Workflow fixes, documentation reorganization, and LibTorch integration - #560
Merged
bHimes merged 24 commits intoOct 2, 2025
Merged
Conversation
… to unblur to make Decolace alignment work. I am going to immediately reset these files in the next commit, but wanted to stage them here for easy comparison/retrieval.
- Modernize additional_programs.m4 with reusable CISTEM_OPTIONAL_PROGRAM macro - Fix calculate_template_pvalue to only build when Eigen library is present - Improve tasks.json with absolute path error messages for IDE navigation - Update CUDA configuration for newer architectures (sm_86, sm_89, sm_90) - Enhance configure.ac with better MKL threading options This fixes compilation issues in master where calculate_template_pvalue required Eigen but was built unconditionally.
…This is not a sustainable solution, suggesting state be tracked only by main_frame.
…ation After hours of debugging a tricky segfault during workflow switching, this commit: Removes temporary debug code: - Remove all wxPrintf debug statements used during troubleshooting - Remove resolved FIXME comments - Clean up unnecessary TODO comments Adds comprehensive documentation explaining the fixes: - Document why panel destructors must nullify global pointers - Explain the critical sequence in SwitchWorkflowPanels - Detail why null checks in Dirty*() methods prevent crashes - Document global pointer lifecycle in workflow headers The segfault was caused by dangling pointers after panel destruction during workflow switching. The fix involves proper pointer nullification in destructors and defensive null checks before any pointer access. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
Replace debug assertion with graceful early return when no project is open. This prevents crashes when switching from Single Particle to Template Matching workflow, where the constructor is called during workflow transition when the project state may be temporarily inconsistent. Also fix syntax error in TemplateMatchQueueLogger.h QM_TRACE_DB_SCHEMA define. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
Document comprehensive guidelines for using STL containers vs wxWidgets legacy containers, memory management patterns for GUI vs non-GUI objects, and smart pointer usage. Key findings from wxWidgets documentation research: - Use STL containers (std::vector, std::deque) for all new code - Use raw pointers for wxWindow-derived objects (parent-child model) - Use smart pointers for non-GUI data structures - Static members for persistent state across dialog instances Also added build system clarifications and emphasis on reading full documentation. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
Split the monolithic root CLAUDE.md into context-specific files: - src/gui/CLAUDE.md: wxWidgets safety, memory patterns, database access - src/core/CLAUDE.md: Core library, image processing, mathematical ops - src/programs/CLAUDE.md: CLI program patterns, independence from GUI/DB - scripts/CLAUDE.md: Build system and utility script guidance Root CLAUDE.md now focuses on high-level project guidance while detailed technical documentation lives closer to relevant code. This improves discoverability and reduces cognitive load when working in specific areas of the codebase. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit implements LibTorch (PyTorch C++ API) integration as an optional dependency for machine learning features such as the blush regularization tool. Key changes: Build system: - Add m4/libtorch.m4 autoconf macro for LibTorch detection - LibTorch is opt-in via --enable-libtorch configure flag - Set AM_CONDITIONAL and AC_DEFINE for conditional compilation - Use LIBTORCH_CXX_FLAGS pattern (not direct CPPFLAGS modification) - Configure RPATH with $ORIGIN for bundled library distribution - Update src/Makefile.am to link LibTorch for enabled programs Container infrastructure: - Add LibTorch 2.5.0 CPU installation to top_image Dockerfile - Add Python 3.10 venv and scientific computing packages - Add documentation tooling (sphinx, doxygen) - Update container version tags Development tools: - Add VS Code task for LibTorch-enabled debug builds - Task automatically copies required .so files to build/src/lib/ - Include libtorch.so, libtorch_cpu.so, libc10.so, libgomp.so Code infrastructure: - Add include/libtorch/cistem_torch_helper.h wrapper header - Helper handles macro conflicts (N_, NONE, TEXT, INTEGER, etc.) - Clean interface for including torch headers in cisTEM code - Add LibTorch test to console_test.cpp (only when enabled) Build system reorganization: - Move m4 macros to m4/ directory (ax_cuda, additional_programs, etc.) - Update .gitignore for m4/ with whitelist pattern - Simplify regenerate_project.b The LibTorch integration uses dynamic linking with RPATH configuration to support easy distribution bundling. Libraries are found via $ORIGIN/lib, $ORIGIN/../lib, and /opt/libtorch/lib paths. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
twagner9
reviewed
Sep 30, 2025
twagner9
left a comment
twagner9
left a comment
Collaborator
There was a problem hiding this comment.
Overall, this looks good, and I'm interested to see how well the persona specifications will improve the copilot output with Claude.
I just had a couple of questions as I was reviewing, once those are cleared up I'm good to approve!
| @@ -0,0 +1,150 @@ | |||
| # CLAUDE.md | |||
|
|
|||
| This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository. | |||
Collaborator
There was a problem hiding this comment.
The fact that you can specify the persona and provide important context to the AI agent is very neat. Based on the other CLAUDE.md files, is it safe to assume that the is required to be CLAUDE.md to be picked up?
I know the placement of each is relevant to the persona set within the file, I was just wondering if it's possible to add details to the name of the .md so it's more clear what it specifies at a glance.
Comment on lines
+17
to
+49
| #pragma push_macro("N_") | ||
| #pragma push_macro("NONE") | ||
| #pragma push_macro("TEXT") | ||
| #pragma push_macro("INTEGER") | ||
| #pragma push_macro("FLOAT") | ||
| #pragma push_macro("BOOL") | ||
| #pragma push_macro("LONG") | ||
| #pragma push_macro("DOUBLE") | ||
| #pragma push_macro("CHAR") | ||
|
|
||
| // Undefine conflicting macros | ||
| #undef N_ | ||
| #undef NONE | ||
| #undef TEXT | ||
| #undef INTEGER | ||
| #undef FLOAT | ||
| #undef BOOL | ||
| #undef LONG | ||
| #undef DOUBLE | ||
| #undef CHAR | ||
|
|
||
| // Include LibTorch headers | ||
| #include <torch/torch.h> | ||
|
|
||
| // Restore all macros for use in cisTEM code | ||
| #pragma pop_macro("CHAR") | ||
| #pragma pop_macro("DOUBLE") | ||
| #pragma pop_macro("LONG") | ||
| #pragma pop_macro("BOOL") | ||
| #pragma pop_macro("FLOAT") | ||
| #pragma pop_macro("INTEGER") | ||
| #pragma pop_macro("TEXT") | ||
| #pragma pop_macro("NONE") |
Collaborator
There was a problem hiding this comment.
Just to make sure I understand correctly, this is storing the cisTEM macros in a stack, un-defing them to avoid the conflicts in the LibTorch version, and then "re-defing" by popping them back off the stack.
Will this cause potential undefined behavior at runtime or compiler errors because the cisTEM defs will be present again, or do the macros that are brought in from the libtorch include become the only ones available because they were the first defined for the source file in which it's included?
Collaborator
Author
There was a problem hiding this comment.
I don't think so. This is what should happen:
- the precompiled objects in the dynamic library are obv unaffected
- the precompileR will set aside those defines, then begin parsing the included statment, basically copy/pasting each line from the torch header into the top of console_test.cpp (well, a copy of that.) and as it goes it swaps out the defines for strings, just like if you typed it in.
- after the pc is done with its copy paste it resets those macros in whatever environment it uses to track defines.
Comment on lines
+130
to
+131
| "chat.tools.terminal.autoApprove": { | ||
| "mkdir": true, |
Collaborator
There was a problem hiding this comment.
I am assuming that this only prevents having to individually approve a series of actions, rather than default approving all changes made by the agent?
Collaborator
Author
There was a problem hiding this comment.
This approves specific tools, and specifically blocks others. So non-destructive actions like grep are good to approve so you don't have to babysit as much. These rules are for github copilot chat specifically, so they can probably be deprecated. I doubt I'll be switching back. Happy to wait to see what you think of claude code first.
| @@ -0,0 +1,170 @@ | |||
| #!/bin/bash | |||
Collaborator
There was a problem hiding this comment.
This is nice modularity for the installation of tools/kits needed in the container; makes it very clear to follow.
Comment on lines
+10
to
+29
| ActionsPanelSpa::~ActionsPanelSpa( ) { | ||
| // CRITICAL: Nullify all global panel pointers to prevent segfaults during workflow switching. | ||
| // | ||
| // When switching workflows (e.g., from Single Particle to Template Matching), the following sequence occurs: | ||
| // 1. The current ActionsPanelSpa and all its child panels are destroyed | ||
| // 2. These panels are wxWidgets children of ActionsBook, so they're automatically deleted | ||
| // 3. However, global pointers to these panels persist and become dangling pointers | ||
| // 4. Various MainFrame::Dirty*() methods may be called during or after workflow switch | ||
| // 5. These methods check panel pointers and try to set dirty flags if non-null | ||
| // 6. Without nullifying here, they would dereference freed memory → segfault | ||
| // | ||
| // This issue was particularly tricky because: | ||
| // - The segfault was inconsistent (depended on memory reuse patterns) | ||
| // - It often occurred several UI operations after the actual workflow switch | ||
| // - The crash location varied (any Dirty*() method could trigger it) | ||
| // | ||
| // By explicitly nullifying these pointers in the destructor, we ensure that: | ||
| // - Dirty*() methods safely skip destroyed panels (null check fails) | ||
| // - The new workflow can create fresh panel instances without conflicts | ||
| // - Memory access violations are prevented during the transition period |
Collaborator
There was a problem hiding this comment.
Nice! I had noticed this error after the merge and was going to spend some time on it. I knew it was erroneous pointer management but hadn't spent enough time to track it down. I'm glad you were able to identify the problem despite the inconsistency, as that was what was throwing me.
Collaborator
Author
There was a problem hiding this comment.
the robot and I had a few shouting matches over it, but we sorted it out : )
Comment on lines
+1
to
+12
| # cisTEM Makefile for src directory | ||
| # | ||
| # LibTorch Support: | ||
| # ----------------- | ||
| # Programs requiring LibTorch (e.g., blush regularization) should add to their LDADD: | ||
| # program_LDADD = libcore.a $(WX_LIBS_BASE) $(MKL_LIBS) $(LIBTORCH_LIBS) | ||
| # | ||
| # And add RPATH flags to their LDFLAGS: | ||
| # program_LDFLAGS = $(LIBTORCH_RPATH) | ||
| # | ||
| # The LIBTORCH_RPATH setting enables bundling of .so files with the distribution. | ||
| # See m4/libtorch.m4 for configuration details. |
Collaborator
There was a problem hiding this comment.
Nice comment, this would be good practice going forward in case any other third-party limited scope libraries are used.
Collaborator
Author
There was a problem hiding this comment.
Yeah, I wasn't sure if it made more sense at the top or to have it near programs that rely on libtorch, but at the end of the day, it is there for a ctrl+shift+f (or better yet for an agent to find.)
Collaborator
Author
|
@twagner9 we could expand the console test, but i figured a simple tensor multiplication was sufficient for now. |
twagner9
previously approved these changes
Sep 30, 2025
This commit implements a three-layer formatting enforcement system: 1. Pre-commit hook that checks C++/CUDA file formatting before commits - Installed via scripts/install_clang_format_hook.sh - Generates convenience script in /tmp to fix all issues at once - Excludes wxFormBuilder files and third-party headers 2. Auto-installation in regenerate_containers.sh - All developers get the hook automatically 3. CI workflow for final safety net - Checks all PRs and pushes to master/*_with_ci branches - Same exclusion rules as pre-commit hook Excluded from formatting checks: - include/ directory (third-party headers) - src/gui/wxformbuilder/ (wxFormBuilder input files) - Files with ProjectX_gui in name (wxFormBuilder generated) - Files with "DO NOT EDIT" warning in headers 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
…weren't caught locally because I cherry-picked the commit and did not run regenerate_containers.sh to actually add the hook locally)
Adds concurrency control to all workflow files to automatically cancel in-progress runs when new commits are pushed to the same branch. This prevents wasting CI resources on outdated builds. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Updates all build workflows to run only after the formatting check passes, preventing wasted CI resources on improperly formatted code. Changes include: - Switch build workflows to use workflow_run trigger that waits for "Check C++ Formatting" to complete - Add conditional to skip builds if formatting check fails - Replace manual clang-format-14 installation with RafikFarhad/clang-format-github-action@v3 for faster execution - Simplify check_formatting.yml by using pre-built action This ensures formatting issues are caught immediately and prevents all build jobs from running unnecessarily when formatting fails. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Apply clang-format-14 to files with formatting violations detected by CI workflow. Changes include alignment adjustments and whitespace normalization. Includes projectx.cpp which had persistent unformatted changes that were never committed. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Build workflows previously only triggered via workflow_run on push events to master and *_with_ci branches. This caused builds to not run for pull requests, even after formatting checks passed. Add pull_request triggers to all build workflows to enable CI builds on PRs targeting master or *_with_ci branches. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Build workflows were being skipped on pull_request events because the conditional 'if: github.event.workflow_run.conclusion == success' only evaluates true for workflow_run events. Update conditional to run on either pull_request events OR successful workflow_run events, enabling builds to run on PRs. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Convert all build workflows from independent triggers to callable workflows that are orchestrated by check_formatting.yml. This ensures formatting always runs first and creates a clean dependency chain. Changes: - Convert all build workflows to use 'workflow_call' trigger - Remove workflow_run and pull_request triggers from build workflows - Update check_formatting.yml to call all build workflows after format check - Remove conditional logic that's no longer needed This fixes the issue where workflow_run triggers don't work on feature branches and eliminates duplicate workflow runs. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Remove concurrency groups from build workflows that are now called as reusable workflows. These were causing deadlocks when the parent workflow (check_formatting) tried to call multiple child workflows with the same concurrency group pattern. The concurrency control is now managed entirely by the parent workflow. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Collaborator
Author
|
Hey @twagner9 This string of commits just affects CI (and local git hooks). The files outside of those that are changed are just those that had escaped formatting with clang-format as we normally do. This is safe to re-approve, but happy to take any suggestions. One of the reasons I've pushed to get a unified build env in vscode is to make sure tooling (like clang-format) is consistent. This commit drew attention to the fact that edits outside vscode escape the format on save. Small changes have been building up, but Claude's EDIT() task made this really come to light. I've done the following in the last string of commits (it took quite a few to sort out an issue with the CI)
|
The formatting fix script generated by the pre-commit hook now uses absolute paths instead of relative paths, making it work correctly from any directory location. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Collaborator
|
Hey @bHimes, What you've laid out makes sense and I reviewed these changes and they look good. Just to make sure I understand, the changes in .github/workflows/check_formatting.yml only apply to the CI on GitHub, while the script created in scripts actually implements and installs the logic in bash for the local pre-commit hook? This way the formatting is checked and applied where appropriate to avoid instances where clang-format didn't execute for whatever reason? Overall this makes sense, only want to clarify since I haven't dug into CI logic too much at this point. |
twagner9
approved these changes
Oct 2, 2025
Collaborator
Author
|
Yep, that's right! A little redundant but I figured it would be good to nip
this in the bud.
…------------------------
Benjamin A. Himes, PhD
Managing Member
Stochastic Analytics, LLC
------------------------
On Thu, Oct 2, 2025 at 3:56 PM Tim W ***@***.***> wrote:
*twagner9* left a comment (timothygrant80/cisTEM#560)
<#560 (comment)>
Hey @bHimes <https://github.com/bHimes>,
What you've laid out makes sense and I reviewed these changes and they
look good.
Just to make sure I understand, the changes in
.github/workflows/check_formatting.yml only apply to the CI on GitHub,
while the script created in scripts actually implements and installs the
logic in bash for the local pre-commit hook? This way the formatting is
checked and applied where appropriate to avoid instances where clang-format
didn't execute for whatever reason?
Overall this makes sense, only want to clarify since I haven't dug into CI
logic too much at this point.
—
Reply to this email directly, view it on GitHub
<#560 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAXPLGATDQJ3HRWXEN3FZBD3VV7NRAVCNFSM6AAAAACH5CTNTCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNRSG4ZDEMJZGI>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
twagner9
added a commit
to twagner9/cisTEM
that referenced
this pull request
Oct 8, 2025
-Rebased onto timothygrant80#560 to take advantage of some of the dev environment changes. -Added submodule_Blush.m4 for conditional compilation of LibTorch/Blush routines. -configure.ac: moved libtorch/blush conditions to submodule_Blush. Added submodule_Blush to configure.ac. -Split cistem_torch_helper.h logic into two headers to allow for multiple includes from torch headers. -refine3d.cpp: WIP; put blush in the refinement rather than the merge logic -- may give better results, and spares users from having to run generate 3D on a volume to get the non-blushed form. -merge3d.cpp: remove blush logic -GUI Panels: shifting blush around to the refine3d logic rather than merge3d; includes arguments changes in the IPC via the JobPackage class. -Files in src/programs/blush_refinement: added the new format for splitting the logic for avoiding macro conflicts. -console_test.cpp: added new format for splitting the logic for avoiding macro conflicts. -Deleted blush_weights.dat
twagner9
added a commit
to twagner9/cisTEM
that referenced
this pull request
Oct 13, 2025
-Rebased onto timothygrant80#560 to take advantage of some of the dev environment changes. -Added submodule_Blush.m4 for conditional compilation of LibTorch/Blush routines. -configure.ac: moved libtorch/blush conditions to submodule_Blush. Added submodule_Blush to configure.ac. -Split cistem_torch_helper.h logic into two headers to allow for multiple includes from torch headers. -refine3d.cpp: WIP; put blush in the refinement rather than the merge logic -- may give better results, and spares users from having to run generate 3D on a volume to get the non-blushed form. -merge3d.cpp: remove blush logic -GUI Panels: shifting blush around to the refine3d logic rather than merge3d; includes arguments changes in the IPC via the JobPackage class. -Files in src/programs/blush_refinement: added the new format for splitting the logic for avoiding macro conflicts. -console_test.cpp: added new format for splitting the logic for avoiding macro conflicts. -Deleted blush_weights.dat
twagner9
added a commit
to twagner9/cisTEM
that referenced
this pull request
May 28, 2026
-Rebased onto timothygrant80#560 to take advantage of some of the dev environment changes. -Added submodule_Blush.m4 for conditional compilation of LibTorch/Blush routines. -configure.ac: moved libtorch/blush conditions to submodule_Blush. Added submodule_Blush to configure.ac. -Split cistem_torch_helper.h logic into two headers to allow for multiple includes from torch headers. -refine3d.cpp: WIP; put blush in the refinement rather than the merge logic -- may give better results, and spares users from having to run generate 3D on a volume to get the non-blushed form. -merge3d.cpp: remove blush logic -GUI Panels: shifting blush around to the refine3d logic rather than merge3d; includes arguments changes in the IPC via the JobPackage class. -Files in src/programs/blush_refinement: added the new format for splitting the logic for avoiding macro conflicts. -console_test.cpp: added new format for splitting the logic for avoiding macro conflicts. -Deleted blush_weights.dat
twagner9
added a commit
to twagner9/cisTEM
that referenced
this pull request
May 28, 2026
-Rebased onto timothygrant80#560 to take advantage of some of the dev environment changes. -Added submodule_Blush.m4 for conditional compilation of LibTorch/Blush routines. -configure.ac: moved libtorch/blush conditions to submodule_Blush. Added submodule_Blush to configure.ac. -Split cistem_torch_helper.h logic into two headers to allow for multiple includes from torch headers. -refine3d.cpp: WIP; put blush in the refinement rather than the merge logic -- may give better results, and spares users from having to run generate 3D on a volume to get the non-blushed form. -merge3d.cpp: remove blush logic -GUI Panels: shifting blush around to the refine3d logic rather than merge3d; includes arguments changes in the IPC via the JobPackage class. -Files in src/programs/blush_refinement: added the new format for splitting the logic for avoiding macro conflicts. -console_test.cpp: added new format for splitting the logic for avoiding macro conflicts. -Deleted blush_weights.dat
twagner9
added a commit
to twagner9/cisTEM
that referenced
this pull request
Jun 16, 2026
-Rebased onto timothygrant80#560 to take advantage of some of the dev environment changes. -Added submodule_Blush.m4 for conditional compilation of LibTorch/Blush routines. -configure.ac: moved libtorch/blush conditions to submodule_Blush. Added submodule_Blush to configure.ac. -Split cistem_torch_helper.h logic into two headers to allow for multiple includes from torch headers. -refine3d.cpp: WIP; put blush in the refinement rather than the merge logic -- may give better results, and spares users from having to run generate 3D on a volume to get the non-blushed form. -merge3d.cpp: remove blush logic -GUI Panels: shifting blush around to the refine3d logic rather than merge3d; includes arguments changes in the IPC via the JobPackage class. -Files in src/programs/blush_refinement: added the new format for splitting the logic for avoiding macro conflicts. -console_test.cpp: added new format for splitting the logic for avoiding macro conflicts. -Deleted blush_weights.dat
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Description
This PR contains multiple improvements to cisTEM organized into three main areas:
1. Workflow System Fixes (GUI Stability)
Critical bug fix: Resolved workflow switching crashes and state management issues in the GUI.
Problem: Switching between workflows (SPA ↔ Template Matching) caused crashes due to improper panel state management and event handling.
Solution:
MatchTemplatePanel::FillGroupComboBox()crash by checking for null pointers before accessing workflow statesrc/gui/CLAUDE.mdImpact: Users can now reliably switch between workflows without crashes.
2. Documentation Reorganization
Restructured CLAUDE.md documentation into hierarchical structure for better maintainability:
CLAUDE.md: Project overview and Claude Code personasrc/core/CLAUDE.md: Core library development guidelinessrc/gui/CLAUDE.md: GUI development patterns and wxWidgets best practicessrc/programs/CLAUDE.md: Command-line program developmentscripts/CLAUDE.md: Build system and container managementAdded comprehensive documentation on:
3. LibTorch Integration (ML Infrastructure)
Added optional LibTorch (PyTorch C++ API) support for implementing machine learning-based tools such as the blush regularization method for cryo-EM density map denoising.
Motivation: The blush tool (PR #541) requires LibTorch for running trained neural networks. Rather than requiring all users to have LibTorch installed, this PR implements LibTorch as an opt-in dependency that can be enabled at build time with
--enable-libtorch.Key Design Decisions
--enable-libtorchLIBTORCH_CXX_FLAGS,LIBTORCH_LIBS) rather than modifying global compiler flags$ORIGINRPATH to support bundling LibTorch .so files with binaries for easy distributioninclude/libtorch/cistem_torch_helper.hwrapper to handle conflicts between cisTEM/wxWidgets macros and LibTorch headersTechnical Implementation
Build System (
m4/libtorch.m4)AX_LIBTORCHdetects LibTorch at/opt/libtorchor$LIBTORCH_ROOTAM_CONDITIONAL([ENABLE_LIBTORCH_AM])for conditional compilation in MakefilescisTEM_USING_LIBTORCHpreprocessor macro for conditional code$ORIGIN/lib,$ORIGIN/../lib,/opt/libtorch/libContainer Infrastructure
/opt/libtorchCode Integration
include/libtorch/cistem_torch_helper.h: Wrapper header that handles macro conflicts using#pragma push_macro/#pragma pop_macroDevelopment Tools
build/dir/src/lib/after buildAdditional Changes
Build System Improvements
m4/directory for better organizationax_cuda.m4,additional_programs.m4,submodule_FastFFT.m4.gitignorewith whitelist pattern form4/directoryTemplate Matching Utilities
scripts/filter_template_matches_to_group.sh: Filter matches to specific groupsscripts/list_template_match_info.sh: List template match informationscripts/README_template_filtering.mdFixes # (workflow switching crashes)
Related to PR #541 (blush tool - depends on LibTorch infrastructure)
I have rebased my feature branch to be current with the master branch
Which compilers were tested
These changes are isolated to the
How has the functionality been tested?
--enable-libtorchflaglddandreadelfbuild/src/lib/[1,2,3,4], squares it, verifies result[1,4,9,16]--enable-libtorch(default behavior unchanged)Checklist:
cistem_torch_helper.hlibtorch.m4--enable-libtorchflag and LibTorch requirementsTesting Instructions
Build with LibTorch (opt-in)
Build without LibTorch (default)
Verify RPATH
Future Work
Related PRs