Stefan Walcz 9b0b7da96e fix(rocm): point *_TENSILE_LIBPATH at the architecture folder (#12411) ...

Newer ROCm releases split the rocBLAS and hipBLASLt TensileLibrary data
into one folder per GPU architecture (lib/hipblaslt/library/gfx1151/...).
The run.sh of the ROCm-capable backends exports ROCBLAS_TENSILE_LIBPATH and
HIPBLASLT_TENSILE_LIBPATH as the parent folder, and both libraries only look
directly in that folder, so they miss every kernel:

  rocblaslt error: Cannot read ".../hipblaslt/library/TensileLibrary_lazy_gfx1151.dat"
  hipModuleLoad failed: .../hipblaslt/library/Kernels.so-000-gfx1151.hsaco

and fall back to slower code paths.

rocm_tensile_dir keeps the folder when it has files at the top level (the
older flat layout) or several architecture folders, and descends into the
architecture folder when the bundle carries exactly one.

Assisted-by: Claude:claude-opus-5-5

Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de>

2026-10-05 01:38:09 +02:00

..

patches

chore: ⬆️ Update TheTom/llama-cpp-turboquant to a3d5603d110bda29222d2011596cdc84d7fa532d (#12232)

2026-09-27 21:05:49 +02:00

apply-patches.sh

feat(backend): add turboquant llama.cpp-fork backend (#9355)

2026-04-15 01:25:04 +02:00

Makefile

chore: ⬆️ Update TheTom/llama-cpp-turboquant to bcb85fc3ae85efa0f5f392c6c880dfc524923860 (#12344)

2026-09-30 16:53:22 +02:00

package.sh

refactor(backends): extract package-system-libs.sh from 31 package.sh (#11095)

2026-07-30 16:30:49 +02:00

patch-grpc-server.sh

chore: ⬆️ Update ggml-org/llama.cpp to f280b26983ad0fdb705a0d9ebf0503e76f2899b0 (#11646)

2026-08-25 12:57:12 +02:00

run.sh

fix(rocm): point *_TENSILE_LIBPATH at the architecture folder (#12411)

2026-10-05 01:38:09 +02:00