update vitis nnet_pooling with some changes from vivado backend

Author: jmitrevs

hls4ml/templates/vitis/nnet_utils/nnet_pooling.h

@@ -70,6 +70,7 @@ struct pooling1d_config {
     static const unsigned n_out = (n_in - pool_width) / stride_width + 1;
     static const unsigned pad_left = 0;
     static const unsigned pad_right = 0;
+    static const bool count_pad = false;
     // Pooling function
     static const Pool_Op pool_op = Max;
 };
@@ -130,6 +131,7 @@ void global_pooling1d_cl(data_T data[CONFIG_T::n_in * CONFIG_T::n_filt], res_T r
 
     for (int ff = 0; ff < CONFIG_T::n_filt; ff++) {
         data_T pool[CONFIG_T::n_in];
+        #pragma HLS ARRAY_PARTITION variable=pool complete dim=0
         for (int jj = 0; jj < CONFIG_T::n_in; jj++) {
             pool[jj] = data[jj * CONFIG_T::n_filt + ff];
         }
@@ -154,6 +156,7 @@ struct pooling2d_config {
     static const unsigned pad_bottom = 0;
     static const unsigned pad_left = 0;
     static const unsigned pad_right = 0;
+    static const bool count_pad = false;
     // Pooling function
     static const Pool_Op pool_op = Max;
     // Reuse factor
@@ -245,6 +248,7 @@ void pooling2d_cf(data_T data[CONFIG_T::in_height * CONFIG_T::in_width * CONFIG_
             // Loop over input image x in steps of stride
             for (int jj = 0; jj < padded_width; jj += CONFIG_T::stride_width) {
                 data_T pool[CONFIG_T::pool_height * CONFIG_T::pool_width];
+                #pragma HLS ARRAY_PARTITION variable=pool complete dim=0
                 // Keep track of number of pixels in image vs padding region
                 unsigned img_overlap = 0;
                 // Loop over pool window y
@@ -255,10 +259,12 @@ void pooling2d_cf(data_T data[CONFIG_T::in_height * CONFIG_T::in_width * CONFIG_
                             jj + ll < CONFIG_T::pad_left || jj + ll >= (padded_width - CONFIG_T::pad_right)) {
                             // Add padding
                             pool[kk * CONFIG_T::stride_width + ll] = pad_val<data_T, CONFIG_T::pool_op>();
+                            if (CONFIG_T::count_pad)
+                                img_overlap++;
                         } else {
                             pool[kk * CONFIG_T::stride_width + ll] =
-                                data[(ii + kk) * CONFIG_T::in_width + ff * CONFIG_T::in_width * CONFIG_T::in_height + ll +
-                                     jj];
+                                data[(ii + kk - CONFIG_T::pad_top) * CONFIG_T::in_width +
+                                     ff * CONFIG_T::in_width * CONFIG_T::in_height + ll + jj - CONFIG_T::pad_left];
                             img_overlap++;
                         }
                     }