bpart: Track whether any binding replacement has happened in image modules #57433
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…dules This implements the optimization proposed in #57426 by keeping track of whether any bindings were replaced in image modules (excluding `Main` as facilitated by #57426). In addition, we augment serialization to keep track of whether a method body contains any GlobalRefs that point to a loaded (system or package) image. If both of these flags are true, we can skip scanning the body of the method, since we know that we neither need to add any additional backedges nor were any of the referenced bindings invalidated. The performance impact on end-to-end load time is small, but measurable. Overall `@time using ModelingToolkit` consistently improves about 5% using this PR. However, I should note that using time is still about 40% slower than 1.11. This is not necessarily an Apples-to-Apples comparison as there were substantial other changes on 1.12 (as well as current load-time-tunings targeting older versions), but I wanted to put the number context.
KristofferC
pushed a commit
that referenced
this pull request
Feb 17, 2025
…dules (#57433) This implements the optimization proposed in #57426 by keeping track of whether any bindings were replaced in image modules (excluding `Main` as facilitated by #57426). In addition, we augment serialization to keep track of whether a method body contains any GlobalRefs that point to a loaded (system or package) image. If both of these flags are true, we can skip scanning the body of the method, since we know that we neither need to add any additional backedges nor were any of the referenced bindings invalidated. The performance impact on end-to-end load time is small, but measurable. Overall `@time using ModelingToolkit` consistently improves about 5% using this PR. However, I should note that using time is still about 40% slower than 1.11. This is not necessarily an Apples-to-Apples comparison as there were substantial other changes on 1.12 (as well as current load-time-tunings targeting older versions), but I wanted to put the number context. (cherry picked from commit f6e2b98)
vtjnash
reviewed
Feb 18, 2025
| @@ -25,7 +25,8 @@ end | |||
| function insert_backedges(edges::Vector{Any}, ext_ci_list::Union{Nothing,Vector{Any}}, extext_methods::Vector{Any}, internal_methods::Vector{Any}) | |||
| # determine which CodeInstance objects are still valid in our image | |||
| # to enable any applicable new codes | |||
| methods_with_invalidated_source = Base.scan_new_methods(extext_methods, internal_methods) | |||
| backedges_only = unsafe_load(cglobal(:jl_first_image_replacement_world, UInt)) == typemax(UInt) | |||
There was a problem hiding this comment.
unsafe_load requires an atomic ordering specifier to access mutable data (this can be strong UB here without it, as we define unsafe_load to be a memcpy and not a unordered load)
KristofferC
added a commit
that referenced
this pull request
Feb 26, 2025
Backported PRs: - [x] #57302 <!-- Add explicit imports for types and fix bugs --> - [x] #57420 <!-- Compiler: Fix check for IRShow definedness --> - [x] #57419 <!-- generated: Switch resolution module back to what it was before --> - [x] #57421 <!-- bpart: Skip implicit import reval if using'd export set is unchanged --> - [x] #57425 <!-- Prohibit binding replacement in closed modules during precompile --> - [x] #57426 <!-- Prohibit `import`ing or `using` Main during incremental compilation --> - [x] #57433 <!-- bpart: Track whether any binding replacement has happened in image modules --> - [x] #57445 <!-- Run all `--sysimage-native-code=no` cmdlineargs tests single-threaded --> - [x] #57386 <!-- Only strip invariant.load from special pointers --> - [x] #57453 <!-- Revert "Make emitted egal code more loopy (#54121)" --> - [x] #57389 <!-- Change memory indexing to use the type as index instead of i8 --> - [x] #57447 <!-- Don't return null pointer when asking for the type of a declared global --> - [x] #57467 <!-- using/import: ensure world update after each observable operation --> - [x] #57471 <!-- staticdata: Don't use `newm` pointer after it has been invalidated --> - [x] #57416 <!-- lowering: Don't mutate lambda in `linearize` -->
KristofferC
removed
the
backport 1.12
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This implements the optimization proposed in #57426 by keeping track of whether any bindings were replaced in image modules (excluding
Mainas facilitated by #57426). In addition, we augment serialization to keep track of whether a method body contains any GlobalRefs that point to a loaded (system or package) image. If both of these flags are true, we can skip scanning the body of the method, since we know that we neither need to add any additional backedges nor were any of the referenced bindings invalidated. The performance impact on end-to-end load time is small, but measurable. Overall@time using ModelingToolkitconsistently improves about 5% using this PR. However, I should note that using time is still about 40% slower than 1.11. This is not necessarily an Apples-to-Apples comparison as there were substantial other changes on 1.12 (as well as current load-time-tunings targeting older versions), but I wanted to put the number context.