AVX gradients (#2878)

* AVX gradients

* Add google test for AVX

* Create fallback implementation, remove fma instruction

* Improved accuracy of AVX exp function

src/common/math.h

@@ -11,6 +11,7 @@
 #include <vector>
 #include <cmath>
 #include <algorithm>
+#include "avx_helpers.h"
 
 namespace xgboost {
 namespace common {
@@ -23,6 +24,10 @@ inline float Sigmoid(float x) {
   return 1.0f / (1.0f + std::exp(-x));
 }
 
+inline avx::Float8 Sigmoid(avx::Float8 x) {
+  return avx::Sigmoid(x);
+}
+
 /*!
  * \brief do inplace softmax transformaton on p_rec
  * \param p_rec the input/output vector of the values.