Improved scheduling

See aidge_core!94 (merged)

include/aidge/backend/cpu/operator/GlobalAveragePoolingImpl_forward_kernels.hpp

@@ -49,12 +49,12 @@ void GlobalAveragePoolingImpl_cpu_forward_kernel(
     for (DimSize_t channel = 0; channel < dims[1]; ++channel) {
       const I *filter_start = std::next(
           input, (batch * in_batch_nb_elems) + (channel * in_channel_nb_elems));
-      I sum = 0;
+      I mean = 0;
       for (size_t i = 0; i < in_channel_nb_elems; ++i) {
-        sum += filter_start[i];
+        // Single pass numerically stable mean, using the fmaf
+        mean = fmaf(filter_start[i] - mean, 1.0f/(i+1), mean);
       }
-      output[batch * out_batch_nb_elems + channel] =
-          sum / static_cast<I>(in_channel_nb_elems);
+      output[batch * out_batch_nb_elems + channel] = mean;
     }
   }
 }