\errorcontextlines999\relax
The various ``XX'' and ``xX'' pairs test whether there are errant spaces
in the macros or the macro files.
X\input mfpextra\relax X
X\input mfpextra\relax X
\def\empty{}
\def\frac#1#2{{#1\over#2}}
\def\cs#1{{\tt \char`\\#1}}
\def\mybreak{\vskip 0pt plus 100pt\penalty 0 \vskip 0pt plus -100pt\relax}
\def\\{\hfil\break\ignorespaces}
\def\y{Rpop\Z\Z\\}
\def\Y{Rpop\Z\Z}
\everymath{\displaystyle}
{\bf Stack-only operations:}\\
Example of a program. Computes the solution of $ax^2 + bx + c = 0$ using
the quadratic formula. If the result is complex, it detects this.
Coefficients stored in \cs{A}, \cs{B} and \cs{C}. First case: $4x^2 +
5y^2 - 1= 0$. Solutions are $x = -\frac{5}{8} \pm \frac{1}{8}\sqrt{41}$
\def\A{4}\def\B{5}\def\C{-1}
X\startMFPprogram % stack
\Rpush\B\Rdup\Rsq % B(B^2)
\Rpush\A\Rpush\C\Rmul\Rdbl\Rdbl % B(B^2)(4AC)
\Rsub % B(B^2-4AC)
\IFneg{\def\I{i}\Rchs}{\def\I{}}% % B(|B^2-4AC|)
\Rsqrt\Rpush\A\Rdbl\Rdiv % B(sqrt(|B^2-4AC|)/2A)
\Rpop\Ypart % B
\Rpush\A\Rdbl\Rdiv\Rchs % (-B/2A)
\Rpop\Xpart %
%\expandafter\show\csname MFP@Rstack\endcsname
\Rpush\Xpart\Rpush\Ypart\Radd\Rpop\Broot
\Rpush\Xpart\Rpush\Ypart\Rsub\Rpop\Sroot
\Export\Xpart
\Export\Ypart
\Export\Broot
\Export\Sroot
\Export\I
\stopMFPprogram X
\indent Solution: $x = \Xpart \pm \I\Ypart = \Broot$ and $\Sroot$.
Second case $2x^2 - 2x + 3 = 0$. Solutions are $x = \frac{1}{2} \pm
\frac{i}{2}\sqrt{5}$.
\def\A{2}\def\B{-2}\def\C{3}
X\startMFPprogram % stack
\Rpush\B\Rdup\Rsq % B(B^2)
\Rpush\A\Rpush\C\Rmul\Rdbl\Rdbl % B(B^2)(4AC)
\Rsub % B(B^2-4AC)
\IFneg{\def\I{i}\Rchs}{\def\I{}}% % B(|B^2-4AC|)
\Rsqrt\Rpush\A\Rdbl\Rdiv % B(sqrt(|B^2-4AC|)/2A)
\Rpop\Ypart % B
\Rpush\A\Rdbl\Rdiv\Rchs % (-B/2A)
\Rpop\Xpart %
\Export\Xpart
\Export\Ypart
\Export\I
\stopMFPprogram X
%\expandafter\show\csname MFP@Rstack\endcsname
\indent Solution: $x = \Xpart \pm \I\Ypart$.
Now try square roots (should be exactly 1234.5678 and 1524):
X\startMFPprogram % stack
\Rpush{1524157.65279684}\Rsqrt\Rpop\X
\Rpush{1524}\Rsq\Rsqrt\Rpop\Y
\Export\X
\Export\Y
\stopMFPprogram X \X { and }\Y.
Below we test for speed and to check for any space characters
accidentally produced. You should see only a few xX pairs with hopefully
no spaces in between them. As curently set up, these tests perform about
34000 numerical operations. It all takes about 31 seconds on a
moderately old (2 years?) Windows 7 running plain tex from
TeX~Live~2012.
This operation count does not distinguish between basic operations like
addition and multiplication, and those operations from mfpextra that are
probably each equivalent to a dozen or more multiplications. Counting
each such operation with the an estimated multiplicity, the tests
probably perform 400 thousand or more basic operations.
Of the basic operations, multiplication is (by measurement) 2--4 times
as lengthy as addition, and division is 2--3 times as lengthy as
multiplication. Actual times depend on the machine, but the ratios
remain pretty much the same. Here is a summary of timings on my fastest
machine; each operation is run 500 times in a loop. (A loop in which
an input number is processed and a value returned, but no calculations are
performed ({\tt\string\MFPzero}), times at $0.0\,$sec.) Timing obtained
with {\tt\string\pdfelapsedtime}.
\medskip
\indent\vtop{\halign{\hfil$#$&\quad$#\,$sec\cr
\noalign{\hrule\smallskip}
2.54321+22432.87654321 &0.016\cr
2.54321\times22432.87654321 &0.046\cr
22432.87654321/2.54321 &0.11\cr
\sqrt{23456789.54321} &0.172\cr
\mathop{\fam0 rand}(23456789.54321) &0.105\cr
1.00001234^{8000} &0.72\cr
\exp(2.54321) &0.42\cr
\sin(2.54321) &0.41\cr
\log(2.54321) &0.73\cr
\mathop{\fam0 angle}(254.321,100) &1.14\cr
\noalign{\smallskip\hrule}
}}
\medskip
Originally, all the tests below combined took 21 seconds on a 4-year-old
Windows XP under TeX Live 2011. But since then I have changed angle and
power computations so that they are considerably more accurate, but with
a possible reduction in speed. I cannot test the speed reduction, since
I no longer have that machine.
For my current machines: On a Windows 7 machine, 64-bit, laptop, it
takes 32 seconds to process this file. On another Windows 7 machine,
32-bit, desktop, it takes about 10 seconds. (This difference could be
explained partly by the fact that the last machine is newer and partly
by the fact that TeX is a 32-bit program and therefore a better match to
the 32-bit hardware.)
\def\testi{% stack forms
\startMFPprogram
\Rpush{0.000 001}\Rpop\X
\Rpush{1.2}\Rpush{-2.3}\Rexch\Rdup
\Rpop\X\Rpop\X
\Rpop\X\Rpush{21.34}\Rchs
\Rpop\X\Rpush{21.34}\Rabs
\Rpop\X\Rpush{21.34}\Rchs
\Rpop\X\Rpush{21.34}\Rint
\Rpop\X\Rpush{21.34}\Rfrac
\Rpop\X\Rpush{21.34}\Rdbl
\Rpop\X\Rpush{21.34}\Rhalve
\Rpop\X\Rpush{21.34}\Rsgn
\Rpop\X\Rpush{21.34}\Rsin
\Rpop\X\Rpush{21.34}\Rcos
\Rpop\X\Rpush{21.34}\Rdeg
\Rpop\X\Rpush{21.34}\Rrad
\Rpop\X\Rpush{21.34}\Rlog
\Rpop\X\Rpush{21.34}\Rln
\Rpop\X\Rpush{-1.34}\Rexp
\Rpop\X\Rpush{3.3}\Rexp
\Rpop\X\Rpush{21.34}\Rsq
\Rpop\X\Rpush{21.34}\Rinv
\Rpop\X\Rpush{21.34}\Rfloor
\Rpop\X\Rpush{21.34}\Rceil
\Rpop\X\Rpush{21.34}\Rsqrt
\Rpop\X\Rpush{21.34}\Rrand
\Rpop\X\Rpush{21.34}\Rpush{12.34}\Rcmp
\IFlt{}{}\IFgt{}{}\IFeq{}{}\Rsub
\IFneg{}{}\IFpos{}{}\IFzero{}{}\Rpop\X
\Rpush{1.2}\Rpush{-2.3}\Radd
\Rpop\X\Rpush{1.2}\Rpush{-2.3}\Rsub
\Rpop\X\Rpush{1.2}\Rpush{-2.3}\Rmul
\Rpop\X\Rpush{1.2}\Rpush{-2.3}\Rdiv
\Rpop\X\Rpush{2.3}\Rpush{17}\Rpow
\Rpop\X\Rpush{2.3}\Rpush{-17}\Rpow
\Rpop\X\Rpush{1.2}\Rpush{-2.3}\Rmax
\Rpop\X\Rpush{1.2}\Rpush{-2.3}\Rmin
\stopMFPprogram}
\def\testii{% unary operand forms, including
\MFPchs\X\Z % extra tests of sin, log, exp and pow
\MFPchs\Y\Z
\MFPabs\X\Z
\MFPabs\Y\Z
\MFPdbl\X\Z
\MFPdbl\Y\Z
\MFPhalve\X\Z
\MFPhalve\Y\Z
\MFPint\X\Z
\MFPint\Y\Z
\MFPsgn\X\Z
\MFPsgn\Y\Z
\MFPsq\X\Z
\MFPsq\Y\Z
\MFPinv\X\Z
\MFPinv\Y\Z
\MFPfrac\X\Z
\MFPfrac\Y\Z
\MFPfloor\X\Z
\MFPfloor\Y\Z
\MFPceil\X\Z
\MFPceil\Y\Z
\MFPsin{30}\Z
\MFPsin{420}\Z
\MFPcos{60}\Z
\MFPcos{390}\Z
\MFPlog\X\Z
\MFPln\X\Z
\MFPexp\X\Z
\MFPexp\Y\Z
\MFPsin{1}\Z
\MFPsin{2}\Z
\MFPsin{3}\Z
\MFPsin{4}\Z
\MFPsin{5}\Z
\MFPsin{6}\Z
\MFPsin{7}\Z
\MFPsin{8}\Z
\MFPsin{9}\Z
\MFPsin{10}\Z
\MFPsin{20}\Z
\MFPsin{30}\Z
\MFPsin{40}\Z
\MFPsin{50}\Z
\MFPsin{60}\Z
\MFPsin{70}\Z
\MFPsin{80}\Z
\MFPsin{90}\Z
\MFPlog{.1}\Z
\MFPlog{.2}\Z
\MFPlog{.3}\Z
\MFPlog{.4}\Z
\MFPlog{.5}\Z
\MFPlog{.6}\Z
\MFPlog{.7}\Z
\MFPlog{.8}\Z
\MFPlog{.9}\Z
\MFPlog{1}\Z
\MFPlog{1.01}\Z
\MFPlog{1.02}\Z
\MFPlog{1.03}\Z
\MFPlog{1.04}\Z
\MFPlog{1.05}\Z
\MFPlog{1.06}\Z
\MFPlog{1.07}\Z
\MFPlog{1.08}\Z
\MFPlog{1.09}\Z
\MFPexp{.000001}\Z
\MFPexp{.00001}\Z
\MFPexp{.0001}\Z
\MFPexp{.001}\Z
\MFPexp{.01}\Z
\MFPexp{.1}\Z
\MFPexp{1}\Z
\MFPexp{2}\Z
\MFPexp{3}\Z
\MFPexp{4}\Z
\MFPexp{5}\Z
\MFPexp{6}\Z
\MFPexp{7}\Z
\MFPexp{8}\Z
\MFPexp{9}\Z
\MFPexp{10}\Z
\MFPsqrt{10}\Z
\MFPrand{10}\Z
\MFPexp{-8.3254}\Z
\MFPpow\MFPe{-10}\Z
\MFPpow\MFPe{-9}\Z
\MFPpow\MFPe{-8}\Z
\MFPpow\MFPe{-7}\Z
\MFPpow\MFPe{-6}\Z
\MFPpow\MFPe{-5}\Z
\MFPpow\MFPe{-4}\Z
\MFPpow\MFPe{-3}\Z
\MFPpow\MFPe{-2}\Z
\MFPpow\MFPe{-1}\Z
\MFPpow\MFPe{0}\Z
\MFPpow\MFPe{1}\Z
\MFPpow\MFPe{2}\Z
\MFPpow\MFPe{3}\Z
\MFPpow\MFPe{4}\Z
\MFPpow\MFPe{5}\Z
\MFPpow\MFPe{6}\Z
\MFPpow\MFPe{7}\Z
\MFPpow\MFPe{8}\Z
\MFPpow\MFPe{9}\Z
\MFPpow\MFPe{10}\Z}
\def\testiii{%% binary operand forms and print formating, plus
\MFPsqrt{0}\Z % additional tests of sqrt
\MFPsqrt{1}\Z
\MFPsqrt{2}\Z
\MFPsqrt{3}\Z
\MFPsqrt{4}\Z
\MFPsqrt{5}\Z
\MFPsqrt{6}\Z
\MFPsqrt{7}\Z
\MFPsqrt{8}\Z
\MFPsqrt{9}\Z
\MFPsqrt{10}\Z
\MFPsqrt{1524157.65279684}\Z
\MFPadd\X\Y\Z
\MFPsub\X\Y\Z
\MFPsub\Y\X\Z
\MFPsub\X\X\Z
\MFPsub\Y\Y\Z
\MFPmul\X\Y\Z
\MFPdiv\X\Y\Z
\MFPdiv\Y\X\Z
\MFPmax\X\Y\Z
\MFPmin\X\Y\Z
\MFPpow\X{5}\Z
\MFPpow\X{-5}\Z
\MFPpow\Y{5}\Z
\MFPpow\Y{-5}\Z
\MFPcmp\X\Y
\IFlt{}{}\IFgt{}{}\IFeq{}{}%
\MFPsub\X\Y\Z
\IFneg{}{}\IFpos{}{}\IFzero{}{}%
\def\T{333.00000000}%
\def\S{1357.12345678}%
\MFPtruncate{4}\T\Z
\MFPtruncate{0}\T\Z
\MFPtruncate{-2}\T\Z
\MFPstrip\T\Z
\MFPstrip*\T\Z
\MFPround{3}\S\Z
\MFPround{5}\S\Z
\MFPround{0}\S\Z
\MFPround{-2}\S\Z
\def\T{-333.00000000}%
\def\S{-1357.12345678}%
\MFPtruncate{4}\T\Z
\MFPtruncate{0}\T\Z
\MFPtruncate{-2}\T\Z
\MFPstrip\T\Z
\MFPstrip*\T\Z
\MFPround{3}\S\Z
\MFPround{5}\S\Z
\MFPround{0}\S\Z
\MFPround{-2}\S\Z}
Three test loops follow. The first repeats 500 times a stack program
that performs each available command followed by popping the result and
repushing the original value(s).
\newcount\n
\def\testloopi{%
\ifnum \n>0
\advance\n -1
\testi
\expandafter
\testloopi
\fi
}
\n=500
x\testloopi X
The second repeats 100 times a sequence in which all the unary operand
commands are performed twice, plus extra of sine, log and exp.
\def\testloopii{%
\ifnum \n>0
\advance\n -1
\testii
\expandafter
\testloopii
\fi
}
\def\X{1.2}
\def\Y{-2.3}
\n=100
x\testloopii X
The third repeats 100 times a sequence in which all the binary operand
commands are performed, plus some extra tests of sqrt and then all the
print-preparation commands.
\def\testloopiii{%
\ifnum \n>0
\advance\n -1
\testiii
\expandafter
\testloopiii
\fi
}
\n=100
x\testloopiii X
\end
\end{document}