pass in partition step parameter

[ieee754fpu.git] / src / ieee754 / part_mul_add / multiply.py

diff --git a/src/ieee754/part_mul_add/multiply.py b/src/ieee754/part_mul_add/multiply.py
index aef5c16e1a1a5db0c402857c41a68876ce5f12f4..b5706014a3b404fbfa72978a46910bfe337d16a9 100644 (file)
--- a/src/ieee754/part_mul_add/multiply.py
+++ b/src/ieee754/part_mul_add/multiply.py
@@ -346,7 +346,9 @@ class FinalAdd(Elaboratable):
     """ Final stage of add reduce
     """
 
-    def __init__(self, n_inputs, output_width, n_parts, partition_points):
+    def __init__(self, n_inputs, output_width, n_parts, partition_points,
+                       partition_step=1):
+        self.partition_step = partition_step
         self.output_width = output_width
         self.n_inputs = n_inputs
         self.n_parts = n_parts
@@ -381,7 +383,7 @@ class FinalAdd(Elaboratable):
             # base case for adding 2 inputs
             assert self.n_inputs == 2
             adder = PartitionedAdder(output_width,
-                                     self.i.part_pts, 2)
+                                     self.i.part_pts, self.partition_step)
             m.submodules.final_adder = adder
             m.d.comb += adder.a.eq(self.i.terms[0])
             m.d.comb += adder.b.eq(self.i.terms[1])
@@ -543,18 +545,19 @@ class AddReduceInternal:
         supported, except for by ``Signal.eq``.
     """
 
-    def __init__(self, inputs, output_width, partition_points,
-                       part_ops):
+    def __init__(self, i, output_width, partition_step=1):
         """Create an ``AddReduce``.
 
         :param inputs: input ``Signal``s to be summed.
         :param output_width: bit-width of ``output``.
         :param partition_points: the input partition points.
         """
-        self.inputs = inputs
-        self.part_ops = part_ops
+        self.i = i
+        self.inputs = i.terms
+        self.part_ops = i.part_ops
         self.output_width = output_width
-        self.partition_points = partition_points
+        self.partition_points = i.part_pts
+        self.partition_step = partition_step
 
         self.create_levels()
 
@@ -580,7 +583,7 @@ class AddReduceInternal:
             part_ops = next_level.i.part_ops
 
         next_level = FinalAdd(ilen, self.output_width, n_parts,
-                              partition_points)
+                              partition_points, self.partition_step)
         mods.append(next_level)
 
         self.levels = mods
@@ -598,8 +601,8 @@ class AddReduce(AddReduceInternal, Elaboratable):
         supported, except for by ``Signal.eq``.
     """
 
-    def __init__(self, inputs, output_width, register_levels, partition_points,
-                       part_ops):
+    def __init__(self, inputs, output_width, register_levels, part_pts,
+                       part_ops, partition_step=1):
         """Create an ``AddReduce``.
 
         :param inputs: input ``Signal``s to be summed.
@@ -608,10 +611,14 @@ class AddReduce(AddReduceInternal, Elaboratable):
             pipeline registers.
         :param partition_points: the input partition points.
         """
-        AddReduceInternal.__init__(self, inputs, output_width,
-                                   partition_points, part_ops)
+        self._inputs = inputs
+        self._part_pts = part_pts
+        self._part_ops = part_ops
         n_parts = len(part_ops)
-        self.o = FinalReduceData(partition_points, output_width, n_parts)
+        self.i = AddReduceData(part_pts, len(inputs),
+                             output_width, n_parts)
+        AddReduceInternal.__init__(self, self.i, output_width, partition_step)
+        self.o = FinalReduceData(part_pts, output_width, n_parts)
         self.register_levels = register_levels
 
     @staticmethod
@@ -629,17 +636,12 @@ class AddReduce(AddReduceInternal, Elaboratable):
         """Elaborate this module."""
         m = Module()
 
+        m.d.comb += self.i.eq_from(self._part_pts, self._inputs, self._part_ops)
+
         for i, next_level in enumerate(self.levels):
             setattr(m.submodules, "next_level%d" % i, next_level)
 
-        partition_points = self.partition_points
-        inputs = self.inputs
-        part_ops = self.part_ops
-        n_parts = len(part_ops)
-        n_inputs = len(inputs)
-        output_width = self.output_width
-        i = AddReduceData(partition_points, n_inputs, output_width, n_parts)
-        m.d.comb += i.eq_from(partition_points, inputs, part_ops)
+        i = self.i
         for idx in range(len(self.levels)):
             mcur = self.levels[idx]
             if idx in self.register_levels:
@@ -1387,7 +1389,8 @@ class Mul8_16_32_64(Elaboratable):
                                128,
                                self.register_levels,
                                t.o.part_pts,
-                               t.o.part_ops)
+                               t.o.part_ops,
+                               partition_step=2)
 
         m.submodules.add_reduce = add_reduce