diff options
author | Kawrakow <iwankawrakow@gmail.com> | 2025-05-17 08:57:26 +0300 |
---|---|---|
committer | GitHub <noreply@github.com> | 2025-05-17 08:57:26 +0300 |
commit | 7abdf2b099ecf9bea156a635a8f22d168483f2b1 (patch) | |
tree | 98a642e178695ba49f0a31a4a4b2700b28cdf69b /ggml/src/ggml-common.h | |
parent | 134d5481737c05421eb1ba7cd7573136e3fdbd69 (diff) |
IQ5_KS_R4: row-interleaved IQ5_KS (#426)
* iq5_ks_r4: basics
* iq5_ks_r4: Zen4 works
* iq5_ks_r4: AVX2 works
* iq5_ks_r4: NEON
* Fix iq5_ks on NEON
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml-common.h')
-rw-r--r-- | ggml/src/ggml-common.h | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ggml/src/ggml-common.h b/ggml/src/ggml-common.h index 1c2d1b17..26041ac2 100644 --- a/ggml/src/ggml-common.h +++ b/ggml/src/ggml-common.h @@ -694,6 +694,13 @@ typedef struct { } block_iq5_ks; static_assert(sizeof(block_iq5_ks) == QK_K/32 + QK_K/2 + QK_K/8, "wrong iq5_ks block size/padding"); +typedef struct { + uint8_t scales[QK_K/8]; + uint8_t qs[QK_K*2]; + uint8_t qh[QK_K/2]; +} block_iq5_ks_r4; +static_assert(sizeof(block_iq5_ks_r4) == 4*sizeof(block_iq5_ks), "wrong iq5_ks_r4 block size/padding"); + #endif // GGML_COMMON_DECL #endif // GGML_COMMON_DECL |