commit e09555ac749b528f9faa1c8644c2fb0b9bfb072e
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:49:31 2026 +0200

sci-libs/hipBLASLt: drop redundant warning comment

commit 962c1d7d3600a49412d7d68644d22bf7834edd6e
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:49:31 2026 +0200

sci-libs/hipBLASLt: drop redundant warning comment

commit d76fed9fc1479fd3a0788e4ad3d8280a35f146fc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 04:34:45 2026 +0200

sci-libs/hipBLASLt: trim 7.2.4 comment prose

commit b31abfdc14ed131def212f0ffc9a1b834d48b617
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 00:26:31 2026 +0200

sci-libs/hipBLASLt: trim 10.0.0 comment prose

commit a9a8d68e6abdddddba20568032724bdff8b048a2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 23:39:38 2026 +0200

sci-libs/hipBLASLt: 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Superseded by 7.2.4, the last release of the 7.x series and the rollback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 4c0ff5f794433749253ef908d81178b33c722d1a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 22:32:01 2026 +0200

sci-libs/hipBLASLt: drop the dead TENSILE_ROCM_ASSEMBLER_PATH export

The vendored TensileLite has zero references to TENSILE_ROCM_ASSEMBLER_PATH --
checked across hipblaslt/, origami/ and stinkytofu/ at therock-10.0. It picks
its assembler through Tensile/Toolchain/Validators.py instead, which searches
ROCM_PATH/bin, ROCM_PATH/lib/llvm/bin, /opt/rocm/{bin,lib/llvm/bin} and then
PATH. So ROCM_PATH (exported just above) is what actually steers it, and this
export only made it look as though something else did.

The neighbouring CMAKE_CXX_COMPILER export is NOT dead and stays: TensileLite
really does read it from the environment, at
Tensile/Common/GlobalParameters.py:869-870. Both facts are now recorded in the
ebuild so the next reader does not have to re-derive them -- and so this one is
not "cleaned up" by mistake.

Build-verified rather than assumed, since removing an export is exactly the
kind of change that looks safe and is not: full build green, 203 of 204
toolchain invocations resolving to /usr/lib/rocm-llvm/bin (AMD's fork, as
intended -- 201 clang++ plus 2 clang; the single exception is
clang-offload-bundler from llvm/23, which is not a compiler),
libhipblaslt.so.1.4 built with gfx1150 code objects.

commit 8f1c8e594764cc8e86e834fe208befd4beec1374
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 19:05:37 2026 +0200

sci-libs/hipBLASLt: require dev-util/hipcc[amd-llvm]

hipBLASLt is the package profiles/package.mask names as the primary reason
dev-util/hipcc grew USE=amd-llvm: TensileLite GENERATES gfx1150 assembly the
vanilla assembler rejects ("operands are not valid for this GPU or mode" on
v_max_f16 with an abs() source modifier), and being generated it cannot be
patched away. The requirement nonetheless lived only in that mask comment,
which is documentation rather than a constraint -- building against a hipcc
without the flag failed deep in compilation with no resolver-level signal.

Unlike composable-kernel and rocBLAS, this ebuild already declared
dev-util/hipcc in BDEPEND, so only the USE-dep was missing.

Unconditional on purpose; the comment records what would justify a narrower
amdgpu_targets_*? form.

sci-libs/hipsparselt deliberately does NOT get this. It builds the same
bundled tensilelite, but its IUSE targets are gfx942/gfx950 only -- CDNA, not
the gfx11xx/gfx12xx family every recorded vanilla-LLVM failure comes from --
so requiring the flag there would constrain on no evidence.

Verified: `emerge -p` resolves unchanged and pkgcheck is clean.

commit ce70aa7ec3c0f70e0d834d7fa3be48c5cec2ccc4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 19:01:50 2026 +0200

sci-libs/hipBLASLt: assert the remaining sed anchors in src_prepare

Three of the six seds had no anchor assertion, while the origami, stinkytofu
and STINKYTOFU_ENABLE_WERROR ones -- added in the same commit that made
asserting anchors the rule here -- did.

Validators.py's amdclang is the load-bearing one. TensileLite resolves its
toolchain through Tensile/Toolchain/Validators.py, whose ToolchainDefaults
name the compiler and assembler by BARE NAME; this sed is what rewrites that
name to the compiler rocm_use_clang selected.

tensilelite/Makefile's $(ROCM_PATH)/bin/amdclang++ is guarded for
consistency, not because the package build needs it. Nothing in the build
references that Makefile: it sets AS/LDD for a `co:` target globbing
$/1_BenchmarkProblems/.../caches, i.e. the kernel-tuning
workflow, and the build log has no hits for it. Worth keeping correct, worth
not overstating.

COMPONENT tests -> COMPONENT tests EXCLUDE_FROM_ALL is the silent one: the
build succeeds and the test binaries are merged into the image. All 13 sites
in this file use the unquoted spelling, so one anchor covers them.

Verified with `ebuild ... clean prepare`: src_prepare completes and no guard
fires, so all three anchors are present in therock-10.0.

commit 71150c703fdebeb5a576fdab4eb22da52a362fbc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 00:51:13 2026 +0200

sci-libs/hipBLASLt: build 10.0.0 with NDEBUG and without stinkytofu's -Werror

Now builds, against llvm-core/rocm-llvm. Two build-configuration fixes were
needed beyond the compiler swap, and they interact:

1. -DNDEBUG + CMAKE_BUILD_TYPE=Release. cmake.eclass defaults to RelWithDebInfo
AND blanks CMAKE_CXX_FLAGS_RELWITHDEBINFO, so the standard -DNDEBUG never
lands and C++ assert()s stay live. TensileLite aborted partway through
generating assembly kernels on
rocisa/src/pass/macro_inline.cpp:315 "macroToInstruction: unexpected item
type in macro body". This aligns our build with AMD's, where that assert
compiles away -- it does NOT fix the rocisa logic, which walks the same path
in AMD's shipped binaries without aborting. sci-libs/composable-kernel
already carries this idiom.

2. Turn off stinkytofu's -Werror, which hipBLASLt force-enables (overriding
stinkytofu's own default of OFF). Fix 1 exposes it immediately:
StinkyAsmIR.cpp declares `auto it` and uses it ONLY inside an assert(), so
with NDEBUG the variable becomes unused and -Wunused-variable becomes an
error. Two individually defensible upstream choices that are jointly
unbuildable. -Werror is wrong for a distribution build regardless -- any
warning from a compiler upstream did not test becomes a hard failure -- so
this is fixed at the -Werror rather than by silencing the one variable.

The original blocker is gone: zero "operands are not valid for this GPU or
mode" assembler errors this build, against a wall of them under vanilla clang.

commit f0f639f53d6d040fa9000c7042cee79d37382646
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Aug 29 20:17:30 2026 +0200

sci-libs/hipBLASLt: add 10.0.0 (requires AMD's LLVM fork)

Landed masked and not yet buildable, for the same reason as
sci-libs/composable-kernel: the packaging work is finished and worth banking,
but the build cannot complete against a vanilla LLVM 23. llvm-core/rocm-llvm,
later in this series, is what makes it build -- via USE=amd-llvm on
dev-util/hipcc.

Packaging is complete:
- tag retarget; LLVM_COMPAT moved to 23.
- $-7.1.0-no-git.patch dropped: cmake/dependencies.cmake no longer calls
find_package(Git) at all.
- rocisa-nanobind regenerated rather than dropped. 10.0 SPLIT the logic: a
ROCISA_STANDALONE branch that already calls find_package(nanobind REQUIRED
CONFIG), and an else() branch -- the one taken when rocisa builds as a
hipblaslt subdirectory, i.e. ours -- that still does a network FetchContent.
Only the else() branch needs patching.
- NEW at 10.0: tensilelite/rocisa requires "stinkytofu", via find_package or a
monorepo sibling at ../../../../shared/stinkytofu, and hard-FATAL_ERRORs
otherwise. Co-unpacked from its own release asset like origami, with the
sibling lookup redirected. It also needed a patch: stinkytofu declares
option(ENABLE_CLANG_TIDY ... OFF) and then unconditionally includes ClangTidy
from the monorepo's shared cmake/modules, which a standalone tarball build
cannot reach.

WHY A VANILLA LLVM IS NOT ENOUGH: TensileLite generates gfx1150 assembly that
vanilla LLVM's assembler rejects -- "operands are not valid for this GPU or
mode" on v_max_f16 with an abs() source modifier. That is AMD-LLVM-fork-specific
assembly emitted by AMD's kernel generator; correcting it would mean changing
the generator, not the packaging -- so the fork is packaged instead.

Also shortens the SRC_URI lines via MY_BASE.

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 b092be0e5c92ffdb21bcdec72f23c58a29e0bcc7
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 29 14:12:06 2026 +0200

sci-libs/hipBLASLt: add 7.2.4

commit 059a510ab7c58104d38e4460fd94950759a9ab3a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 8 20:18:18 2026 +0200

sci-libs/hipBLASLt: drop stale patches superseded by 7.2.3

commit 9cb07a7e21de52d7e3442d77035678d50e83e1ab
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 10:56:27 2026 +0200

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

The 7.2.3 hipblaslt.tar.gz and origami.tar.gz from ROCm/rocm-libraries both unpack
flat — they share top-level cmake/, CMakeLists.txt, etc., so extracting them into
$ together would collide. Add src_unpack to extract each into its own
hipblaslt/ and origami/ subdir; ORIGAMI_S already points at $/origami.

commit 10ce5e413267259322e5781610f1de784ba396d4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 01:58:15 2026 +0200

sci-libs/hipBLASLt: add 7.2.3

commit 8a7082f1ba65632398134de525d263330415ac96
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Dec 6 21:51:11 2025 +0100

*/*: drop rocm/hip as newer version is in main tree

Signed-off-by: Ivan S. Titov <iohann.s.titov@gmail.com>

commit cfa997ed467a25f0a98e30e69523013fb12b4136
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Oct 18 14:26:09 2025 +0200

added rocm and HIP v. 6.4.4 for now, as main tree lags