wehub-resource-sync
47 lines
1.6 KiB
C++
47 lines
1.6 KiB
C++
//
|
|
// MNNAvgPoolInt8.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2026/04/01.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
#include <riscv_vector.h>
|
|
#include <stdint.h>
|
|
#include <sys/types.h>
|
|
#include <limits.h>
|
|
#include "../../compute/Int8FunctionsOpt.h"
|
|
|
|
void MNNAvgPoolInt8_RVV(int8_t* dst, int8_t* src, size_t outputWidth, size_t inputWidth, size_t kernelx, size_t kernely,
|
|
size_t stridesx, ssize_t paddingx, ssize_t factor) {
|
|
const size_t vl = __riscv_vsetvl_e8m1(16);
|
|
const size_t pack = 16;
|
|
|
|
int8_t* dstPtr = dst;
|
|
const int8_t* srcPtr = src;
|
|
|
|
vint32m4_t v_factor = __riscv_vmv_v_x_i32m4(factor, vl);
|
|
|
|
for (size_t ox = 0; ox < outputWidth; ++ox) {
|
|
vint32m4_t vec_sum = __riscv_vmv_v_x_i32m4(0, vl);
|
|
|
|
for (size_t y = 0; y < kernely; ++y) {
|
|
for (size_t x = 0; x < kernelx; ++x) {
|
|
const int8_t* inputPtr = srcPtr + pack * (x + inputWidth * y);
|
|
vint8m1_t vec_input = __riscv_vle8_v_i8m1(inputPtr, vl);
|
|
vint32m4_t vec_input_ext = __riscv_vsext_vf4_i32m4(vec_input, vl);
|
|
vec_sum = __riscv_vadd_vv_i32m4(vec_sum, vec_input_ext, vl);
|
|
}
|
|
}
|
|
|
|
vint32m4_t v_mul = __riscv_vmul_vv_i32m4(vec_sum, v_factor, vl);
|
|
vint32m4_t v_shr = __riscv_vsra_vx_i32m4(v_mul, 24, vl);
|
|
vint16m2_t v_temp = __riscv_vncvt_x_x_w_i16m2(v_shr, vl);
|
|
vint8m1_t v_result = __riscv_vncvt_x_x_w_i8m1(v_temp, vl);
|
|
|
|
__riscv_vse8_v_i8m1(dstPtr, v_result, vl);
|
|
|
|
dstPtr += pack;
|
|
srcPtr += pack * stridesx;
|
|
}
|
|
}
|