vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ggml: copy q->f32 assumes some contiguity in the destination

Browse source
This commit is contained in:
Jeff Bolz 2025-01-11 08:57:32 -06:00
parent 305dc66649
commit 0a982a414e
1 changed files with 5 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml/src/ggml-cpu/ggml-cpu.c
View file

@ -3982,6 +3982,11 @@ static void ggml_compute_forward_dup_q(
size_t qk = ggml_blck_size(type); size_t qk = ggml_blck_size(type);
const int64_t nr = ggml_nelements(src1) / qk; const int64_t nr = ggml_nelements(src1) / qk;
// destination must be contiguous in the first dimension
GGML_ASSERT(nb10 == ggml_type_size(dst->type));
// must either have first dimension large enough to hold a row, or fully contiguous
GGML_ASSERT((ne10 % qk) == 0 || ggml_is_contiguous(dst));
const int ith = params->ith; const int ith = params->ith;
const int nth = params->nth; const int nth = params->nth;