aboutsummaryrefslogtreecommitdiff
path: root/manual/math.texi
diff options
context:
space:
mode:
Diffstat (limited to 'manual/math.texi')
-rw-r--r--manual/math.texi64
1 files changed, 62 insertions, 2 deletions
diff --git a/manual/math.texi b/manual/math.texi
index 2b7cb82f5e..15a2075d33 100644
--- a/manual/math.texi
+++ b/manual/math.texi
@@ -54,6 +54,7 @@ in case of double using @code{double} is a good compromise.
constant.
* FP Comparison Functions:: Special functions to compare floating-point
numbers.
+* FP Function Optimizations:: Fast code or small code.
* Trig Functions:: Sine, cosine, and tangent.
* Inverse Trig Functions:: Arc sine, arc cosine, and arc tangent.
* Exponents and Logarithms:: Also includes square root.
@@ -704,6 +705,35 @@ that the comparison for equality and unequality do @emph{not} throw an
exception if one of the arguments is an unordered value.
+@node FP Function Optimizations
+@section Is Fast Code or Small Code preferred?
+@cindex Optimization
+
+If an application uses many floating point function it is often the case
+that the costs for the function calls itseld are not neglectable.
+Modern processor implementation often can execute the operation itself
+very fast but the call means a disturbance of the control flow.
+
+For this reason the GNU C Library provides optimizations for many of the
+frequently used math functions. When the GNU CC is used and the user
+activates the optimizer several new inline functions and macros get
+defined. These new functions and macros have the same names as the
+library function and so get used instead of the later. In case of
+inline functions the compiler will decide whether it is reasonable to
+use the inline function and this decision is usually correct.
+
+For the generated code this means that no calls to the library functions
+are necessary. This increases the speed significantly. But the
+drawback is that the code size increases and this increase is not always
+neglectable.
+
+In cases where the inline functions and macros are not wanted the symbol
+@code{__NO_MATH_INLINES} should be defined before any system header is
+included. This will make sure only library functions are used. Of
+course it can be determined for each single file in the project whether
+giving this option is preferred or not.
+
+
@node Trig Functions
@section Trigonometric Functions
@cindex trigonometric functions
@@ -1430,8 +1460,14 @@ the same seed, used in different C libraries or on different CPU types,
will give you different random numbers.
The GNU library supports the standard @w{ISO C} random number functions
-plus another set derived from BSD. We recommend you use the standard
-ones, @code{rand} and @code{srand}.
+plus two other sets derived from BSD and SVID. We recommend you use the
+standard ones, @code{rand} and @code{srand} if only a small number of
+random bits are required. The SVID functions provide an interface which
+allows better randon number generator algorithms and they return up to
+48 random bits in one calls and they also return random floating-point
+numbers if wanted. The SVID function might not be available on some BSD
+derived systems but since they are required in the XPG they are
+available on all Unix-conformant systems.
@menu
* ISO Random:: @code{rand} and friends.
@@ -1478,6 +1514,30 @@ To produce truly random numbers (not just pseudo-random), do @code{srand
(time (0))}.
@end deftypefun
+A completely broken interface was designed by the POSIX.1 committee to
+support reproducible random numbers in multi-threaded programs.
+
+@comment stdlib.h
+@comment POSIX.1
+@deftypefun int rand_r (unsigned int *@var{seed})
+This function returns a random number in the range 0 to @code{RAND_MAX}
+just as @code{rand} does. But this function does not keep an internal
+state for the RNG. Instead the @code{unsigned int} variable pointed to
+by the argument @var{seed} is the only state. Before the value is
+returned the state will be updated so that the next call will return a
+new number.
+
+I.e., the state of the RNG can only have as much bits as the type
+@code{unsigned int} has. This is far too few to provide a good RNG.
+This interface is broken by design.
+
+If the program requires reproducible random numbers in multi-threaded
+programs the reentrant SVID functions are probably a better choice. But
+these functions are GNU extensions and therefore @code{rand_r}, as being
+standardized in POSIX.1, should always be kept as a default method.
+@end deftypefun
+
+
@node BSD Random
@subsection BSD Random Number Functions