The remainder is the left-hand-side of the comparison minus the
right-hand-side of the comparison in the above formulas.
"""
-from nmigen import (Elaboratable, Module, Signal, Const, Mux)
+from nmigen import (Elaboratable, Module, Signal, Const, Mux, Cat)
import enum
return (self.bit_width + self.log2_radix - 1) // self.log2_radix
-class DivPipeCoreOperation(enum.IntEnum):
+class DivPipeCoreOperation(enum.Enum):
""" Operation for ``DivPipeCore``.
:attribute UDivRem: unsigned divide/remainder.
SqrtRem = 1
RSqrtRem = 2
+ def __int__(self):
+ """ Convert to int. """
+ return self.value
+
@classmethod
def create_signal(cls, *, src_loc_at=0, **kwargs):
""" Create a signal that can contain a ``DivPipeCoreOperation``. """
- return Signal(min=int(min(cls)),
- max=int(max(cls)),
+ return Signal(min=min(map(int, cls)),
+ max=max(map(int, cls)),
src_loc_at=(src_loc_at + 1),
- decoder=cls,
+ decoder=lambda v: str(cls(v)),
**kwargs)
m.d.comb += self.o.quotient_root.eq(0)
m.d.comb += self.o.root_times_radicand.eq(0)
- with m.If(self.i.operation == DivPipeCoreOperation.UDivRem):
+ with m.If(self.i.operation == int(DivPipeCoreOperation.UDivRem)):
m.d.comb += self.o.compare_lhs.eq(self.i.dividend
<< self.core_config.fract_width)
- with m.Elif(self.i.operation == DivPipeCoreOperation.SqrtRem):
+ with m.Elif(self.i.operation == int(DivPipeCoreOperation.SqrtRem)):
m.d.comb += self.o.compare_lhs.eq(
self.i.divisor_radicand << (self.core_config.fract_width * 2))
with m.Else(): # DivPipeCoreOperation.RSqrtRem
trial_compare_rhs_values = []
pass_flags = []
for trial_bits in range(radix):
- shifted_trial_bits = Const(trial_bits, log2_radix) << current_shift
- shifted_trial_bits_sqrd = shifted_trial_bits * shifted_trial_bits
+ tb = trial_bits << current_shift
+ tb_width = log2_radix + current_shift
+ shifted_trial_bits = Const(tb, tb_width)
+ shifted_trial_bits2 = Const(tb*2, tb_width+1)
+ shifted_trial_bits_sqrd = Const(tb * tb, tb_width * 2)
# UDivRem
div_rhs = self.i.compare_rhs
- div_factor1 = self.i.divisor_radicand * shifted_trial_bits
- div_rhs += div_factor1 << self.core_config.fract_width
+ if tb != 0: # no point adding stuff that's multiplied by zero
+ div_factor1 = self.i.divisor_radicand * shifted_trial_bits2
+ div_rhs += div_factor1 << self.core_config.fract_width
# SqrtRem
sqrt_rhs = self.i.compare_rhs
- sqrt_factor1 = self.i.quotient_root * (shifted_trial_bits << 1)
- sqrt_rhs += sqrt_factor1 << self.core_config.fract_width
- sqrt_factor2 = shifted_trial_bits_sqrd
- sqrt_rhs += sqrt_factor2 << self.core_config.fract_width
+ if tb != 0: # no point adding stuff that's multiplied by zero
+ sqrt_factor1 = self.i.quotient_root * shifted_trial_bits2
+ sqrt_rhs += sqrt_factor1 << self.core_config.fract_width
+ sqrt_factor2 = shifted_trial_bits_sqrd
+ sqrt_rhs += sqrt_factor2 << self.core_config.fract_width
# RSqrtRem
rsqrt_rhs = self.i.compare_rhs
- rsqrt_rhs += self.i.root_times_radicand * (shifted_trial_bits << 1)
- rsqrt_rhs += self.i.divisor_radicand * shifted_trial_bits_sqrd
+ if tb != 0: # no point adding stuff that's multiplied by zero
+ rsqrt_rhs += self.i.root_times_radicand * shifted_trial_bits2
+ rsqrt_rhs += self.i.divisor_radicand * shifted_trial_bits_sqrd
trial_compare_rhs = Signal.like(
- self.o.compare_rhs, name=f"trial_compare_rhs_{trial_bits}")
+ self.o.compare_rhs, name=f"trial_compare_rhs_{trial_bits}",
+ reset_less=True)
- with m.If(self.i.operation == DivPipeCoreOperation.UDivRem):
+ with m.If(self.i.operation == int(DivPipeCoreOperation.UDivRem)):
m.d.comb += trial_compare_rhs.eq(div_rhs)
- with m.Elif(self.i.operation == DivPipeCoreOperation.SqrtRem):
+ with m.Elif(self.i.operation == int(DivPipeCoreOperation.SqrtRem)):
m.d.comb += trial_compare_rhs.eq(sqrt_rhs)
with m.Else(): # DivPipeCoreOperation.RSqrtRem
m.d.comb += trial_compare_rhs.eq(rsqrt_rhs)
trial_compare_rhs_values.append(trial_compare_rhs)
- pass_flag = Signal(name=f"pass_flag_{trial_bits}")
+ pass_flag = Signal(name=f"pass_flag_{trial_bits}", reset_less=True)
m.d.comb += pass_flag.eq(self.i.compare_lhs >= trial_compare_rhs)
pass_flags.append(pass_flag)
# Assumes that pass_flag[0] is always set (since
# compare_lhs >= compare_rhs is a pipeline invariant).
- next_bits = Signal(log2_radix)
+ next_bits = Signal(log2_radix, reset_less=True)
for i in range(log2_radix):
bit_value = 1
for j in range(0, radix, 1 << i):
bit_value ^= pass_flags[j]
m.d.comb += next_bits.part(i, 1).eq(bit_value)
- next_compare_rhs = 0
+ next_compare_rhs = Signal(radix, reset_less=True)
+ l = []
for i in range(radix):
- next_flag = pass_flags[i + 1] if i + 1 < radix else 0
- next_compare_rhs |= Mux(pass_flags[i] & ~next_flag,
- trial_compare_rhs_values[i],
- 0)
-
- m.d.comb += self.o.compare_rhs.eq(next_compare_rhs)
+ next_flag = pass_flags[i + 1] if (i + 1 < radix) else Const(0)
+ flag = Signal(reset_less=True, name=f"flag{i}")
+ test = Signal(reset_less=True, name=f"test{i}")
+ # XXX TODO: check the width on this
+ m.d.comb += test.eq((pass_flags[i] & ~next_flag))
+ m.d.comb += flag.eq(Mux(test, trial_compare_rhs_values[i], 0))
+ l.append(flag)
+
+ m.d.comb += next_compare_rhs.eq(Cat(*l))
+ m.d.comb += self.o.compare_rhs.eq(next_compare_rhs.bool())
m.d.comb += self.o.root_times_radicand.eq(self.i.root_times_radicand
+ ((self.i.divisor_radicand
* next_bits)
projects / ieee754fpu.git / blobdiff