Reapply "Move non-sysdep third-party projects after base and compiler" by ScottTodd · Pull Request #3440 · ROCm/TheRock

ScottTodd · 2026-02-17T00:22:08Z

Motivation

This reverts #3439 to reapply #2045 and fix #2003.

Technical Details

The original PR was missing changes in BUILD_TOPOLOGY.toml. If we switch more third-party subprojects to use the amd-llvm toolchain they will need similar changes.

I'm also testing a follow-up change to migrate (nearly) all other third-party subprojects to use amd-llvm.

We could re-evaluate the build stages as "Foundation" gets smaller and "Compiler Runtime" gets larger (could merge them or split out a new stage for third party deps that use the compiler 🤔).

Test Plan

Run multi-arch CI on this branch: https://github.com/ROCm/TheRock/actions/runs/22112365715

check logs for "Foundation" stage (should NOT build spdlog)
check logs for "Compiler Runtime" stage (should build spdlog and upload the artifact)
check "fetch inbound artifacts" logs for "Profiler Apps" stage (should fetch spdlog artifact)

Submission Checklist

Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

#2045)" This reverts commit 5af3094.

The spdlog add_subdirectory is now guarded by if(THEROCK_ENABLE_SPDLOG), so configure_stage.py must know about the dependency to enable the feature in downstream stages. Without this, math-libs, profiler-apps, and iree-libs stages fail to configure because the therock-spdlog target is never created. Changes: - Add "spdlog" to artifact_deps of blas, hipdnn, miopen-plugin, hipblaslt-plugin, hipdnn-samples, fusilli-plugin, and rocprofiler-systems 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

marbre · 2026-02-18T23:13:53Z

 artifact_group = "math-libs"
 type = "target-specific"
-artifact_deps = ["core-runtime", "core-hip", "core-amdsmi", "host-blas", "host-suite-sparse", "rocprofiler-sdk"]
+artifact_deps = ["core-runtime", "core-hip", "core-amdsmi", "host-blas", "host-suite-sparse", "rocprofiler-sdk", "spdlog"]


So far spdlog belonged to third-party-libs which covers optional deps. If we strictly want to pull spdlog in, we should rather consider moving it to third-party-sysdeps. However, it might just wrong how the components depend on it if this is optional?

I don't understand what "optional deps" mean in the context of this file. There are subprojects that require these dependencies. This is also not a sysdep (it isn't under https://github.com/ROCm/TheRock/tree/main/third-party/sysdeps), so it shouldn't go in third-party-sysdeps, unless I'm missing some detail in how deps are defined...

Yeah I think the descriptions here are just wrong:

TheRock/BUILD_TOPOLOGY.toml

Lines 195 to 202 in 1f5412f

[artifact_groups.third-party-sysdeps]

description = "Critical system dependencies (always needed)"

type = "generic"

[artifact_groups.third-party-libs]

description = "Optional third-party libraries (for tests/specific features)"

type = "generic"

source_sets = [] # No submodules - uses vendored/in-tree deps

third-party-sysdeps are libraries that we build for portable distribution

third-party-libs are build dependencies. they include some truly optional libraries (HOST_MATH: host-blas, suite-sparse, fftw3)

I can send a separate PR to update the docs. Raw notes from Claude Code: https://gist.github.com/ScottTodd/bb0ca24ff2a90698df3229a886d2611c

Okay, that might have trapped me here. Doc update can be a follow up for sure if needed but it feels there might be more wrong. Will try to take a closer look.

Sent doc updates as #3548

ScottTodd · 2026-02-19T00:47:40Z

(#3474 is stacked after this and it moves nearly all third-party subprojects to build with amd-llvm)

marbre · 2026-02-20T06:03:50Z

 artifact_group = "math-libs"
 type = "target-specific"
-artifact_deps = ["core-runtime", "core-hip", "core-amdsmi", "host-blas", "host-suite-sparse", "rocprofiler-sdk"]
+artifact_deps = ["core-runtime", "core-hip", "core-amdsmi", "host-blas", "host-suite-sparse", "rocprofiler-sdk", "spdlog"]


Okay, that might have trapped me here. Doc update can be a follow up for sure if needed but it feels there might be more wrong. Will try to take a closer look.

…l-forward-2045

ScottTodd · 2026-02-20T16:43:31Z

Synced and resolved some merge conflicts in the topology file. Triggered a multi-arch CI test run just to be sure: https://github.com/ROCm/TheRock/actions/runs/22232603395

ScottTodd · 2026-02-20T17:45:02Z

CI runs seem to be far enough along to trust that they will succeed.

) ## Motivation Following up on discussion at #3440 (comment). The descriptions in `BUILD_TOPOLOGY.toml` for `third-party-sysdeps` and `third-party-libs` were misleading. This adds more specific background details explaining what qualifies as a sysdep. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

## Motivation Follow-up to #2045 (technically #3440) and based on comments at #1284 (comment). The general goal here is to build _only_ sysdeps with the system toolchain/compiler and to build the rest of the third-party deps (and ROCm) using our own amd-llvm toolchain. This will let more subprojects build with sanitizers as those are generally not portable across toolchains. ## Technical Details * This required some additional features in the build topology system to add a `THEROCK_ENABLE_THIRD_PARTY_LIBS` CMake option that gets automatically enabled when an artifact includes the `third-party-libs` feature via `group_deps`. The existing `artifact_deps` was not sufficient for this as the third party libraries do not produce artifacts and are instead pure source dependencies, which now take a dependency on the amd-llvm (compiler) artifact. * I wasn't able to build `therock-eigen` or `therock-host-blas` on Windows using the amd-llvm toolchain without deeper changes (see the logs snippets in the source). I did not test those builds on Linux. ## Test Plan * Local build on Windows * Regular CI on this PR * Multi-arch CI manually triggered: * (recent) https://github.com/ROCm/TheRock/actions/runs/22598994883 * (older) https://github.com/ROCm/TheRock/actions/runs/22235154958 * (even older) https://github.com/ROCm/TheRock/actions/runs/22119457085 ## Test Result * Local build succeeded and used Clang from amd-llvm as expected for each of the third-party subprojects. * CI build https://github.com/ROCm/TheRock/actions/runs/22121852952?pr=3474 passed builds and most tests (preexisting flakes?) * Multi-arch CI build https://github.com/ROCm/TheRock/actions/runs/22119457085 passed builds and most tests (preexisting flakes/failures?) ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Marius Brehler <marius.brehler@amd.com>

ScottTodd added 2 commits February 16, 2026 16:16

Reapply "Move non-sysdep third-party projects after base and compiler. (

bddac93

#2045)" This reverts commit 5af3094.

Adjust topology and CMake inclusion for spdlog.

3994080

github-project-automation Bot added this to TheRock Triage Feb 17, 2026

github-project-automation Bot moved this to TODO in TheRock Triage Feb 17, 2026

ScottTodd commented Feb 17, 2026

View reviewed changes

Comment thread BUILD_TOPOLOGY.toml

ScottTodd mentioned this pull request Feb 17, 2026

Bump rocm-libraries from 8f32542 to 30c3715 #3424

Merged

ScottTodd marked this pull request as ready for review February 17, 2026 21:42

ScottTodd requested a review from marbre February 17, 2026 21:42

ScottTodd commented Feb 17, 2026

View reviewed changes

Comment thread third-party/CMakeLists.txt

ScottTodd mentioned this pull request Feb 18, 2026

Switch most third-party subprojects to the amd-llvm toolchain #3474

Merged

1 task

marbre reviewed Feb 18, 2026

View reviewed changes

ScottTodd requested a review from marbre February 19, 2026 23:43

marbre approved these changes Feb 20, 2026

View reviewed changes

Merge remote-tracking branch 'upstream/main' into users/scotttodd/rol…

d80efe9

…l-forward-2045

ScottTodd merged commit 295368e into main Feb 20, 2026
29 checks passed

ScottTodd deleted the users/scotttodd/roll-forward-2045 branch February 20, 2026 17:45

github-project-automation Bot moved this from TODO to Done in TheRock Triage Feb 20, 2026

ScottTodd mentioned this pull request Feb 21, 2026

Clarify the difference between sysdeps and other third-party deps #3548

Merged

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Reapply "Move non-sysdep third-party projects after base and compiler"#3440

Reapply "Move non-sysdep third-party projects after base and compiler"#3440
ScottTodd merged 4 commits into
mainfrom
users/scotttodd/roll-forward-2045

ScottTodd commented Feb 17, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

marbre Feb 18, 2026

Uh oh!

ScottTodd Feb 19, 2026

Uh oh!

ScottTodd Feb 19, 2026

Uh oh!

marbre Feb 20, 2026

Uh oh!

ScottTodd Feb 21, 2026

Uh oh!

ScottTodd commented Feb 19, 2026

Uh oh!

marbre Feb 20, 2026

Uh oh!

ScottTodd commented Feb 20, 2026

Uh oh!

ScottTodd commented Feb 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	[artifact_groups.third-party-sysdeps]
	description = "Critical system dependencies (always needed)"
	type = "generic"

	[artifact_groups.third-party-libs]
	description = "Optional third-party libraries (for tests/specific features)"
	type = "generic"
	source_sets = [] # No submodules - uses vendored/in-tree deps

Conversation

ScottTodd commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Technical Details

Test Plan

Submission Checklist

Uh oh!

Uh oh!

Uh oh!

marbre Feb 18, 2026

Choose a reason for hiding this comment

Uh oh!

ScottTodd Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

ScottTodd Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

marbre Feb 20, 2026

Choose a reason for hiding this comment

Uh oh!

ScottTodd Feb 21, 2026

Choose a reason for hiding this comment

Uh oh!

ScottTodd commented Feb 19, 2026

Uh oh!

marbre Feb 20, 2026

Choose a reason for hiding this comment

Uh oh!

ScottTodd commented Feb 20, 2026

Uh oh!

ScottTodd commented Feb 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ScottTodd commented Feb 17, 2026 •

edited

Loading