]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
Test support for alternative conv layout (#25617)
authorpmaybank <redacted>
Thu, 30 Jul 2026 17:14:16 +0000 (18:14 +0100)
committerGitHub <redacted>
Thu, 30 Jul 2026 17:14:16 +0000 (01:14 +0800)
commit958d9c0b61d0025ca8e2fff32d66a1576173fde9
treec3a53dcdb2328f1f179aacbc3ccdd1bf9fcc8ee3
parent432d7ffe2c3b4e539f3d0d4ae0a4893090a018d6
Test support for alternative conv layout (#25617)

* add  bool cwhn = true to conv_2d test cases

* add layout check at graph building time

* extend layout checks for conv2d.cu kernel

* in CPU back-end kernel needs to be stored contiguously to prevent test failures with cwhn=1

* trim white space

* do op support check in vulkan backend

* fix CI failure and vulkan run-time assert failure by introducing new graph build-time check in ggml_backend_vk_device_supports_op

* add additional check in support_op function for Vulkan to fix run-time assert failure
ggml/src/ggml-cpu/ggml-cpu.cpp
ggml/src/ggml-cuda/conv2d.cu
ggml/src/ggml-cuda/ggml-cuda.cu
ggml/src/ggml-vulkan/ggml-vulkan.cpp
tests/test-backend-ops.cpp