forked from M-Labs/artiq
1
0
Fork 0

sawg: extract spline

This commit is contained in:
Robert Jördens 2016-12-04 16:52:08 +01:00
parent 39becd0b4e
commit 695eb705b3
2 changed files with 216 additions and 215 deletions

View File

@ -1,218 +1,4 @@
from numpy import int32, int64
from artiq.language.core import kernel, now_mu, portable, delay
from artiq.coredevice.rtio import rtio_output, rtio_output_wide
from artiq.language.types import TInt32, TInt64, TFloat
class Spline:
"""Spline interpolating RTIO channel.
One knot of a polynomial basis spline (B-spline) :math:`u(t)`
is defined by the coefficients :math:`u_n` up to order :math:`n = k`.
If the knot is evaluated starting at time :math:`t_0`, the output
:math:`u(t)` for :math:`t > t_0, t_0` is:
.. math::
u(t) = \sum_{n=0}^k \frac{u_n}{n!} (t - t_0)^n
= u_0 + u_1 (t - t_0) + \frac{u_2}{2} (t - t_0)^2 + \dots
"""
kernel_invariants = {"channel", "core", "scale", "width",
"time_width", "time_scale"}
def __init__(self, width, time_width, channel, core_device, scale=1.):
self.core = core_device
self.channel = channel
self.width = width
self.scale = float((int64(1) << width) * scale)
self.time_width = time_width
self.time_scale = float((1 << time_width) *
core_device.coarse_ref_period)
@portable(flags={"fast-math"})
def to_mu(self, value: TFloat) -> TInt32:
"""Convert floating point `value` from physical units to 32 bit
integer machine units."""
return int32(round(value*self.scale))
@portable(flags={"fast-math"})
def from_mu(self, value: TInt32) -> TFloat:
"""Convert 32 bit integer `value` from machine units to floating point
physical units."""
return value/self.scale
@portable(flags={"fast-math"})
def to_mu64(self, value: TFloat) -> TInt64:
"""Convert floating point `value` from physical units to 64 bit
integer machine units."""
return int64(round(value*self.scale))
@kernel
def set_mu(self, value: TInt32):
"""Set spline value (machine units).
:param value: Spline value in integer machine units.
"""
rtio_output(now_mu(), self.channel, 0, value)
@kernel(flags={"fast-math"})
def set(self, value: TFloat):
"""Set spline value.
:param value: Spline value relative to full-scale.
"""
if self.width > 32:
l = [int32(0)] * 2
self.pack_coeff_mu([self.to_mu64(value)], l)
rtio_output_wide(now_mu(), self.channel, 0, l)
else:
rtio_output(now_mu(), self.channel, 0, self.to_mu(value))
@kernel
def set_coeff_mu(self, value): # TList(TInt32)
"""Set spline raw values.
:param value: Spline packed raw values.
"""
rtio_output_wide(now_mu(), self.channel, 0, value)
@portable(flags={"fast-math"})
def pack_coeff_mu(self, coeff, packed): # TList(TInt64), TList(TInt32)
"""Pack coefficients into RTIO data
:param coeff: TList(TInt64) list of machine units spline coefficients.
Lowest (zeroth) order first. The coefficient list is zero-extended
by the RTIO gateware.
:param packed: TList(TInt32) list for packed RTIO data. Must be
pre-allocated. Length in bits is
`n*width + (n - 1)*n//2*time_width`
"""
pos = 0
for i in range(len(coeff)):
wi = self.width + i*self.time_width
ci = coeff[i]
while wi != 0:
j = pos//32
used = pos - 32*j
avail = 32 - used
if avail > wi:
avail = wi
cij = int32(ci)
if avail != 32:
cij &= (1 << avail) - 1
packed[j] |= cij << used
ci >>= avail
wi -= avail
pos += avail
@portable(flags={"fast-math"})
def coeff_to_mu(self, coeff, coeff64): # TList(TFloat), TList(TInt64)
"""Convert a floating point list of coefficients into a 64 bit
integer (preallocated).
:param coeff: TList(TFloat) list of coefficients in physical units.
:param coeff64: TList(TInt64) preallocated list of coefficients in
machine units.
"""
for i in range(len(coeff)):
vi = coeff[i] * self.scale
for j in range(i):
vi *= self.time_scale
ci = int64(round(vi))
coeff64[i] = ci
# artiq.wavesynth.coefficients.discrete_compensate:
if i == 2:
coeff64[1] += ci >> self.time_width + 1
elif i == 3:
coeff64[2] += ci >> self.time_width
coeff64[1] += ci // 6 >> 2*self.time_width
def coeff_as_packed_mu(self, coeff64):
"""Pack 64 bit integer machine units coefficients into 32 bit integer
RTIO data list.
This is a host-only method that can be used to generate packed
spline knot data to be frozen into kernels at compile time.
"""
n = len(coeff64)
width = n*self.width + (n - 1)*n//2*self.time_width
packed = [int32(0)] * ((width + 31)//32)
self.pack_coeff_mu(coeff64, packed)
return packed
def coeff_as_packed(self, coeff):
"""Convert floating point spline coefficients into 32 bit integer
packed data.
This is a host-only method that can be used to generate packed
spline knot data to be frozen into kernels at compile time.
"""
coeff64 = [int64(0)] * len(coeff)
self.coeff_to_mu(coeff, coeff64)
return self.coeff_as_packed_mu(coeff64)
@kernel(flags={"fast-math"})
def set_coeff(self, coeff): # TList(TFloat)
"""Set spline coefficients.
Missing coefficients (high order) are zero-extended byt the RTIO
gateware.
If more coefficients are supplied than the gateware supports the extra
coefficients are ignored.
:param value: List of floating point spline knot coefficients,
lowest order (constant) coefficient first. Units are the
unit of this spline's value times increasing powers of 1/s.
"""
n = len(coeff)
coeff64 = [int64(0)] * n
self.coeff_to_mu(coeff, coeff64)
width = n*self.width + (n - 1)*n//2*self.time_width
packed = [int32(0)] * ((width + 31)//32)
self.pack_coeff_mu(coeff64, packed)
self.set_coeff_mu(packed)
@kernel(flags={"fast-math"})
def smooth(self, start: TFloat, stop: TFloat, duration: TFloat,
order: TInt32):
"""Initiate an interpolated value change.
For zeroth order (step) interpolation, the step is at
`start + duration/2`.
First order interpolation corresponds to a linear value ramp from
`start` to `stop` over `duration`.
The third order interpolation is constrained to have zero first
order derivative at both `start` and `stop`.
For first order and third order interpolation (linear and cubic)
the interpolator needs to be stopped (or fed a new spline knot)
explicitly at the stop time.
This method advances the timeline by `duration`.
:param start: Initial value of the change. In physical units.
:param stop: Final value of the change. In physical units.
:param duration: Duration of the interpolation. In physical units.
:param order: Order of the interpolation. Only 0, 1,
and 3 are valid: step, linear, cubic.
"""
if order == 0:
delay(duration/2.)
self.set_coeff([stop])
delay(duration/2.)
elif order == 1:
self.set_coeff([start, (stop - start)/duration])
delay(duration)
elif order == 3:
v2 = 6.*(stop - start)/(duration*duration)
self.set_coeff([start, 0., v2, -2.*v2/duration])
delay(duration)
else:
raise ValueError("Invalid interpolation order. "
"Supported orders are: 0, 1, 3.")
from artiq.coredevice.spline import Spline
class SAWG:

215
artiq/coredevice/spline.py Normal file
View File

@ -0,0 +1,215 @@
from numpy import int32, int64
from artiq.language.core import kernel, now_mu, portable, delay
from artiq.coredevice.rtio import rtio_output, rtio_output_wide
from artiq.language.types import TInt32, TInt64, TFloat
class Spline:
"""Spline interpolating RTIO channel.
One knot of a polynomial basis spline (B-spline) :math:`u(t)`
is defined by the coefficients :math:`u_n` up to order :math:`n = k`.
If the knot is evaluated starting at time :math:`t_0`, the output
:math:`u(t)` for :math:`t > t_0, t_0` is:
.. math::
u(t) = \sum_{n=0}^k \frac{u_n}{n!} (t - t_0)^n
= u_0 + u_1 (t - t_0) + \frac{u_2}{2} (t - t_0)^2 + \dots
"""
kernel_invariants = {"channel", "core", "scale", "width",
"time_width", "time_scale"}
def __init__(self, width, time_width, channel, core_device, scale=1.):
self.core = core_device
self.channel = channel
self.width = width
self.scale = float((int64(1) << width) * scale)
self.time_width = time_width
self.time_scale = float((1 << time_width) *
core_device.coarse_ref_period)
@portable(flags={"fast-math"})
def to_mu(self, value: TFloat) -> TInt32:
"""Convert floating point `value` from physical units to 32 bit
integer machine units."""
return int32(round(value*self.scale))
@portable(flags={"fast-math"})
def from_mu(self, value: TInt32) -> TFloat:
"""Convert 32 bit integer `value` from machine units to floating point
physical units."""
return value/self.scale
@portable(flags={"fast-math"})
def to_mu64(self, value: TFloat) -> TInt64:
"""Convert floating point `value` from physical units to 64 bit
integer machine units."""
return int64(round(value*self.scale))
@kernel
def set_mu(self, value: TInt32):
"""Set spline value (machine units).
:param value: Spline value in integer machine units.
"""
rtio_output(now_mu(), self.channel, 0, value)
@kernel(flags={"fast-math"})
def set(self, value: TFloat):
"""Set spline value.
:param value: Spline value relative to full-scale.
"""
if self.width > 32:
l = [int32(0)] * 2
self.pack_coeff_mu([self.to_mu64(value)], l)
rtio_output_wide(now_mu(), self.channel, 0, l)
else:
rtio_output(now_mu(), self.channel, 0, self.to_mu(value))
@kernel
def set_coeff_mu(self, value): # TList(TInt32)
"""Set spline raw values.
:param value: Spline packed raw values.
"""
rtio_output_wide(now_mu(), self.channel, 0, value)
@portable(flags={"fast-math"})
def pack_coeff_mu(self, coeff, packed): # TList(TInt64), TList(TInt32)
"""Pack coefficients into RTIO data
:param coeff: TList(TInt64) list of machine units spline coefficients.
Lowest (zeroth) order first. The coefficient list is zero-extended
by the RTIO gateware.
:param packed: TList(TInt32) list for packed RTIO data. Must be
pre-allocated. Length in bits is
`n*width + (n - 1)*n//2*time_width`
"""
pos = 0
for i in range(len(coeff)):
wi = self.width + i*self.time_width
ci = coeff[i]
while wi != 0:
j = pos//32
used = pos - 32*j
avail = 32 - used
if avail > wi:
avail = wi
cij = int32(ci)
if avail != 32:
cij &= (1 << avail) - 1
packed[j] |= cij << used
ci >>= avail
wi -= avail
pos += avail
@portable(flags={"fast-math"})
def coeff_to_mu(self, coeff, coeff64): # TList(TFloat), TList(TInt64)
"""Convert a floating point list of coefficients into a 64 bit
integer (preallocated).
:param coeff: TList(TFloat) list of coefficients in physical units.
:param coeff64: TList(TInt64) preallocated list of coefficients in
machine units.
"""
for i in range(len(coeff)):
vi = coeff[i] * self.scale
for j in range(i):
vi *= self.time_scale
ci = int64(round(vi))
coeff64[i] = ci
# artiq.wavesynth.coefficients.discrete_compensate:
if i == 2:
coeff64[1] += ci >> self.time_width + 1
elif i == 3:
coeff64[2] += ci >> self.time_width
coeff64[1] += ci // 6 >> 2*self.time_width
def coeff_as_packed_mu(self, coeff64):
"""Pack 64 bit integer machine units coefficients into 32 bit integer
RTIO data list.
This is a host-only method that can be used to generate packed
spline knot data to be frozen into kernels at compile time.
"""
n = len(coeff64)
width = n*self.width + (n - 1)*n//2*self.time_width
packed = [int32(0)] * ((width + 31)//32)
self.pack_coeff_mu(coeff64, packed)
return packed
def coeff_as_packed(self, coeff):
"""Convert floating point spline coefficients into 32 bit integer
packed data.
This is a host-only method that can be used to generate packed
spline knot data to be frozen into kernels at compile time.
"""
coeff64 = [int64(0)] * len(coeff)
self.coeff_to_mu(coeff, coeff64)
return self.coeff_as_packed_mu(coeff64)
@kernel(flags={"fast-math"})
def set_coeff(self, coeff): # TList(TFloat)
"""Set spline coefficients.
Missing coefficients (high order) are zero-extended byt the RTIO
gateware.
If more coefficients are supplied than the gateware supports the extra
coefficients are ignored.
:param value: List of floating point spline knot coefficients,
lowest order (constant) coefficient first. Units are the
unit of this spline's value times increasing powers of 1/s.
"""
n = len(coeff)
coeff64 = [int64(0)] * n
self.coeff_to_mu(coeff, coeff64)
width = n*self.width + (n - 1)*n//2*self.time_width
packed = [int32(0)] * ((width + 31)//32)
self.pack_coeff_mu(coeff64, packed)
self.set_coeff_mu(packed)
@kernel(flags={"fast-math"})
def smooth(self, start: TFloat, stop: TFloat, duration: TFloat,
order: TInt32):
"""Initiate an interpolated value change.
For zeroth order (step) interpolation, the step is at
`start + duration/2`.
First order interpolation corresponds to a linear value ramp from
`start` to `stop` over `duration`.
The third order interpolation is constrained to have zero first
order derivative at both `start` and `stop`.
For first order and third order interpolation (linear and cubic)
the interpolator needs to be stopped (or fed a new spline knot)
explicitly at the stop time.
This method advances the timeline by `duration`.
:param start: Initial value of the change. In physical units.
:param stop: Final value of the change. In physical units.
:param duration: Duration of the interpolation. In physical units.
:param order: Order of the interpolation. Only 0, 1,
and 3 are valid: step, linear, cubic.
"""
if order == 0:
delay(duration/2.)
self.set_coeff([stop])
delay(duration/2.)
elif order == 1:
self.set_coeff([start, (stop - start)/duration])
delay(duration)
elif order == 3:
v2 = 6.*(stop - start)/(duration*duration)
self.set_coeff([start, 0., v2, -2.*v2/duration])
delay(duration)
else:
raise ValueError("Invalid interpolation order. "
"Supported orders are: 0, 1, 3.")