New version: Blobs v1.1.0 #123214
New version: Blobs v1.1.0 #123214
Conversation
JuliaRegistrator
commented
Jan 17, 2025
•
JuliaRegistrator
commented
- Registering package: Blobs
- Repository: https://github.com/RelationalAI-oss/Blobs.jl
- Created by: @NHDaly
- Version: v1.1.0
- Commit: 8026b8d4e6ea63a92c7fa7554da35f880236b888
- Reviewed by: @NHDaly
- Reference: RelationalAI-oss/Blobs.jl@8026b8d#commitcomment-151473602
- Description: Binary blobs with on-the-fly pointer patching
- Release notes:
* Convert getindex(::Blob, ::Val{field}) from generated func.
- The produced code is unchanged, and the perf remains the same.
- This is tested by a new testitem
* Eliminate another generated function: self_size!
Managed via compiler annotations
This new function is ~10x faster than the older `@generated` function:
- ~10ms down to ~1ms
* Reorganize to hopefully minimize the generated part of unsafe_store!
* Recursion for computing blob_offset
* Get unsafe_load() un-generated as well! :)
* Bump to v1.1
* Switch unsafe_store! to non-generated as well.
2x slower, but less compile time so worth it.
* Improve compile-time perf on self_size computation
* De-val getindex, which fixes getproperty codegen. (Unclear why)
```
julia> @code_llvm debuginfo=:none (b->b.x)(b)
define void @"julia_#35_644"([3 x i64]* noalias nocapture noundef nonnull sret([3 x i64]) align 8 dereferenceable(24) %0, [3 x i64]* nocapture noundef nonnull readonly align 8 dereferenceable(24) %1) #0 {
top:
%memcpy_refined_src1 = getelementptr inbounds [3 x i64], [3 x i64]* %1, i64 0, i64 2
%2 = load i64, i64* %memcpy_refined_src1, align 8
%3 = bitcast [3 x i64]* %1 to <2 x i64>*
%4 = load <2 x i64>, <2 x i64>* %3, align 8
%5 = bitcast [3 x i64]* %0 to <2 x i64>*
store <2 x i64> %4, <2 x i64>* %5, align 8
%newstruct.sroa.3.0..sroa_idx4 = getelementptr inbounds [3 x i64], [3 x i64]* %0, i64 0, i64 2
store i64 %2, i64* %newstruct.sroa.3.0..sroa_idx4, align 8
ret void
}
```
* Force fieldidx to const-fold, even for large (100 field) structs
According to @benchmark, this made compiling getproperty and setproperty
for 100-field structs take 2ms longer (7ms -> 9ms)
* Add tests for allocations
* Noinline the function to actually throw the assertion
---------
Co-authored-by: Sacha Verweij <[email protected]>
JuliaRegistrator
temporarily deployed
to
stopwatch
GitHub Actions
Inactive
JuliaRegistrator
temporarily deployed
to
stopwatch
GitHub Actions
Inactive
JuliaRegistrator
temporarily deployed
to
stopwatch
GitHub Actions
Inactive
UUID: 163b9779-6631-5f90-a265-3de947924de8 Repo: https://github.com/RelationalAI-oss/Blobs.jl.git Tree: e57267f4b0f92329f879add30e1badd4073b21d9 Registrator tree SHA: 17aec322677d9b81cdd6b9b9236b09a3f1374c6a
JuliaRegistrator
force-pushed
the
registrator-blobs-163b9779-v1.1.0-3acc3b7384
branch
from
6c8c2cb to
eeda611
Compare
latest release only supports 1.10, because of new use of compiler assume_effects
JuliaRegistrator
temporarily deployed
to
stopwatch
GitHub Actions
Inactive
|
Hello, I am an automated registration bot. I help manage the registration process by checking your registration against a set of AutoMerge guidelines. If all these guidelines are met, this pull request will be merged automatically, completing your registration. It is strongly recommended to follow the guidelines, since otherwise the pull request needs to be manually reviewed and merged by a human. 1. AutoMerge Guidelines which are not met ❌
2. Needs action: here's what to do next
If you need help fixing the AutoMerge issues, or want your pull request to be manually merged instead, please post a comment explaining what you need help with or why you would like this pull request to be manually merged. Then, send a message to the 3. To pause or stop registrationIf you want to prevent this pull request from being auto-merged, simply leave a comment. If you want to post a comment without blocking auto-merging, you must include the text Tip: You can edit blocking comments to add |
