From 4ade4c568c331acad22537f7b9519c740c7a06d0 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Tue, 17 Dec 2024 10:18:33 +0100 Subject: IQ2_K_R4 (#146) * iq2_k_r4: Zen4 * iq2_k_r4: NEON * iq2_k_r4: better matrix x vector multiplication on NEON --------- Co-authored-by: Iwan Kawrakow --- ggml/src/ggml-common.h | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'ggml/src/ggml-common.h') diff --git a/ggml/src/ggml-common.h b/ggml/src/ggml-common.h index ca56704c..03cc3460 100644 --- a/ggml/src/ggml-common.h +++ b/ggml/src/ggml-common.h @@ -521,6 +521,14 @@ typedef struct { } block_iq2_k; static_assert(sizeof(block_iq2_k) == sizeof(ggml_half) + sizeof(uint16_t) + QK_K/32 + QK_K/4, "wrong iq2_k block size/padding"); +typedef struct { + ggml_half d[4]; + uint8_t extra[8]; + uint8_t scales[QK_K/8]; + uint8_t qs[QK_K]; +} block_iq2_k_r4; +static_assert(sizeof(block_iq2_k_r4) == 4*sizeof(block_iq2_k), "wrong iq2_k_r4 block size/padding"); + typedef struct { uint16_t extra; uint8_t scales[QK_K/64]; -- cgit v1.2.3