Quartus test for Softmax

fastmachinelearning · Apr 21, 2022 · 0b7eb38 · 0b7eb38
1 parent df206d2
commit 0b7eb38
Showing 1 changed file with 22 additions and 14 deletions.
diff --git a/test/pytest/test_softmax.py b/test/pytest/test_softmax.py
@@ -25,31 +25,39 @@ def generate_data(function, input_shape):
 
 
 # TODO: include latency strategy with flat_distribution when it can be made to pass
-@pytest.mark.parametrize('strategy,function,input_shape,io_type', [#('latency', flat_distribution, (8,), 'io_parallel'),
-                                                                   #('latency', flat_distribution, (8, 8, 3), 'io_stream'),
-                                                                   ('stable', flat_distribution, (8,), 'io_parallel'),
-                                                                   ('stable', high_accuracy_distribution, (8,), 'io_parallel'),
-                                                                   ('stable', flat_distribution, (8,), 'io_stream'),
-                                                                   ('stable', high_accuracy_distribution, (8,), 'io_stream'),
-                                                                   # Multi-dimensional tests, only for io_stream for now
-                                                                   ('stable', flat_distribution, (8, 8, 3), 'io_stream'),
-                                                                   ('stable', high_accuracy_distribution, (8, 8, 3), 'io_stream')])
-def test_softmax(strategy, generate_data, input_shape, io_type):
+@pytest.mark.parametrize('backend,strategy,function,input_shape,io_type', [   
+                            #('latency', flat_distribution, (8,), 'io_parallel'),
+                            #('latency', flat_distribution, (8, 8, 3), 'io_stream'),
+                            ('Vivado', 'stable', flat_distribution, (8,), 'io_parallel'),
+                            ('Vivado', 'stable', high_accuracy_distribution, (8,), 'io_parallel'),
+                            ('Quartus', 'resource', flat_distribution, (8,), 'io_parallel'),
+                            ('Quartus', 'resource', high_accuracy_distribution, (8,), 'io_parallel'),
+                            ('Vivado', 'stable', flat_distribution, (8,), 'io_stream'),
+                            ('Vivado', 'stable', high_accuracy_distribution, (8,), 'io_stream'),
+                            # Multi-dimensional tests, only for io_stream for now
+                            ('Vivado', 'stable', flat_distribution, (8, 8, 3), 'io_stream'),
+                            ('Vivado', 'stable', high_accuracy_distribution, (8, 8, 3), 'io_stream')
+                        
+                        ])
+def test_softmax(backend, strategy, generate_data, input_shape, io_type):
     X = generate_data
     model = tf.keras.models.Sequential()
     model.add(tf.keras.layers.Activation(input_shape=input_shape, activation='softmax', name='softmax'))
     model.compile()
+
+    f_type = 'ac_fixed<18,8,true,AC_RND,AC_SAT>' if backend == 'Quartus' else 'ap_fixed<18,8,AP_RND,AP_SAT>'
     cfg = hls4ml.utils.config_from_keras_model(model, granularity='name')
     cfg['LayerName']['softmax']['Strategy'] = strategy
-    cfg['LayerName']['softmax']['inv_table_t'] = 'ap_fixed<18,8,AP_RND,AP_SAT>'
-    cfg['LayerName']['softmax']['exp_table_t'] = 'ap_fixed<18,8,AP_RND,AP_SAT>'
+    cfg['LayerName']['softmax']['inv_table_t'] = f_type
+    cfg['LayerName']['softmax']['exp_table_t'] = f_type
+
     odir = str(test_root_path / 'hls4mlprj_softmax_{}'.format(strategy))
     hls_model = hls4ml.converters.convert_from_keras_model(model, hls_config=cfg, io_type=io_type,
-                                                           output_dir=odir)
+                                                           output_dir=odir, backend=backend)
     hls_model.compile()
+
     y_keras = model.predict(X)
     y_hls4ml = hls_model.predict(X).reshape(y_keras.shape)
-
     acc_hls4ml = accuracy_score(np.argmax(y_keras, axis=-1).ravel(), np.argmax(y_hls4ml, axis=-1).ravel())
 
     print('Accuracy hls4ml relative to keras: {}'.format(acc_hls4ml))