nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 18 Packages Projects Releases Wiki Activity

ggml-cpu:add RISC-V RVV (Zvfh) optimization for FP16 vector scaling (#17314)

Browse Source 
* ggml-cpu:add RISC-V RVV (Zvfh) optimization for FP16 vector scaling

Signed-off-by: Wang Yang <yangwang@iscas.ac.cn>

* fix comment

* fix comment 2

---------

Signed-off-by: Wang Yang <yangwang@iscas.ac.cn>
This commit is contained in:
ixgbe
2025-11-20 08:09:18 +02:00
committed by GitHub
parent 7d77f07325
commit 5be353ec4a
1 changed files with 50 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/vec.h
+10 -9
View File
@@ -698,8 +698,7 @@ inline static void ggml_vec_scale_f32(const int n, float * y, const float v) {
} }
inline static void ggml_vec_scale_f16(const int n, ggml_fp16_t * y, const float v) { inline static void ggml_vec_scale_f16(const int n, ggml_fp16_t * y, const float v) {
#if defined(GGML_SIMD) #if defined(GGML_SIMD) && defined(__ARM_FEATURE_SVE)
#if defined(__ARM_FEATURE_SVE)
const int sve_register_length = svcntb() * 8; const int sve_register_length = svcntb() * 8;
const int ggml_f16_epr = sve_register_length / 16; const int ggml_f16_epr = sve_register_length / 16;
const int ggml_f16_step = 2 * ggml_f16_epr; const int ggml_f16_step = 2 * ggml_f16_epr;
@@ -725,13 +724,16 @@ inline static void ggml_vec_scale_f16(const int n, ggml_fp16_t * y, const float
svfloat16_t out = svmul_f16_m(pg, hy, vx); svfloat16_t out = svmul_f16_m(pg, hy, vx);
svst1_f16(pg, (__fp16 *)(y + np), out); svst1_f16(pg, (__fp16 *)(y + np), out);
} }
#elif defined(__riscv_v_intrinsic) #elif defined(__riscv_v_intrinsic) && defined(__riscv_zvfh)
// todo: RVV impl for (int i = 0, vl; i < n; i += vl) {
// scalar vl = __riscv_vsetvl_e16m2(n - i);
for (int i = 0; i < n; ++i) { vfloat16m2_t vy = __riscv_vle16_v_f16m2((_Float16 *)&y[i], vl);
y[i] = GGML_CPU_FP32_TO_FP16(GGML_CPU_FP16_TO_FP32(y[i])*v); vfloat32m4_t vy32 = __riscv_vfwcvt_f_f_v_f32m4(vy, vl);
vy32 = __riscv_vfmul_vf_f32m4(vy32, v, vl);
vy = __riscv_vfncvt_f_f_w_f16m2(vy32, vl);
__riscv_vse16_v_f16m2((_Float16 *)&y[i], vy, vl);
} }
#else #elif defined(GGML_SIMD)
const int np = (n & ~(GGML_F16_STEP - 1)); const int np = (n & ~(GGML_F16_STEP - 1));
GGML_F16_VEC vx = GGML_F16_VEC_SET1(v); GGML_F16_VEC vx = GGML_F16_VEC_SET1(v);
@@ -751,7 +753,6 @@ inline static void ggml_vec_scale_f16(const int n, ggml_fp16_t * y, const float
for (int i = np; i < n; ++i) { for (int i = np; i < n; ++i) {
y[i] = GGML_CPU_FP32_TO_FP16(GGML_CPU_FP16_TO_FP32(y[i])*v); y[i] = GGML_CPU_FP32_TO_FP16(GGML_CPU_FP16_TO_FP32(y[i])*v);
} }
#endif
#else #else
// scalar // scalar
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {