Corrected 1x1 kernel edge case, updated testbenches for additional layer filenames

Author: Brmanzo

rtl/blocks/conv_layer.sv

@@ -57,8 +57,8 @@ module conv_layer #(
 
    ,localparam int unsigned PaddedWidth  = LineWidthPx + (2 * Padding)
    ,localparam int unsigned PaddedHeight = LineCountPx + (2 * Padding)
-   ,localparam int XBits = (LineWidthPx <= 1) ? 1 : $clog2(PaddedWidth + 1)
-   ,localparam int YBits = (LineCountPx <= 1) ? 1 : $clog2(PaddedHeight + 1)
+   ,localparam int XBits = $clog2(PaddedWidth + 1)
+   ,localparam int YBits = $clog2(PaddedHeight + 1)
 
    ,localparam int unsigned WeightIndex = InChannels * KernelArea * WeightBits
    ,parameter logic signed [OutChannels*InChannels*KernelWidth*KernelWidth*WeightBits-1:0] Weights = '0

sim/functional_models/cnn_model.py

@@ -203,15 +203,28 @@ def step(self, x: List[int]):
 
         # Return flattened results (e.g. [class_id1, class_id2, ...])
         # For classifier, res item is (class_id, logits)
-        return [item[0] for item in current_burst]
+        final_res = [item[0] for item in current_burst]
+        if final_res:
+            print("DEBUG CNNModel step returned:", final_res)
+        return final_res
 
     def consume(self):
         """Standard ModelRunner interface for top-level DUT."""
         from util.bitwise import unpack_terms
         packed = int(self._dut.data_i.value.integer)
         is_unsigned = self._InBits > 2
         raw_val = unpack_terms(packed, self._InBits, self._InChannels, signed=not is_unsigned)
-        return self.step(raw_val)
+        
+        if not hasattr(self, '_consume_count'):
+            self._consume_count = 0
+        self._consume_count += 1
+        
+        res = self.step(raw_val)
+        if res is not None:
+            print(f"DEBUG: CNNModel.consume() produced {res} on call {self._consume_count}")
+        elif self._consume_count == 784:
+            print(f"DEBUG: CNNModel.consume() produced None on call 784!!!")
+        return res
 
     def produce(self, expected):
         """Standard ModelRunner interface for top-level DUT."""

sim/integration_testing/cnn/test_cnn.py

@@ -221,7 +221,12 @@ def _write_vh(path: str, name: str, total_bits: int, raw_val: int) -> None:
     # 4. Run simulator
     os.environ["CHECK_INFERENCE"] = os.environ.get("CHECK_INFERENCE", "1")
     os.environ["SAMPLE_IDX"]      = os.environ.get("SAMPLE_IDX", "10")
-    
+
+    # Remove the single-layer fallback keys added by inject_weights_and_biases;
+    # cnn.sv only has FileName_0…FileName_N, not plain FileName / FileName_hi.
+    params.pop("FileName", None)
+    params.pop("FileName_hi", None)
+
     runner(
         simulator=simulator,
         timescale="1ps/1ps",

sim/integration_testing/cnn_framed/tb_cnn_framed.sv

@@ -7,10 +7,10 @@ module cnn_framed #(
   parameter int unsigned BusBits = 8
   ,parameter              FileName_0    = "nn/data/roms/hex/zeros.hex"
   ,parameter              FileName_1    = "nn/data/roms/hex/zeros.hex"
-  ,parameter              FileName_1_hi = "nn/data/roms/hex/zeros.hex"
   ,parameter              FileName_2    = "nn/data/roms/hex/zeros.hex"
-  ,parameter              FileName_2_hi = "nn/data/roms/hex/zeros.hex"
   ,parameter              FileName_3    = "nn/data/roms/hex/zeros.hex"
+  ,parameter              FileName_4    = "nn/data/roms/hex/zeros.hex"
+  ,parameter              FileName_5    = "nn/data/roms/hex/zeros.hex"
 
   ,parameter  int unsigned WidthIn   = 320
   ,parameter  int unsigned HeightIn  = 240
@@ -61,10 +61,10 @@ module cnn_framed #(
      .BusBits      (BusBits)
     ,.FileName_0   (FileName_0)
     ,.FileName_1   (FileName_1)
-    ,.FileName_1_hi(FileName_1_hi)
     ,.FileName_2   (FileName_2)
-    ,.FileName_2_hi(FileName_2_hi)
     ,.FileName_3   (FileName_3)
+    ,.FileName_4   (FileName_4)
+    ,.FileName_5   (FileName_5)
   ) dut (
      .clk_i   (clk_i)
     ,.rst_i   (rst_i)

sim/integration_testing/cnn_framed/test_cnn_framed.py

@@ -252,6 +252,9 @@ def _write_vh(path: str, name: str, total_bits: int, raw_val: int) -> None:
     os.environ["CHECK_INFERENCE"] = os.environ.get("CHECK_INFERENCE", "1")
     os.environ["SAMPLE_IDX"]      = os.environ.get("SAMPLE_IDX", "10")
 
+    params.pop("FileName", None)
+    params.pop("FileName_hi", None)
+
     runner(
         simulator=simulator,
         timescale="1ps/1ps",

sim/integration_testing/cnn_uart/test_cnn_uart.py

@@ -253,6 +253,9 @@ def _write_vh(path: str, name: str, total_bits: int, raw_val: int) -> None:
 
     os.environ["SAMPLE_IDX"] = os.environ.get("SAMPLE_IDX", "10")
 
+    params.pop("FileName", None)
+    params.pop("FileName_hi", None)
+
     runner(
         simulator=simulator,
         timescale="1ps/1ps",

sim/util/components.py

@@ -100,22 +100,24 @@ async def _run_input(self):
             if self._rst_i.value.is_resolvable and int(self._rst_i.value) == 1:
                 continue
 
-            # Evaluate Input Handshake
             v = self._valid_i.value == 1 if self._valid_i is not None else True
             r = self._ready_o.value == 1 if self._ready_o is not None else True
 
-            if not (v and r):
-                continue
-
-            expected = self._model.consume()
-
-            # Restore backward compatibility for single items vs lists
-            if expected is not None:
-                if isinstance(expected, (list, tuple)):
-                    for item in expected:
-                        self._events.put(item)
-                else:
-                    self._events.put(expected)
+            if not hasattr(self, '_run_input_calls'):
+                self._run_input_calls = 0
+            
+            if v and r:
+                self._run_input_calls += 1
+                if self._run_input_calls % 100 == 0 or self._run_input_calls == 1 or self._run_input_calls >= 780:
+                    print(f"DEBUG: ModelRunner v and r is True on call {self._run_input_calls}")
+                
+                expected = self._model.consume()
+                if expected is not None:
+                    if isinstance(expected, (list, tuple)):
+                        for item in expected:
+                            self._events.put(item)
+                    else:
+                        self._events.put(expected)
 
     async def _run_output(self):
         from decimal import Decimal