test cleanup and bringing up to updated finn-hlslib commit. Potential…

… failing cases might exist due to removal of extra iteration in inWidth < OutWidth c

ases

fetch-repos.sh

@@ -32,7 +32,7 @@ FINN_EXP_COMMIT="0724be21111a21f0d81a072fccc1c446e053f851"
 BREVITAS_COMMIT="d4834bd2a0fad3c1fbc0ff7e1346d5dcb3797ea4"
 PYVERILATOR_COMMIT="ce0a08c20cb8c1d1e84181d6f392390f846adbd1"
 CNPY_COMMIT="4e8810b1a8637695171ed346ce68f6984e585ef4"
-HLSLIB_COMMIT="16e5847a5e3ef76cffe84c8fad2f010d593457d3"
+HLSLIB_COMMIT="35a04fcfc58044cbbbdd6ef07a38a247aa76efb6"
 OMX_COMMIT="0b59762f9e4c4f7e5aa535ee9bc29f292434ca7a"
 AVNET_BDF_COMMIT="2d49cfc25766f07792c0b314489f21fe916b639b"
 XIL_BDF_COMMIT="8cf4bb674a919ac34e3d99d8d71a9e60af93d14e"
@@ -45,7 +45,7 @@ FINN_EXP_URL="https://github.com/Xilinx/finn-experimental.git"
 BREVITAS_URL="https://github.com/Xilinx/brevitas.git"
 PYVERILATOR_URL="https://github.com/maltanar/pyverilator.git"
 CNPY_URL="https://github.com/rogersce/cnpy.git"
-HLSLIB_URL="https://github.com/Xilinx/finn-hlslib.git"
+HLSLIB_URL="https://github.com/lstasytis/finn-hlslib.git"
 OMX_URL="https://github.com/maltanar/oh-my-xilinx.git"
 AVNET_BDF_URL="https://github.com/Avnet/bdf.git"
 XIL_BDF_URL="https://github.com/Xilinx/XilinxBoardStore.git"

src/finn/custom_op/fpgadataflow/hls/iodma_hls.py

@@ -224,7 +224,7 @@ def get_ap_int_max_w(self):
     def docompute(self):
         direction = self.get_nodeattr("direction")
         mode = self.get_nodeattr("burstMode")
-        dwc_func = "StreamingDataWidthConverter_Batch"
+        dwc_func = "StreamingDataWidthConverterGeneralized_Batch"
         if direction == "in":
             if mode == "wrap":
                 func = "Mem2Stream_Batch_external_wmem"
@@ -236,7 +236,7 @@ def docompute(self):
             raise ValueError("Invalid IODMA direction, please set to in or out")
         # define templates for instantiation
         dma_inst_template = func + "<DataWidth1, NumBytes1>(%s, %s, numReps);"
-        dwc_inst_template = dwc_func + "<%d, %d, %d, %d, %d>(%s, %s, numReps);"
+        dwc_inst_template = dwc_func + "<%d, %d, %d, %d>(%s, %s, numReps);"
         # do stream infrastructure and instantiations
         intfw = self.get_nodeattr("intfWidth")
         strmw = self.get_nodeattr("streamWidth")
@@ -252,10 +252,10 @@ def docompute(self):
 
             numInWords = total_bits // inWidth
             numOutWords = total_bits // outWidth
-            totalIters = max(numInWords, numOutWords)
+            # totalIters = max(numInWords, numOutWords)
 
-            if outWidth > inWidth:
-                totalIters += int(np.floor(outWidth / inWidth) + 1) - 1
+            # if outWidth > inWidth:
+            #    totalIters += int(np.floor(outWidth / inWidth) + 1) - 1
 
             # AXI MM -> IODMA -> (DWCs) -> out
             # DWCs depend on AXI MM and out interface width
@@ -277,7 +277,6 @@ def docompute(self):
                         outWidth,
                         numInWords,
                         numOutWords,
-                        totalIters,
                         "dma2dwc",
                         "out_" + self.hls_sname(),
                     ),
@@ -289,10 +288,10 @@ def docompute(self):
 
             numInWords = total_bits // inWidth
             numOutWords = total_bits // outWidth
-            totalIters = max(numInWords, numOutWords)
+            # totalIters = max(numInWords, numOutWords)
 
-            if outWidth > inWidth:
-                totalIters += int(np.floor(outWidth / inWidth) + 1) - 1
+            # if outWidth > inWidth:
+            #    totalIters += int(np.floor(outWidth / inWidth) + 1) - 1
 
             # in0 -> (DWCs) -> IODMA -> AXI MM
             # DWCs depend on AXI MM and out interface width
@@ -313,7 +312,6 @@ def docompute(self):
                         outWidth,
                         numInWords,
                         numOutWords,
-                        totalIters,
                         "in0_" + self.hls_sname(),
                         "dwc2dma",
                     ),

src/finn/custom_op/fpgadataflow/hls/streamingdatawidthconverter_hls.py

@@ -26,10 +26,11 @@
 # OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
+import math
 import numpy as np
 import os
 from qonnx.core.datatype import DataType
-import math
+
 from finn.custom_op.fpgadataflow.hlsbackend import HLSBackend
 from finn.custom_op.fpgadataflow.streamingdatawidthconverter import (
     StreamingDataWidthConverter,
@@ -60,30 +61,19 @@ def defines(self, var):
         # so we use numReps to represent the first 2 dimensions
         # + batching if shape[0] != 1
         numReps = int(np.prod(self.get_folded_input_shape()[:-2]))
-        # numReps = 1
 
         # assuming folded shapes are at least 2 dim-long
         numInWords = int(np.prod(self.get_folded_input_shape()[-2:-1]))
         numOutWords = int(np.prod(self.get_folded_output_shape()[-2:-1]))
 
-        # numInWords = int(np.prod(self.get_folded_input_shape()[-2:]))
-        # numOutWords = int(np.prod(self.get_folded_output_shape()[-2:]))
-
         inWidth = self.get_nodeattr("inWidth")
         outWidth = self.get_nodeattr("outWidth")
-        totalIters = max(numInWords, numOutWords)
-
-        # if we are building up a word, the overall loop count is longer
-        if outWidth > inWidth:
-            totalIters += int(np.floor(outWidth / inWidth) + 1) - 1
-
 
         self.code_gen_dict["$DEFINES$"] = [
             "#define InWidth %d " % inWidth,
             "#define OutWidth %d " % outWidth,
             "#define NumInWords %d " % numInWords,
             "#define NumOutWords %d " % numOutWords,
-            "#define totalIters %d " % totalIters,
             "#define numReps %d" % numReps,
         ]
 
@@ -106,8 +96,8 @@ def docompute(self):
         op = "StreamingDataWidthConverterGeneralized_Batch"
 
         self.code_gen_dict["$DOCOMPUTE$"] = [
-            "%s<InWidth, OutWidth, NumInWords,NumOutWords," % op
-            + " totalIters>(in0_%s, out_%s, numReps);" % (self.hls_sname(), self.hls_sname())
+            "%s<InWidth, OutWidth, NumInWords,NumOutWords" % op
+            + ">(in0_%s, out_%s, numReps);" % (self.hls_sname(), self.hls_sname())
         ]
 
     def blackboxfunction(self):
@@ -243,7 +233,6 @@ def execute_node(self, context, graph):
         ), """Output
         shape doesn't match expected shape, should be same as input shape"""
 
-
     def lut_estimation(self):
         """Calculates resource estimations for LUTs"""
 
@@ -270,6 +259,6 @@ def lut_estimation(self):
         cset_luts += intw + outw
 
         # generalized DWC cost penalty, this value is temporary
-        cnt_luts *=8
+        cnt_luts *= 8
 
-        return int(cnt_luts + cset_luts)
+        return int(cnt_luts + cset_luts)