Skip to content

ggml-cpu: optimize q4_0_q8_0 scales using Zvfhmin #19196

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ixgbe wants to merge 1 commit into
base: master
Choose a base branch
from
+27 −16
Open

ggml-cpu: optimize q4_0_q8_0 scales using Zvfhmin #19196

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 27 additions & 16 deletions ggml/src/ggml-cpu/arch/riscv/repack.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,21 +90,26 @@ void ggml_gemv_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
const vint32m1_t sumi_h8 = __riscv_vwadd_vv_i32m1(sumi_h8_0, sumi_h8_1, vl / 4);
const vfloat32m1_t facc = __riscv_vfcvt_f_x_v_f32m1(sumi_h8, vl / 4);

// vector version needs Zvfhmin extension
const float a_scale = GGML_CPU_FP16_TO_FP32(a_ptr[l].d);
const float b_scales[8] = {
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[0]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[1]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[2]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[3]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[4]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[5]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[6]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[7])
};
const vfloat32m1_t b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scale, vl / 4);
sumf = __riscv_vfmacc_vv_f32m1(sumf, tmp1, b_scales_vec, vl / 4);
vfloat32m1_t b_scales_vec;
#if defined(__riscv_zvfhmin)
vfloat16mf2_t b_scales_f16 = __riscv_vle16_v_f16mf2((const _Float16*)b_ptr[l].d, vl / 4);
b_scales_vec = __riscv_vfwcvt_f_f_v_f32m1(b_scales_f16, vl / 4);
#else
const float b_scales[8] = {
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[0]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[1]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[2]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[3]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[4]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[5]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[6]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[7])
};
b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
#endif
const vfloat32m1_t tmp1 = __riscv_vfmul_vf_f32m1(facc, a_scale, vl / 4);
sumf = __riscv_vfmacc_vv_f32m1(sumf, tmp1, b_scales_vec, vl / 4);
}
__riscv_vse32_v_f32m1(s + x * ncols_interleaved, sumf, vl / 4);
}
Expand Down Expand Up @@ -156,13 +161,18 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
const vint8m2_t rhs_vec_hi_0 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 0);
const vint8m2_t rhs_vec_hi_1 = __riscv_vget_v_i8m4_i8m2(rhs_vec_hi, 1);

// vector version needs Zvfhmin extension
const float a_scales[4] = {
GGML_CPU_FP16_TO_FP32(a_ptr[l].d[0]),
GGML_CPU_FP16_TO_FP32(a_ptr[l].d[1]),
GGML_CPU_FP16_TO_FP32(a_ptr[l].d[2]),
GGML_CPU_FP16_TO_FP32(a_ptr[l].d[3])
};

vfloat32m1_t b_scales_vec;
#if defined(__riscv_zvfhmin)
vfloat16mf2_t b_scales_f16 = __riscv_vle16_v_f16mf2((const _Float16*)b_ptr[l].d, vl / 4);
b_scales_vec = __riscv_vfwcvt_f_f_v_f32m1(b_scales_f16, vl / 4);
#else
const float b_scales[8] = {
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[0]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[1]),
Expand All @@ -173,7 +183,8 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[6]),
GGML_CPU_FP16_TO_FP32(b_ptr[l].d[7])
};
const vfloat32m1_t b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
b_scales_vec = __riscv_vle32_v_f32m1(b_scales, vl / 4);
#endif

const int64_t A0 = *(const int64_t *)&a_ptr[l].qs[0];
const int64_t A4 = *(const int64_t *)&a_ptr[l].qs[32];
Expand Down
Loading