""" Julia code printer The `JuliaCodePrinter` converts SymPy expressions into Julia expressions. A complete code generator, which uses `julia_code` extensively, can be found in `sympy.utilities.codegen`. The `codegen` module can be used to generate complete source code files. """ from typing import Any, Dict as tDict from sympy.core import Mul, Pow, S, Rational from sympy.core.mul import _keep_coeff from sympy.printing.codeprinter import CodePrinter from sympy.printing.precedence import precedence, PRECEDENCE from re import search # List of known functions. First, those that have the same name in # SymPy and Julia. This is almost certainly incomplete! known_fcns_src1 = ["sin", "cos", "tan", "cot", "sec", "csc", "asin", "acos", "atan", "acot", "asec", "acsc", "sinh", "cosh", "tanh", "coth", "sech", "csch", "asinh", "acosh", "atanh", "acoth", "asech", "acsch", "sinc", "atan2", "sign", "floor", "log", "exp", "cbrt", "sqrt", "erf", "erfc", "erfi", "factorial", "gamma", "digamma", "trigamma", "polygamma", "beta", "airyai", "airyaiprime", "airybi", "airybiprime", "besselj", "bessely", "besseli", "besselk", "erfinv", "erfcinv"] # These functions have different names ("SymPy": "Julia"), more # generally a mapping to (argument_conditions, julia_function). known_fcns_src2 = { "Abs": "abs", "ceiling": "ceil", "conjugate": "conj", "hankel1": "hankelh1", "hankel2": "hankelh2", "im": "imag", "re": "real" } class JuliaCodePrinter(CodePrinter): """ A printer to convert expressions to strings of Julia code. """ printmethod = "_julia" language = "Julia" _operators = { 'and': '&&', 'or': '||', 'not': '!', } _default_settings = { 'order': None, 'full_prec': 'auto', 'precision': 17, 'user_functions': {}, 'human': True, 'allow_unknown_functions': False, 'contract': True, 'inline': True, } # type: tDict[str, Any] # Note: contract is for expressing tensors as loops (if True), or just # assignment (if False). FIXME: this should be looked a more carefully # for Julia. def __init__(self, settings={}): super().__init__(settings) self.known_functions = dict(zip(known_fcns_src1, known_fcns_src1)) self.known_functions.update(dict(known_fcns_src2)) userfuncs = settings.get('user_functions', {}) self.known_functions.update(userfuncs) def _rate_index_position(self, p): return p*5 def _get_statement(self, codestring): return "%s" % codestring def _get_comment(self, text): return "# {}".format(text) def _declare_number_const(self, name, value): return "const {} = {}".format(name, value) def _format_code(self, lines): return self.indent_code(lines) def _traverse_matrix_indices(self, mat): # Julia uses Fortran order (column-major) rows, cols = mat.shape return ((i, j) for j in range(cols) for i in range(rows)) def _get_loop_opening_ending(self, indices): open_lines = [] close_lines = [] for i in indices: # Julia arrays start at 1 and end at dimension var, start, stop = map(self._print, [i.label, i.lower + 1, i.upper + 1]) open_lines.append("for %s = %s:%s" % (var, start, stop)) close_lines.append("end") return open_lines, close_lines def _print_Mul(self, expr): # print complex numbers nicely in Julia if (expr.is_number and expr.is_imaginary and expr.as_coeff_Mul()[0].is_integer): return "%sim" % self._print(-S.ImaginaryUnit*expr) # cribbed from str.py prec = precedence(expr) c, e = expr.as_coeff_Mul() if c < 0: expr = _keep_coeff(-c, e) sign = "-" else: sign = "" a = [] # items in the numerator b = [] # items that are in the denominator (if any) pow_paren = [] # Will collect all pow with more than one base element and exp = -1 if self.order not in ('old', 'none'): args = expr.as_ordered_factors() else: # use make_args in case expr was something like -x -> x args = Mul.make_args(expr) # Gather args for numerator/denominator for item in args: if (item.is_commutative and item.is_Pow and item.exp.is_Rational and item.exp.is_negative): if item.exp != -1: b.append(Pow(item.base, -item.exp, evaluate=False)) else: if len(item.args[0].args) != 1 and isinstance(item.base, Mul): # To avoid situations like #14160 pow_paren.append(item) b.append(Pow(item.base, -item.exp)) elif item.is_Rational and item is not S.Infinity: if item.p != 1: a.append(Rational(item.p)) if item.q != 1: b.append(Rational(item.q)) else: a.append(item) a = a or [S.One] a_str = [self.parenthesize(x, prec) for x in a] b_str = [self.parenthesize(x, prec) for x in b] # To parenthesize Pow with exp = -1 and having more than one Symbol for item in pow_paren: if item.base in b: b_str[b.index(item.base)] = "(%s)" % b_str[b.index(item.base)] # from here it differs from str.py to deal with "*" and ".*" def multjoin(a, a_str): # here we probably are assuming the constants will come first r = a_str[0] for i in range(1, len(a)): mulsym = '*' if a[i-1].is_number else '.*' r = r + mulsym + a_str[i] return r if not b: return sign + multjoin(a, a_str) elif len(b) == 1: divsym = '/' if b[0].is_number else './' return sign + multjoin(a, a_str) + divsym + b_str[0] else: divsym = '/' if all(bi.is_number for bi in b) else './' return (sign + multjoin(a, a_str) + divsym + "(%s)" % multjoin(b, b_str)) def _print_Relational(self, expr): lhs_code = self._print(expr.lhs) rhs_code = self._print(expr.rhs) op = expr.rel_op return "{} {} {}".format(lhs_code, op, rhs_code) def _print_Pow(self, expr): powsymbol = '^' if all(x.is_number for x in expr.args) else '.^' PREC = precedence(expr) if expr.exp == S.Half: return "sqrt(%s)" % self._print(expr.base) if expr.is_commutative: if expr.exp == -S.Half: sym = '/' if expr.base.is_number else './' return "1" + sym + "sqrt(%s)" % self._print(expr.base) if expr.exp == -S.One: sym = '/' if expr.base.is_number else './' return "1" + sym + "%s" % self.parenthesize(expr.base, PREC) return '%s%s%s' % (self.parenthesize(expr.base, PREC), powsymbol, self.parenthesize(expr.exp, PREC)) def _print_MatPow(self, expr): PREC = precedence(expr) return '%s^%s' % (self.parenthesize(expr.base, PREC), self.parenthesize(expr.exp, PREC)) def _print_Pi(self, expr): if self._settings["inline"]: return "pi" else: return super()._print_NumberSymbol(expr) def _print_ImaginaryUnit(self, expr): return "im" def _print_Exp1(self, expr): if self._settings["inline"]: return "e" else: return super()._print_NumberSymbol(expr) def _print_EulerGamma(self, expr): if self._settings["inline"]: return "eulergamma" else: return super()._print_NumberSymbol(expr) def _print_Catalan(self, expr): if self._settings["inline"]: return "catalan" else: return super()._print_NumberSymbol(expr) def _print_GoldenRatio(self, expr): if self._settings["inline"]: return "golden" else: return super()._print_NumberSymbol(expr) def _print_Assignment(self, expr): from sympy.codegen.ast import Assignment from sympy.functions.elementary.piecewise import Piecewise from sympy.tensor.indexed import IndexedBase # Copied from codeprinter, but remove special MatrixSymbol treatment lhs = expr.lhs rhs = expr.rhs # We special case assignments that take multiple lines if not self._settings["inline"] and isinstance(expr.rhs, Piecewise): # Here we modify Piecewise so each expression is now # an Assignment, and then continue on the print. expressions = [] conditions = [] for (e, c) in rhs.args: expressions.append(Assignment(lhs, e)) conditions.append(c) temp = Piecewise(*zip(expressions, conditions)) return self._print(temp) if self._settings["contract"] and (lhs.has(IndexedBase) or rhs.has(IndexedBase)): # Here we check if there is looping to be done, and if so # print the required loops. return self._doprint_loops(rhs, lhs) else: lhs_code = self._print(lhs) rhs_code = self._print(rhs) return self._get_statement("%s = %s" % (lhs_code, rhs_code)) def _print_Infinity(self, expr): return 'Inf' def _print_NegativeInfinity(self, expr): return '-Inf' def _print_NaN(self, expr): return 'NaN' def _print_list(self, expr): return 'Any[' + ', '.join(self._print(a) for a in expr) + ']' def _print_tuple(self, expr): if len(expr) == 1: return "(%s,)" % self._print(expr[0]) else: return "(%s)" % self.stringify(expr, ", ") _print_Tuple = _print_tuple def _print_BooleanTrue(self, expr): return "true" def _print_BooleanFalse(self, expr): return "false" def _print_bool(self, expr): return str(expr).lower() # Could generate quadrature code for definite Integrals? #_print_Integral = _print_not_supported def _print_MatrixBase(self, A): # Handle zero dimensions: if S.Zero in A.shape: return 'zeros(%s, %s)' % (A.rows, A.cols) elif (A.rows, A.cols) == (1, 1): return "[%s]" % A[0, 0] elif A.rows == 1: return "[%s]" % A.table(self, rowstart='', rowend='', colsep=' ') elif A.cols == 1: # note .table would unnecessarily equispace the rows return "[%s]" % ", ".join([self._print(a) for a in A]) return "[%s]" % A.table(self, rowstart='', rowend='', rowsep=';\n', colsep=' ') def _print_SparseRepMatrix(self, A): from sympy.matrices import Matrix L = A.col_list(); # make row vectors of the indices and entries I = Matrix([k[0] + 1 for k in L]) J = Matrix([k[1] + 1 for k in L]) AIJ = Matrix([k[2] for k in L]) return "sparse(%s, %s, %s, %s, %s)" % (self._print(I), self._print(J), self._print(AIJ), A.rows, A.cols) def _print_MatrixElement(self, expr): return self.parenthesize(expr.parent, PRECEDENCE["Atom"], strict=True) \ + '[%s,%s]' % (expr.i + 1, expr.j + 1) def _print_MatrixSlice(self, expr): def strslice(x, lim): l = x[0] + 1 h = x[1] step = x[2] lstr = self._print(l) hstr = 'end' if h == lim else self._print(h) if step == 1: if l == 1 and h == lim: return ':' if l == h: return lstr else: return lstr + ':' + hstr else: return ':'.join((lstr, self._print(step), hstr)) return (self._print(expr.parent) + '[' + strslice(expr.rowslice, expr.parent.shape[0]) + ',' + strslice(expr.colslice, expr.parent.shape[1]) + ']') def _print_Indexed(self, expr): inds = [ self._print(i) for i in expr.indices ] return "%s[%s]" % (self._print(expr.base.label), ",".join(inds)) def _print_Idx(self, expr): return self._print(expr.label) def _print_Identity(self, expr): return "eye(%s)" % self._print(expr.shape[0]) def _print_HadamardProduct(self, expr): return '.*'.join([self.parenthesize(arg, precedence(expr)) for arg in expr.args]) def _print_HadamardPower(self, expr): PREC = precedence(expr) return '.**'.join([ self.parenthesize(expr.base, PREC), self.parenthesize(expr.exp, PREC) ]) # Note: as of 2015, Julia doesn't have spherical Bessel functions def _print_jn(self, expr): from sympy.functions import sqrt, besselj x = expr.argument expr2 = sqrt(S.Pi/(2*x))*besselj(expr.order + S.Half, x) return self._print(expr2) def _print_yn(self, expr): from sympy.functions import sqrt, bessely x = expr.argument expr2 = sqrt(S.Pi/(2*x))*bessely(expr.order + S.Half, x) return self._print(expr2) def _print_Piecewise(self, expr): if expr.args[-1].cond != True: # We need the last conditional to be a True, otherwise the resulting # function may not return a result. raise ValueError("All Piecewise expressions must contain an " "(expr, True) statement to be used as a default " "condition. Without one, the generated " "expression may not evaluate to anything under " "some condition.") lines = [] if self._settings["inline"]: # Express each (cond, expr) pair in a nested Horner form: # (condition) .* (expr) + (not cond) .* () # Expressions that result in multiple statements won't work here. ecpairs = ["({}) ? ({}) :".format (self._print(c), self._print(e)) for e, c in expr.args[:-1]] elast = " (%s)" % self._print(expr.args[-1].expr) pw = "\n".join(ecpairs) + elast # Note: current need these outer brackets for 2*pw. Would be # nicer to teach parenthesize() to do this for us when needed! return "(" + pw + ")" else: for i, (e, c) in enumerate(expr.args): if i == 0: lines.append("if (%s)" % self._print(c)) elif i == len(expr.args) - 1 and c == True: lines.append("else") else: lines.append("elseif (%s)" % self._print(c)) code0 = self._print(e) lines.append(code0) if i == len(expr.args) - 1: lines.append("end") return "\n".join(lines) def indent_code(self, code): """Accepts a string of code or a list of code lines""" # code mostly copied from ccode if isinstance(code, str): code_lines = self.indent_code(code.splitlines(True)) return ''.join(code_lines) tab = " " inc_regex = ('^function ', '^if ', '^elseif ', '^else$', '^for ') dec_regex = ('^end$', '^elseif ', '^else$') # pre-strip left-space from the code code = [ line.lstrip(' \t') for line in code ] increase = [ int(any(search(re, line) for re in inc_regex)) for line in code ] decrease = [ int(any(search(re, line) for re in dec_regex)) for line in code ] pretty = [] level = 0 for n, line in enumerate(code): if line in ('', '\n'): pretty.append(line) continue level -= decrease[n] pretty.append("%s%s" % (tab*level, line)) level += increase[n] return pretty def julia_code(expr, assign_to=None, **settings): r"""Converts `expr` to a string of Julia code. Parameters ========== expr : Expr A SymPy expression to be converted. assign_to : optional When given, the argument is used as the name of the variable to which the expression is assigned. Can be a string, ``Symbol``, ``MatrixSymbol``, or ``Indexed`` type. This can be helpful for expressions that generate multi-line statements. precision : integer, optional The precision for numbers such as pi [default=16]. user_functions : dict, optional A dictionary where keys are ``FunctionClass`` instances and values are their string representations. Alternatively, the dictionary value can be a list of tuples i.e. [(argument_test, cfunction_string)]. See below for examples. human : bool, optional If True, the result is a single string that may contain some constant declarations for the number symbols. If False, the same information is returned in a tuple of (symbols_to_declare, not_supported_functions, code_text). [default=True]. contract: bool, optional If True, ``Indexed`` instances are assumed to obey tensor contraction rules and the corresponding nested loops over indices are generated. Setting contract=False will not generate loops, instead the user is responsible to provide values for the indices in the code. [default=True]. inline: bool, optional If True, we try to create single-statement code instead of multiple statements. [default=True]. Examples ======== >>> from sympy import julia_code, symbols, sin, pi >>> x = symbols('x') >>> julia_code(sin(x).series(x).removeO()) 'x.^5/120 - x.^3/6 + x' >>> from sympy import Rational, ceiling >>> x, y, tau = symbols("x, y, tau") >>> julia_code((2*tau)**Rational(7, 2)) '8*sqrt(2)*tau.^(7/2)' Note that element-wise (Hadamard) operations are used by default between symbols. This is because its possible in Julia to write "vectorized" code. It is harmless if the values are scalars. >>> julia_code(sin(pi*x*y), assign_to="s") 's = sin(pi*x.*y)' If you need a matrix product "*" or matrix power "^", you can specify the symbol as a ``MatrixSymbol``. >>> from sympy import Symbol, MatrixSymbol >>> n = Symbol('n', integer=True, positive=True) >>> A = MatrixSymbol('A', n, n) >>> julia_code(3*pi*A**3) '(3*pi)*A^3' This class uses several rules to decide which symbol to use a product. Pure numbers use "*", Symbols use ".*" and MatrixSymbols use "*". A HadamardProduct can be used to specify componentwise multiplication ".*" of two MatrixSymbols. There is currently there is no easy way to specify scalar symbols, so sometimes the code might have some minor cosmetic issues. For example, suppose x and y are scalars and A is a Matrix, then while a human programmer might write "(x^2*y)*A^3", we generate: >>> julia_code(x**2*y*A**3) '(x.^2.*y)*A^3' Matrices are supported using Julia inline notation. When using ``assign_to`` with matrices, the name can be specified either as a string or as a ``MatrixSymbol``. The dimensions must align in the latter case. >>> from sympy import Matrix, MatrixSymbol >>> mat = Matrix([[x**2, sin(x), ceiling(x)]]) >>> julia_code(mat, assign_to='A') 'A = [x.^2 sin(x) ceil(x)]' ``Piecewise`` expressions are implemented with logical masking by default. Alternatively, you can pass "inline=False" to use if-else conditionals. Note that if the ``Piecewise`` lacks a default term, represented by ``(expr, True)`` then an error will be thrown. This is to prevent generating an expression that may not evaluate to anything. >>> from sympy import Piecewise >>> pw = Piecewise((x + 1, x > 0), (x, True)) >>> julia_code(pw, assign_to=tau) 'tau = ((x > 0) ? (x + 1) : (x))' Note that any expression that can be generated normally can also exist inside a Matrix: >>> mat = Matrix([[x**2, pw, sin(x)]]) >>> julia_code(mat, assign_to='A') 'A = [x.^2 ((x > 0) ? (x + 1) : (x)) sin(x)]' Custom printing can be defined for certain types by passing a dictionary of "type" : "function" to the ``user_functions`` kwarg. Alternatively, the dictionary value can be a list of tuples i.e., [(argument_test, cfunction_string)]. This can be used to call a custom Julia function. >>> from sympy import Function >>> f = Function('f') >>> g = Function('g') >>> custom_functions = { ... "f": "existing_julia_fcn", ... "g": [(lambda x: x.is_Matrix, "my_mat_fcn"), ... (lambda x: not x.is_Matrix, "my_fcn")] ... } >>> mat = Matrix([[1, x]]) >>> julia_code(f(x) + g(x) + g(mat), user_functions=custom_functions) 'existing_julia_fcn(x) + my_fcn(x) + my_mat_fcn([1 x])' Support for loops is provided through ``Indexed`` types. With ``contract=True`` these expressions will be turned into loops, whereas ``contract=False`` will just print the assignment expression that should be looped over: >>> from sympy import Eq, IndexedBase, Idx >>> len_y = 5 >>> y = IndexedBase('y', shape=(len_y,)) >>> t = IndexedBase('t', shape=(len_y,)) >>> Dy = IndexedBase('Dy', shape=(len_y-1,)) >>> i = Idx('i', len_y-1) >>> e = Eq(Dy[i], (y[i+1]-y[i])/(t[i+1]-t[i])) >>> julia_code(e.rhs, assign_to=e.lhs, contract=False) 'Dy[i] = (y[i + 1] - y[i])./(t[i + 1] - t[i])' """ return JuliaCodePrinter(settings).doprint(expr, assign_to) def print_julia_code(expr, **settings): """Prints the Julia representation of the given expression. See `julia_code` for the meaning of the optional arguments. """ print(julia_code(expr, **settings))