vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 17:14:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	Makefile
#	tests/test-grad0.cpp
#	tests/test-quantize-perf.cpp
This commit is contained in:
Concedo 2023-12-13 14:49:03 +08:00
commit c2c238b4f3
20 changed files with 44 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-quants.c
View file

@ -3116,7 +3116,7 @@ void ggml_vec_dot_q5_0_q8_0(const int n, float * restrict s, const void * restri
size_t vl = __riscv_vsetvl_e8m1(qk/2);
// These tempory registers are for masking and shift operations
// These temporary registers are for masking and shift operations
vuint32m2_t vt_1 = __riscv_vid_v_u32m2(vl);
vuint32m2_t vt_2 = __riscv_vsll_vv_u32m2(__riscv_vmv_v_x_u32m2(1, vl), vt_1, vl);
@ -4759,7 +4759,7 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
vl = 16;
// retreive lane to multiply with scale
// retrieve lane to multiply with scale
vint32m2_t aux0_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a0, 0), (scale[0]), vl);
vint32m2_t aux0_1 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a0, 1), (scale[1]), vl);
vint32m2_t aux1_0 = __riscv_vwmul_vx_i32m2(__riscv_vget_v_i16m2_i16m1(a1, 0), (scale[2]), vl);