diff --git a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
index 6685f3c17b6a2299480c6fa24118db854330eda3..519d3466ee8ac2274b851c39fe6f254ade777f48 100644
--- a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
@@ -1,8 +1,9 @@
import os
import subprocess
-import numpy as np
import tempfile as tmp
+import numpy as np
+
import finn.core.utils as utils
from finn.backend.fpgadataflow.utils import numpy_to_hls_code
from finn.core.datatype import DataType
@@ -22,11 +23,22 @@ class StreamingFCLayer_Batch(HLSCustomOp):
pass
def execute_node(self, node, context, graph):
+ # make temporary directory for generated files
self.tmp_dir = tmp.mkdtemp(prefix=str(node.op_type) + "_")
+
+ # create empty list for temporary files to enable the option
+ # to delete the files after the execution
+ temp_files = []
+
+ # get attributes for correct packing of weights and thresholds
self.get_attributes(node)
+
+ # create a npy file fore each input of the node (in_ind is input index)
in_ind = 0
- temp_files = []
for inputs in node.input:
+ # it is assumed that the first input of the node is the data input
+ # the second input are the weights
+ # the third input are the thresholds
if in_ind == 0:
np.save(
os.path.join(self.tmp_dir, "input_{}.npy".format(in_ind)),
@@ -36,12 +48,15 @@ class StreamingFCLayer_Batch(HLSCustomOp):
elif in_ind == 1:
weights = context[inputs]
self.WMEM = weights.shape[2]
+ # transpose and expand the weights to get the right shape
+ # for the code generation
weights = np.transpose(weights, (1, 2, 0))
weights = np.expand_dims(weights, 0)
weights = numpy_to_hls_code(
weights, DataType.BINARY, "weights", True, True
)
+ # write weights into params.h
f_weights = open("{}/params.h".format(self.tmp_dir), "w")
f_weights.write(
"static BinaryWeights<{},{},{}> weights = ".format(
@@ -55,11 +70,15 @@ class StreamingFCLayer_Batch(HLSCustomOp):
else:
thresholds = context[inputs]
self.TMEM = thresholds.shape[0]
+ # transpose and expand the thresholds to get the right shape
+ # for the code generation
thresholds = np.transpose(thresholds, (1, 0, 2))
thresholds = np.expand_dims(thresholds, 0)
thresholds = numpy_to_hls_code(
thresholds, DataType.BINARY, "thresholds", True, True
)
+
+ # write weights into thresh.h
f_thresh = open("{}/thresh.h".format(self.tmp_dir), "w")
f_thresh.write(
"""static ThresholdsActivation<{},{},1,ap_uint<16>,
@@ -73,7 +92,10 @@ class StreamingFCLayer_Batch(HLSCustomOp):
in_ind += 1
+ # code generation
self.code_generation(node)
+
+ # c++ compilation and execution flow
temp_files.append("{}/execute_{}.cpp".format(self.tmp_dir, node.op_type))
bash_compile = """g++ -o {}/execute_{} {}/execute_{}.cpp
/workspace/cnpy/cnpy.cpp -I/workspace/cnpy/
@@ -88,6 +110,8 @@ class StreamingFCLayer_Batch(HLSCustomOp):
process_execute.communicate()
temp_files.append("{}/execute_{}".format(self.tmp_dir, node.op_type))
temp_files.append("{}/output.npy".format(self.tmp_dir))
+
+ # load output npy file
output = np.load("{}/output.npy".format(self.tmp_dir))
context[node.output[0]] = output
# deleting temporary files
@@ -121,6 +145,8 @@ class StreamingFCLayer_Batch(HLSCustomOp):
]
def read_npy_data(self, node):
+ # c++ code to read out an npy file
+ # and put it in hls::stream in the correct order
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
"""cnpy::NpyArray arr0 = cnpy::npy_load("{}/input_0.npy");\n
diff --git a/src/finn/custom_op/fpgadataflow/streamingmaxpool.py b/src/finn/custom_op/fpgadataflow/streamingmaxpool.py
index 3387f15b4bb54da131998fc2507579d400681b8d..e137951eaa63e9ec50c4cbf173fa4bf040d70a07 100644
--- a/src/finn/custom_op/fpgadataflow/streamingmaxpool.py
+++ b/src/finn/custom_op/fpgadataflow/streamingmaxpool.py
@@ -1,8 +1,8 @@
import os
-import numpy as np
import subprocess
import tempfile as tmp
+import numpy as np
from finn.core.utils import get_by_name
from finn.custom_op.fpgadataflow import HLSCustomOp
@@ -16,22 +16,34 @@ class StreamingMaxPool(HLSCustomOp):
pass
def execute_node(self, node, context, graph):
+ # make temporary directory for generated files
self.tmp_dir = tmp.mkdtemp(prefix=str(node.op_type) + "_")
- in_ind = 0
+
+ # create empty list for temporary files to enable the option
+ # to delete the files after the execution
temp_files = []
+
+ # create a npy file fore each input of the node (in_ind is input index)
+ in_ind = 0
for inputs in node.input:
- np.save(os.path.join(self.tmp_dir, "input_{}.npy".format(in_ind)),
- context[inputs],)
+ np.save(
+ os.path.join(self.tmp_dir, "input_{}.npy".format(in_ind)),
+ context[inputs],
+ )
temp_files.append("{}/input_{}.npy".format(self.tmp_dir, in_ind))
in_ind += 1
+
+ # code generation
self.code_generation(node)
+
+ # c++ compilation and execution flow
temp_files.append("{}/execute_{}.cpp".format(self.tmp_dir, node.op_type))
bash_compile = """g++ -o {}/execute_{} {}/execute_{}.cpp
/workspace/cnpy/cnpy.cpp -I/workspace/cnpy/
-I/workspace/finn-hlslib -I/workspace/vivado-hlslib
--std=c++11 -lz""".format(
self.tmp_dir, node.op_type, self.tmp_dir, node.op_type
- )
+ )
process_compile = subprocess.Popen(bash_compile.split(), stdout=subprocess.PIPE)
process_compile.communicate()
bash_execute = "{}/execute_{}".format(self.tmp_dir, node.op_type)
@@ -39,13 +51,14 @@ class StreamingMaxPool(HLSCustomOp):
process_execute.communicate()
temp_files.append("{}/execute_{}".format(self.tmp_dir, node.op_type))
temp_files.append("{}/output.npy".format(self.tmp_dir))
- output = np.load("{}/output.npy".format(self.tmp_dir))
+
+ # load output npy file
+ output = np.load("{}/output.npy".format(self.tmp_dir))
context[node.output[0]] = output
# deleting temporary files
- #for temp_file in temp_files:
+ # for temp_file in temp_files:
# os.remove(temp_file)
-
def get_attributes(self, node):
self.ImgDim = get_by_name(node.attribute, "ImgDim").i
self.PoolDim = get_by_name(node.attribute, "PoolDim").i
@@ -62,6 +75,8 @@ class StreamingMaxPool(HLSCustomOp):
]
def read_npy_data(self, node):
+ # c++ code to read out an npy file
+ # and put it in hls::stream in the correct order
self.code_gen_dict["$READNPYDATA$"] = []
input_ind = 0
input_file_names = []
diff --git a/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py b/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
index a89c4a07ec22f27b7b2cb3855105621e896cb650..c9b5b0b37726173c6a35179ec9107fb11fcdd8a3 100644
--- a/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
@@ -1,8 +1,8 @@
import os
-import numpy as np
import subprocess
import tempfile as tmp
+import numpy as np
from finn.core.utils import get_by_name
from finn.custom_op.fpgadataflow import HLSCustomOp
@@ -14,24 +14,36 @@ class StreamingMaxPool_Batch(HLSCustomOp):
def infer_node_datatype(self, node, model):
pass
-
+
def execute_node(self, node, context, graph):
+ # make temporary directory for generated files
self.tmp_dir = tmp.mkdtemp(prefix=str(node.op_type) + "_")
- in_ind = 0
+
+ # create empty list for temporary files to enable the option
+ # to delete the files after the execution
temp_files = []
+
+ # create a npy file fore each input of the node (in_ind is input index)
+ in_ind = 0
for inputs in node.input:
- np.save(os.path.join(self.tmp_dir, "input_{}.npy".format(in_ind)),
- context[inputs],)
+ np.save(
+ os.path.join(self.tmp_dir, "input_{}.npy".format(in_ind)),
+ context[inputs],
+ )
temp_files.append("{}/input_{}.npy".format(self.tmp_dir, in_ind))
in_ind += 1
+
+ # code generation
self.code_generation(node)
+
+ # c++ compilation and execution flow
temp_files.append("{}/execute_{}.cpp".format(self.tmp_dir, node.op_type))
bash_compile = """g++ -o {}/execute_{} {}/execute_{}.cpp
/workspace/cnpy/cnpy.cpp -I/workspace/cnpy/
-I/workspace/finn-hlslib -I/workspace/vivado-hlslib
--std=c++11 -lz""".format(
self.tmp_dir, node.op_type, self.tmp_dir, node.op_type
- )
+ )
process_compile = subprocess.Popen(bash_compile.split(), stdout=subprocess.PIPE)
process_compile.communicate()
bash_execute = "{}/execute_{}".format(self.tmp_dir, node.op_type)
@@ -39,10 +51,12 @@ class StreamingMaxPool_Batch(HLSCustomOp):
process_execute.communicate()
temp_files.append("{}/execute_{}".format(self.tmp_dir, node.op_type))
temp_files.append("{}/output.npy".format(self.tmp_dir))
- output = np.load("{}/output.npy".format(self.tmp_dir))
+
+ # load output npy file
+ output = np.load("{}/output.npy".format(self.tmp_dir))
context[node.output[0]] = output
# deleting temporary files
- #for temp_file in temp_files:
+ # for temp_file in temp_files:
# os.remove(temp_file)
def get_attributes(self, node):
@@ -63,6 +77,8 @@ class StreamingMaxPool_Batch(HLSCustomOp):
]
def read_npy_data(self, node):
+ # c++ code to read out an npy file
+ # and put it in hls::stream in the correct order
self.code_gen_dict["$READNPYDATA$"] = []
input_ind = 0
input_file_names = []
@@ -153,7 +169,7 @@ class StreamingMaxPool_Batch(HLSCustomOp):
self.code_gen_dict["$SAVEASCNPY$"] = [
"""cnpy::npy_save("{}/output.npy",&output_data_vector[0],
{{{},{},{}}},"w");""".format(
- self.tmp_dir,
+ self.tmp_dir,
numReps,
self.NumChannels,
int(self.ImgDim / self.PoolDim),