Always call cnn_load_bias() even when not using bias (#124)

* Improve error message when running out of bias memory

Author: Robert Muchsel

README.md

@@ -1,6 +1,6 @@
 # MAX78000 Model Training and Synthesis
 
-_April 8, 2021_
+_April 20, 2021_
 
 The Maxim Integrated AI project is comprised of four repositories:
 
@@ -257,6 +257,10 @@ By default, the main branch is checked out. This branch has been tested more rig
 (ai8x-training) $ git checkout develop
 ```
 
+###### TensorFlow / Keras
+
+Support for TensorFlow / Keras is currently in the `develop-tf` branch.
+
 ##### Updates
 
 After additional testing, `develop` is merged into the main branch at regular intervals.
@@ -728,6 +732,8 @@ The MAX78000 hardware does not support arbitrary network parameters. Specificall
 
 * The number of output channels must not exceed 1024 per layer.
 
+  * Bias is supported for up to 512 output channels per layer.
+
 * The number of layers must not exceed 32 (where pooling and element-wise operations do not add to the count when preceding a convolution).
 
 * The maximum dimension (number of rows or columns) for input or output data is 1023.
@@ -817,7 +823,7 @@ The following table describes the most important command line arguments for `tra
 | `--pr-curves`              | Generate precision-recall curves                             |                                 |
 | `--embedding`              | Display embedding (using projector)                          |                                 |
 | *Hardware*                 |                                                              |                                 |
-| `--use-bias`               | The `bias=True` parameter is passed to the model. The effect of this parameter is model dependent (the parameter is either ignored, effective for some operations, or all operations). |                                 |
+| `--use-bias`               | The `bias=True` parameter is passed to the model. The effect of this parameter is model dependent (the parameter does nothing, effects some operations, or all operations). |                                 |
 | `--avg-pool-rounding`      | Use rounding for AvgPool                                     |                                 |
 | *Evaluation*               |                                                              |                                 |
 | `-e`, `--evaluate`         | Evaluate previously trained model                            |                                 |

README.pdf

@@ -196,8 +196,9 @@ def bias_sort(e):
 
         if group_bias_max[group] + blen > tc.dev.BIAS_SIZE:
             eprint(f'Layer {ll}: bias memory capacity exceeded - available groups: '
-                   f'{gmap}, used so far: {group_bias_max}, needed: {blen}, '
-                   f'best available: group {group}.')
+                   f'{gmap}, used so far: {group_bias_max}, needed: {blen} bytes, '
+                   f'best available: group {group} with '
+                   f'{tc.dev.BIAS_SIZE - group_bias_max[group]} bytes available.')
         bias_group[ll] = group
         for i in range(tc.dev.P_NUMGROUPS):
             bias_offs[ll][i] = group_bias_max[group]

izer/kbias.py

@@ -600,7 +600,7 @@ def main(
             if bias:
                 memfile.write('  cnn_load_bias();\n')
             else:
-                memfile.write('  // cnn_load_bias(); // Not used in this network\n')
+                memfile.write('  cnn_load_bias(); // Not used in this network\n')
             memfile.write('  cnn_configure(); // Configure state machine\n')
             if not measure_energy:
                 if not fifo: