Quartus Clone Optimizer

hls4ml/backends/vivado/passes/clone.py → hls4ml/backends/fpga/passes/clone.py

@@ -3,7 +3,6 @@
 from hls4ml.model.optimizer import OptimizerPass
 
 from hls4ml.model.layers import Layer, register_layer
-from hls4ml.backends import get_backend
 from hls4ml.backends.template import FunctionCallTemplate
 
 class Clone(Layer):

hls4ml/backends/quartus/quartus_backend.py

@@ -1,17 +1,9 @@
 import numpy as np
-import math
 import os
-import copy
-import webbrowser
-from calmjs.parse import es5
-from calmjs.parse import asttypes
-from tabulate import tabulate
-from ast import literal_eval
 from contextlib import contextmanager
-
 from hls4ml.model.types import NamedType, IntegerPrecisionType, FixedPrecisionType
-from hls4ml.model.layers import Embedding, Layer, Dense, BatchNormalization, Activation, ParametrizedActivation, PReLU, Softmax
-from hls4ml.model.optimizer import get_backend_passes, layer_optimizer, model_optimizer
+from hls4ml.model.layers import Layer, Dense, Activation, Softmax, Embedding
+from hls4ml.model.optimizer import get_backend_passes, layer_optimizer
 from hls4ml.model.flow import register_flow
 from hls4ml.backends import FPGABackend
 from hls4ml.report import parse_quartus_report
@@ -34,6 +26,11 @@ def _register_flows(self):
         initializers = self._get_layer_initializers()
         init_flow = register_flow('init_layers', initializers, requires=['optimize'], backend=self.name)
 
+        streaming_passes = [
+            'quartus:clone_output'
+        ]
+        streaming_flow = register_flow('streaming', streaming_passes, requires=[init_flow], backend=self.name)
+
         quartus_types = [
             'quartus:transform_types',
         ]
@@ -46,7 +43,6 @@ def _register_flows(self):
         ]
         quantization_flow = register_flow('quantization', quantization_passes, requires=[init_flow], backend=self.name)
 
-
         templates = self._get_layer_templates()
         template_flow = register_flow('apply_templates', templates, requires=[init_flow], backend=self.name)
 
@@ -69,7 +65,7 @@ def _register_flows(self):
         else:
             extras_flow = None
 
-        ip_flow_requirements = ['optimize', init_flow, quantization_flow, quartus_types_flow, extras_flow, template_flow]
+        ip_flow_requirements = ['optimize', init_flow, streaming_flow, quantization_flow, quartus_types_flow, extras_flow, template_flow]
         ip_flow_requirements = list(filter(None, ip_flow_requirements))
 
         self._default_flow = register_flow('ip', None, requires=ip_flow_requirements, backend=self.name)

hls4ml/templates/quartus/firmware/nnet_utils/nnet_common.h

@@ -37,11 +37,7 @@ typedef ac_fixed<16,6> table_default_t;
 namespace nnet {
 
 // Common type definitions
-<<<<<<< HEAD
 enum io_type {io_parallel = 0, io_stream};
-=======
-enum io_type {io_parallel = 0, io_serial, io_stream};
->>>>>>> ea9eb66a (Quartus Stream Variable Converter & Backend io_stream enabled)
 
 // Default data types (??) TODO: Deprecate
 typedef ac_fixed<16,4>  weight_t_def;

hls4ml/templates/quartus/firmware/nnet_utils/nnet_stream.h

@@ -0,0 +1,36 @@
+#ifndef NNET_STREAM_H
+#define NNET_STREAM_H
+
+namespace nnet {
+
+struct broadcast_config {
+  static const unsigned in_height = 10;
+  static const unsigned in_width = 10;
+  static const unsigned n_chan = 1;
+  static const unsigned n_dupl = 2;
+};
+
+template<class data_T, class res_T, int N>
+void clone_stream(stream<data_T> &data, stream<res_T> &res1, stream<res_T> &res2) {
+    CloneLoop: 
+    #pragma ii 1
+    for (int i = 0; i < N / data_T::size; i++) {
+        data_T in_data = data.read();
+        res_T out_data1;
+        res_T out_data2;
+
+        ClonePack:
+        #pragma unroll
+        for (int j = 0; j < data_T::size; j++) {
+            out_data1[j] = in_data[j];
+            out_data2[j] = in_data[j];
+        }
+
+        res1.write(out_data1);
+        res2.write(out_data2);
+    }
+}
+
+}
+
+#endif