]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: e7e6487)
Allow all RDNA2 archs to use sdot4 intrinsic (#8629)
authorJeroen Mostert <redacted>
Tue, 23 Jul 2024 08:50:40 +0000 (10:50 +0200)
committerGitHub <redacted>
Tue, 23 Jul 2024 08:50:40 +0000 (10:50 +0200)
commit46e47417aa4f18c08738afd4d9a3e838e97ca03f
treef8aa3c9afdb77f394d284d9b4c144c17f2eb0309tree
parente7e6487ba06634edf58dfdf9673bad9df41b445acommit | diff
Allow all RDNA2 archs to use sdot4 intrinsic (#8629)

The check gating the use of `__builtin_amdgc_sdot4` specifically checks for gfx1030. This causes a severe perf regression for anything gfx103? that's not gfx1030 and not using `HSA_OVERRIDE_GFX_VERSION` (if you've built ROCm to support it). We already have a generic RDNA2 define, let's use it.
ggml/src/ggml-cuda/common.cuh diff | blob | history
Packaging of ggml-org/llama.cpp