artiq/artiq/compiler/math_fns.py
David Nadlinger 9ff47bacab compiler: Provide libm special functions (erf, Bessel functions, …)
Tests hard-depend on SciPy to make sure this is exercised
during CI.
2020-11-11 19:15:30 +01:00

133 lines
3.9 KiB
Python

r"""
The :mod:`math_fns` module lists math-related functions from NumPy recognized
by the ARTIQ compiler so host function objects can be :func:`match`\ ed to
the compiler type metadata describing their core device analogue.
"""
from collections import OrderedDict
import numpy
from . import builtins, types
# Some special mathematical functions are exposed via their scipy.special
# equivalents. Since the rest of the ARTIQ core does not depend on SciPy,
# gracefully handle it not being present, making the functions simply not
# available.
try:
import scipy.special as scipy_special
except ImportError:
scipy_special = None
#: float -> float numpy.* math functions for which llvm.* intrinsics exist.
unary_fp_intrinsics = [(name, "llvm." + name + ".f64") for name in [
"sin",
"cos",
"exp",
"exp2",
"log",
"log10",
"log2",
"fabs",
"floor",
"ceil",
"trunc",
"sqrt",
]] + [
# numpy.rint() seems to (NumPy 1.19.0, Python 3.8.5, Linux x86_64)
# implement round-to-even, but unfortunately, rust-lang/libm only
# provides round(), which always rounds away from zero.
#
# As there is no equivalent of the latter in NumPy (nor any other
# basic rounding function), expose round() as numpy.rint anyway,
# even if the rounding modes don't match up, so there is some way
# to do rounding on the core device. (numpy.round() has entirely
# different semantics; it rounds to a configurable number of
# decimals.)
("rint", "llvm.round.f64"),
]
#: float -> float numpy.* math functions lowered to runtime calls.
unary_fp_runtime_calls = [
("tan", "tan"),
("arcsin", "asin"),
("arccos", "acos"),
("arctan", "atan"),
("sinh", "sinh"),
("cosh", "cosh"),
("tanh", "tanh"),
("arcsinh", "asinh"),
("arccosh", "acosh"),
("arctanh", "atanh"),
("expm1", "expm1"),
("cbrt", "cbrt"),
]
#: float -> float numpy.* math functions lowered to runtime calls.
unary_fp_runtime_calls = [
("tan", "tan"),
("arcsin", "asin"),
("arccos", "acos"),
("arctan", "atan"),
("sinh", "sinh"),
("cosh", "cosh"),
("tanh", "tanh"),
("arcsinh", "asinh"),
("arccosh", "acosh"),
("arctanh", "atanh"),
("expm1", "expm1"),
("cbrt", "cbrt"),
]
scipy_special_unary_runtime_calls = [
("erf", "erf"),
("erfc", "erfc"),
("gamma", "tgamma"),
("gammaln", "lgamma"),
("j0", "j0"),
("j1", "j1"),
("y0", "y0"),
("y1", "y1"),
]
# Not mapped: jv/yv, libm only supports integer orders.
#: (float, float) -> float numpy.* math functions lowered to runtime calls.
binary_fp_runtime_calls = [
("arctan2", "atan2"),
("copysign", "copysign"),
("fmax", "fmax"),
("fmin", "fmin"),
# ("ldexp", "ldexp"), # One argument is an int; would need a bit more plumbing.
("hypot", "hypot"),
("nextafter", "nextafter"),
]
#: Array handling builtins (special treatment due to allocations).
numpy_builtins = ["transpose"]
def fp_runtime_type(name, arity):
args = [("arg{}".format(i), builtins.TFloat()) for i in range(arity)]
return types.TExternalFunction(
OrderedDict(args),
builtins.TFloat(),
name,
# errno isn't observable from ARTIQ Python.
flags={"nounwind", "nowrite"},
broadcast_across_arrays=True)
math_fn_map = {
getattr(numpy, symbol): fp_runtime_type(mangle, arity=1)
for symbol, mangle in (unary_fp_intrinsics + unary_fp_runtime_calls)
}
for symbol, mangle in binary_fp_runtime_calls:
math_fn_map[getattr(numpy, symbol)] = fp_runtime_type(mangle, arity=2)
for name in numpy_builtins:
math_fn_map[getattr(numpy, name)] = types.TBuiltinFunction("numpy." + name)
if scipy_special is not None:
for symbol, mangle in scipy_special_unary_runtime_calls:
math_fn_map[getattr(scipy_special, symbol)] = fp_runtime_type(mangle, arity=1)
def match(obj):
return math_fn_map.get(obj, None)