fix conv1d io_parallel resource (#403)

* fix conv1d io_parallel resource

* replace tabs with spaces in modified code

* Add test for Conv1D example model

Co-authored-by: Sioni Summers <[email protected]>

Author: jmitrevs

hls4ml/templates/vivado/nnet_utils/nnet_conv1d_resource.h

@@ -41,7 +41,7 @@ void conv_1d_full(
     data_T data_conv[CONFIG_T::filt_width * CONFIG_T::n_chan * CONFIG_T::out_width];
     data_T data_col[CONFIG_T::filt_width * CONFIG_T::n_chan];
     res_T res_col[CONFIG_T::n_filt];
-    
+
     //#pragma HLS ARRAY_PARTITION variable=data_conv complete
     #pragma HLS ARRAY_PARTITION variable=data_col complete
     #pragma HLS ARRAY_PARTITION variable=res_col complete
@@ -142,17 +142,17 @@ void conv_1d_resource_cf(
 template<class data_T, typename CONFIG_T>
 void im2col_1d_cl(data_T data[CONFIG_T::in_width * CONFIG_T::n_chan], data_T data_col[CONFIG_T::filt_width * CONFIG_T::n_chan], const int col) {
     int index = 0;
-    ChannelLoop:
-    for (int channel = CONFIG_T::n_chan; channel--; data++) {
-		#pragma HLS UNROLL
-        KernelLoop:
-        for (int kernel_col = 0; kernel_col < CONFIG_T::filt_width; kernel_col++) {
-            int input_col = -CONFIG_T::pad_left + kernel_col * CONFIG_T::dilation + col * CONFIG_T::stride_width;
-            if (input_col >= 0 && input_col < CONFIG_T::in_width) {
-                //*(data_col++) = data[input_col * CONFIG_T::n_chan];
-                data_col[index] = data[input_col * CONFIG_T::n_chan];
+    KernelLoop:
+    for (int kernel_col = 0; kernel_col < CONFIG_T::filt_width; kernel_col++) {
+        #pragma HLS UNROLL
+
+        ChannelLoop:
+        for (int channel = 0; channel < CONFIG_T::n_chan; channel++) {
+            int index_data = (col*CONFIG_T::stride_width+kernel_col-CONFIG_T::pad_left) * CONFIG_T::n_chan + channel;
+
+            if (index_data >= 0 && index_data < CONFIG_T::in_width*CONFIG_T::n_chan) {
+                data_col[index] = data[index_data];
             } else {
-                //*(data_col++) = 0;
                 data_col[index] = 0;
             }
             index++;
@@ -191,7 +191,6 @@ void conv_1d_resource_cl(
         dense_resource<data_T, res_T, typename CONFIG_T::mult_config>(data_col, res_col, weights, biases);
         for (int j = 0; j < CONFIG_T::n_filt; j++) {
             res[i * CONFIG_T::n_filt + j] = res_col[j];
-            //res[j * CONFIG_T::out_width + i] = res_col[j]; // Transposed order
         }
     }
 }

test/pytest/test_conv1d.py

@@ -0,0 +1,64 @@
+from hls4ml.converters.keras_to_hls import keras_to_hls
+import pytest
+import hls4ml
+import numpy as np
+from sklearn.metrics import accuracy_score
+import tensorflow as tf
+from tensorflow.keras.models import model_from_json
+import yaml
+
+@pytest.fixture(scope='module')
+def data():
+    X = np.random.rand(100,100,7)
+    return X
+
+@pytest.fixture(scope='module')
+def keras_model():
+    jsons = open('../../example-models/keras/KERAS_conv1d.json','r').read()
+    model = model_from_json(jsons)
+    model.load_weights('../../example-models/keras/KERAS_conv1d_weights.h5')
+    return model
+
+@pytest.fixture      
+@pytest.mark.parametrize('settings', [('io_parallel', 'latency'),
+                                      ('io_parallel', 'resource'),
+                                      ('io_stream', 'latency'),
+                                      ('io_stream', 'resource')])
+def hls_model(settings):
+    io_type = settings[0]
+    strategy = settings[1]
+    config = hls4ml.converters.create_config(output_dir = 'hls4mlprj_conv1d_{}_{}'.format(io_type, strategy))
+    config['KerasJson'] = '../../example-models/keras/KERAS_conv1d.json'
+    config['KerasH5'] = '../../example-models/keras/KERAS_conv1d_weights.h5'
+    config['OutputDir'] = 'hls4mlprj_conv1d_{}_{}'.format(io_type, strategy)
+    config['IOType'] = io_type
+    
+    hls_config = {'Model' : {'Strategy' : strategy,
+                             'ReuseFactor' : 1,
+                             'Precision' : 'ap_fixed<16,3,AP_RND_CONV,AP_SAT>'}}
+    # Some model specific precision tuning
+    config['LayerName'] = {}
+    config['LayerName']['fc1_relu'] = {'Precision':{'weight' : 'ap_fixed<16,3>', 'result' : 'ap_fixed<16,6,AP_RND_CONV,AP_SAT>'}}
+    config['LayerName']['output_softmax'] = {'Precision':{'weight' : 'ap_fixed<16,6>', 'result' : 'ap_fixed<16,6,AP_RND_CONV,AP_SAT>'}}
+    config['LayerName']['output_softmax_softmax'] = {'Strategy':'Stable'}
+    config['HLSConfig'] = hls_config
+    hls_model = keras_to_hls(config)
+    hls_model.compile()
+    return hls_model
+
+@pytest.mark.parametrize('settings', [('io_parallel', 'latency'),
+                                      ('io_parallel', 'resource'),
+                                      ('io_stream', 'latency'),
+                                      ('io_stream', 'resource')])
+def test_accuracy(data, keras_model, hls_model):
+    X = data
+    model = keras_model
+    # model under test predictions and accuracy
+    y_keras = model.predict(X)
+    y_hls4ml   = hls_model.predict(X)
+    # "accuracy" of hls4ml predictions vs keras
+    rel_acc = accuracy_score(np.argmax(y_keras, axis=1), np.argmax(y_hls4ml, axis=1))
+
+    print('hls4ml accuracy relative to keras: {}'.format(rel_acc))
+
+    assert rel_acc > 0.98