Fix quant fused conv op snippet bug

uTensor · Jul 11, 2019 · f25eac0 · f25eac0
1 parent f56b40b
commit f25eac0
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 8 deletions.
diff --git a/utensor_cgen/backend/operators.py b/utensor_cgen/backend/operators.py
@@ -494,6 +494,7 @@ class _QuantizedFusedConv2DMaxpoolOperator(_Operator):
 
   def __init__(self, op_info, **kwargs):
     _Operator.__init__(self)
+    # import pdb; pdb.set_trace()
     inputs = [tensor_info.name for tensor_info in op_info.input_tensors]
     outputs = [tensor_info.name for tensor_info in op_info.output_tensors]
     in_dtype, filter_dtype = (op_info.input_tensors[0].dtype,
@@ -504,12 +505,12 @@ def __init__(self, op_info, **kwargs):
     padding = op_info.op_attr['_utensor_conv']["padding"].value.decode('utf8')
     parser = NamescopedKWArgsParser(RefCntOptimizer.KWARGS_NAMESCOPE,
                                     op_info.op_attr)
-    ref_count = parser.get('ref_counts', [0])[0]
+    ref_counts = parser.get('ref_counts', None)
     to_eval = parser.get('to_eval', False)
     self._snippet = QuantizedFusedConv2DMaxpoolOpSnippet(
       inputs, outputs, strides, ksize, padding,
       in_dtype=in_dtype, filter_dtype=filter_dtype, out_dtypes=out_dtypes,
-      ref_count=ref_count, to_eval=to_eval
+      ref_counts=ref_counts, to_eval=to_eval
     )
 
 

diff --git a/utensor_cgen/backend/snippets/_snippets.py b/utensor_cgen/backend/snippets/_snippets.py
@@ -671,18 +671,19 @@ class QuantizedFusedConv2DMaxpoolOpSnippet(Snippet):
 
   def __init__(self, inputs, outputs, strides, ksize, padding,
                in_dtype, filter_dtype, out_dtypes,
-               ref_count=0,
+               ref_counts=None,
                to_eval=False):
+    # import pdb; pdb.set_trace()
     Snippet.__init__(self)
-    if ref_count:
-      self.template_vars["ref_count"] = ref_count
+    if ref_counts:
+      self.template_vars["ref_counts"] = ref_counts
     self.template_vars["inputs"] = inputs
     self.template_vars["outputs"] = outputs
     self.template_vars["in_dtype"] = NP_TYPES_MAP[in_dtype].tensor_type_str
     self.template_vars["filter_dtype"] = NP_TYPES_MAP[filter_dtype].tensor_type_str
     self.template_vars["out_dtypes"] = [
-      NP_TYPES_MAP[out_dtype].tensor_type_str
-      for out_dtype in out_dtypes
+      NP_TYPES_MAP[dtype].tensor_type_str
+      for dtype in out_dtypes
     ]
     self.template_vars["strides"] = strides
     self.template_vars["ksize"] = ksize

diff --git a/utensor_cgen/backend/snippets/templates/snippets/quantized_fused_conv2d_maxpool_op.cpp b/utensor_cgen/backend/snippets/templates/snippets/quantized_fused_conv2d_maxpool_op.cpp
@@ -8,7 +8,7 @@
     ctx.add(new RamTensor<{{out_dtypes[1]}}>({1}), "{{outputs[1]}}");
     ctx.add(new RamTensor<{{out_dtypes[2]}}>({1}), "{{outputs[2]}}");
     {% endif %}
-    ctx.push(new QuantizedFusedConvMaxpoolOp<{{in_dtype}}, {{filter_dtype}}, {{out_dtype}}>({ {% for s in strides[:-1]%}{{s}}, {%endfor%}{{strides[-1]}} }, { {% for s in ksize[:-1]%}{{s}}, {%endfor%}{{ksize[-1]}} },{{padding}}),
+    ctx.push(new QuantizedFusedConvMaxpoolOp<{{in_dtype}}, {{filter_dtype}}, {{out_dtypes[0]}}>({ {% for s in strides[:-1]%}{{s}}, {%endfor%}{{strides[-1]}} }, { {% for s in ksize[:-1]%}{{s}}, {%endfor%}{{ksize[-1]}} },{{padding}}),
              { {% for tname in inputs[:-1]%}"{{tname}}", {%endfor%}"{{inputs[-1]}}" },
              { {% for tname in outputs[:-1]%}"{{tname}}", {%endfor%}"{{outputs[-1]}}" });
     {% if to_eval %}