]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
sycl : Add DMMV ESIMD Q3_K kernel (llama/26251)
authorTodd Malsbary <redacted>
Thu, 13 Aug 2026 08:32:36 +0000 (01:32 -0700)
committerGeorgi Gerganov <redacted>
Fri, 14 Aug 2026 19:16:06 +0000 (22:16 +0300)
commit90d4ed185961b76ea411148c34a7653fc431752a
tree84a20b328bf3d3d4b286c407b5d80a230bacbd13
parentf7011606ad0d84205104556eb2b34de0d6a73364
sycl : Add DMMV ESIMD Q3_K kernel (llama/26251)

* Add DMMV Q4_K and Q6_K ESIMD kernels

Configure cmake build with -DGGML_SYCL_ESIMD=ON to enable.

Signed-off-by: Todd Malsbary <redacted>
* Refactor ESIMD kernels to share common code

Signed-off-by: Todd Malsbary <redacted>
* Move control of ESIMD from compile to runtime

Signed-off-by: Todd Malsbary <redacted>
* Use ESIMD by default when available

Signed-off-by: Todd Malsbary <redacted>
* Fix possible error when using ESIMD by default

While not an issue in the current version, this will become an
issue when additional QK ESIMD kernels are added (such as Q2_K).

Signed-off-by: Todd Malsbary <redacted>
* Add explicit unroll to ESIMD kernels

Signed-off-by: Todd Malsbary <redacted>
* Tidy up ESIMD kernels a bit

Signed-off-by: Todd Malsbary <redacted>
* Add DMMV Q3_K ESIMD kernel

Signed-off-by: Todd Malsbary <redacted>
---------

Signed-off-by: Todd Malsbary <redacted>
ggml/src/ggml-sycl/common.hpp
ggml/src/ggml-sycl/dmmv.cpp
ggml/src/ggml-sycl/esimd.hpp [new file with mode: 0644]
ggml/src/ggml-sycl/ggml-sycl.cpp