diff --git a/requirements.txt b/requirements.txt
index a9e691fea28c0fc8f633a68905dbbe74cdd79a7a..e3f74c23f9701280812d26b9caa720d037bc26ab 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -2,6 +2,7 @@ bitstring==3.1.7
clize==4.1.1
dataclasses-json==0.5.7
gspread==3.6.0
+ipython==8.12.2
numpy==1.24.1
onnx==1.13.0
onnxoptimizer
diff --git a/src/finn/custom_op/fpgadataflow/addstreams_batch.py b/src/finn/custom_op/fpgadataflow/addstreams_batch.py
index af106d9c0698d2d49bbd8f8998f57cad0b2e781e..8fbdf9c452ab356e2623ef7d3f00f077682961e2 100644
--- a/src/finn/custom_op/fpgadataflow/addstreams_batch.py
+++ b/src/finn/custom_op/fpgadataflow/addstreams_batch.py
@@ -268,37 +268,60 @@ class AddStreams_Batch(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
npy_in = "%s/input_1.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in1);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in1_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in1 ("in1");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in1_{} ("in1_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
node = self.onnx_node
self.code_gen_dict["$DOCOMPUTE$"] = [
- """{}<{}, {}, {}, {}, {}> (in0, in1, out, 1);""".format(
+ """{}<{}, {}, {}, {}, {}> (in0_{}, in1_{}, out_{}, 1);""".format(
node.op_type,
self.get_nodeattr("PE"),
self.get_input_datatype().get_hls_datatype_str(),
self.get_input_datatype().get_hls_datatype_str(),
self.get_output_datatype().get_hls_datatype_str(),
self.get_number_output_values(),
+ self.hls_sname(),
+ self.hls_sname(),
+ self.hls_sname(),
)
]
@@ -315,12 +338,13 @@ class AddStreams_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -331,24 +355,27 @@ class AddStreams_Batch(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0, hls::stream<ap_uint<{}>> &in1,
- hls::stream<ap_uint<{}>> &out)""".format(
+ """void {}(hls::stream<ap_uint<{}>> &in0_{}, hls::stream<ap_uint<{}>> &in1_{},
+ hls::stream<ap_uint<{}>> &out_{})""".format(
self.onnx_node.name,
self.get_nodeattr("PE") * self.get_input_datatype().bitwidth(),
+ self.hls_sname(),
self.get_nodeattr("PE") * self.get_input_datatype().bitwidth(),
+ self.hls_sname(),
self.get_nodeattr("PE") * self.get_output_datatype().bitwidth(),
+ self.hls_sname(),
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=in1 name=in1_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in1_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/channelwise_op_batch.py b/src/finn/custom_op/fpgadataflow/channelwise_op_batch.py
index 7791647abf1c5a51c0425ae835471ec9ceb7c21c..71fc37b184a27ac707bc4dfacd860f2c5256c60c 100644
--- a/src/finn/custom_op/fpgadataflow/channelwise_op_batch.py
+++ b/src/finn/custom_op/fpgadataflow/channelwise_op_batch.py
@@ -486,17 +486,28 @@ class ChannelwiseOp_Batch(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
# note: the innermost dim is reversed for the input
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0, false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -512,10 +523,12 @@ class ChannelwiseOp_Batch(HLSCustomOp):
raise Exception("""Unexpeted input shape""")
self.code_gen_dict["$DOCOMPUTE$"] = [
"""Thresholding_Batch<{}, NumChannels1, PE1, {}, {}>
- (in0, out, threshs, numReps);""".format(
+ (in0_{}, out_{}, threshs, numReps);""".format(
spatial_dim,
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
+ self.hls_sname(),
+ self.hls_sname(),
)
]
@@ -536,12 +549,13 @@ class ChannelwiseOp_Batch(HLSCustomOp):
# note: the innermost dim is not reversed for the output
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
shape_cpp_str,
npy_out,
)
@@ -552,21 +566,23 @@ class ChannelwiseOp_Batch(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &out
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/checksum.py b/src/finn/custom_op/fpgadataflow/checksum.py
index 99646274fa1bc5b710b23ea42a25d0fed0da529c..c9d16c0011b65e63b8d7965083e5be9f1c4415c1 100644
--- a/src/finn/custom_op/fpgadataflow/checksum.py
+++ b/src/finn/custom_op/fpgadataflow/checksum.py
@@ -241,17 +241,28 @@ class CheckSum(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
# note: the innermost dim is reversed for the input
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0, false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append("ap_uint<32> chk;")
# set drain = false for cppsim
@@ -259,7 +270,8 @@ class CheckSum(HLSCustomOp):
def docompute(self):
self.code_gen_dict["$DOCOMPUTE$"] = [
- """checksum<WORDS_PER_FRAME, ITEMS_PER_WORD>(in0, out, chk, drain);"""
+ """checksum<WORDS_PER_FRAME, ITEMS_PER_WORD>(in0_%s, out_%s, chk, drain);"""
+ % (self.hls_sname(), self.hls_sname())
]
def dataoutstrm(self):
@@ -279,12 +291,13 @@ class CheckSum(HLSCustomOp):
# note: the innermost dim is not reversed for the output
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
shape_cpp_str,
npy_out,
),
@@ -299,18 +312,18 @@ class CheckSum(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """using T = ap_uint<WORD_SIZE>;\n void {}(hls::stream<T> &in0,
- hls::stream<T> &out, ap_uint<32> &chk, ap_uint<1> &drain)""".format(
- self.onnx_node.name
+ """using T = ap_uint<WORD_SIZE>;\n void {}(hls::stream<T> &in0_{},
+ hls::stream<T> &out_{}, ap_uint<32> &chk, ap_uint<1> &drain)""".format(
+ self.onnx_node.name, self.hls_sname(), self.hls_sname()
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS interface axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS interface axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS interface axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS interface axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS interface s_axilite port=chk bundle=checksum"
diff --git a/src/finn/custom_op/fpgadataflow/concat.py b/src/finn/custom_op/fpgadataflow/concat.py
index 8b655b570d0396e253a1c98231702f816072da20..c43e88d59df4f80765680bd09025198ce3ea9adb 100644
--- a/src/finn/custom_op/fpgadataflow/concat.py
+++ b/src/finn/custom_op/fpgadataflow/concat.py
@@ -278,8 +278,16 @@ class StreamingConcat(HLSCustomOp):
packed_hls_type = "ap_uint<%d>" % packed_bits
npy_in = "%s/input_%d.npy" % (code_gen_dir, i)
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in%d);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in, i)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in%d_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ i,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
@@ -288,21 +296,28 @@ class StreamingConcat(HLSCustomOp):
for i in range(n_inputs):
packed_bits = self.get_instream_width(i)
packed_hls_type = "ap_uint<%d>" % packed_bits
- stream_name = "in%d" % i
+ stream_name = "in%d_%s" % (i, self.hls_sname())
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
'hls::stream<%s> %s ("%s");'
% (packed_hls_type, stream_name, stream_name)
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
self.code_gen_dict["$DOCOMPUTE$"] = []
n_inputs = self.get_n_inputs()
- in_stream_names = ["in%d" % x for x in range(n_inputs)]
- in_stream_names = ",".join(in_stream_names)
- comp_call = "StreamingConcat(%s, out, NumReps);" % (in_stream_names)
+ in_streams = []
+ for i in range(n_inputs):
+ in_streams.append("in%d_%s" % (i, self.hls_sname()))
+ in_stream_names = ",".join(in_streams)
+ comp_call = "StreamingConcat(%s, out_%s, NumReps);" % (
+ in_stream_names,
+ self.hls_sname(),
+ )
self.code_gen_dict["$DOCOMPUTE$"] = [comp_call]
def dataoutstrm(self):
@@ -318,12 +333,13 @@ class StreamingConcat(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -337,10 +353,15 @@ class StreamingConcat(HLSCustomOp):
in_streams = []
for i in range(n_inputs):
iwidth = self.get_instream_width(i)
- in_streams.append("hls::stream<ap_uint<%d>> &in%d" % (iwidth, i))
+ in_streams.append(
+ "hls::stream<ap_uint<%d>> &in%d_%s" % (iwidth, i, self.hls_sname())
+ )
in_streams = ",".join(in_streams)
total_width = self.get_input_datatype().bitwidth() * self.get_total_elems()
- out_stream = "hls::stream<ap_uint<%d>> &out" % (total_width)
+ out_stream = "hls::stream<ap_uint<%d>> &out_%s" % (
+ total_width,
+ self.hls_sname(),
+ )
blackbox_hls = "void %s(%s, %s)" % (self.onnx_node.name, in_streams, out_stream)
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [blackbox_hls]
@@ -349,12 +370,11 @@ class StreamingConcat(HLSCustomOp):
pragmas = []
for i in range(n_inputs):
pragmas.append(
- "#pragma HLS INTERFACE axis port=in%d name=in%d_%s"
- % (i, i, self.hls_sname())
+ "#pragma HLS INTERFACE axis port=in%d_%s" % (i, self.hls_sname())
)
self.code_gen_dict["$PRAGMAS$"] = pragmas
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
index 6cc9208bb81ff68fe941c8d8d006c65b635eb437..c80f79a8c970ce7f55024d1ad7cb57001dc7ec22 100644
--- a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
+++ b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator.py
@@ -401,17 +401,28 @@ class ConvolutionInputGenerator(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -436,15 +447,15 @@ class ConvolutionInputGenerator(HLSCustomOp):
if self.get_nodeattr("depthwise") == 1:
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}_dws<ConvKernelDim1, IFMChannels1, Input_precision1, IFMDim1,
- OFMDim1, SIMD1, Stride1> (in0, out, numReps, {});""".format(
- hls_call, hls_ram_style
+ OFMDim1, SIMD1, Stride1> (in0_{}, out_{}, numReps, {});""".format(
+ hls_call, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
else:
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1, IFMChannels1, Input_precision1, IFMDim1,
- OFMDim1, SIMD1, Stride1> (in0, out, numReps, {});""".format(
- hls_call, hls_ram_style
+ OFMDim1, SIMD1, Stride1> (in0_{}, out_{}, numReps, {});""".format(
+ hls_call, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
@@ -464,12 +475,13 @@ class ConvolutionInputGenerator(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -480,18 +492,18 @@ class ConvolutionInputGenerator(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0,
- hls::stream<ap_uint<SIMD1*Input_precision1>> &out)""".format(
- self.onnx_node.name
+ """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0_{},
+ hls::stream<ap_uint<SIMD1*Input_precision1>> &out_{})""".format(
+ self.onnx_node.name, self.hls_sname(), self.hls_sname()
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator1d.py b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator1d.py
index 6e792ca585718ff9690b0a2430fc09ba46e0a2ba..43e8df17b45c8f7f2c804b1257a01ef717929b12 100644
--- a/src/finn/custom_op/fpgadataflow/convolutioninputgenerator1d.py
+++ b/src/finn/custom_op/fpgadataflow/convolutioninputgenerator1d.py
@@ -601,17 +601,28 @@ class ConvolutionInputGenerator1D(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -630,40 +641,40 @@ class ConvolutionInputGenerator1D(HLSCustomOp):
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1_x, IFMChannels1, Input_precision1,
IFMDim1_x, OFMDim1_x, Stride1_x, SIMD1>
- (in0, out, numReps, {});""".format(
- swu_variant, hls_ram_style
+ (in0_{}, out_{}, numReps, {});""".format(
+ swu_variant, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
if swu_variant == "ConvolutionInputGenerator_1D":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1_x, IFMChannels1, Input_precision1,
IFMDim1_x, OFMDim1_x, Stride1_x, SIMD1>
- (in0, out, numReps, {});""".format(
- swu_variant, hls_ram_style
+ (in0_{}, out_{}, numReps, {});""".format(
+ swu_variant, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
if swu_variant == "ConvolutionInputGenerator_1D_dws":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1_x, IFMChannels1, Input_precision1,
IFMDim1_x, OFMDim1_x, SIMD1>
- (in0, out, numReps, {});""".format(
- swu_variant, hls_ram_style
+ (in0_{}, out_{}, numReps, {});""".format(
+ swu_variant, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
if swu_variant == "ConvolutionInputGenerator_1D_dws_stride":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1_x, IFMChannels1, Input_precision1,
IFMDim1_x, OFMDim1_x, Stride1_x, SIMD1>
- (in0, out, numReps, {});""".format(
- swu_variant, hls_ram_style
+ (in0_{}, out_{}, numReps, {});""".format(
+ swu_variant, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
if swu_variant == "ConvolutionInputGenerator_1D_dws_naive":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ConvKernelDim1_x, IFMChannels1, Input_precision1,
IFMDim1_x, OFMDim1_x, Stride1_x, Dilation1_x, SIMD1>
- (in0, out, numReps, {});""".format(
- swu_variant, hls_ram_style
+ (in0_{}, out_{}, numReps, {});""".format(
+ swu_variant, self.hls_sname(), self.hls_sname(), hls_ram_style
)
]
@@ -690,12 +701,13 @@ class ConvolutionInputGenerator1D(HLSCustomOp):
multi_pixel_out = 1
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", true, 1, %d);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", true, 1, %d);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
multi_pixel_out,
@@ -708,26 +720,26 @@ class ConvolutionInputGenerator1D(HLSCustomOp):
def blackboxfunction(self):
if self.use_parallel_window_output():
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0,
+ """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0_{},
hls::stream<ap_uint<ConvKernelDim1_x*SIMD1*Input_precision1>>
- &out)""".format(
- self.onnx_node.name
+ &out_{})""".format(
+ self.onnx_node.name, self.hls_sname(), self.hls_sname()
)
]
else:
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0,
- hls::stream<ap_uint<SIMD1*Input_precision1>> &out)""".format(
- self.onnx_node.name
+ """void {}(hls::stream<ap_uint<SIMD1*Input_precision1>> &in0_{},
+ hls::stream<ap_uint<SIMD1*Input_precision1>> &out_{})""".format(
+ self.onnx_node.name, self.hls_sname(), self.hls_sname()
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/downsampler.py b/src/finn/custom_op/fpgadataflow/downsampler.py
index 255606ee7f1998586c2b357904bd32b9a5590c96..d42a076c3004290c398da5888c2b178ec9f761f9 100644
--- a/src/finn/custom_op/fpgadataflow/downsampler.py
+++ b/src/finn/custom_op/fpgadataflow/downsampler.py
@@ -212,24 +212,36 @@ class DownSampler(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
dim_var = "1D" if (self.get_nodeattr("is1D") == 1) else "2D"
+ sname = self.hls_sname()
self.code_gen_dict["$DOCOMPUTE$"] = [
f"""ConvolutionInputGenerator_{dim_var}_kernel1<IFMChannels, Input_precision,
- IFMDim, SIMD,Stride> (in0, out, numReps);"""
+ IFMDim, SIMD,Stride> (in0_{sname}, out_{sname}, numReps);"""
]
def dataoutstrm(self):
@@ -248,12 +260,13 @@ class DownSampler(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -266,16 +279,22 @@ class DownSampler(HLSCustomOp):
packed_bits = self.get_instream_width()
packed_hls_type = "ap_uint<%d>" % packed_bits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_hls_type, packed_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type,
+ self.hls_sname(),
+ packed_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/duplicatestreams_batch.py b/src/finn/custom_op/fpgadataflow/duplicatestreams_batch.py
index 312f5e7e4a799d75aa0b9b7cd82b83c1b0e51dd9..0d5d806dc5e6df4bb71c0298c4305d45558ac586 100644
--- a/src/finn/custom_op/fpgadataflow/duplicatestreams_batch.py
+++ b/src/finn/custom_op/fpgadataflow/duplicatestreams_batch.py
@@ -309,18 +309,27 @@ class DuplicateStreams_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
n_outputs = self.get_num_output_streams()
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
for i in range(n_outputs):
- out_name = "out%d" % i
+ out_name = "out%d_%s" % (i, self.hls_sname())
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
'hls::stream<ap_uint<%d>> %s ("%s");'
% (self.get_outstream_width(), out_name, out_name)
@@ -328,8 +337,13 @@ class DuplicateStreams_Batch(HLSCustomOp):
def docompute(self):
n_outputs = self.get_num_output_streams()
- ostreams = ["out%d" % x for x in range(n_outputs)]
- dc = "DuplicateStreamsCustom(in0, %s);" % (",".join(ostreams))
+ ostreams = []
+ for i in range(n_outputs):
+ ostreams.append("out%d_%s" % (i, self.hls_sname()))
+ dc = "DuplicateStreamsCustom(in0_%s, %s);" % (
+ self.hls_sname(),
+ ",".join(ostreams),
+ )
self.code_gen_dict["$DOCOMPUTE$"] = [dc]
def dataoutstrm(self):
@@ -346,7 +360,7 @@ class DuplicateStreams_Batch(HLSCustomOp):
outstrm_code = []
for i in range(n_outputs):
- out_name = "out%d" % i
+ out_name = "out%d_%s" % (i, self.hls_sname())
npy_out = "%s/output%d.npy" % (code_gen_dir, i)
outstrm_code.append(
'apintstream2npy<%s, %s, %d, %s>(%s, %s, "%s");'
@@ -371,10 +385,14 @@ class DuplicateStreams_Batch(HLSCustomOp):
inp_streams = []
o_stream_w = self.get_outstream_width()
i_stream_w = self.get_instream_width()
- in_stream = "hls::stream<ap_uint<%d> > &in0" % (i_stream_w)
+ in_stream = "hls::stream<ap_uint<%d> > &in0_%s" % (i_stream_w, self.hls_sname())
inp_streams.append(in_stream)
for i in range(n_outputs):
- out_stream = "hls::stream<ap_uint<%d> > &out%d" % (o_stream_w, i)
+ out_stream = "hls::stream<ap_uint<%d> > &out%d_%s" % (
+ o_stream_w,
+ i,
+ self.hls_sname(),
+ )
inp_streams.append(out_stream)
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
@@ -387,12 +405,11 @@ class DuplicateStreams_Batch(HLSCustomOp):
def pragmas(self):
n_outputs = self.get_num_output_streams()
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
for i in range(n_outputs):
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out%d name=out%d_%s"
- % (i, i, self.hls_sname())
+ "#pragma HLS INTERFACE axis port=out%d_%s" % (i, self.hls_sname())
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/eltwise.py b/src/finn/custom_op/fpgadataflow/eltwise.py
index c96f12f06bb1104152cecc6f5c6cdf5c0cc215f1..348e3147928631d5d8abc8586641900ad653e20e 100644
--- a/src/finn/custom_op/fpgadataflow/eltwise.py
+++ b/src/finn/custom_op/fpgadataflow/eltwise.py
@@ -354,25 +354,45 @@ class StreamingEltwise(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type_0, elem_hls_type_0, elem_bits_0, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type_0,
+ elem_hls_type_0,
+ elem_bits_0,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
npy_in = "%s/input_1.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in1);'
- % (packed_hls_type_1, elem_hls_type_1, elem_bits_1, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in1_%s);'
+ % (
+ packed_hls_type_1,
+ elem_hls_type_1,
+ elem_bits_1,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width(0))
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(0), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in1 ("in1");'.format(self.get_instream_width(1))
+ 'hls::stream<ap_uint<{}>> in1_{} ("in1_{}");'.format(
+ self.get_instream_width(1), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -394,7 +414,7 @@ class StreamingEltwise(HLSCustomOp):
out_hls_type,
)
self.code_gen_dict["$DOCOMPUTE$"] = [
- """{}<{}, {}, {}, {}, {}, {}>(in0, in1, out, {});""".format(
+ """{}<{}, {}, {}, {}, {}, {}>(in0_{}, in1_{}, out_{}, {});""".format(
"StreamingEltwise",
self.get_nodeattr("NumChannels"),
self.get_nodeattr("PE"),
@@ -402,6 +422,9 @@ class StreamingEltwise(HLSCustomOp):
slice_in0,
slice_in1,
slice_out,
+ self.hls_sname(),
+ self.hls_sname(),
+ self.hls_sname(),
eltwise_op_str,
)
]
@@ -419,12 +442,13 @@ class StreamingEltwise(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -435,24 +459,27 @@ class StreamingEltwise(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0, hls::stream<ap_uint<{}>> &in1,
- hls::stream<ap_uint<{}>> &out)""".format(
+ """void {}(hls::stream<ap_uint<{}>> &in0_{}, hls::stream<ap_uint<{}>> &in1_{},
+ hls::stream<ap_uint<{}>> &out_{})""".format(
self.onnx_node.name,
self.get_nodeattr("PE") * self.get_input_datatype(0).bitwidth(),
+ self.hls_sname(),
self.get_nodeattr("PE") * self.get_input_datatype(1).bitwidth(),
+ self.hls_sname(),
self.get_nodeattr("PE") * self.get_output_datatype().bitwidth(),
+ self.hls_sname(),
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=in1 name=in1_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in1_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/fmpadding_batch.py b/src/finn/custom_op/fpgadataflow/fmpadding_batch.py
index bdb5775c3eea84b09297025501f0116438b09ae7..ea9028d925c82efa24238d9e64a3b4556fb90e02 100644
--- a/src/finn/custom_op/fpgadataflow/fmpadding_batch.py
+++ b/src/finn/custom_op/fpgadataflow/fmpadding_batch.py
@@ -228,17 +228,28 @@ class FMPadding_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -256,8 +267,8 @@ class FMPadding_Batch(HLSCustomOp):
hls_call = node.op_type
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ImgDim1, OutputDim1, PaddingBefore1, PaddingBehind1, NumChannels1, SIMD1,
- {}> (in0, out, numReps);""".format(
- hls_call, in_t
+ {}> (in0_{}, out_{}, numReps);""".format(
+ hls_call, in_t, self.hls_sname(), self.hls_sname()
)
]
else:
@@ -265,8 +276,8 @@ class FMPadding_Batch(HLSCustomOp):
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<OutputDim1_x, OutputDim1_y, PaddingLeft1, PaddingRight1,
PaddingTop1, PaddingBottom1, NumChannels1,
- SIMD1, {}> (in0, out, numReps);""".format(
- hls_call, in_t
+ SIMD1, {}> (in0_{}, out_{}, numReps);""".format(
+ hls_call, in_t, self.hls_sname(), self.hls_sname()
)
]
@@ -286,12 +297,13 @@ class FMPadding_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -304,16 +316,22 @@ class FMPadding_Batch(HLSCustomOp):
packed_bits = self.get_instream_width()
packed_hls_type = "ap_uint<%d>" % packed_bits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_hls_type, packed_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type,
+ self.hls_sname(),
+ packed_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/globalaccpool_batch.py b/src/finn/custom_op/fpgadataflow/globalaccpool_batch.py
index 220856922c1ed805ccfa60213dc0cf32f45573a1..e5185070342e81af17d7fd49b0bbc5a4df495738 100644
--- a/src/finn/custom_op/fpgadataflow/globalaccpool_batch.py
+++ b/src/finn/custom_op/fpgadataflow/globalaccpool_batch.py
@@ -267,27 +267,40 @@ class GlobalAccPool_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
self.code_gen_dict["$DOCOMPUTE$"] = [
- """AccPool_Batch<{}, {}, {}, {}, {}> (in0, out, 1);""".format(
+ """AccPool_Batch<{}, {}, {}, {}, {}> (in0_{}, out_{}, 1);""".format(
self.get_normal_input_shape()[1],
self.get_nodeattr("NumChannels"),
self.get_input_datatype().get_hls_datatype_str(),
self.get_nodeattr("PE"),
self.get_output_datatype().get_hls_datatype_str(),
+ self.hls_sname(),
+ self.hls_sname(),
)
]
@@ -304,12 +317,13 @@ class GlobalAccPool_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -320,20 +334,22 @@ class GlobalAccPool_Batch(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &out)""".format(
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &out_{})""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/iodma.py b/src/finn/custom_op/fpgadataflow/iodma.py
index 8a756b630ddbd25d5740f0e46297a4ae6f686d2b..4b4ad28defc5e9832316143702f34d82735abf38 100644
--- a/src/finn/custom_op/fpgadataflow/iodma.py
+++ b/src/finn/custom_op/fpgadataflow/iodma.py
@@ -47,7 +47,7 @@ from finn.custom_op.fpgadataflow.hlscustomop import HLSCustomOp
# Interfaces
# - AXI-MM name specified by intfName unless this is set to "" (empty, the default)
-# in which case output AXI-MM are named "out" and input AXI-MM are named "in0"
+# in which case output AXI-MM are named "out_V" and input AXI-MM are named "in0_V"
# - AXI-MM interface width (in bits) is specified by intfWidth
# - AXI-Stream interface width (in bits) is specified by streamWidth
# - If inftWidth and streamWidth are not equal, the DMA core performs
@@ -254,15 +254,24 @@ class IODMA(HLSCustomOp):
# DWCs depend on AXI MM and out interface width
if strmw == intfw:
# case 0: AXI MM width = out width, no DWCs needed
- self.code_gen_dict["$DOCOMPUTE$"] = [dma_inst_template % ("in0", "out")]
+ self.code_gen_dict["$DOCOMPUTE$"] = [
+ dma_inst_template
+ % ("in0_" + self.hls_sname(), "out_" + self.hls_sname())
+ ]
elif (strmw % intfw == 0) or (intfw % strmw == 0):
# case 1: AXI MM width divisible by out width or vice versa
# single DWC + single extra stream needed
self.code_gen_dict["$DOCOMPUTE$"] = [
"hls::stream<ap_uint<%d> > dma2dwc;" % intfw,
- dma_inst_template % ("in0", "dma2dwc"),
+ dma_inst_template % ("in0_" + self.hls_sname(), "dma2dwc"),
dwc_inst_template
- % (intfw, strmw, total_bits // intfw, "dma2dwc", "out"),
+ % (
+ intfw,
+ strmw,
+ total_bits // intfw,
+ "dma2dwc",
+ "out_" + self.hls_sname(),
+ ),
]
else:
# case 2: AXI MM width not divisible by out width or vice versa
@@ -271,26 +280,41 @@ class IODMA(HLSCustomOp):
self.code_gen_dict["$DOCOMPUTE$"] = [
"hls::stream<ap_uint<%d> > dma2lcm;" % intfw,
"hls::stream<ap_uint<%d> > lcm2out;" % width_lcm,
- dma_inst_template % ("in0", "dma2lcm"),
+ dma_inst_template % ("in0_" + self.hls_sname(), "dma2lcm"),
dwc_inst_template
% (intfw, width_lcm, total_bits // intfw, "dma2lcm", "lcm2out"),
dwc_inst_template
- % (width_lcm, strmw, total_bits // width_lcm, "lcm2out", "out"),
+ % (
+ width_lcm,
+ strmw,
+ total_bits // width_lcm,
+ "lcm2out",
+ "out_" + self.hls_sname(),
+ ),
]
elif direction == "out":
# in0 -> (DWCs) -> IODMA -> AXI MM
# DWCs depend on AXI MM and out interface width
if strmw == intfw:
# case 0: in width = AXI MM width, no DWCs needed
- self.code_gen_dict["$DOCOMPUTE$"] = [dma_inst_template % ("in0", "out")]
+ self.code_gen_dict["$DOCOMPUTE$"] = [
+ dma_inst_template
+ % ("in0_" + self.hls_sname(), "out_" + self.hls_sname())
+ ]
elif (strmw % intfw == 0) or (intfw % strmw == 0):
# case 1: AXI MM width divisible by in width or vice versa
# single DWC + single extra stream needed
self.code_gen_dict["$DOCOMPUTE$"] = [
"hls::stream<ap_uint<%d> > dwc2dma;" % intfw,
dwc_inst_template
- % (strmw, intfw, total_bits // strmw, "in0", "dwc2dma"),
- dma_inst_template % ("dwc2dma", "out"),
+ % (
+ strmw,
+ intfw,
+ total_bits // strmw,
+ "in0_" + self.hls_sname(),
+ "dwc2dma",
+ ),
+ dma_inst_template % ("dwc2dma", "out_" + self.hls_sname()),
]
else:
# case 2: AXI MM width not divisible by out width or vice versa
@@ -300,10 +324,16 @@ class IODMA(HLSCustomOp):
"hls::stream<ap_uint<%d> > in2lcm;" % width_lcm,
"hls::stream<ap_uint<%d> > lcm2dma;" % intfw,
dwc_inst_template
- % (strmw, width_lcm, total_bits // strmw, "in0", "in2lcm"),
+ % (
+ strmw,
+ width_lcm,
+ total_bits // strmw,
+ "in0_" + self.hls_sname(),
+ "in2lcm",
+ ),
dwc_inst_template
% (width_lcm, intfw, total_bits // width_lcm, "in2lcm", "lcm2dma"),
- dma_inst_template % ("lcm2dma", "out"),
+ dma_inst_template % ("lcm2dma", "out_" + self.hls_sname()),
]
else:
raise Exception("Unknown IODMA direction: %s" % direction)
@@ -316,13 +346,25 @@ class IODMA(HLSCustomOp):
direction = self.get_nodeattr("direction")
if direction == "in":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(%s *in0, hls::stream<%s > &out, unsigned int numReps)"
- % (self.onnx_node.name, packed_hls_type_in, packed_hls_type_out)
+ "void %s(%s *in0_%s, hls::stream<%s > &out_%s, unsigned int numReps)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type_in,
+ self.hls_sname(),
+ packed_hls_type_out,
+ self.hls_sname(),
+ )
]
elif direction == "out":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, %s *out, unsigned int numReps)"
- % (self.onnx_node.name, packed_hls_type_in, packed_hls_type_out)
+ "void %s(hls::stream<%s > &in0_%s, %s *out_%s, unsigned int numReps)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type_in,
+ self.hls_sname(),
+ packed_hls_type_out,
+ self.hls_sname(),
+ )
]
else:
raise ValueError("Invalid IODMA direction, please set to in or out")
@@ -339,32 +381,36 @@ class IODMA(HLSCustomOp):
if direction == "in":
if intfname == "":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE m_axi offset=slave port=in0"
+ "#pragma HLS INTERFACE m_axi offset=slave port=in0_"
+ + self.hls_sname()
)
else:
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE m_axi offset=slave port=%s" % (intfname)
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE s_axilite port=in0 bundle=control"
+ "#pragma HLS INTERFACE s_axilite port=in0_%s bundle=control"
+ % (self.hls_sname())
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
elif direction == "out":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
)
if intfname == "":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE m_axi offset=slave port=out"
+ "#pragma HLS INTERFACE m_axi offset=slave port=out_"
+ + self.hls_sname()
)
else:
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE m_axi offset=slave port=%s" % (intfname)
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE s_axilite port=out bundle=control"
+ "#pragma HLS INTERFACE s_axilite port=out_%s bundle=control"
+ % (self.hls_sname())
)
else:
raise ValueError("Invalid IODMA direction, please set to in or out")
diff --git a/src/finn/custom_op/fpgadataflow/labelselect_batch.py b/src/finn/custom_op/fpgadataflow/labelselect_batch.py
index 492cd0107321f3abbfe02d5e456ee3732da982d0..12a88dacd4dfb366780c5ef5e42b355a0ca51b59 100644
--- a/src/finn/custom_op/fpgadataflow/labelselect_batch.py
+++ b/src/finn/custom_op/fpgadataflow/labelselect_batch.py
@@ -275,29 +275,42 @@ class LabelSelect_Batch(HLSCustomOp):
# Also notice that StreamingDataWidthConverter_Batch performs LE packing
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0,false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
node = self.onnx_node
self.code_gen_dict["$DOCOMPUTE$"] = [
- """{}<{}, {}, {}, {}, {} > (in0, out, 1);""".format(
+ """{}<{}, {}, {}, {}, {} > (in0_{}, out_{}, 1);""".format(
node.op_type,
self.get_nodeattr("Labels"),
self.get_nodeattr("PE"),
self.get_nodeattr("K"),
self.get_input_datatype().get_hls_datatype_str(),
self.get_output_datatype().get_hls_datatype_str(),
+ self.hls_sname(),
+ self.hls_sname(),
)
]
@@ -314,12 +327,13 @@ class LabelSelect_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -330,21 +344,23 @@ class LabelSelect_Batch(HLSCustomOp):
def blackboxfunction(self):
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}*{}>> &in0,
- hls::stream<ap_uint<{}> > &out)""".format(
+ """void {}(hls::stream<ap_uint<{}*{}>> &in0_{},
+ hls::stream<ap_uint<{}> > &out_{})""".format(
self.onnx_node.name,
self.get_nodeattr("PE"),
self.get_input_datatype().bitwidth(),
+ self.hls_sname(),
self.get_output_datatype().bitwidth(),
+ self.hls_sname(),
)
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/lookup.py b/src/finn/custom_op/fpgadataflow/lookup.py
index ed560ac962477965bae39d296287c09eb077eca0..ecf630ef7f10b32b4ccea319aaddda603ad43bec 100644
--- a/src/finn/custom_op/fpgadataflow/lookup.py
+++ b/src/finn/custom_op/fpgadataflow/lookup.py
@@ -206,8 +206,15 @@ class Lookup(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def dataoutstrm(self):
@@ -226,12 +233,13 @@ class Lookup(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", %s);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", %s);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
"false",
@@ -244,10 +252,14 @@ class Lookup(HLSCustomOp):
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -255,12 +267,14 @@ class Lookup(HLSCustomOp):
if mem_mode == "const":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""StreamingLookup<NumEmbeddings, EmbeddingDim, NumInputs,
- InputType, EmbeddingType >(in0, out, embeddings);"""
+ InputType, EmbeddingType >(in0_%s, out_%s, embeddings);"""
+ % (self.hls_sname(), self.hls_sname())
]
elif mem_mode == "external":
self.code_gen_dict["$DOCOMPUTE$"] = [
- """StreamingLookup_ext<EmbeddingSize>(in0, out, mem, size, oob_count,
+ """StreamingLookup_ext<EmbeddingSize>(in0_%s, out_%s, mem, size, oob_count,
oob_irq);"""
+ % (self.hls_sname(), self.hls_sname())
]
def blackboxfunction(self):
@@ -271,26 +285,29 @@ class Lookup(HLSCustomOp):
packed_output_hls_type = "ap_uint<%d>" % obits
if mem_mode == "const":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_input_hls_type, packed_output_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_input_hls_type,
+ self.hls_sname(),
+ packed_output_hls_type,
+ self.hls_sname(),
+ )
]
elif mem_mode == "external":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
"void "
+ self.onnx_node.name
- + "(hls::stream<T_SRC> &in0, hls::stream<T_DST> &out, "
+ + "(hls::stream<T_SRC> &in0_%s, hls::stream<T_DST> &out_%s, "
+ % (self.hls_sname(), self.hls_sname())
+ "T_DST const *const mem, unsigned const size, "
+ "unsigned &oob_count, bool &oob_irq)"
]
def pragmas(self):
mem_mode = self.get_nodeattr("mem_mode")
- my_pragmas = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
- ]
- my_pragmas.append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
- )
+ my_pragmas = ["#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()]
+ my_pragmas.append("#pragma HLS INTERFACE axis port=out_" + self.hls_sname())
my_pragmas.append("#pragma HLS INTERFACE ap_ctrl_none port=return")
if mem_mode == "const":
my_pragmas.append(
diff --git a/src/finn/custom_op/fpgadataflow/matrixvectoractivation.py b/src/finn/custom_op/fpgadataflow/matrixvectoractivation.py
index 899bce98d2dd9572d1adf2f20910a0463f9d5994..fae2d86d887ce3ae0523f3a99681315af12dd645 100644
--- a/src/finn/custom_op/fpgadataflow/matrixvectoractivation.py
+++ b/src/finn/custom_op/fpgadataflow/matrixvectoractivation.py
@@ -1097,8 +1097,15 @@ class MatrixVectorActivation(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
# note: the innermost dim is reversed for the input
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0, false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
mem_mode = self.get_nodeattr("mem_mode")
@@ -1112,24 +1119,35 @@ class MatrixVectorActivation(HLSCustomOp):
npy_in = "%s/weights.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", weights, false, numReps);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", weights_%s, false, numReps);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
mem_mode = self.get_nodeattr("mem_mode")
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
if mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> weights ("weights");'.format(
- self.get_weightstream_width()
+ 'hls::stream<ap_uint<{}>> weights_{} ("weights_{}");'.format(
+ self.get_weightstream_width(), self.hls_sname(), self.hls_sname()
)
)
@@ -1149,10 +1167,12 @@ class MatrixVectorActivation(HLSCustomOp):
if mem_mode == "const":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""Matrix_Vector_Activate_Batch<MW1, MH1, SIMD1, PE1, 1, {}, {}, {}>
- (in0, out, weights, {}, numReps, {});""".format(
+ (in0_{}, out_{}, weights, {}, numReps, {});""".format(
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
tmpl_args["TWeightI"],
+ self.hls_sname(),
+ self.hls_sname(),
threshs,
map_to_hls_mult_style[self.get_nodeattr("resType")],
)
@@ -1166,11 +1186,14 @@ class MatrixVectorActivation(HLSCustomOp):
wdtype_hls_str = export_wdt.get_hls_datatype_str()
self.code_gen_dict["$DOCOMPUTE$"] = [
"""Matrix_Vector_Activate_Stream_Batch<MW1, MH1, SIMD1, PE1, {}, {}, {}, {} >
- (in0, out, weights, {}, numReps, {});""".format(
+ (in0_{}, out_{}, weights_{}, {}, numReps, {});""".format(
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
tmpl_args["TWeightI"],
wdtype_hls_str,
+ self.hls_sname(),
+ self.hls_sname(),
+ self.hls_sname(),
threshs,
map_to_hls_mult_style[self.get_nodeattr("resType")],
)
@@ -1199,12 +1222,13 @@ class MatrixVectorActivation(HLSCustomOp):
# note: the innermost dim is not reversed for the output
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
shape_cpp_str,
npy_out,
)
@@ -1217,25 +1241,30 @@ class MatrixVectorActivation(HLSCustomOp):
mem_mode = self.get_nodeattr("mem_mode")
if mem_mode == "const":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &out
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
elif mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
"""void {}(
- hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &weights,
- hls::stream<ap_uint<{}>> &out
+ hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &weights_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_weightstream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
@@ -1249,10 +1278,10 @@ class MatrixVectorActivation(HLSCustomOp):
mem_mode = self.get_nodeattr("mem_mode")
ram_style_thresholds = self.get_nodeattr("ram_style_thresholds")
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
@@ -1270,11 +1299,10 @@ class MatrixVectorActivation(HLSCustomOp):
)
elif mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=weights name=weights_"
- + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=weights_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS stream depth=8 variable=weights"
+ "#pragma HLS stream depth=8 variable=weights_" + self.hls_sname()
)
else:
diff --git a/src/finn/custom_op/fpgadataflow/pool_batch.py b/src/finn/custom_op/fpgadataflow/pool_batch.py
index 813f13e504eae181f4398eccbe40ad66b6e3bf16..8ccfce78209d82a77aa0d018e43ed4c4e80d7ebc 100644
--- a/src/finn/custom_op/fpgadataflow/pool_batch.py
+++ b/src/finn/custom_op/fpgadataflow/pool_batch.py
@@ -239,17 +239,28 @@ class Pool_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0,false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -281,8 +292,8 @@ class Pool_Batch(HLSCustomOp):
self.code_gen_dict["$DOCOMPUTE$"] += [
"""Pool_batch<Channels, PE, KernelSize,Slice<{} >, Slice< {} > >
- (in0,out, pool_fxn, OFMDimTotal*numReps);""".format(
- i_hls_dt, o_hls_dt
+ (in0_{}, out_{}, pool_fxn, OFMDimTotal*numReps);""".format(
+ i_hls_dt, o_hls_dt, self.hls_sname(), self.hls_sname()
)
]
@@ -302,12 +313,13 @@ class Pool_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s",false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -323,16 +335,22 @@ class Pool_Batch(HLSCustomOp):
packed_obits = self.get_outstream_width()
packed_out_hls_type = "ap_uint<%d>" % packed_obits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_in_hls_type, packed_out_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_in_hls_type,
+ self.hls_sname(),
+ packed_out_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/streamingdatawidthconverter_batch.py b/src/finn/custom_op/fpgadataflow/streamingdatawidthconverter_batch.py
index a80d2bbefac96e8ec2a48e04179d3d285e78cef7..dc905658b199e3d0d13db3cea2d24ab7f5aed92c 100644
--- a/src/finn/custom_op/fpgadataflow/streamingdatawidthconverter_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingdatawidthconverter_batch.py
@@ -236,14 +236,23 @@ class StreamingDataWidthConverter_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
if self.needs_lcm():
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
@@ -252,7 +261,9 @@ class StreamingDataWidthConverter_Batch(HLSCustomOp):
)
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -263,13 +274,15 @@ class StreamingDataWidthConverter_Batch(HLSCustomOp):
'hls::stream<ap_uint<{}>> intermediate ("intermediate");'.format(
self.get_iowidth_lcm()
),
- "%s<InWidth, LCMWidth, NumInWords>(in0, intermediate, numReps);" % (op),
- "%s<LCMWidth, OutWidth, NumLCMToOut>(intermediate, out, numReps);"
- % (op),
+ "%s<InWidth, LCMWidth, NumInWords>(in0_%s, intermediate, numReps);"
+ % (op, self.hls_sname()),
+ "%s<LCMWidth, OutWidth, NumLCMToOut>(intermediate, out_%s, numReps);"
+ % (op, self.hls_sname()),
]
else:
self.code_gen_dict["$DOCOMPUTE$"] = [
- "%s<InWidth, OutWidth, NumInWords>(in0, out, numReps);" % (op)
+ "%s<InWidth, OutWidth, NumInWords>(in0_%s, out_%s, numReps);"
+ % (op, self.hls_sname(), self.hls_sname())
]
def dataoutstrm(self):
@@ -288,12 +301,13 @@ class StreamingDataWidthConverter_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -308,16 +322,22 @@ class StreamingDataWidthConverter_Batch(HLSCustomOp):
out_packed_bits = self.get_outstream_width()
out_packed_hls_type = "ap_uint<%d>" % out_packed_bits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, in_packed_hls_type, out_packed_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ in_packed_hls_type,
+ self.hls_sname(),
+ out_packed_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py b/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
index a0e60931edd8590aaebc0560c4bd28d61d62e8ea..78f4095cbeaad0987fe44f2fcdfbd7f5652eb173 100755
--- a/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingmaxpool_batch.py
@@ -254,17 +254,28 @@ class StreamingMaxPool_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -275,7 +286,8 @@ class StreamingMaxPool_Batch(HLSCustomOp):
else:
op = "StreamingMaxPool"
self.code_gen_dict["$DOCOMPUTE$"] = [
- "%s<ImgDim, PoolDim, NumChannels>(in0, out);" % (op)
+ "%s<ImgDim, PoolDim, NumChannels>(in0_%s, out_%s);"
+ % (op, self.hls_sname(), self.hls_sname())
]
else:
dtype = self.get_input_datatype()
@@ -285,14 +297,14 @@ class StreamingMaxPool_Batch(HLSCustomOp):
op = "StreamingMaxPool_Precision_1d"
self.code_gen_dict["$DOCOMPUTE$"] = [
"""%s<ImgDim, PoolDim, NumChannels, PE,
- OutputSize, %s, %s>(in0, out);"""
- % (op, dtype_hls, minval_str)
+ OutputSize, %s, %s>(in0_%s, out_%s);"""
+ % (op, dtype_hls, minval_str, self.hls_sname(), self.hls_sname())
]
else:
op = "StreamingMaxPool_Precision"
self.code_gen_dict["$DOCOMPUTE$"] = [
- "%s<ImgDim, PoolDim, NumChannels, %s, %s>(in0, out);"
- % (op, dtype_hls, minval_str)
+ "%s<ImgDim, PoolDim, NumChannels, %s, %s>(in0_%s, out_%s);"
+ % (op, dtype_hls, minval_str, self.hls_sname(), self.hls_sname())
]
def dataoutstrm(self):
@@ -311,12 +323,13 @@ class StreamingMaxPool_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -329,16 +342,22 @@ class StreamingMaxPool_Batch(HLSCustomOp):
packed_bits = self.get_instream_width()
packed_hls_type = "ap_uint<%d>" % packed_bits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_hls_type, packed_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type,
+ self.hls_sname(),
+ packed_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/thresholding_batch.py b/src/finn/custom_op/fpgadataflow/thresholding_batch.py
index 12e635b3d612235f6b464e6f54d0f24011c7c907..fc5aa61d6669fc566b2454b1ea18d3a0a42f7d9c 100644
--- a/src/finn/custom_op/fpgadataflow/thresholding_batch.py
+++ b/src/finn/custom_op/fpgadataflow/thresholding_batch.py
@@ -613,8 +613,15 @@ class Thresholding_Batch(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
# note: the innermost dim is reversed for the input
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0, false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
mem_mode = self.get_nodeattr("mem_mode")
if mem_mode == "decoupled":
@@ -627,23 +634,34 @@ class Thresholding_Batch(HLSCustomOp):
npy_in = "%s/thresholds.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", weights, false, ImgDim1);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", weights_%s, false, ImgDim1);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
mem_mode = self.get_nodeattr("mem_mode")
if mem_mode == "decoupled":
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> weights ("weights");'.format(
- self.get_weightstream_width()
+ 'hls::stream<ap_uint<{}>> weights_{} ("weights_{}");'.format(
+ self.get_weightstream_width(), self.hls_sname(), self.hls_sname()
)
)
@@ -654,10 +672,12 @@ class Thresholding_Batch(HLSCustomOp):
if mem_mode == "const":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ImgDim1, NumChannels1, PE1, {}, {}>
- (in0, out, threshs, numReps);""".format(
+ (in0_{}, out_{}, threshs, numReps);""".format(
node.op_type,
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
+ self.hls_sname(),
+ self.hls_sname(),
)
]
elif mem_mode == "decoupled":
@@ -666,10 +686,13 @@ class Thresholding_Batch(HLSCustomOp):
# - for synth the unit runs continuously anyway (ap_ctrl_none)
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<ImgDim1, NumChannels1, PE1, {}, {}, ActVal1, ThresType1, NumSteps1>
- (in0, out, weights, numReps);""".format(
+ (in0_{}, out_{}, weights_{}, numReps);""".format(
"Thresholding_Stream_Batch",
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
+ self.hls_sname(),
+ self.hls_sname(),
+ self.hls_sname(),
)
]
else:
@@ -692,12 +715,13 @@ class Thresholding_Batch(HLSCustomOp):
# note: the innermost dim is not reversed for the output
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
shape_cpp_str,
npy_out,
)
@@ -709,24 +733,29 @@ class Thresholding_Batch(HLSCustomOp):
def blackboxfunction(self):
if self.get_nodeattr("mem_mode") == "const":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &out
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
elif self.get_nodeattr("mem_mode") == "decoupled":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &weights,
- hls::stream<ap_uint<{}>> &out
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &weights_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_weightstream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
else:
@@ -734,10 +763,10 @@ class Thresholding_Batch(HLSCustomOp):
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
@@ -789,8 +818,7 @@ class Thresholding_Batch(HLSCustomOp):
)
elif self.get_nodeattr("mem_mode") == "decoupled":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=weights name=weights_"
- + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=weights_" + self.hls_sname()
)
def code_generation_ipi(self):
diff --git a/src/finn/custom_op/fpgadataflow/tlastmarker.py b/src/finn/custom_op/fpgadataflow/tlastmarker.py
index 895a2eedab51cee6322c7307ea1944d49a0dade5..6eaf03ab16d323129dcbf1ddba0fd133b105b476 100644
--- a/src/finn/custom_op/fpgadataflow/tlastmarker.py
+++ b/src/finn/custom_op/fpgadataflow/tlastmarker.py
@@ -130,9 +130,11 @@ class TLastMarker(HLSCustomOp):
self.code_gen_dict["$DOCOMPUTE$"] = [
"for(unsigned int i=0; i<NumItersPerImg; i++) {",
"#pragma HLS PIPELINE II=1",
- "out.write(in0.read().get_data());"
+ "out_%s.write(in0_%s.read().get_data());"
+ % (self.hls_sname(), self.hls_sname())
if use_qdma_axis
- else "out.write(in0.read().data);",
+ else "out_%s.write(in0_%s.read().data);"
+ % (self.hls_sname(), self.hls_sname()),
"}",
]
@@ -146,17 +148,21 @@ class TLastMarker(HLSCustomOp):
"#pragma HLS protocol fixed",
"// do a first read from stream before we decide on numIters",
"// giving software a chance to set up the numIters prior to startup",
- "t.set_data(in0.read());" if use_qdma_axis else "t.data = in0.read();",
+ "t.set_data(in0_%s.read());" % self.hls_sname()
+ if use_qdma_axis
+ else "t.data = in0_%s.read();" % self.hls_sname(),
"n = (numIters == 0 ? NumItersPerImg : numIters);",
"t.set_last(n==1);" if use_qdma_axis else "t.last = (n==1);",
- "out.write(t);",
+ "out_%s.write(t);" % self.hls_sname(),
"} // end of cycle accurate region",
"// do one less iteration than spec since we already did one",
"for(unsigned int i=1; i<n; i++) {",
"#pragma HLS PIPELINE II=1",
- "t.set_data(in0.read());" if use_qdma_axis else "t.data = in0.read();",
+ "t.set_data(in0_%s.read());" % self.hls_sname()
+ if use_qdma_axis
+ else "t.data = in0_%s.read();" % self.hls_sname(),
"t.set_last(i==(n-1));" if use_qdma_axis else "t.last = (i==(n-1));",
- "out.write(t);",
+ "out_%s.write(t);" % self.hls_sname(),
"}",
]
@@ -168,11 +174,13 @@ class TLastMarker(HLSCustomOp):
"t.set_keep(-1);" if use_qdma_axis else "t.keep = -1;",
"for(unsigned int i=0; i<NumItersPerImg; i++) {",
"#pragma HLS PIPELINE II=1",
- "t.set_data(in0.read());" if use_qdma_axis else "t.data = in0.read();",
+ "t.set_data(in0_%s.read());" % self.hls_sname()
+ if use_qdma_axis
+ else "t.data = in0_%s.read();" % self.hls_sname(),
"t.set_last(i==(NumItersPerImg-1));"
if use_qdma_axis
else "t.last = (i==(NumItersPerImg-1));",
- "out.write(t);",
+ "out_%s.write(t);" % self.hls_sname(),
"}",
]
@@ -187,22 +195,23 @@ class TLastMarker(HLSCustomOp):
if dyn_iters == 1:
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void %s(hls::stream<InDType> &in0,
- hls::stream<OutDType> &out, unsigned int numIters)"""
- % self.onnx_node.name
+ """void %s(hls::stream<InDType> &in0_%s,
+ hls::stream<OutDType> &out_%s, unsigned int numIters)"""
+ % (self.onnx_node.name, self.hls_sname(), self.hls_sname())
]
else:
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void %s(hls::stream<InDType> &in0, hls::stream<OutDType> &out)"""
- % self.onnx_node.name
+ """void %s(hls::stream<InDType> &in0_%s,
+ hls::stream<OutDType> &out_%s)"""
+ % (self.onnx_node.name, self.hls_sname(), self.hls_sname())
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
dyn_iters = self.get_nodeattr("DynIters")
@@ -239,10 +248,12 @@ class TLastMarker(HLSCustomOp):
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<InDType> in0 ("in0");'
+ 'hls::stream<InDType> in0_%s ("in0_%s");'
+ % (self.hls_sname(), self.hls_sname())
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<OutDType> out ("out");'
+ 'hls::stream<OutDType> out_%s ("out_%s");'
+ % (self.hls_sname(), self.hls_sname())
)
def get_verilog_top_module_intf_names(self):
diff --git a/src/finn/custom_op/fpgadataflow/upsampler.py b/src/finn/custom_op/fpgadataflow/upsampler.py
index b653b9386e940dd2220fa1fb0d198e63b81a356d..ab5a734e7c20a17603583a62bdb2650b11319307 100644
--- a/src/finn/custom_op/fpgadataflow/upsampler.py
+++ b/src/finn/custom_op/fpgadataflow/upsampler.py
@@ -187,17 +187,28 @@ class UpsampleNearestNeighbour_Batch(HLSCustomOp):
npy_in = "%s/input_0.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"] = []
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
def docompute(self):
@@ -206,13 +217,15 @@ class UpsampleNearestNeighbour_Batch(HLSCustomOp):
if is_2d:
self.code_gen_dict["$DOCOMPUTE$"] = [
"""UpsampleNearestNeighbour_Batch<OFMDim, IFMDim, IFMChannels,
- ap_uint<Input_precision> > (in0, out, numReps);"""
+ ap_uint<Input_precision> > (in0_%s, out_%s, numReps);"""
+ % (self.hls_sname(), self.hls_sname())
]
else:
assert batch == 1, "1D upsampler currently needs numReps=1"
self.code_gen_dict["$DOCOMPUTE$"] = [
"""UpsampleNearestNeighbour_1D<OFMDim, IFMDim, IFMChannels,
- ap_uint<Input_precision> > (in0, out);"""
+ ap_uint<Input_precision> > (in0_%s, out_%s);"""
+ % (self.hls_sname(), self.hls_sname())
]
def dataoutstrm(self):
@@ -231,12 +244,13 @@ class UpsampleNearestNeighbour_Batch(HLSCustomOp):
oshape_cpp_str = str(oshape).replace("(", "{").replace(")", "}")
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s");'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s");'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
oshape_cpp_str,
npy_out,
)
@@ -249,16 +263,22 @@ class UpsampleNearestNeighbour_Batch(HLSCustomOp):
packed_bits = self.get_instream_width()
packed_hls_type = "ap_uint<%d>" % packed_bits
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- "void %s(hls::stream<%s > &in0, hls::stream<%s > &out)"
- % (self.onnx_node.name, packed_hls_type, packed_hls_type)
+ "void %s(hls::stream<%s > &in0_%s, hls::stream<%s > &out_%s)"
+ % (
+ self.onnx_node.name,
+ packed_hls_type,
+ self.hls_sname(),
+ packed_hls_type,
+ self.hls_sname(),
+ )
]
def pragmas(self):
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
diff --git a/src/finn/custom_op/fpgadataflow/vectorvectoractivation.py b/src/finn/custom_op/fpgadataflow/vectorvectoractivation.py
index ede572f1a4c5fcf60476aa84beba8244eb9ec5a8..64fb5dcbe17adf726e6e982dd2203683bb86d1a8 100644
--- a/src/finn/custom_op/fpgadataflow/vectorvectoractivation.py
+++ b/src/finn/custom_op/fpgadataflow/vectorvectoractivation.py
@@ -802,8 +802,15 @@ class VectorVectorActivation(HLSCustomOp):
self.code_gen_dict["$READNPYDATA$"] = []
# note: the innermost dim is reversed for the input
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", in0, false);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", in0_%s, false);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
mem_mode = self.get_nodeattr("mem_mode")
@@ -817,23 +824,34 @@ class VectorVectorActivation(HLSCustomOp):
npy_in = "%s/weights.npy" % code_gen_dir
self.code_gen_dict["$READNPYDATA$"].append(
- 'npy2apintstream<%s, %s, %d, %s>("%s", weights, false, numReps);'
- % (packed_hls_type, elem_hls_type, elem_bits, npy_type, npy_in)
+ 'npy2apintstream<%s, %s, %d, %s>("%s", weights_%s, false, numReps);'
+ % (
+ packed_hls_type,
+ elem_hls_type,
+ elem_bits,
+ npy_type,
+ npy_in,
+ self.hls_sname(),
+ )
)
def strm_decl(self):
mem_mode = self.get_nodeattr("mem_mode")
self.code_gen_dict["$STREAMDECLARATIONS$"] = []
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> in0 ("in0");'.format(self.get_instream_width())
+ 'hls::stream<ap_uint<{}>> in0_{} ("in0_{}");'.format(
+ self.get_instream_width(), self.hls_sname(), self.hls_sname()
+ )
)
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> out ("out");'.format(self.get_outstream_width())
+ 'hls::stream<ap_uint<{}>> out_{} ("out_{}");'.format(
+ self.get_outstream_width(), self.hls_sname(), self.hls_sname()
+ )
)
if mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$STREAMDECLARATIONS$"].append(
- 'hls::stream<ap_uint<{}>> weights ("weights");'.format(
- self.get_weightstream_width()
+ 'hls::stream<ap_uint<{}>> weights_{} ("weights_{}");'.format(
+ self.get_weightstream_width(), self.hls_sname(), self.hls_sname()
)
)
@@ -854,10 +872,12 @@ class VectorVectorActivation(HLSCustomOp):
if mem_mode == "const":
self.code_gen_dict["$DOCOMPUTE$"] = [
"""Vector_Vector_Activate_Batch<Channels1, InnerProdDim, SIMD1, PE1, 1, {}, {}, {}>
- (in0, out, weights, {}, numReps, {});""".format(
+ (in0_{}, out_{}, weights, {}, numReps, {});""".format(
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
tmpl_args["TWeightI"],
+ self.hls_sname(),
+ self.hls_sname(),
threshs,
map_to_hls_mult_style[self.get_nodeattr("resType")],
)
@@ -871,12 +891,15 @@ class VectorVectorActivation(HLSCustomOp):
wdtype_hls_str = export_wdt.get_hls_datatype_str()
self.code_gen_dict["$DOCOMPUTE$"] = [
"""{}<Channels1, InnerProdDim, SIMD1, PE1, 1, {}, {}, {}, {}>
- (in0, out, weights, {}, numReps, {});""".format(
+ (in0_{}, out_{}, weights_{}, {}, numReps, {});""".format(
"Vector_Vector_Activate_Stream_Batch",
tmpl_args["TSrcI"],
tmpl_args["TDstI"],
tmpl_args["TWeightI"],
wdtype_hls_str,
+ self.hls_sname(),
+ self.hls_sname(),
+ self.hls_sname(),
threshs,
map_to_hls_mult_style[self.get_nodeattr("resType")],
)
@@ -904,12 +927,13 @@ class VectorVectorActivation(HLSCustomOp):
# note: the innermost dim is not reversed for the output
self.code_gen_dict["$DATAOUTSTREAM$"] = [
- 'apintstream2npy<%s, %s, %d, %s>(out, %s, "%s", false);'
+ 'apintstream2npy<%s, %s, %d, %s>(out_%s, %s, "%s", false);'
% (
packed_hls_type,
elem_hls_type,
elem_bits,
npy_type,
+ self.hls_sname(),
shape_cpp_str,
npy_out,
)
@@ -922,25 +946,30 @@ class VectorVectorActivation(HLSCustomOp):
mem_mode = self.get_nodeattr("mem_mode")
if mem_mode == "const":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
- """void {}(hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &out
+ """void {}(hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
elif mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$BLACKBOXFUNCTION$"] = [
"""void {}(
- hls::stream<ap_uint<{}>> &in0,
- hls::stream<ap_uint<{}>> &weights,
- hls::stream<ap_uint<{}>> &out
+ hls::stream<ap_uint<{}>> &in0_{},
+ hls::stream<ap_uint<{}>> &weights_{},
+ hls::stream<ap_uint<{}>> &out_{}
)""".format(
self.onnx_node.name,
self.get_instream_width(),
+ self.hls_sname(),
self.get_weightstream_width(),
+ self.hls_sname(),
self.get_outstream_width(),
+ self.hls_sname(),
)
]
else:
@@ -952,10 +981,10 @@ class VectorVectorActivation(HLSCustomOp):
def pragmas(self):
mem_mode = self.get_nodeattr("mem_mode")
self.code_gen_dict["$PRAGMAS$"] = [
- "#pragma HLS INTERFACE axis port=in0 name=in0_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=in0_" + self.hls_sname()
]
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=out name=out_" + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=out_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
@@ -973,11 +1002,10 @@ class VectorVectorActivation(HLSCustomOp):
)
elif mem_mode == "decoupled" or mem_mode == "external":
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS INTERFACE axis port=weights name=weights_"
- + self.hls_sname()
+ "#pragma HLS INTERFACE axis port=weights_" + self.hls_sname()
)
self.code_gen_dict["$PRAGMAS$"].append(
- "#pragma HLS stream depth=8 variable=weights"
+ "#pragma HLS stream depth=8 variable=weights_" + self.hls_sname()
)
else:
raise Exception(