Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: fix mmq out of bounds reads (#17108)

Browse Source 
* vulkan: fix mmq out of bounds reads, streamline outdated matmul host code

* fix mul_mat_id quantization call

* Fix compiler warnings
This commit is contained in:
Ruben Ortlam
2025-11-09 09:52:57 +01:00
committed by GitHub
parent 80a6cf6347
commit 8a3519b708
5 changed files with 113 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp
+1 -1
View File
@@ -61,7 +61,7 @@ void quantize() {
const uint a_idx = ib * 8 + iqs;
vec4 vals = a_idx < p.ne ? data_a[a_idx] : vec4(0.0f);
vec4 vals = a_idx < p.ne / 4 ? data_a[a_idx] : vec4(0.0f);
const vec4 abs_vals = abs(vals);
// Find absolute max for each block