happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ggml-cpu: change redsum to lmul 4, fix leftover

This commit is contained in:
taimur-10x 2025-12-15 13:29:11 +05:00
parent 2786a97ef0
commit e5c8adbbce
2 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-cpu/vec.cpp
View File

@ -232,11 +232,8 @@ void ggml_vec_dot_bf16(int n, float * GGML_RESTRICT s, size_t bs, ggml_bf16_t *
}
// reduce
vl = __riscv_vsetvlmax_e32m2();
vfloat32m2_t acc0 = __riscv_vfadd_vv_f32m2(__riscv_vget_v_f32m4_f32m2(vsum0, 0), __riscv_vget_v_f32m4_f32m2(vsum0, 1), vl);
vl = __riscv_vsetvlmax_e32m1();
vfloat32m1_t acc1 = __riscv_vfadd_vv_f32m1(__riscv_vget_v_f32m2_f32m1(acc0, 0), __riscv_vget_v_f32m2_f32m1(acc0, 1), vl);
vfloat32m1_t redsum = __riscv_vfredusum_vs_f32m1_f32m1(acc1, __riscv_vfmv_v_f_f32m1(0.0f, 1), vl);
vl = __riscv_vsetvlmax_e32m4();
vfloat32m1_t redsum = __riscv_vfredusum_vs_f32m4_f32m1(vsum0, __riscv_vfmv_v_f_f32m1(0.0f, 1), vl);
sumf += __riscv_vfmv_f_s_f32m1_f32(redsum);
#endif

5
ggml/src/ggml-cpu/vec.h
View File

@ -539,7 +539,7 @@ inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * GGML_RESTRICT y,
// calculate step size
const int epr = __riscv_vsetvlmax_e16m4();
const int step = epr * 2;
const int np = (n & ~(step - 1));
int np = (n & ~(step - 1));
// unroll by 2
for (int i = 0; i < np; i += step) {
@ -560,11 +560,12 @@ inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * GGML_RESTRICT y,
int vl;
for (int i = np; i < n; i += vl) {
vl = __riscv_vsetvl_e16m4(n - i);
vfloat16m4_t ax0 = __riscv_vle16_v_f16m4((const _Float16*)x + i , vl);
vfloat16m4_t ax0 = __riscv_vle16_v_f16m4((const _Float16*)x + i, vl);
vfloat16m4_t ay0 = __riscv_vle16_v_f16m4((const _Float16*)y + i, vl);
ay0 = __riscv_vfmacc_vf_f16m4(ay0, scale, ax0, vl);
__riscv_vse16_v_f16m4((_Float16*)y + i, ay0, vl);
}
np = n;
#elif defined(GGML_SIMD)
const int np = (n & ~(GGML_F16_STEP - 1));