happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix wrong memcpy length for block_interleave == 4 (#19575)

This commit is contained in:
Alberto Cabrera Pérez 2026-02-13 12:32:14 +00:00 committed by GitHub
parent 0e21991472
commit cc2aa81513
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-cpu/repack.cpp
View File

@ -1916,9 +1916,10 @@ static block_q4_Kx8 make_block_q4_Kx8(block_q4_K * in, unsigned int blck_size_in
int src_offset = (i / 8) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;
// buffer large enough for the max interleave block size (8 bytes)
uint64_t elems;
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint64_t));
memcpy(&out.qs[dst_offset], &elems, sizeof(uint64_t));
memcpy(&elems, &in[src_id].qs[src_offset], blck_size_interleave);
memcpy(&out.qs[dst_offset], &elems, blck_size_interleave);
}
// The below logic is designed so as to unpack and rearrange scales and mins values in Q4_K