Mercurial > hg > octave-lyh
view libinterp/corefcn/gammainc.cc @ 17335:fe6518a1d87c
Move TeX symbol decoding into the lexer (bug #39831).
* libinterp/Makefile.am (BUILT_SOURCES): Add corefcn/oct-tex-symbols.cc.
(BUILT_DISTFILES): Add corefcn/oct-tex-lexer.ll and
corefcn/oct-tex-symbols.cc.
(EXTRA_DIST): Add corefcn/oct-tex-lexer.in.ll and
corefcn/oct-tex-symbols.in.
(ULT_DIST_SRC): Filter out corefcn/oct-tex-lexer.ll from DIST_SRC and
add corefcn/oct-tex-lexer.in.ll instead.
(CLEAN_FILES): Add corefcn/oct-tex-parser.output.
* libinterp/corefcn/modules.mk (corefcn/oct-tex-lexer.ll,
corefcn/oct-tex-symbols.cc): New rules to build the TeX lexer.
(corefcn/txt-eng.cc): Add dependency on corefcn/oct-tex-symbols.cc.
* libinterp/corefcn/oct-tex-lexer.in.ll: Renamed from oct-tex.lexer.ll.
Remove COMMAND state. Remove ID regex. Replace rules for symbols with
tag @SYMBOL_RULES@.
* libinterp/corefcn/oct-tex-parser.yy (ID, CMD, identifier): Remove
tokens.
(SYM, sym): New token and value.
(symbol_element): Build from SYM.
* libinterp/corefcn/oct-tex-symbols.in: New file with supported TeX
symbols and corresponding codes (unicode and MS symbols).
* libinterp/corefcn/txt-eng.h (class text_element_symbol): Make it
inherit from text_element.
(text_element_symbol::code): Removed member.
(text_element_symbol::symbol): New member.
(text_element_symbol::text_element_symbol): Adapt constructor.
(text_element_symbol::get_symbol): New method.
(text_element_symbol::get_symbol_code): Make const.
* libinterp/corefcn/txt-eng.cc (symbol_names, symbol_codes): Remove
static variables, now auto-generated from oct-tex-symbols.in.
(oct-tex-symbols.cc): New include.
(text_element_symbol::get_symbol_code): Change implementation to simply
index into auto-generated symbol_codes array.
* libinterp/corefcn/txt-eng-ft.cc
(ft_render::visit(text_element_symbol)): Don't use
text_element_symbol::string_value(), use
text_element_symbol::get_symbol() instead.
author | Michael Goffioul <michael.goffioul@gmail.com> |
---|---|
date | Sat, 24 Aug 2013 14:27:09 -0400 |
parents | bc924baa2c4e |
children |
line wrap: on
line source
/* Copyright (C) 1997-2012 John W. Eaton This file is part of Octave. Octave is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. Octave is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with Octave; see the file COPYING. If not, see <http://www.gnu.org/licenses/>. */ #ifdef HAVE_CONFIG_H #include <config.h> #endif #include "lo-specfun.h" #include "defun.h" #include "error.h" #include "gripes.h" #include "oct-obj.h" #include "utils.h" DEFUN (gammainc, args, , "-*- texinfo -*-\n\ @deftypefn {Mapping Function} {} gammainc (@var{x}, @var{a})\n\ @deftypefnx {Mapping Function} {} gammainc (@var{x}, @var{a}, \"lower\")\n\ @deftypefnx {Mapping Function} {} gammainc (@var{x}, @var{a}, \"upper\")\n\ Compute the normalized incomplete gamma function,\n\ @tex\n\ $$\n\ \\gamma (x, a) = {1 \\over {\\Gamma (a)}}\\displaystyle{\\int_0^x t^{a-1} e^{-t} dt}\n\ $$\n\ @end tex\n\ @ifnottex\n\ \n\ @example\n\ @group\n\ x\n\ 1 /\n\ gammainc (x, a) = --------- | exp (-t) t^(a-1) dt\n\ gamma (a) /\n\ t=0\n\ @end group\n\ @end example\n\ \n\ @end ifnottex\n\ with the limiting value of 1 as @var{x} approaches infinity.\n\ The standard notation is @math{P(a,x)}, e.g., Abramowitz and Stegun (6.5.1).\n\ \n\ If @var{a} is scalar, then @code{gammainc (@var{x}, @var{a})} is returned\n\ for each element of @var{x} and vice versa.\n\ \n\ If neither @var{x} nor @var{a} is scalar, the sizes of @var{x} and\n\ @var{a} must agree, and @code{gammainc} is applied element-by-element.\n\ \n\ By default the incomplete gamma function integrated from 0 to @var{x} is\n\ computed. If @qcode{\"upper\"} is given then the complementary function\n\ integrated from @var{x} to infinity is calculated. It should be noted that\n\ \n\ @example\n\ gammainc (@var{x}, @var{a}) @equiv{} 1 - gammainc (@var{x}, @var{a}, \"upper\")\n\ @end example\n\ @seealso{gamma, lgamma}\n\ @end deftypefn") { octave_value retval; bool lower = true; int nargin = args.length (); if (nargin == 3) { if (args(2).is_string ()) { std::string s = args(2).string_value (); std::transform (s.begin (), s.end (), s.begin (), tolower); if (s == "upper") lower = false; else if (s != "lower") error ("gammainc: third argument must be \"lower\" or \"upper\""); } else error ("gammainc: third argument must be \"lower\" or \"upper\""); } if (!error_state && nargin >= 2 && nargin <= 3) { octave_value x_arg = args(0); octave_value a_arg = args(1); // FIXME Can we make a template version of the duplicated code below if (x_arg.is_single_type () || a_arg.is_single_type ()) { if (x_arg.is_scalar_type ()) { float x = x_arg.float_value (); if (! error_state) { if (a_arg.is_scalar_type ()) { float a = a_arg.float_value (); if (! error_state) retval = lower ? gammainc (x, a) : static_cast<float>(1) - gammainc (x, a); } else { FloatNDArray a = a_arg.float_array_value (); if (! error_state) retval = lower ? gammainc (x, a) : static_cast<float>(1) - gammainc (x, a); } } } else { FloatNDArray x = x_arg.float_array_value (); if (! error_state) { if (a_arg.is_scalar_type ()) { float a = a_arg.float_value (); if (! error_state) retval = lower ? gammainc (x, a) : static_cast<float>(1) - gammainc (x, a); } else { FloatNDArray a = a_arg.float_array_value (); if (! error_state) retval = lower ? gammainc (x, a) : static_cast<float>(1) - gammainc (x, a); } } } } else { if (x_arg.is_scalar_type ()) { double x = x_arg.double_value (); if (! error_state) { if (a_arg.is_scalar_type ()) { double a = a_arg.double_value (); if (! error_state) retval = lower ? gammainc (x, a) : 1. - gammainc (x, a); } else { NDArray a = a_arg.array_value (); if (! error_state) retval = lower ? gammainc (x, a) : 1. - gammainc (x, a); } } } else { NDArray x = x_arg.array_value (); if (! error_state) { if (a_arg.is_scalar_type ()) { double a = a_arg.double_value (); if (! error_state) retval = lower ? gammainc (x, a) : 1. - gammainc (x, a); } else { NDArray a = a_arg.array_value (); if (! error_state) retval = lower ? gammainc (x, a) : 1. - gammainc (x, a); } } } } } else print_usage (); return retval; } /* %!test %! a = [.5 .5 .5 .5 .5]; %! x = [0 1 2 3 4]; %! v1 = sqrt (pi)*erf (x)./gamma (a); %! v3 = gammainc (x.*x, a); %! assert (v1, v3, sqrt (eps)); %!assert (gammainc (0:4,0.5, "upper"), 1-gammainc (0:4,0.5), 1e-10) %!test %! a = single ([.5 .5 .5 .5 .5]); %! x = single ([0 1 2 3 4]); %! v1 = sqrt (pi ("single"))*erf (x)./gamma (a); %! v3 = gammainc (x.*x, a); %! assert (v1, v3, sqrt (eps ("single"))); %!assert (gammainc (single (0:4), single (0.5), "upper"), %! single (1)-gammainc (single (0:4), single (0.5)), %! single (1e-7)) */