def __init__(self, part_pts, n_inputs, output_width, n_parts):
self.part_ops = [Signal(2, name=f"part_ops_{i}", reset_less=True)
for i in range(n_parts)]
- self.terms = [Signal(output_width, name=f"inputs_{i}",
+ self.terms = [Signal(output_width, name=f"terms_{i}",
reset_less=True)
for i in range(n_inputs)]
self.part_pts = part_pts.like()
return self.eq_from(rhs.part_pts, rhs.output, rhs.part_ops)
-class FinalAdd(Elaboratable):
+class FinalAdd(PipeModBase):
""" Final stage of add reduce
"""
- def __init__(self, lidx, n_inputs, output_width, n_parts, partition_points,
+ def __init__(self, pspec, lidx, n_inputs, partition_points,
partition_step=1):
self.lidx = lidx
self.partition_step = partition_step
- self.output_width = output_width
+ self.output_width = pspec.width * 2
self.n_inputs = n_inputs
- self.n_parts = n_parts
+ self.n_parts = pspec.n_parts
self.partition_points = PartitionPoints(partition_points)
- if not self.partition_points.fits_in_width(output_width):
+ if not self.partition_points.fits_in_width(self.output_width):
raise ValueError("partition_points doesn't fit in output_width")
- self.i = self.ispec()
- self.o = self.ospec()
+ super().__init__(pspec, "finaladd")
def ispec(self):
return AddReduceData(self.partition_points, self.n_inputs,
return FinalReduceData(self.partition_points,
self.output_width, self.n_parts)
- def setup(self, m, i):
- m.submodules.finaladd = self
- m.d.comb += self.i.eq(i)
-
- def process(self, i):
- return self.o
-
def elaborate(self, platform):
"""Elaborate this module."""
m = Module()
class AddReduceInternal:
- """Recursively Add list of numbers together.
+ """Iteratively Add list of numbers together.
:attribute inputs: input ``Signal``s to be summed. Modification not
supported, except for by ``Signal.eq``.
supported, except for by ``Signal.eq``.
"""
- def __init__(self, i, pspec, partition_step=1):
+ def __init__(self, pspec, n_inputs, part_pts, 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.i = i
self.pspec = pspec
- self.inputs = i.terms
- self.part_ops = i.part_ops
+ self.n_inputs = n_inputs
self.output_width = pspec.width * 2
- self.partition_points = i.part_pts
+ self.partition_points = part_pts
self.partition_step = partition_step
self.create_levels()
mods = []
partition_points = self.partition_points
- part_ops = self.part_ops
- n_parts = len(part_ops)
- inputs = self.inputs
- ilen = len(inputs)
+ ilen = self.n_inputs
while True:
- groups = AddReduceSingle.full_adder_groups(len(inputs))
+ groups = AddReduceSingle.full_adder_groups(ilen)
if len(groups) == 0:
break
lidx = len(mods)
self.partition_step)
mods.append(next_level)
partition_points = next_level.i.part_pts
- inputs = next_level.o.terms
- ilen = len(inputs)
- part_ops = next_level.i.part_ops
+ ilen = len(next_level.o.terms)
lidx = len(mods)
- next_level = FinalAdd(lidx, ilen, self.output_width, n_parts,
+ next_level = FinalAdd(self.pspec, lidx, ilen,
partition_points, self.partition_step)
mods.append(next_level)
n_parts = len(part_ops)
self.i = AddReduceData(part_pts, len(inputs),
output_width, n_parts)
- AddReduceInternal.__init__(self, self.i, output_width, partition_step)
+ AddReduceInternal.__init__(self, pspec, n_inputs, part_pts,
+ partition_step)
self.o = FinalReduceData(part_pts, output_width, n_parts)
self.register_levels = register_levels
return m
-class FinalOut(Elaboratable):
+class FinalOut(PipeModBase):
""" selects the final output based on the partitioning.
each byte is selectable independently, i.e. it is possible
that some partitions requested 8-bit computation whilst others
requested 16 or 32 bit.
"""
- def __init__(self, output_width, n_parts, part_pts):
+ def __init__(self, pspec, part_pts):
+
self.part_pts = part_pts
- self.output_width = output_width
- self.n_parts = n_parts
- self.out_wid = output_width//2
+ self.output_width = pspec.width * 2
+ self.n_parts = pspec.n_parts
+ self.out_wid = pspec.width
- self.i = self.ispec()
- self.o = self.ospec()
+ super().__init__(pspec, "finalout")
def ispec(self):
return IntermediateData(self.part_pts, self.output_width, self.n_parts)
def ospec(self):
return OutputData()
- def setup(self, m, i):
- m.submodules.finalout = self
- m.d.comb += self.i.eq(i)
-
- def process(self, i):
- return self.o
-
def elaborate(self, platform):
m = Module()
return m
-class Intermediates(Elaboratable):
+class Intermediates(PipeModBase):
""" Intermediate output modules
"""
- def __init__(self, output_width, n_parts, part_pts):
+ def __init__(self, pspec, part_pts):
self.part_pts = part_pts
- self.output_width = output_width
- self.n_parts = n_parts
+ self.output_width = pspec.width * 2
+ self.n_parts = pspec.n_parts
- self.i = self.ispec()
- self.o = self.ospec()
+ super().__init__(pspec, "intermediates")
def ispec(self):
return FinalReduceData(self.part_pts, self.output_width, self.n_parts)
def ospec(self):
return IntermediateData(self.part_pts, self.output_width, self.n_parts)
- def setup(self, m, i):
- m.submodules.intermediates = self
- m.d.comb += self.i.eq(i)
-
- def process(self, i):
- return self.o
-
def elaborate(self, platform):
m = Module()
class Mul8_16_32_64(Elaboratable):
"""Signed/Unsigned 8/16/32/64-bit partitioned integer multiplier.
+ XXX NOTE: this class is intended for unit test purposes ONLY.
+
Supports partitioning into any combination of 8, 16, 32, and 64-bit
partitions on naturally-aligned boundaries. Supports the operation being
set for each partition independently.
part_pts = self.part_pts
- n_parts = self.pspec.n_parts
n_inputs = 64 + 4
- output_width = self.pspec.width * 2
t = AllTerms(self.pspec, n_inputs)
t.setup(m, self.i)
terms = t.o.terms
- at = AddReduceInternal(t.process(self.i), self.pspec, partition_step=2)
+ at = AddReduceInternal(self.pspec, n_inputs, part_pts, partition_step=2)
- i = at.i
+ i = t.o
for idx in range(len(at.levels)):
mcur = at.levels[idx]
mcur.setup(m, i)
m.d.comb += o.eq(mcur.process(i))
i = o # for next loop
- interm = Intermediates(128, 8, part_pts)
+ interm = Intermediates(self.pspec, part_pts)
interm.setup(m, i)
o = interm.process(interm.i)
# final output
- finalout = FinalOut(128, 8, part_pts)
+ finalout = FinalOut(self.pspec, part_pts)
finalout.setup(m, o)
m.d.comb += self.o.eq(finalout.process(o))
projects / ieee754fpu.git / blobdiff