EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

Vulkan: Fix mmq int dot float cache size (#12722)

Browse Source
This commit is contained in:
0cc4m
2025-04-02 19:12:30 +02:00
committed by GitHub
parent 833e2b7409
commit 92e3006bb6
2 changed files with 4 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp
View File

@@ -17,7 +17,7 @@ i32vec2 repack(uint ib, uint iqs) {
}
ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) {
return ACC_TYPE(da * (float(q_sum) * dsb.x - 8.0 * dsb.y));
return ACC_TYPE(da * (float(q_sum) * dsb.x - 8.0f * dsb.y));
}
#endif
@@ -51,7 +51,7 @@ i32vec2 repack(uint ib, uint iqs) {
}
ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) {
return ACC_TYPE(da * (float(q_sum) * dsb.x - 16.0 * dsb.y));
return ACC_TYPE(da * (float(q_sum) * dsb.x - 16.0f * dsb.y));
}
#endif