finch.fill_range!($(lvl.task), $(ctx(pos_start)), $(ctx(pos_stop)), 0) 
        
           end)

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Lines 220 to 221 in 2289266

    
           return Thunk( 
        
               body = (ctx) -> begin

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Lines 231 to 232 in 2289266

    
           return Thunk( 
        
               body = (ctx) -> begin

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 241 in 2289266

end

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 244 in 2289266

end

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 271 in 2289266

return Thunk(

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 279 in 2289266

    
           $(contain(ctx_2->assemble_level!(ctx_2, lvl.lvl, value(qos_fill, Tp), value(qos_stop, Tp)), ctx))

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 288 in 2289266

body = (ctx) -> VirtualHollowSubFiber(lvl.lvl, value(qos), dirty),

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 295 in 2289266

end

[JuliaFormatter] _{reported by reviewdog 🐶}

Finch.jl/src/tensors/levels/shard_levels.jl

Line 297 in 2289266

end

CONTRIBUTING.md

docs/src/docs/internals/parallel.md

src/architecture.jl

src/tensors/levels/shard_levels.jl

codecov · 2025-03-18T18:29:25Z

Codecov Report

Attention: Patch coverage is 25.18892% with 297 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/tensors/levels/shard_levels.jl	0.00%	259 Missing ⚠️
src/architecture.jl	0.00%	16 Missing ⚠️
src/tensors/levels/sparse_rle_levels.jl	52.00%	12 Missing ⚠️
ext/SparseArraysExt.jl	57.89%	8 Missing ⚠️
src/tensors/levels/dense_levels.jl	50.00%	2 Missing ⚠️

Files with missing lines	Coverage Δ
src/Finch.jl	`94.68% <ø> (ø)`
src/tensors/levels/dense_rle_levels.jl	`79.55% <100.00%> (-4.00%)`	⬇️
src/tensors/levels/separate_levels.jl	`69.93% <ø> (-4.20%)`	⬇️
src/tensors/levels/sparse_band_levels.jl	`75.40% <100.00%> (ø)`
src/tensors/levels/sparse_bytemap_levels.jl	`79.44% <100.00%> (-4.75%)`	⬇️
src/tensors/levels/sparse_coo_levels.jl	`83.40% <100.00%> (-4.81%)`	⬇️
src/tensors/levels/sparse_dict_levels.jl	`79.48% <100.00%> (-3.42%)`	⬇️
src/tensors/levels/sparse_interval_levels.jl	`70.23% <100.00%> (-7.74%)`	⬇️
src/tensors/levels/sparse_list_levels.jl	`89.09% <100.00%> (-3.80%)`	⬇️
src/tensors/levels/sparse_vbl_levels.jl	`77.02% <100.00%> (ø)`
... and 5 more

... and 27 files with indirect coverage changes

🚀 New features to boost your workflow:

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

Paramuths · 2025-05-03T23:24:49Z

src/tensors/levels/shard_levels.jl

+    lvl = transfer(MultiChannelMemory(device, get_num_tasks(device)), lvl)
+    ShardLevel{Device}(
+        device,
+        transfer(MultiChannelMemory(device, get_num_tasks(device)), lvl),


Why do we call transfer on lvl twice?

Paramuths · 2025-05-03T23:31:43Z

src/tensors/levels/shard_levels.jl

+    task_2 = transfer(device, lvl.task)
+    qos_fill_2 = transfer(device, lvl.used)
+    qos_stop_2 = transfer(device, lvl.alloc)
+    return ShardLevel(lvl_2, ptr_2, task_2, qos_fill_2, qos_stop_2)


Do we need device argument?

github-actions bot reviewed Feb 6, 2025

View reviewed changes

willow-ahrens force-pushed the wma/shard_levels branch from e0133b8 to 311c781 Compare February 6, 2025 01:16

willow-ahrens mentioned this pull request Mar 16, 2025

Improve parallel data transfer mechanics #712

Merged

willow-ahrens marked this pull request as draft March 18, 2025 17:52

add Shard Levels

5ce43dd

willow-ahrens force-pushed the wma/shard_levels branch from fc3e6be to 5ce43dd Compare April 9, 2025 18:48

willow-ahrens and others added 4 commits April 9, 2025 20:16

fixing

71efac0

fixes

759f668

consistency

9c964a1

Merge branch 'main' into wma/shard_levels

258d75e

willow-ahrens requested a review from Paramuths April 22, 2025 21:01

willow-ahrens and others added 2 commits April 22, 2025 17:17

reduce noise

7dab8a1

Merge branch 'main' into wma/shard_levels

1f9f7f6

Paramuths reviewed May 3, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Wma/shard levels #697

Wma/shard levels #697

Uh oh!

willow-ahrens commented Feb 6, 2025

Uh oh!

github-actions bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codecov bot commented Mar 18, 2025 •

edited

Loading

Uh oh!

Paramuths May 3, 2025

Uh oh!

Paramuths May 3, 2025

Uh oh!

Uh oh!

	push_preamble!(ctx, quote
	Finch.resize_if_smaller!($(lvl.task), $(ctx(pos_stop)))
	Finch.resize_if_smaller!($(lvl.ptr), $(ctx(pos_stop)))
	Finch.fill_range!($(lvl.task), $(ctx(pos_start)), $(ctx(pos_stop)), 0)
	end)

Wma/shard levels #697

Are you sure you want to change the base?

Wma/shard levels #697

Uh oh!

Conversation

willow-ahrens commented Feb 6, 2025

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

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

Uh oh!

Codecov Report

Uh oh!

Paramuths May 3, 2025

Choose a reason for hiding this comment

Uh oh!

Paramuths May 3, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

codecov bot commented Mar 18, 2025 •

edited

Loading