.. _program_listing_file_src_tensors_cpu_integer_common.cpp:

Program Listing for File integer_common.cpp
===========================================

|exhale_lsh| :ref:`Return to documentation for file <file_src_tensors_cpu_integer_common.cpp>` (``src/tensors/cpu/integer_common.cpp``)

.. |exhale_lsh| unicode:: U+021B0 .. UPWARDS ARROW WITH TIP LEFTWARDS

.. code-block:: cpp

   #include "integer_common.h"
   
   namespace marian {
   namespace cpu {
   namespace integer {
   // This operates on floats after processing so doesn't care about int8_t vs int16_t.
   void AddBias(marian::Tensor C, const marian::Tensor Bias) {
     float* y = C->data();
     const float* x = C->data();
     const float* bias = Bias->data();
   
     const int m = C->shape().elements() / C->shape()[-1];
     const int n = C->shape()[-1];
   
     for(int j = 0; j < m; ++j) {
       int i = 0;
   #ifdef __AVX512F__
       int n16 = n & ~15;
       for(; i < n16; i += 16) {
         __m512 ai = _mm512_loadu_ps(x + j * n + i);
         __m512 bi = _mm512_loadu_ps(bias + i);
         __m512 yi = _mm512_add_ps(ai, bi);
         _mm512_storeu_ps(y + j * n + i, yi);
       }
   #else
       int n4 = (n / 4) * 4;
       for(; i < n4; i += 4) {
         __m128 ai = _mm_loadu_ps(x + j * n + i);
         __m128 bi = _mm_loadu_ps(bias + i);
         __m128 yi = _mm_add_ps(ai, bi);
         _mm_storeu_ps(y + j * n + i, yi);
       }
   #endif
       for(; i < n; i++) {
         y[j * n + i] = x[j * n + i] + bias[i];
       }
     }
   }
   
   //template void prepareAndTranspose<intgemm8>;//(io::Item& item, const char * input);
   //template void prepareAndTranspose<intgemm16>(io::Item&, const char *);
   
   } //integer
   } //cpu
   } //marian