Packaging changes

MANIFEST.in

@@ -3,4 +3,4 @@ include README.md
 graft example-prjs
 graft example-models
 graft test
-recursive-include hls4ml/hls-templates *
+recursive-include hls4ml/templates *

example-models/keras-config.yml

@@ -1,7 +1,7 @@
-KerasJson: example-keras-model-files/KERAS_3layer.json
-KerasH5:   example-keras-model-files/KERAS_3layer_weights.h5
-#InputData: example-keras-model-files/KERAS_3layer_input_features.dat
-#OutputPredictions: example-keras-model-files/KERAS_3layer_predictions.dat
+KerasJson: keras/KERAS_3layer.json
+KerasH5:   keras/KERAS_3layer_weights.h5
+#InputData: keras/KERAS_3layer_input_features.dat
+#OutputPredictions: keras/KERAS_3layer_predictions.dat
 OutputDir: my-hls-test
 ProjectName: myproject
 XilinxPart: xcku115-flvb2104-2-i

example-models/onnx-config.yml

@@ -1,9 +1,18 @@
-OnnxModel: example-onnx-model-files/three_layer_keras.onnx
+OnnxModel: onnx/three_layer_keras.onnx
+#InputData: keras/KERAS_3layer_input_features.dat
+#OutputPredictions: keras/KERAS_3layer_predictions.dat
 OutputDir: my-hls-test
 ProjectName: myproject
 XilinxPart: xcku115-flvb2104-2-i
 ClockPeriod: 5
 
 IOType: io_parallel # options: io_serial/io_parallel
-ReuseFactor: 1
-DefaultPrecision: ap_fixed<16,6> 
+HLSConfig:
+  Model:
+    Precision: ap_fixed<16,6>
+    ReuseFactor: 1
+#  LayerType:
+#    Dense:
+#      ReuseFactor: 2
+#      Strategy: Resource
+#      Compression: True

example-models/pytorch-config.yml

@@ -1,10 +1,18 @@
-#PytorchModel: example-models/two_layer_model.pt
-PytorchModel: example-models/three_layer_model.pt
-OutputDir: my-hls-dir-3L
+PytorchModel: pytorch/three_layer_model.pt
+#InputData: keras/KERAS_3layer_input_features.dat
+#OutputPredictions: keras/KERAS_3layer_predictions.dat
+OutputDir: my-hls-test
 ProjectName: myproject
-XilinxPart: xc7vx690tffg1927-2
+XilinxPart: xcku115-flvb2104-2-i
 ClockPeriod: 5
 
 IOType: io_parallel # options: io_serial/io_parallel
-ReuseFactor: 1
-DefaultPrecision: ap_fixed<18,8> 
+HLSConfig:
+  Model:
+    Precision: ap_fixed<16,6>
+    ReuseFactor: 1
+#  LayerType:
+#    Dense:
+#      ReuseFactor: 2
+#      Strategy: Resource
+#      Compression: True 

hls4ml/converters/keras_to_hls.py

@@ -1,23 +1,13 @@
 from __future__ import print_function
 import numpy as np
 import h5py
-import os
-import tarfile
 import json
-import argparse
-import yaml
-import sys
-from shutil import copyfile
 import math
 
-MAXMULT = 4096
+from hls4ml.model import HLSModel
+from hls4ml.model.optimizer import optimize_model
 
-filedir = os.path.dirname(os.path.abspath(__file__))
-sys.path.insert(0,os.path.join(filedir, "..", "hls-writer"))
-from hls_writer import parse_config, write_hls
-sys.path.insert(0,os.path.join(filedir, "..", "hls-writer/optimizer"))
-from optimizer import optimize_model
-from hls_model import HLSModel
+MAXMULT = 4096
 
 class KerasDataReader:
     def __init__(self, config):
@@ -49,42 +39,11 @@ def h5_visitor_func(name):
 
     return shape
 
-############################################################################################
-## M A I N
-############################################################################################
-def main():
-
-    # Parse command line arguments
-    parser = argparse.ArgumentParser(description='')
-    parser.add_argument("-c", action='store', dest='config',
-                        help="Configuration file.")
-    args = parser.parse_args()
-    if not args.config: parser.error('A configuration file needs to be specified.')
-
-    configDir  = os.path.abspath(os.path.dirname(args.config))
-    yamlConfig = parse_config(args.config)
-    if not os.path.isabs(yamlConfig['OutputDir']):
-        yamlConfig['OutputDir'] = os.path.join(configDir, yamlConfig['OutputDir'])
-    if not os.path.isabs(yamlConfig['KerasH5']):
-        yamlConfig['KerasH5'] = os.path.join(configDir, yamlConfig['KerasH5'])
-    if not os.path.isabs(yamlConfig['KerasJson']):
-        yamlConfig['KerasJson'] = os.path.join(configDir, yamlConfig['KerasJson'])
-    if 'InputData' in yamlConfig and not os.path.isabs(yamlConfig['InputData']):
-        yamlConfig['InputData'] = os.path.join(configDir, yamlConfig['InputData'])
-    if 'OutputPredictions' in yamlConfig and not os.path.isabs(yamlConfig['OutputPredictions']):
-        yamlConfig['OutputPredictions'] = os.path.join(configDir, yamlConfig['OutputPredictions'])
-
-    if not (yamlConfig["IOType"] == "io_parallel" or yamlConfig["IOType"] == "io_serial"):
-        raise Exception('ERROR: Invalid IO type')
-    return yamlConfig
-
-def keras_to_hls_model(yamlConfig):
+def keras_to_hls(yamlConfig):
 
     ######################
     ##  Do translation
     ######################
-    if not os.path.isdir("{}/firmware/weights".format(yamlConfig['OutputDir'])):
-        os.makedirs("{}/firmware/weights".format(yamlConfig['OutputDir']))
 
     #This is a list of dictionaries to hold all the layer info we need to generate HLS
     layer_list = []
@@ -376,14 +335,8 @@ def keras_to_hls_model(yamlConfig):
     #################
 
     reader = KerasDataReader(yamlConfig)
+    print('Creating HLS model')
     hls_model = HLSModel(yamlConfig, reader, layer_list, input_layers, output_layers)
     optimizers = ['eliminate_linear_activation', 'merge_batch_norm_quantized_tanh', 'quantize_dense_output']
     optimize_model(hls_model, optimizers)
-    #write_hls(hls_model)
     return hls_model
-
-
-if __name__ == "__main__":
-    yamlConfig = main()
-    hls_model = keras_to_hls_model(yamlConfig)
-    write_hls(hls_model)

hls4ml/converters/onnx_to_hls.py

@@ -1,23 +1,14 @@
 from __future__ import print_function
 import numpy as np
-import h5py
-import os
-import tarfile
-import json
-import argparse
-import yaml
-import sys
-from shutil import copyfile
 import math
 from onnx import ModelProto, GraphProto, NodeProto, TensorProto
 from onnx import optimizer, helper, numpy_helper, shape_inference
 
-MAXMULT = 4096
+from hls4ml.writer.vivado_writer import write_hls
+from hls4ml.model import HLSModel
+from hls4ml.model.optimizer import optimize_model
 
-filedir = os.path.dirname(os.path.abspath(__file__))
-sys.path.insert(0,os.path.join(filedir, "..", "hls-writer"))
-from hls_writer import parse_config, write_hls
-from hls_model import HLSModel
+MAXMULT = 4096
 
 class ONNXDataReader:
     def __init__(self, model):
@@ -126,33 +117,11 @@ def compute_pads_2d(operation, layer):
 
     return pads
 
-############################################################################################
-## M A I N
-############################################################################################
-def main():
-
-    # Parse command line arguments
-    parser = argparse.ArgumentParser(description='')
-    parser.add_argument("-c", action='store', dest='config',
-                        help="Configuration file.")
-    args = parser.parse_args()
-    if not args.config: parser.error('A configuration file needs to be specified.')
-
-    configDir  = os.path.abspath(os.path.dirname(args.config))
-    yamlConfig = parse_config(args.config)
-    if not os.path.isabs(yamlConfig['OutputDir']):
-        yamlConfig['OutputDir'] = os.path.join(configDir, yamlConfig['OutputDir'])
-    if not os.path.isabs(yamlConfig['OnnxModel']):
-        yamlConfig['OnnxModel'] = os.path.join(configDir, yamlConfig['OnnxModel'])
-
-    if not (yamlConfig["IOType"] == "io_parallel" or yamlConfig["IOType"] == "io_serial"): 
-        raise Exception('ERROR: Invalid IO type')
+def onnx_to_hls(yamlConfig):
 
     ######################
     ##  Do translation
     ######################
-    if not os.path.isdir("{}/firmware/weights".format(yamlConfig['OutputDir'])):
-        os.makedirs("{}/firmware/weights".format(yamlConfig['OutputDir']))
 
     #This is a list of dictionaries to hold all the layer info we need to generate HLS
     layer_list = []
@@ -402,8 +371,8 @@ def main():
     ## Generate HLS
     #################
 
+    print('Creating HLS model')
     hls_model = HLSModel(yamlConfig, reader, layer_list, input_layers, output_layers)
-    write_hls(hls_model)
-
-if __name__ == "__main__":
-    main()
+    optimizers = ['eliminate_linear_activation', 'merge_batch_norm_quantized_tanh', 'quantize_dense_output']
+    optimize_model(hls_model, optimizers)
+    return hls_model

hls4ml/model/hls_model.py

@@ -5,7 +5,7 @@
 from enum import Enum
 from collections import OrderedDict
 
-from templates import get_config_template, get_function_template
+from .templates import get_config_template, get_function_template
 
 class HLSConfig(object):
     def __init__(self, config):

hls4ml/model/optimizer/__init__.py

@@ -0,0 +1,11 @@
+from __future__ import absolute_import
+
+from .optimizer import OptimizerPass, register_pass, get_optimizer, optimize_model
+
+
+from .passes.nop import EliminateLinearActivation
+from .passes.bn_quant import MergeBatchNormAndQuantizedTanh, QuantizeDenseOutput
+
+register_pass('eliminate_linear_activation', EliminateLinearActivation)
+register_pass('merge_batch_norm_quantized_tanh', MergeBatchNormAndQuantizedTanh)
+register_pass('quantize_dense_output', QuantizeDenseOutput)

hls4ml/model/optimizer/optimizer.py

@@ -41,10 +41,3 @@ def optimize_model(model, passes=None):
             break
         else:
             optimization_done = True
-
-from passes.nop import EliminateLinearActivation
-from passes.bn_quant import MergeBatchNormAndQuantizedTanh, QuantizeDenseOutput
-
-register_pass('eliminate_linear_activation', EliminateLinearActivation)
-register_pass('merge_batch_norm_quantized_tanh', MergeBatchNormAndQuantizedTanh)
-register_pass('quantize_dense_output', QuantizeDenseOutput)

hls4ml/model/optimizer/passes/bn_quant.py

@@ -1,12 +1,9 @@
 import numpy as np
-import sys
 import re
 
-sys.path.insert(0, '../')
-from optimizer import OptimizerPass
-sys.path.insert(0, '../..')
-import hls_model
-import templates
+from ..optimizer import OptimizerPass
+import hls4ml.model.hls_model as hls_model
+import hls4ml.model.templates as templates
 
 class BatchNormalizationQuantizedTanh(hls_model.Layer):
     ''' Merged Batch Normalization and quantized (binary or ternary) Tanh layer.

hls4ml/model/optimizer/passes/nop.py

@@ -1,6 +1,4 @@
-import sys
-sys.path.insert(0, '../')
-from optimizer import OptimizerPass
+from ..optimizer import OptimizerPass
 
 class EliminateLinearActivation(OptimizerPass):
     def match(self, node):

hls4ml/templates/vivado/firmware/parameters.h

@@ -4,16 +4,16 @@
 #include <complex>
 #include "ap_int.h"
 #include "ap_fixed.h"
-#include "nnet_dense.h"
-#include "nnet_dense_large.h"
-#include "nnet_dense_compressed.h"
-#include "nnet_conv.h"
-#include "nnet_conv2d.h"
-#include "nnet_activation.h"
-#include "nnet_common.h"
-#include "nnet_batchnorm.h"
-#include "nnet_pooling.h"
-#include "nnet_merge.h"
+#include "nnet_utils/nnet_dense.h"
+#include "nnet_utils/nnet_dense_large.h"
+#include "nnet_utils/nnet_dense_compressed.h"
+#include "nnet_utils/nnet_conv.h"
+#include "nnet_utils/nnet_conv2d.h"
+#include "nnet_utils/nnet_activation.h"
+#include "nnet_utils/nnet_common.h"
+#include "nnet_utils/nnet_batchnorm.h"
+#include "nnet_utils/nnet_pooling.h"
+#include "nnet_utils/nnet_merge.h"
 
 //hls-fpga-machine-learning insert numbers
 

hls4ml/templates/vivado/myproject_test.cpp

@@ -24,7 +24,6 @@
 
 #include "firmware/parameters.h"
 #include "firmware/myproject.h"
-#include "nnet_helpers.h"
 
 #define CHECKPOINT 5000