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

ggml : Format gemm rvv code

Browse source
This commit is contained in:
Xiongchuan Tan 2024-10-25 16:32:13 +08:00
parent 78c78e2af2
commit 37057a0fcf
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml/src/ggml-aarch64.c
View file

@ -3371,10 +3371,10 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
sumi_l2 = sumi_hi_m;
}
{
const vuint32m4_t sumi_i32 = __riscv_vreinterpret_v_i32m4_u32m4(__riscv_vreinterpret_v_i16m4_i32m4(sumi_l2));
const vuint16m2_t sumi_h2_0 = __riscv_vnsrl_wx_u16m2(sumi_i32, 0, vl);
@ -3409,7 +3409,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
const vint16m4_t sumi_lo_1 = __riscv_vwmacc_vv_i16m4(sumi_lo_0, rhs_vec_lo_1, lhs_1_8, vl * 2);
const vint16m4_t sumi_hi_0 = __riscv_vwmacc_vv_i16m4(sumi_lo_1, rhs_vec_hi_0, lhs_2_8, vl * 2);
const vint16m4_t sumi_hi_m = __riscv_vwmacc_vv_i16m4(sumi_hi_0, rhs_vec_hi_1, lhs_3_8, vl * 2);
sumi_l3 = sumi_hi_m;
}