remove Hash Table from FinchAssembly#541
Conversation
Merging this PR will degrade performance by 70.6%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | WallTime | test_ops_reduction[interpret_galley-sum] |
419.5 ms | 1,835.2 ms | -77.14% |
| ❌ | WallTime | test_jit_creation[compile_numba] |
4.7 ms | 20.1 ms | -76.6% |
| ❌ | WallTime | test_jit_creation[interpret_galley] |
4.7 ms | 20.1 ms | -76.44% |
| ❌ | WallTime | test_galley_matmul_chain[dense_last-downstream] |
60.2 ms | 247.1 ms | -75.66% |
| ❌ | WallTime | test_ops_binary[interpret_galley-multiply] |
786.2 ms | 3,213.1 ms | -75.53% |
| ❌ | WallTime | test_ops_binary[interpret_galley-add] |
787 ms | 3,207.1 ms | -75.46% |
| ❌ | WallTime | test_galley_matmul_chain[empty_last-downstream] |
60.8 ms | 247.2 ms | -75.4% |
| ❌ | WallTime | test_galley_matmul_chain[dense_last-optimize] |
4.7 s | 18.5 s | -74.54% |
| ❌ | WallTime | test_galley_matmul_chain[empty_last-optimize] |
249.5 ms | 976.7 ms | -74.46% |
| ❌ | WallTime | test_ops_binary[interpret_galley-matmul] |
58.9 s | 220.5 s | -73.31% |
| ❌ | WallTime | test_f1_jit[interpret_galley] |
11.2 s | 41.5 s | -72.95% |
| ❌ | WallTime | test_f1_jit[compile_numba] |
1.4 s | 4.9 s | -72.26% |
| ❌ | WallTime | test_ops_binary[compile_numba-matmul] |
431.1 ms | 1,528.8 ms | -71.8% |
| ❌ | WallTime | test_ops_reduction[compile_numba-sum] |
247.4 ms | 870.6 ms | -71.58% |
| ❌ | WallTime | test_ops_binary[compile_numba-add] |
394.9 ms | 1,378.4 ms | -71.35% |
| ❌ | WallTime | test_ops_binary[compile_numba-multiply] |
397.9 ms | 1,377.1 ms | -71.1% |
| ❌ | Memory | test_ops_reduction[compile_numba-sum] |
10.8 MB | 26.7 MB | -59.46% |
| ❌ | Memory | test_ops_binary[compile_numba-matmul] |
18 MB | 41.4 MB | -56.47% |
| ❌ | Memory | test_ops_binary[compile_numba-add] |
18.1 MB | 40 MB | -54.83% |
| ❌ | Memory | test_ops_binary[compile_numba-multiply] |
18.2 MB | 39.5 MB | -53.86% |
| ... | ... | ... | ... | ... | ... |
ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing wma/rm_hash (f0ffcff) with main (64775fa)
kylebd99
left a comment
kylebd99
left a comment
There was a problem hiding this comment.
This looks good to me! I love a PR that removes nodes from a language.
|
@willow-ahrens @hameerabbasi codspeed seems a bit sensitive/jittery on wallclock mode, would it make sense to shift the tolerance to like 25% for the moment? Edit: Ahh, nevermind, I thought it was blocking PRs. This is fine. |
|
Yes, lets.
…________________________________
From: Kyle Deeds ***@***.***>
Sent: Tuesday, 30 June 2026 18:26:13
To: finch-tensor/finch-tensor-lite ***@***.***>
Cc: Willow Ahrens ***@***.***>; Mention ***@***.***>
Subject: Re: [finch-tensor/finch-tensor-lite] remove Hash Table from FinchAssembly (PR #541)
[https://avatars.githubusercontent.com/u/27299171?s=20&v=4]kylebd99 left a comment (finch-tensor/finch-tensor-lite#541)<#541 (comment)>
@willow-ahrens<https://github.com/willow-ahrens> @hameerabbasi<https://github.com/hameerabbasi> codspeed seems a bit sensitive/jittery on wallclock mode, would it make sense to shift the tolerance to like 25% for the moment?
—
Reply to this email directly, view it on GitHub<#541?email_source=notifications&email_token=ABMAYDZSPGIBNQG7TAMVHBT5CQ5ALA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOBUHA2DCMJRGA22M4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4848411105>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABMAYD2P4CUJOFC2VGJISJD5CQ5ALAVCNFSNUABFKJSXA33TNF2G64TZHM4TOMRXGQ3DANRXHNEXG43VMU5TINZXHE4DCNZXGA22C5QC>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
| jobs: | ||
| benchmark: | ||
| runs-on: ubuntu-latest | ||
| runs-on: codspeed-macro |
There was a problem hiding this comment.
Sadly, this feature is actually paid.
I'll look into switching to simulation mode; and bringing down the overhead.
There was a problem hiding this comment.
https://codspeed.io/docs/features/macro-runners#pricing-and-usage
@willow-ahrens They might be willing to support us if we contact them.
3 participants
In the future, we will build hash tables out of FinchAssembly. fixes #406