From 698f46508f7ad93fdaf4525979904c6aa275c6f2 Mon Sep 17 00:00:00 2001
From: auphelia <jakobapk@web.de>
Date: Fri, 13 Mar 2020 11:44:11 +0000
Subject: [PATCH] [StreamingFC] Generation of .dat file into generate_params
function
---
.../fpgadataflow/streamingfclayer_batch.py | 55 +++++++++----------
1 file changed, 27 insertions(+), 28 deletions(-)
diff --git a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
index 4b00549d3..af8fe9302 100644
--- a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
@@ -40,6 +40,7 @@ from finn.util.data_packing import (
npy_to_rtlsim_input,
numpy_to_hls_code,
rtlsim_output_to_npy,
+ pack_innermost_dim_as_hex_string,
)
# ONNX i/o tensor shape assumptions for StreamingFCLayer:
@@ -428,14 +429,14 @@ class StreamingFCLayer_Batch(HLSCustomOp):
# convert weights into hlslib-compatible format
weight_tensor = self.get_hls_compatible_weight_tensor(weights)
export_wdt = self.get_weight_datatype()
+ # we have converted bipolar weights to binary for export,
+ # so use it as such for weight generation
+ if self.get_weight_datatype() == DataType.BIPOLAR:
+ export_wdt = DataType.BINARY
code_gen_dir = path
if mem_mode == "const":
"""Saves weights into params.h"""
- # we have converted bipolar weights to binary for export,
- # so use it as such for weight generation
- if self.get_weight_datatype() == DataType.BIPOLAR:
- export_wdt = DataType.BINARY
weight_hls_code = numpy_to_hls_code(
weight_tensor, export_wdt, "weights", True, True
)
@@ -477,6 +478,23 @@ class StreamingFCLayer_Batch(HLSCustomOp):
np.save(
os.path.join(code_gen_dir, "weights.npy"), weight_tensor,
)
+ """Saves weights into .dat file"""
+ # convert weight value sinto hexstring
+ weight_width = self.get_weightstream_width()
+ weight_tensor = pack_innermost_dim_as_hex_string(
+ weight_tensor, export_wdt, weight_width
+ )
+ weight_pad = np.zeros((1024), int).astype(str)
+ weight_tensor = weight_tensor.flatten()
+ # delete "0x" in the beginning of the hexstring
+ for i in range(len(weight_tensor)):
+ weight_tensor[i] = weight_tensor[i][2:]
+ weight_pad[: weight_tensor.shape[0]] = weight_tensor
+ f = open("{}/memblock_0.dat".format(code_gen_dir), "w+")
+ for val in weight_pad:
+ f.write(val + "\n")
+ f.close()
+
else:
raise Exception(
"""Please set mem_mode to "const"i or "decoupled", currently no other
@@ -597,31 +615,12 @@ class StreamingFCLayer_Batch(HLSCustomOp):
# reshape output to have expected shape
context[node.output[0]] = context[node.output[0]].reshape(1, mh)
elif mode == "rtlsim":
- # check mem mode if set to "decoupled" to convert
- # weights.npy file into .dat file
+ # set top name depending on mem_mode
mem_mode = self.get_nodeattr("mem_mode")
- if mem_mode == "decoupled":
- # iterate over array and save in memblock_0.dat
- weights = np.load("{}/weights.npy".format(code_gen_dir))
- weights = weights.flatten()
- wdt = self.get_weight_datatype()
- f = open(
- "{}/project_{}/sol1/impl/verilog/memblock_0.dat".format(
- code_gen_dir, self.onnx_node.name
- ),
- "w+",
- )
- for val in weights:
- # convert signed values into unsigned integer
- if wdt.name.startswith("INT"):
- bitwidth = wdt.bitwidth()
- if val < 0:
- val = int(bin(int(val) + 2 ** bitwidth), 2)
-
- f.write(str(int(val)) + "\n")
- f.close()
-
- prefixed_top_name = "%s_%s" % (node.name, node.name)
+ if mem_mode == "const":
+ prefixed_top_name = "%s_%s" % (node.name, node.name)
+ elif mem_mode == "decoupled":
+ prefixed_top_name == "%s_memstream" % (node.name)
# check if needed file exists
verilog_file = "{}/project_{}/sol1/impl/verilog/{}.v".format(
code_gen_dir, node.name, prefixed_top_name
--
GitLab