rocwmma (7.2.4-1~exp1) experimental; urgency=medium

  * New upstream version 7.2.4
  * d/{ci/*,salsa-ci.yml}: update ci configuration for debian experimental
  * d/p/0001-remove-unknown-compiler-options.patch: some compiler options not
    available in upstream llvm-21
  * debian/control: update for initial package build
    - update build-depends
    - update build-depends-indep for documentation build with rocm-docs
  * debian/rules: use hipcc to build with hardened flags
  * debian/watch: add watch file
  * d{bin/run-tests, librocwmma-tests.install}: add librocwmma-tests files
  * d/librocwmma-tests.install: add files to be installed
  * d/librocwmma-dev.install: add development files to be installed
  * d/{librocwmma-docs.docs, librocwmma-docs.doc-base}: add documentation
    configuration files
  * debian/not-installed: do not install internal headers and Test CMake files
  * d/p/0002-do-not-treat-doxygen-warnings-as-errors.patch: to fix missing docs
    issue
  * d/rules: add version information to docs
  * d/copyright: update copyright to include debian files
  * d/upstream/metadata: update for rocWMMA
  * d/tests/control: add autopkgtest for rocwmma
  * d/rules: use filtered GPU_TARGETS instead of rocm-target-arch
    not all GPU_TARGETS are supported by upstream rocWMMA project we need to
    use `library/include/rocwmma/internal/config.hpp` as reference for
    supported gpu architectures
  * d/p/0003-fix-arm64-mcmodel-large-failure.patch: no mcmodel for arm64
  * d/rules: limit the number of jobs based on available memory
  * d/control: build librocwmma-tests for amd64 and arm64 only
  * d/rules: turn off build samples and turn on offload compress
  * d/{control, librocwmma-tests-validate.install}: split librocwmma-tests
  * d/gbp.conf: create initial config
  * d/control: bump ROCm dependencies to minimum 7.2.0
    hipcc needs to be bumped to 7.2.4 due to hipconfig change
  * d/control: bump Standards-Version to 4.7.4
  * d/{control, rules}: add XB-X-ROCm-GPU-Architecture to the packages that has
    binaries
  * d/{control, rules}: bump llvm dependencies to 22 and update paths
  * d/{librocwmma-dev.install,not-installed,rules}: update for new cmake files
  * d/{control, rules}: use lld from llvm-22
    default ld has 16-bit section header cap and with all the architectures
    we have enabled in rocwmma we are hitting that limit. Opted for lld
    since it doesn't have the cap.

 -- Talha Can Havadar <havadartalha@gmail.com>  Mon, 08 Jun 2026 08:57:05 +0200
