finish reduce sum

source/tensor/core/reduce/VectorBuffer.h

+#include <cstring>
+#include <cmath>
+#include "../../XGlobal.h"
+
+
+class VectorBuffer{
+    private:
+        DTYPE values[32 / sizeof(DTYPE)] = {0};
+    public:
+        static int size() {
+            return 32 / sizeof(DTYPE);
+        }
+        VectorBuffer() {}
+        VectorBuffer(DTYPE val) {
+            for (int i = 0; i != size(); i++) {
+              values[i] = val;
+            }
+        }
+        static VectorBuffer loadu(const DTYPE* ptr, bool isExp = false, DTYPE power = (DTYPE)1.0F, DTYPE* bias = NULL) {
+            int count = 32 / sizeof(DTYPE);
+            VectorBuffer vec;
+            if(isExp){
+                if(bias == 0){
+                    if(power == (DTYPE)1.0){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(*(ptr + i));
+                        }
+                    } else if(power == (DTYPE)2.0){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp((*(ptr + i)) * (*(ptr + i)));
+                        }
+                    } else if(power == (DTYPE)0.5){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(std::sqrt(*(ptr + i)));
+                        }
+                    } else{
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(std::pow(*(ptr + i), power));
+                        }
+                    }
+                }//is bias == 0
+                else{
+                    if(power == (DTYPE)1.0){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(*(ptr + i) - bias[i]);
+                        }
+                    } else if(power == (DTYPE)2.0){
+                        for (int i = 0; i != count; i++) {
+                          DTYPE value = *(ptr + i) - bias[i];
+                          vec.values[i] = (DTYPE)std::exp(value * value);
+                        }
+                    } else if(power == (DTYPE)0.5){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(std::sqrt(*(ptr + i) - bias[i]));
+                        }
+                    } else{
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::exp(std::pow(*(ptr + i) - bias[i], power));
+                        }
+                    }
+                }
+            }//isExp
+            else{
+                if(bias == 0){
+                    if(power == (DTYPE)1.0){
+                        std::memcpy(vec.values, ptr, count * sizeof(DTYPE));
+                    } else if(power == (DTYPE)2.0){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (*(ptr + i)) * (*(ptr + i));
+                        }
+                    } else if(power == (DTYPE)0.5){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::sqrt(*(ptr + i));
+                        }
+                    } else{
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::pow(*(ptr + i), power);
+                        }
+                    }
+                }// if bias == 0
+                else{
+                    if(power == (DTYPE)1.0){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = *(ptr + i) - bias[i];
+                        }
+                    } else if(power == (DTYPE)2.0){
+                        for (int i = 0; i != count; i++) {
+                        DTYPE value = *(ptr + i) - bias[i];
+                          vec.values[i] = value * value;
+                        }
+                    } else if(power == (DTYPE)0.5){
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::sqrt(*(ptr + i) - bias[i]);
+                        }
+                    } else{
+                        for (int i = 0; i != count; i++) {
+                          vec.values[i] = (DTYPE)std::pow(*(ptr + i) - bias[i], power);
+                        }
+                    }
+                }
+            }
+            return vec;
+        }
+        const DTYPE& operator[](int idx) const {
+            return values[idx];
+        }
+        VectorBuffer operator+(const VectorBuffer &a) {
+            for (int i = 0; i != a.size(); i++) {
+                this->values[i] = a[i] + this->values[i];
+            }
+            return *this;
+        }
+};
\ No newline at end of file