[ubuntu/resolute-proposed] pytorch-rocm 2.9.1+dfsg-1~exp2ubuntu2 (Accepted)

[Talha Can Havadar]

pytorch-rocm (2.9.1+dfsg-1~exp2ubuntu2) resolute; urgency=medium

  [ Mario Limonciello (AMD) ]
  * d/control.rocm: Add librocsolver-dev to b-d
  * Add a patch to search for HIP in multiarch search paths as well

  [ Talha Can Havadar ]
  * Fix ROCm build see the changes below for details (LP: #2144078):
    - d/p/0002-add-algorithm-in-header.patch: fix the build failure
    - d/p/1070-add-missing-algorithm-include-for-std-for_each.patch
    - d/p/0003-fix-typo-in-gloo_hip-library-name.patch
    - d/control.rocm: add librocm-smi-dev and librocm-core-dev as b-d
    - d/rules: turn off composable kernel features for rocm
      composable-kernel not available in debian yet so turning
      it off until then.
    - d/rules: removing not used USE_MIOPEN for rocm
    - d/control.rocm: add missing hipblaslt and hipsolver-fortran into b-d
    - d/rules: explicitly state enablement of MPI for ROCm
    - d/rules: fix missing HIPConfig.h error during build for rocm
    - d/control.rocm: add missing libroctx-dev b-d
    - d/rules: disabling fbgemm_genai due to missing ck_extensions
    - d/control.rocm: add clang to b-d as hipconfig uses it for builds
    - d/rules: add HIP_DEVICE_LIB_PATH to hint location for device libs
    - d/p/2330-gloo-skip-half-bfloat16-without-torch-dtypes.patch
    - d/rules: disable KINETO and FLASH ATTN due to aotrition lack of support for rocm7.1
    - d/rules: skip gfx1100 as it fails to build due to long branch size

Date: Thu, 09 Apr 2026 21:26:44 +0200
Changed-By: Talha Can Havadar <talha.can.havadar at canonical.com>
Maintainer: Ubuntu Developers <ubuntu-devel-discuss at lists.ubuntu.com>
Signed-By: Ken VanDine <ken.vandine at canonical.com>
https://launchpad.net/ubuntu/+source/pytorch-rocm/2.9.1+dfsg-1~exp2ubuntu2
-------------- next part --------------
Format: 1.8
Date: Thu, 09 Apr 2026 21:26:44 +0200
Source: pytorch-rocm
Built-For-Profiles: derivative.ubuntu noudeb
Architecture: source
Version: 2.9.1+dfsg-1~exp2ubuntu2
Distribution: resolute
Urgency: medium
Maintainer: Ubuntu Developers <ubuntu-devel-discuss at lists.ubuntu.com>
Changed-By: Talha Can Havadar <talha.can.havadar at canonical.com>
Launchpad-Bugs-Fixed: 2144078
Changes:
 pytorch-rocm (2.9.1+dfsg-1~exp2ubuntu2) resolute; urgency=medium
 .
   [ Mario Limonciello (AMD) ]
   * d/control.rocm: Add librocsolver-dev to b-d
   * Add a patch to search for HIP in multiarch search paths as well
 .
   [ Talha Can Havadar ]
   * Fix ROCm build see the changes below for details (LP: #2144078):
     - d/p/0002-add-algorithm-in-header.patch: fix the build failure
     - d/p/1070-add-missing-algorithm-include-for-std-for_each.patch
     - d/p/0003-fix-typo-in-gloo_hip-library-name.patch
     - d/control.rocm: add librocm-smi-dev and librocm-core-dev as b-d
     - d/rules: turn off composable kernel features for rocm
       composable-kernel not available in debian yet so turning
       it off until then.
     - d/rules: removing not used USE_MIOPEN for rocm
     - d/control.rocm: add missing hipblaslt and hipsolver-fortran into b-d
     - d/rules: explicitly state enablement of MPI for ROCm
     - d/rules: fix missing HIPConfig.h error during build for rocm
     - d/control.rocm: add missing libroctx-dev b-d
     - d/rules: disabling fbgemm_genai due to missing ck_extensions
     - d/control.rocm: add clang to b-d as hipconfig uses it for builds
     - d/rules: add HIP_DEVICE_LIB_PATH to hint location for device libs
     - d/p/2330-gloo-skip-half-bfloat16-without-torch-dtypes.patch
     - d/rules: disable KINETO and FLASH ATTN due to aotrition lack of support for rocm7.1
     - d/rules: skip gfx1100 as it fails to build due to long branch size
Checksums-Sha1:
 bfc16b3abb471ed3e76c6fa76f819088b84f4b68 3692 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.dsc
 5166b5943990ef8d1084cb36b047017062cdf865 214556 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.debian.tar.xz
 409ef6aa55f8e1b03f9aee9e76807827061c71ba 17064 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2_source.buildinfo
Checksums-Sha256:
 cfda11d83413f5e5e38c28c91b61f83fef11c472771d3c34b12f49733f1c087c 3692 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.dsc
 b7946511dd35034331f54ffdc1754ee53fb4985bb4c5e4b221ec8dd23802b48c 214556 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.debian.tar.xz
 463abac9d70c07686eae1163d08538fe2bddd482b9d0c5f7c1036d6e1a0fabb5 17064 pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2_source.buildinfo
Files:
 9a8b106daaa9973e50368e7010c1e9e8 3692 science optional pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.dsc
 2e50a3268c59df5b8f02f304d2d6e853 214556 science optional pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2.debian.tar.xz
 53f1d40a6bc6a841fd0e61f1f3e5337d 17064 science optional pytorch-rocm_2.9.1+dfsg-1~exp2ubuntu2_source.buildinfo
Original-Maintainer: Debian Deep Learning Team <debian-ai at lists.debian.org>
Vcs-Git: https://git.launchpad.net/~ken-vandine/ubuntu/+source/pytorch-rocm
Vcs-Git-Commit: c814f128d35e018e3aada4de3badb0b82b1d7cb5
Vcs-Git-Ref: refs/heads/fix-rocm-build