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

hexagon: add SOLVE_TRI op (#21974)

Browse Source 
* hexagon: add SOLVE_TRI op

* ggml: fix TODO description for solve_tri

* hexagon: rm unused variable/function warnings

* hexagon: chunk vs batch processingfor better thread utilization

* hexagon: vectorize partial f32 loads

* hexagon: move HVX f32 add/sub/mul wrappers to hvx-base.h

---------

Co-authored-by: Todor Boinovski <todorb@qti.qualcomm.com>
This commit is contained in:
Mengsheng Wu
2026-04-23 18:39:13 -07:00
committed by GitHub
co-authored by Todor Boinovski
parent e5f070a1dc
commit 8bc492ebb4
7 changed files with 335 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/htp/hvx-base.h
+24
View File
@@ -256,6 +256,18 @@ static inline HVX_Vector hvx_vec_mul_f16_f16(HVX_Vector a, HVX_Vector b)
return Q6_Vhf_equals_Wqf32(Q6_Wqf32_vmpy_VhfVhf(a, b));
}
static inline HVX_Vector hvx_vec_add_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_equals_Vqf32(Q6_Vqf32_vadd_VsfVsf(a, b));
}
static inline HVX_Vector hvx_vec_sub_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_equals_Vqf32(Q6_Vqf32_vsub_VsfVsf(a, b));
}
static inline HVX_Vector hvx_vec_mul_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_equals_Vqf32(Q6_Vqf32_vmpy_VsfVsf(a, b));
}
#else
static inline HVX_Vector hvx_vec_add_f16_f16(HVX_Vector a, HVX_Vector b)
@@ -273,6 +285,18 @@ static inline HVX_Vector hvx_vec_mul_f16_f16(HVX_Vector a, HVX_Vector b)
return Q6_Vhf_vmpy_VhfVhf(a, b);
}
static inline HVX_Vector hvx_vec_add_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_vadd_VsfVsf(a, b);
}
static inline HVX_Vector hvx_vec_sub_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_vsub_VsfVsf(a, b);
}
static inline HVX_Vector hvx_vec_mul_f32_f32(HVX_Vector a, HVX_Vector b) {
return Q6_Vsf_vmpy_VsfVsf(a, b);
}
#endif // __HVX_ARCH__ < 79
#endif /* HVX_BASE_H */