gpo.zugaina.org

Search Portage & Overlays:

sci-libs/hipsparselt

Sparse GEMM operations library for AMD Instinct accelerators

Screenshots

  • hipsparselt-7.2.0
    ~amd64
    +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 benchmark roctracer test +llvm_slot_22

    View      Download      Browse     License: MIT   
    Overlay: gentoo

ChangeLog

commit a4a66ec664fc9318d8e1afd082ae0035e62e8ab9
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:49:31 2026 +0200

sci-libs/hipsparselt: drop redundant warning comment

commit 6e3bfa92d792813fefa561d797adbbbad34e5fee
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:49:31 2026 +0200

sci-libs/hipsparselt: drop redundant warning comment

commit d33f8adb4b15724396a9314393a76bb1c3162293
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 04:50:09 2026 +0200

sci-libs/hipsparselt: trim 7.2.4 comment prose

commit b7ed90d211b4c935655337e6b97806b075620c83
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 00:28:50 2026 +0200

sci-libs/hipsparselt: trim 10.0.0 comment prose

commit 96e6f71c1586b35138ccec733dd62a534a5921b1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 2 10:27:42 2026 +0200

sci-libs/hipsparselt: require hipcc[amd-llvm], matching hipBLASLt

hipsparselt builds the same TensileLite/rocisa generator as
sci-libs/hipBLASLt, from the bundled hipblaslt.tar.gz and origami.tar.gz
release assets it unpacks, and that generator emits gfx11xx assembly the
vanilla assembler rejects. hipBLASLt already carries
dev-util/hipcc:$[amd-llvm] for exactly that reason; this package
had the bare atom, so a build against a hipcc without the flag would
have died deep in compilation with no resolver-level signal.

Landing it now because the ROCm 10.0 cohort mask is about to be lifted.
Until now that requirement was recorded only in the mask comment, which
is going away, so anything it covered has to live in the ebuilds first.

Honest limit: this is NOT build-verified. hipsparselt excludes gfx1150,
the only GPU on this host, so it has never been compiled here -- the
constraint is carried across by symmetry with the generator the two
packages share, not demonstrated. Noted in the ebuild as the line to
revisit if hipsparselt's slice of TensileLite turns out not to emit the
offending instructions.

BDEPEND-only, so no installed file changes and no revbump; the package
is mask-blocked at present, so no one can have it merged. Verified the
dep is satisfiable -- it still resolves, against the hipcc already built
with amd-llvm.

commit 241a3911e00523a2c07f8a6939ec1dfc7a2fd4cd
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 23:39:39 2026 +0200

sci-libs/hipsparselt: drop 7.2.3

Superseded by 7.2.4, the last release of the 7.x series and the rollback
target for the ROCm 10.0 migration. Version retention keeps the last two
plus the last of each previous major series; 7.2.4 and 10.0.0 satisfy both,
so the third version is redundant.

commit 6ce50218d37af956384400fb941331790ba44079
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 19:02:02 2026 +0200

sci-libs/hipsparselt: keep test artifacts out of the image, and assert the anchors

Upstream spells the install component two ways in CMakeLists.txt and both
install. Anchoring only the quoted form patched one site of four:

:286 TARGETS hipsparselt-test COMPONENT "tests" ...
:297 FILES $ ... COMPONENT tests
:302 FILES $ ... COMPONENT tests
:307 PROGRAMS $ ... COMPONENT tests

USE=test sets both -DHIPSPARSELT_ENABLE_CLIENT=ON and
-DHIPSPARSELT_BUILD_TESTING=ON, so the yaml, data and program payload
installed into /usr/bin. sci-libs/hipBLASLt needs only the unquoted form --
all 13 of its sites use it -- which is why the two ebuilds differ here rather
than sharing one anchor.

The end-of-line anchor is load-bearing, not cosmetic:
rocm_package_add_rpm_dependencies() and its deb twin also take COMPONENT tests
mid-line at :233-234. Those are packaging metadata, not install() calls, and
appending EXCLUDE_FROM_ALL to them would be wrong.

Also asserts the two tensilelite anchors beside it -- the Validators.py
amdclang name and the tensilelite/Makefile compiler path -- which had no guard
while the origami, stinkytofu and STINKYTOFU_ENABLE_WERROR seds did.

Verified by running src_prepare with USE=test: all four install sites carry
EXCLUDE_FROM_ALL, the two rocm_package_add_* lines are untouched, and no guard
fires.

commit c3dd8ab2f06810a4fa4eaeb49e33fac6fe5c8bb4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 18:46:48 2026 +0200

sci-libs/hipsparselt: build the bundled rocisa with -DNDEBUG

hipsparselt co-unpacks hipblaslt.tar.gz and builds the same
tensilelite/rocisa that sci-libs/hipBLASLt does -- same nanobind patch, same
co-unpacked stinkytofu, same STINKYTOFU_ENABLE_WERROR sed -- but it never
picked up the -DNDEBUG / CMAKE_BUILD_TYPE=Release alignment that hipBLASLt
carries. cmake.eclass defaults to RelWithDebInfo AND blanks
CMAKE_CXX_FLAGS_RELWITHDEBINFO, so the standard -DNDEBUG never landed here
and the asserts stayed live. With a GPU target enabled, TensileLite would
abort partway through kernel generation on

rocisa/src/pass/macro_inline.cpp:315: Assertion `false &&
"macroToInstruction: unexpected item type in macro body"' failed
-> Fatal Python error: Aborted

which is exactly what hipBLASLt's block exists to prevent. The -Werror sed
this ebuild already copied from hipBLASLt was itself predicated on NDEBUG
being set, so the two were inconsistent as shipped.

Build-verified on gfx1150, but only partly, and the comment says which part:
rocisa and stinkytofu do compile -- with NDEBUG, through the -Werror sed --
while the codegen pass that trips the assert does not run, since IUSE_TARGETS
is gfx942/gfx950 only and no target is enabled on this host.

commit 0e9eefeddcef61d09538c91754f7a9949edac5fa
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 13:52:42 2026 +0200

sci-libs/hipsparselt: add 10.0.0

Tag moves to therock-10.0 for all three co-unpacked assets, and LLVM_COMPAT
goes 22 -> 23 to join the rest of the subslot-pinned cohort.

Co-unpack stinkytofu, new at 10.0. The bundled hipblaslt tensilelite/rocisa now
requires it via find_package or a monorepo sibling directory, and hard-fails
when neither is present:

CMake Error at .../tensilelite/rocisa/CMakeLists.txt:141 (message):
stinkytofu not found via find_package or sibling directory

It is not packaged separately, so fetch the release asset and repoint rocisa at
$/stinkytofu -- the same treatment sci-libs/hipBLASLt already gives it,
including STINKYTOFU_ENABLE_WERROR=OFF and the optional-clang-tidy patch.

Bump the rocisa-nanobind patch to the 10.0 regeneration in lockstep with
sci-libs/hipBLASLt; the 7.1.0 copy no longer applied to the bundled hipblaslt.

Assert the origami and stinkytofu sibling-path sed anchors: sed exits 0 on
no-match, and a miss means a network fetch (origami) or that FATAL_ERROR.

commit d26e488d095396a8343a1053601f18f59982a8df
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 27 10:04:12 2026 +0200

*/*: drop dead Python targets from PYTHON_COMPAT

139 ebuilds across 65 packages still listed python3_10, python3_11 or
python3_13t. None of the three is a supported implementation any more:
gentoo's python-utils-r1 sets _PYTHON_ALL_IMPLS to python3_ plus
python3_t, and everything older falls into the branch commented
"implementations deprecated prior to EAPI 9 are fatal".

Inert today, which is why nothing caught it. Targets outside
_PYTHON_ALL_IMPLS are silently dropped rather than rejected, so they
never became USE flags and never produced a dependency atom -- an ebuild
declaring python3_ was only ever offering 12, 13 and 14.

Two reasons to clean them up anyway. The eclass branch they land in
dies for any EAPI other than 7 or 8, so all 139 are waiting to break on
the first EAPI 9 bump. And pkgcheck's OldPythonCompat is a git-scoped
check that only inspects ebuilds touched in a commit range, so it never
reports these on a full-tree scan and would instead fire on whichever
unrelated commit next touches one of them.

No revbump, because nothing observable changes. Verified on a sample
spanning every transformation shape that the python USE flags portage
offers are byte-identical before and after, and that a pkgcheck run
over all 65 packages returns a set-identical list of findings.

The eight python2_7 ebuilds are deliberately untouched: that target is
also absent from _PYTHON_ALL_IMPLS, but they reach it through the
_PYTHON_ALLOW_PY27 hatch in the overlay's forked python-utils-r1_py2.

commit 7c384c8cac70105c06770e0ccb7af0fc295f2471
Author: Raukaan Cogbrother <cogbrother@raukaan.local>
Date: Sun May 31 03:29:35 2026 +0200

metadata: normalize tab → 2-space indent in 42 imported files

42 metadata.xml files inherited tab indentation from their origin
overlays (mostly the ROCm cluster, plus a handful of dev-python and
AMD-tooling imports). The overlay convention is 2-space; this brings
them in line with the other 498 files. Maintainer attributions
preserved verbatim.

commit 94c24af27389243e8a0016222738616305dfb23e
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 29 16:49:08 2026 +0200

sci-libs: fix hipBLASLt/hipsparselt double-wrap unpack for ROCm 7.2.4

Both ebuilds kept a custom src_unpack that mkdir'd hipblaslt/, origami/
(and hipsparselt/) wrapper dirs and unpacked into them — a workaround for
7.2.3's flat release-asset tarballs. The rocm-7.2.4 tarballs now carry
those top-level dirs themselves, so the manual wrap double-nested them
(work/hipblaslt/hipblaslt/...) and src_prepare died: S/ORIGAMI_S/
HIPBLASLT_S pointed at the outer wrapper, missing tensilelite/Tensile.
Unpack the self-rooting tarballs directly, matching the dev-util/hip and
dev-libs/rocm-opencl-runtime fix.

commit 3ab8f927677379afe02227c61c8a4d7414de5c71
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 29 14:12:08 2026 +0200

sci-libs/hipsparselt: add 7.2.4

commit 8be3c26f22e0fafeb30a067919d44f961297545c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 10:56:30 2026 +0200

sci-libs/hipsparselt: 7.2.3 add src_unpack for flat upstream tarballs

The 7.2.3 hipsparselt.tar.gz, origami.tar.gz, and hipblaslt.tar.gz from ROCm/rocm-
libraries all unpack flat and share top-level entries. Add src_unpack to extract each
into its own subdir, matching the hipsparselt / origami / hipblaslt paths the ebuild
expects.

commit 58fcf0494096ef327172fdf2e41bd0663ce77014
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 01:58:16 2026 +0200

sci-libs/hipsparselt: add 7.2.3