Expose HIPSPARSELT_ENABLE_THEROCK through the build drivers by talumbau · Pull Request #12374 · ROCm/rocm-libraries

talumbau · 2026-09-21T16:09:05Z

Why

HIPSPARSELT_ENABLE_THEROCK selects the path where hipSPARSELt consumes a staged hipBLASLt package via find_package and links the shared libtensilelite-host.so, instead of pulling hipBLASLt in with add_subdirectory and recompiling TensileLite into libhipsparselt.so.

The option defaults OFF and there is currently no way to turn it on from outside the source tree:

install-host.sh builds cmake_common_options from a closed set of flags — no $@, no --cmake-args passthrough
tasks.py build() has a fixed keyword list with no corresponding entry

So any packaging build always takes the duplicating path. On a 7.0.2 drop build that means libhipsparselt.so carries 8,787 hidden TensileLite symbols and 5,774 ContractionProblemGemm strings, with no DT_NEEDED on libtensilelite-host.so.1.

What this unblocks

This is the switch that lets the hipBLASLt/hipSPARSELt drop pipeline pick up the TensileLite deduplication from #12888. Measured on gfx942:

	current drop build	with `--therock`
`libhipsparselt.so`	7.43 MB	0.92 MB
TensileLite symbols inside it	8,787	127
`DT_NEEDED` on tensilelite	none	`libtensilelite-host.so.1`

It also activates the hipSPARSELt packaging work already merged in #12888. That dependency is gated on

if(HIPSPARSELT_ENABLE_HIP AND (HIPSPARSELT_ENABLE_THEROCK OR NOT BUILD_SHARED_LIBS))

which is false in the current drop build, so no tensilelite-host dependency is emitted at all today — the shipped hipsparselt RPM requires only rocm-core. Without this PR, #12888's hipSPARSELt half never runs in the RPM path.

What it changes

Adds a --therock flag to install-host.sh and a matching therock= keyword to tasks.py build(), so a CI driver can select the shared path. 21 lines, purely plumbing.

Additive and opt-in. No defaults change — the add_subdirectory path remains the default for standalone builds and is unaffected.

Ordering

Should land after #12373. Enabling THEROCK=ON without those fixes produces a hipSPARSELt whose clients fail to link (dlange_, _gfortran_*) and whose hipsparselt-test cannot find libomp.so, because THEROCK=ON also forces TheRock's cblas target — which outside TheRock is netlib CBLAS with no LAPACK. #12373 separates those two concerns.

The companion rocJenkins change that actually passes the flag is held until this merges.

TheRock impact: none

TheRock configures CMake directly via therock_cmake_subproject_declare and does not use install-host.sh or tasks.py.

Refs AIHPBLAS-3522, AIHPBLAS-4434.

therock-pr-bot · 2026-09-21T16:43:22Z

✅ All Checks Passed — Ready for Review

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	⚠️ Warning	Error: Source/code files changed without an accompanying unit test. Expected: add at least one test file named like `test_<name>.py` / `test_<name>.cpp` (or `<name>_test.`). Current:* code file(s) changed: `projects/hipsparselt/tasks.py`; no test file found
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—
🤖 therock-pr-bot	✅ Pass	—

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

therock-pr-bot · 2026-09-21T16:43:24Z

🎉 All checks passed! This PR is ready for review.

HIPSPARSELT_ENABLE_THEROCK selects the path where hipSPARSELt consumes a staged hipBLASLt package via find_package and links the shared libtensilelite-host.so, instead of pulling hipBLASLt in with add_subdirectory and recompiling TensileLite into libhipsparselt.so. The option defaults OFF and there was no way to turn it on from outside the source tree: install-host.sh builds cmake_common_options from a closed set of flags with no passthrough, and tasks.py build() has a fixed keyword list. Packaging builds therefore always took the duplicating path -- a 7.0.2 drop build shows 8,787 hidden TensileLite symbols and 5,774 ContractionProblemGemm strings inside libhipsparselt.so, and no DT_NEEDED on libtensilelite-host.so.1. Add a --therock flag to install-host.sh and a matching therock= keyword to tasks.py build(), so a CI driver can select the shared path. Additive and opt-in; no defaults change. The add_subdirectory path remains the default for standalone builds and is unaffected. Refs AIHPBLAS-3522, AIHPBLAS-4434. Co-Authored-By: Claude <noreply@anthropic.com>

…12373) ## Why hipSPARSELt has accumulated several undeclared dependencies on hipBLASLt's **configure-time side effects**. When hipBLASLt is pulled in via `add_subdirectory()`, its `CMakeLists.txt` runs in the same configure and hipSPARSELt silently inherits settings from it. Consuming hipBLASLt as a package instead removes those, and nothing replaces them. **Nothing shipping today is broken.** Both current build paths work: | Build | `THEROCK` | BLAS it resolves to | clients | |---|---|---|---| | ROCm drop RPMs | OFF | BLIS + netlib LAPACK, statically linked | fine | | TheRock nightly | ON | `librocm-openblas.so.0` | fine | | staged-package build outside TheRock | ON | netlib CBLAS — no LAPACK | **fails** | That third configuration is the one #12374 is about to enable for the drop pipeline, in order to get the TensileLite deduplication into the RPMs. This PR makes it work before it gets switched on. These are three independent fixes, each defensible on its own: - a Fortran/LAPACK setup gated on the wrong condition, - a BLAS selector coupled to an unrelated flag, - a project relying on another project's cache variable. ## What **1. Scope the Fortran/LAPACK setup to configurations that need it.** It was gated on `NOT HIPSPARSELT_ENABLE_THEROCK` while `add_subdirectory(clients)` ran unconditionally. That works in TheRock because its `cblas` target is `rocm-openblas`, which bundles BLAS, CBLAS, LAPACK and the Fortran runtime in one library — verified against a TheRock nightly, where `hipsparselt-test` and `hipsparselt-bench` both carry `DT_NEEDED` on `librocm-openblas.so.0` and resolve `dlange_` from it. Outside TheRock the same branch resolves to the netlib CBLAS built by `fetch_lapack`, which has no LAPACK and does not pull the Fortran runtime: ``` ld.lld: error: undefined symbol: _gfortran_st_write ld.lld: error: undefined symbol: dlange_ ``` Now gated on the same selector the BLAS choice uses: `NOT THEROCK OR BLIS`. > ⚠️ Deliberately **not** widened to "whenever clients are built". `HIPSPARSELT_ENABLE_CLIENT` defaults `ON` and TheRock does not override it, so an unconditional `enable_language(Fortran)` + `find_package(LAPACK)` would run in TheRock builds — where `HIPSPARSELT_ENABLE_FETCH=OFF` sends a failed `find_package(LAPACK)` into the `FATAL_ERROR` branch of `fetch_lapack.cmake`. An earlier revision of this commit did exactly that and would have broken TheRock. **2. Decouple the client BLAS choice from `HIPSPARSELT_ENABLE_THEROCK`.** That option selects two unrelated things: how hipBLASLt is consumed, *and* which CPU reference BLAS the clients link. `hipsparselt_link_blas_libraries()` linked a bare `cblas` target whenever THEROCK was on, which is only correct where `cblas` is OpenBLAS. The explicit BLAS selection is now checked first. **3. Declare hipSPARSELt's install RPATH.** hipSPARSELt never set `CMAKE_INSTALL_RPATH`; it inherited hipBLASLt's, set as a **cache** variable at `projects/hipblaslt/CMakeLists.txt:114`. On the staged-package path that never runs: ``` add_subdirectory : $ORIGIN/../lib:$ORIGIN/../llvm/lib:$ORIGIN/../lib:... staged package : $ORIGIN/../lib:$ORIGIN/../lib/hipsparselt/lib:... ``` `libomp.so` lives in `../llvm/lib`, so `hipsparselt-test` failed at startup with `error while loading shared libraries: libomp.so`. ## TheRock impact Commits 1 and 2: **none**, verified. TheRock's flags (`math-libs/BLAS/CMakeLists.txt:366`) leave `HIPSPARSELT_ENABLE_BLIS` at its `OFF` default, so it still takes the `cblas` branch and still skips the Fortran setup. Configure-only runs against a staged hipBLASLt package: | Configuration | Fortran block | `fetch_lapack` FATAL | BLAS chosen | configure | |---|---|---|---|---| | `THEROCK=ON, BLIS unset, FETCH=OFF` *(TheRock's actual flags)* | no | no | TheRock OpenBLAS | OK | | `THEROCK=ON, BLIS=true, FETCH=ON` | yes | no | BLIS + LAPACK | OK | | `THEROCK=OFF, BLIS=true` *(current drop default)* | yes | no | BLIS + LAPACK | OK | Row 1 matches TheRock's behaviour before this series exactly. ⚠️ **Commit 3 is a real behaviour change for TheRock and wants a sign-off.** TheRock passes no `-DCMAKE_INSTALL_RPATH`, so where its hipSPARSELt binaries previously had no install RPATH they will now get `$ORIGIN/../lib:$ORIGIN/../llvm/lib`. Low risk — relative RPATH entries that do not resolve are skipped by the loader, and TheRock can override by passing the variable, since the `set()` uses `CACHE` without `FORCE`. If you would rather not take it, the fallback is to gate the `set()` on `NOT HIPSPARSELT_ENABLE_THEROCK`. ## Testing gfx942, ROCm 7.0.2, RHEL 9.5. - `hipsparselt-test --gtest_filter=*smoke*:*quick*:*pre_checkin*-*known_bug*`: **77,993 passed** - `hipblaslt-test --gtest_filter=*smoke*`: **3,118 passed** - **`THEROCK=OFF` path re-verified** after these fixes, against the original drop RPMs: `DT_NEEDED` on tensilelite 0 → 0, `Requires` `rocm-core` → `rocm-core`, RUNPATH still contains `../llvm/lib`. Unchanged. Refs AIHPBLAS-4434, AIHPBLAS-3522. --------- Co-authored-by: Claude <noreply@anthropic.com>

hcman2

OK for me. better to have Vin's approval

talumbau · 2026-10-28T16:06:17Z

@vin-huang I see @hcman2 's comment now. Please advise on follow ups if needed. Trying to get this in before next scheduled drop on 10/29.

assistant-librarian · 2026-10-30T04:11:28Z

TheRock Submodule Bump Activity

Newest first

2026-10-30 — Included in TheRock via ROCm/TheRock#9594.

github-actions Bot added the project: hipsparselt label Sep 21, 2026

therock-pr-bot Bot added the Not ready to Review label Sep 21, 2026

assistant-librarian Bot added the organization: ROCm label Sep 21, 2026

talumbau requested a review from davidd-amd September 23, 2026 23:18

talumbau marked this pull request as ready for review September 23, 2026 23:18

talumbau requested a review from a team as a code owner September 23, 2026 23:18

talumbau force-pushed the tja/tlh-r4-therock-flag branch from de70062 to 62e1e73 Compare September 23, 2026 23:22

talumbau force-pushed the tja/tlh-r4-therock-flag branch from 62e1e73 to 8f45c98 Compare October 16, 2026 18:22

talumbau mentioned this pull request Oct 16, 2026

Stop inheriting hipBLASLt's configure-time settings in hipSPARSELt #12373

Merged

talumbau force-pushed the tja/tlh-r4-therock-flag branch from 8f45c98 to 40dc3f9 Compare October 16, 2026 21:51

therock-pr-bot Bot removed the Not ready to Review label Oct 16, 2026

davidd-amd approved these changes Oct 19, 2026

View reviewed changes

talumbau requested review from hcman2 and vin-huang October 20, 2026 21:36

hcman2 approved these changes Oct 28, 2026

View reviewed changes

talumbau merged commit 8cd2084 into develop Oct 28, 2026
142 of 151 checks passed

talumbau deleted the tja/tlh-r4-therock-flag branch October 28, 2026 16:05

Navigation Menu

Repository navigation

Expose HIPSPARSELT_ENABLE_THEROCK through the build drivers - #12374

Expose HIPSPARSELT_ENABLE_THEROCK through the build drivers#12374
talumbau merged 1 commit into
developfrom
tja/tlh-r4-therock-flag

talumbau commented Sep 21, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Sep 21, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Sep 21, 2026 •

edited

Loading

Uh oh!

hcman2 left a comment

Uh oh!

Uh oh!

talumbau commented Oct 28, 2026

Uh oh!

assistant-librarian Bot commented Oct 30, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

talumbau commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

What this unblocks

What it changes

Ordering

TheRock impact: none

Uh oh!

therock-pr-bot Bot commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ All Checks Passed — Ready for Review

Uh oh!

therock-pr-bot Bot commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hcman2 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

talumbau commented Oct 28, 2026

Uh oh!

assistant-librarian Bot commented Oct 30, 2026

TheRock Submodule Bump Activity

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

talumbau commented Sep 21, 2026 •

edited

Loading

therock-pr-bot Bot commented Sep 21, 2026 •

edited

Loading

therock-pr-bot Bot commented Sep 21, 2026 •

edited

Loading