cvsroot/libecb/ecb.pod

=head1 LIBECB - e-C-Builtins

=head2 ABOUT LIBECB

Libecb is currently a simple header file that doesn't require any
configuration to use or include in your project.

It's part of the e-suite of libraries, other members of which include
libev and libeio.

Its homepage can be found here:

    http://software.schmorp.de/pkg/libecb

It mainly provides a number of wrappers around GCC built-ins, together
with replacement functions for other compilers. In addition to this,
it provides a number of other lowlevel C utilities, such as endianness
detection, byte swapping or bit rotations.

Or in other words, things that should be built into any standard C system,
but aren't, implemented as efficient as possible with GCC, and still
correct with other compilers.

More might come.

=head2 ABOUT THE HEADER

At the moment, all you have to do is copy F<ecb.h> somewhere where your
compiler can find it and include it:

   #include <ecb.h>

The header should work fine for both C and C++ compilation, and gives you
all of F<inttypes.h> in addition to the ECB symbols.

There are currently no object files to link to - future versions might
come with an (optional) object code library to link against, to reduce
code size or gain access to additional features.

It also currently includes everything from F<inttypes.h>.

=head2 ABOUT THIS MANUAL / CONVENTIONS

This manual mainly describes each (public) function available after
including the F<ecb.h> header. The header might define other symbols than
these, but these are not part of the public API, and not supported in any
way.

When the manual mentions a "function" then this could be defined either as
as inline function, a macro, or an external symbol.

When functions use a concrete standard type, such as C<int> or
C<uint32_t>, then the corresponding function works only with that type. If
only a generic name is used (C<expr>, C<cond>, C<value> and so on), then
the corresponding function relies on C to implement the correct types, and
is usually implemented as a macro. Specifically, a "bool" in this manual
refers to any kind of boolean value, not a specific type.

=head2 TYPES / TYPE SUPPORT

ecb.h makes sure that the following types are defined (in the expected way):

   int8_t   uint8_t   int16_t  uint16_t
   int32_t  uint32_t  int64_t  uint64_t
   intptr_t uintptr_t

The macro C<ECB_PTRSIZE> is defined to the size of a pointer on this
platform (currently C<4> or C<8>) and can be used in preprocessor
expressions.

For C<ptrdiff_t> and C<size_t> use C<stddef.h>.

=head2 LANGUAGE/ENVIRONMENT/COMPILER VERSIONS

All the following symbols expand to an expression that can be tested in
preprocessor instructions as well as treated as a boolean (use C<!!> to
ensure it's either C<0> or C<1> if you need that).

=over 4

=item ECB_C

True if the implementation defines the C<__STDC__> macro to a true value,
while not claiming to be C++.

=item ECB_C99

True if the implementation claims to be compliant to C99 (ISO/IEC
9899:1999) or any later version, while not claiming to be C++.

Note that later versions (ECB_C11) remove core features again (for
example, variable length arrays).

=item ECB_C11

True if the implementation claims to be compliant to C11 (ISO/IEC
9899:2011) or any later version, while not claiming to be C++.

=item ECB_CPP

True if the implementation defines the C<__cplusplus__> macro to a true
value, which is typically true for C++ compilers.

=item ECB_CPP11

True if the implementation claims to be compliant to ISO/IEC 14882:2011
(C++11) or any later version.

=item ECB_GCC_VERSION (major, minor)

Expands to a true value (suitable for testing in by the preprocessor)
if the compiler used is GNU C and the version is the given version, or
higher.

This macro tries to return false on compilers that claim to be GCC
compatible but aren't.

=item ECB_EXTERN_C

Expands to C<extern "C"> in C++, and a simple C<extern> in C.

This can be used to declare a single external C function:

   ECB_EXTERN_C int printf (const char *format, ...);

=item ECB_EXTERN_C_BEG / ECB_EXTERN_C_END

These two macros can be used to wrap multiple C<extern "C"> definitions -
they expand to nothing in C.

They are most useful in header files:

   ECB_EXTERN_C_BEG

   int mycfun1 (int x);
   int mycfun2 (int x);

   ECB_EXTERN_C_END

=item ECB_STDFP

If this evaluates to a true value (suitable for testing in by the
preprocessor), then C<float> and C<double> use IEEE 754 single/binary32
and double/binary64 representations internally I<and> the endianness of
both types match the endianness of C<uint32_t> and C<uint64_t>.

This means you can just copy the bits of a C<float> (or C<double>) to an
C<uint32_t> (or C<uint64_t>) and get the raw IEEE 754 bit representation
without having to think about format or endianness.

This is true for basically all modern platforms, although F<ecb.h> might
not be able to deduce this correctly everywhere and might err on the safe
side.

=item ECB_AMD64, ECB_AMD64_X32

These two macros are defined to C<1> on the x86_64/amd64 ABI and the X32
ABI, respectively, and undefined elsewhere.

The designers of the new X32 ABI for some inexplicable reason decided to
make it look exactly like amd64, even though it's completely incompatible
to that ABI, breaking about every piece of software that assumed that
C<__x86_64> stands for, well, the x86-64 ABI, making these macros
necessary.

=back

=head2 MACRO TRICKERY

=over 4

=item ECB_CONCAT (a, b)

Expands any macros in C<a> and C<b>, then concatenates the result to form
a single token. This is mainly useful to form identifiers from components,
e.g.:

   #define S1 str
   #define S2 cpy

   ECB_CONCAT (S1, S2)(dst, src); // == strcpy (dst, src);

=item ECB_STRINGIFY (arg)

Expands any macros in C<arg> and returns the stringified version of
it. This is mainly useful to get the contents of a macro in string form,
e.g.:

   #define SQL_LIMIT 100
   sql_exec ("select * from table limit " ECB_STRINGIFY (SQL_LIMIT));

=back

=head2 ATTRIBUTES

A major part of libecb deals with additional attributes that can be
assigned to functions, variables and sometimes even types - much like
C<const> or C<volatile> in C. They are implemented using either GCC
attributes or other compiler/language specific features. Attributes
declarations must be put before the whole declaration:

   ecb_const int mysqrt (int a);
   ecb_unused int i;

=over 4

=item ecb_unused

Marks a function or a variable as "unused", which simply suppresses a
warning by GCC when it detects it as unused. This is useful when you e.g.
declare a variable but do not always use it:

  {
    ecb_unused int var;

    #ifdef SOMECONDITION
       var = ...;
       return var;
    #else
       return 0;
    #endif
  }

=item ecb_deprecated

Similar to C<ecb_unused>, but marks a function, variable or type as
deprecated. This makes some compilers warn when the type is used.

=item ecb_deprecated_message (message)

Same as C<ecb_deprecated>, but if possible, supply a diagnostic that is
used instead of a generic depreciation message when the object is being
used.

=item ecb_inline

Expands either to C<static inline> or to just C<static>, if inline
isn't supported. It should be used to declare functions that should be
inlined, for code size or speed reasons.

Example: inline this function, it surely will reduce codesize.

   ecb_inline int
   negmul (int a, int b)
   {
     return - (a * b);
   }

=item ecb_noinline

Prevent a function from being inlined - it might be optimised away, but
not inlined into other functions. This is useful if you know your function
is rarely called and large enough for inlining not to be helpful.

=item ecb_noreturn

Marks a function as "not returning, ever". Some typical functions that
don't return are C<exit> or C<abort> (which really works hard to not
return), and now you can make your own:

   ecb_noreturn void
   my_abort (const char *errline)
   {
     puts (errline);
     abort ();
   }

In this case, the compiler would probably be smart enough to deduce it on
its own, so this is mainly useful for declarations.

=item ecb_restrict

Expands to the C<restrict> keyword or equivalent on compilers that support
them, and to nothing on others. Must be specified on a pointer type or
an array index to indicate that the memory doesn't alias with any other
restricted pointer in the same scope.

Example: multiply a vector, and allow the compiler to parallelise the
loop, because it knows it doesn't overwrite input values.

   void
   multiply (ecb_restrict float *src,
             ecb_restrict float *dst,
             int len, float factor)
   {
     int i;

     for (i = 0; i < len; ++i)
       dst [i] = src [i] * factor;
   }

=item ecb_const

Declares that the function only depends on the values of its arguments,
much like a mathematical function. It specifically does not read or write
any memory any arguments might point to, global variables, or call any
non-const functions. It also must not have any side effects.

Such a function can be optimised much more aggressively by the compiler -
for example, multiple calls with the same arguments can be optimised into
a single call, which wouldn't be possible if the compiler would have to
expect any side effects.

It is best suited for functions in the sense of mathematical functions,
such as a function returning the square root of its input argument.

Not suited would be a function that calculates the hash of some memory
area you pass in, prints some messages or looks at a global variable to
decide on rounding.

See C<ecb_pure> for a slightly less restrictive class of functions.

=item ecb_pure

Similar to C<ecb_const>, declares a function that has no side
effects. Unlike C<ecb_const>, the function is allowed to examine global
variables and any other memory areas (such as the ones passed to it via
pointers).

While these functions cannot be optimised as aggressively as C<ecb_const>
functions, they can still be optimised away in many occasions, and the
compiler has more freedom in moving calls to them around.

Typical examples for such functions would be C<strlen> or C<memcmp>. A
function that calculates the MD5 sum of some input and updates some MD5
state passed as argument would I<NOT> be pure, however, as it would modify
some memory area that is not the return value.

=item ecb_hot

This declares a function as "hot" with regards to the cache - the function
is used so often, that it is very beneficial to keep it in the cache if
possible.

The compiler reacts by trying to place hot functions near to each other in
memory.

Whether a function is hot or not often depends on the whole program,
and less on the function itself. C<ecb_cold> is likely more useful in
practise.

=item ecb_cold

The opposite of C<ecb_hot> - declares a function as "cold" with regards to
the cache, or in other words, this function is not called often, or not at
speed-critical times, and keeping it in the cache might be a waste of said
cache.

In addition to placing cold functions together (or at least away from hot
functions), this knowledge can be used in other ways, for example, the
function will be optimised for size, as opposed to speed, and codepaths
leading to calls to those functions can automatically be marked as if
C<ecb_expect_false> had been used to reach them.

Good examples for such functions would be error reporting functions, or
functions only called in exceptional or rare cases.

=item ecb_artificial

Declares the function as "artificial", in this case meaning that this
function is not really meant to be a function, but more like an accessor
- many methods in C++ classes are mere accessor functions, and having a
crash reported in such a method, or single-stepping through them, is not
usually so helpful, especially when it's inlined to just a few instructions.

Marking them as artificial will instruct the debugger about just this,
leading to happier debugging and thus happier lives.

Example: in some kind of smart-pointer class, mark the pointer accessor as
artificial, so that the whole class acts more like a pointer and less like
some C++ abstraction monster.

  template<typename T>
  struct my_smart_ptr
  {
    T *value;

    ecb_artificial
    operator T *()
    {
      return value;
    }
  };

=back

=head2 OPTIMISATION HINTS

=over 4

=item bool ecb_is_constant (expr)

Returns true iff the expression can be deduced to be a compile-time
constant, and false otherwise.

For example, when you have a C<rndm16> function that returns a 16 bit
random number, and you have a function that maps this to a range from
0..n-1, then you could use this inline function in a header file:

  ecb_inline uint32_t
  rndm (uint32_t n)
  {
    return (n * (uint32_t)rndm16 ()) >> 16;
  }

However, for powers of two, you could use a normal mask, but that is only
worth it if, at compile time, you can detect this case. This is the case
when the passed number is a constant and also a power of two (C<n & (n -
1) == 0>):

  ecb_inline uint32_t
  rndm (uint32_t n)
  {
    return is_constant (n) && !(n & (n - 1))
      ? rndm16 () & (num - 1)
      : (n * (uint32_t)rndm16 ()) >> 16;
  }

=item ecb_expect (expr, value)

Evaluates C<expr> and returns it. In addition, it tells the compiler that
the C<expr> evaluates to C<value> a lot, which can be used for static
branch optimisations.

Usually, you want to use the more intuitive C<ecb_expect_true> and
C<ecb_expect_false> functions instead.

=item bool ecb_expect_true (cond)

=item bool ecb_expect_false (cond)

These two functions expect a expression that is true or false and return
C<1> or C<0>, respectively, so when used in the condition of an C<if> or
other conditional statement, it will not change the program:

  /* these two do the same thing */
  if (some_condition) ...;
  if (ecb_expect_true (some_condition)) ...;

However, by using C<ecb_expect_true>, you tell the compiler that the
condition is likely to be true (and for C<ecb_expect_false>, that it is
unlikely to be true).

For example, when you check for a null pointer and expect this to be a
rare, exceptional, case, then use C<ecb_expect_false>:

  void my_free (void *ptr)
  {
    if (ecb_expect_false (ptr == 0))
      return;
  }

Consequent use of these functions to mark away exceptional cases or to
tell the compiler what the hot path through a function is can increase
performance considerably.

You might know these functions under the name C<likely> and C<unlikely>
- while these are common aliases, we find that the expect name is easier
to understand when quickly skimming code. If you wish, you can use
C<ecb_likely> instead of C<ecb_expect_true> and C<ecb_unlikely> instead of
C<ecb_expect_false> - these are simply aliases.

A very good example is in a function that reserves more space for some
memory block (for example, inside an implementation of a string stream) -
each time something is added, you have to check for a buffer overrun, but
you expect that most checks will turn out to be false:

  /* make sure we have "size" extra room in our buffer */
  ecb_inline void
  reserve (int size)
  {
    if (ecb_expect_false (current + size > end))
      real_reserve_method (size); /* presumably noinline */
  }

=item ecb_assume (cond)

Try to tell the compiler that some condition is true, even if it's not
obvious.

This can be used to teach the compiler about invariants or other
conditions that might improve code generation, but which are impossible to
deduce form the code itself.

For example, the example reservation function from the C<ecb_expect_false>
description could be written thus (only C<ecb_assume> was added):

  ecb_inline void
  reserve (int size)
  {
    if (ecb_expect_false (current + size > end))
      real_reserve_method (size); /* presumably noinline */

    ecb_assume (current + size <= end);
  }

If you then call this function twice, like this:

  reserve (10);
  reserve (1);

Then the compiler I<might> be able to optimise out the second call
completely, as it knows that C<< current + 1 > end >> is false and the
call will never be executed.

=item ecb_unreachable ()

This function does nothing itself, except tell the compiler that it will
never be executed. Apart from suppressing a warning in some cases, this
function can be used to implement C<ecb_assume> or similar functions.

=item ecb_prefetch (addr, rw, locality)

Tells the compiler to try to prefetch memory at the given C<addr>ess
for either reading (C<rw> = 0) or writing (C<rw> = 1). A C<locality> of
C<0> means that there will only be one access later, C<3> means that
the data will likely be accessed very often, and values in between mean
something... in between. The memory pointed to by the address does not
need to be accessible (it could be a null pointer for example), but C<rw>
and C<locality> must be compile-time constants.

An obvious way to use this is to prefetch some data far away, in a big
array you loop over. This prefetches memory some 128 array elements later,
in the hope that it will be ready when the CPU arrives at that location.

  int sum = 0;

  for (i = 0; i < N; ++i)
    {
      sum += arr [i]
      ecb_prefetch (arr + i + 128, 0, 0);
    }

It's hard to predict how far to prefetch, and most CPUs that can prefetch
are often good enough to predict this kind of behaviour themselves. It
gets more interesting with linked lists, especially when you do some fair
processing on each list element:

  for (node *n = start; n; n = n->next)
    {
      ecb_prefetch (n->next, 0, 0);
      ... do medium amount of work with *n
    }

After processing the node, (part of) the next node might already be in
cache.

=back

=head2 BIT FIDDLING / BIT WIZARDRY

=over 4

=item bool ecb_big_endian ()

=item bool ecb_little_endian ()

These two functions return true if the byte order is big endian
(most-significant byte first) or little endian (least-significant byte
first) respectively.

On systems that are neither, their return values are unspecified.

=item int ecb_ctz32 (uint32_t x)

=item int ecb_ctz64 (uint64_t x)

Returns the index of the least significant bit set in C<x> (or
equivalently the number of bits set to 0 before the least significant bit
set), starting from 0. If C<x> is 0 the result is undefined.

For smaller types than C<uint32_t> you can safely use C<ecb_ctz32>.

For example:

  ecb_ctz32 (3) = 0
  ecb_ctz32 (6) = 1

=item bool ecb_is_pot32 (uint32_t x)

=item bool ecb_is_pot64 (uint32_t x)

Return true iff C<x> is a power of two or C<x == 0>.

For smaller types then C<uint32_t> you can safely use C<ecb_is_pot32>.

=item int ecb_ld32 (uint32_t x)

=item int ecb_ld64 (uint64_t x)

Returns the index of the most significant bit set in C<x>, or the number
of digits the number requires in binary (so that C<< 2**ld <= x <
2**(ld+1) >>). If C<x> is 0 the result is undefined. A common use case is
to compute the integer binary logarithm, i.e. C<floor (log2 (n))>, for
example to see how many bits a certain number requires to be encoded.

This function is similar to the "count leading zero bits" function, except
that that one returns how many zero bits are "in front" of the number (in
the given data type), while C<ecb_ld> returns how many bits the number
itself requires.

For smaller types than C<uint32_t> you can safely use C<ecb_ld32>.

=item int ecb_popcount32 (uint32_t x)

=item int ecb_popcount64 (uint64_t x)

Returns the number of bits set to 1 in C<x>.

For smaller types than C<uint32_t> you can safely use C<ecb_popcount32>.

For example:

  ecb_popcount32 (7) = 3
  ecb_popcount32 (255) = 8

=item uint8_t  ecb_bitrev8  (uint8_t  x)

=item uint16_t ecb_bitrev16 (uint16_t x)

=item uint32_t ecb_bitrev32 (uint32_t x)

Reverses the bits in x, i.e. the MSB becomes the LSB, MSB-1 becomes LSB+1
and so on.

Example:

   ecb_bitrev8 (0xa7) = 0xea
   ecb_bitrev32 (0xffcc4411) = 0x882233ff

=item uint32_t ecb_bswap16 (uint32_t x)

=item uint32_t ecb_bswap32 (uint32_t x)

=item uint64_t ecb_bswap64 (uint64_t x)

These functions return the value of the 16-bit (32-bit, 64-bit) value
C<x> after reversing the order of bytes (0x11223344 becomes 0x44332211 in
C<ecb_bswap32>).

=item uint8_t  ecb_rotl8  (uint8_t  x, unsigned int count)

=item uint16_t ecb_rotl16 (uint16_t x, unsigned int count)

=item uint32_t ecb_rotl32 (uint32_t x, unsigned int count)

=item uint64_t ecb_rotl64 (uint64_t x, unsigned int count)

=item uint8_t  ecb_rotr8  (uint8_t  x, unsigned int count)

=item uint16_t ecb_rotr16 (uint16_t x, unsigned int count)

=item uint32_t ecb_rotr32 (uint32_t x, unsigned int count)

=item uint64_t ecb_rotr64 (uint64_t x, unsigned int count)

These two families of functions return the value of C<x> after rotating
all the bits by C<count> positions to the right (C<ecb_rotr>) or left
(C<ecb_rotl>).

Current GCC versions understand these functions and usually compile them
to "optimal" code (e.g. a single C<rol> or a combination of C<shld> on
x86).

=back

=head2 FLOATING POINT FIDDLING

=over 4

=item ECB_INFINITY

Evaluates to positive infinity if supported by the platform, otherwise to
a truly huge number.

=item ECB_NON

Evaluates to a quiet NAN if supported by the platform, otherwise to
C<ECB_INFINITY>.

=item float ecb_ldexpf (float x, int exp)

Same as C<ldexpf>, but always available.

=item uint32_t ecb_float_to_binary32  (float  x) [-UECB_NO_LIBM]

=item uint64_t ecb_double_to_binary64 (double x) [-UECB_NO_LIBM]

These functions each take an argument in the native C<float> or C<double>
type and return the IEEE 754 bit representation of it.

The bit representation is just as IEEE 754 defines it, i.e. the sign bit
will be the most significant bit, followed by exponent and mantissa.

This function should work even when the native floating point format isn't
IEEE compliant, of course at a speed and code size penalty, and of course
also within reasonable limits (it tries to convert NaNs, infinities and
denormals, but will likely convert negative zero to positive zero).

On all modern platforms (where C<ECB_STDFP> is true), the compiler should
be able to optimise away this function completely.

These functions can be helpful when serialising floats to the network - you
can serialise the return value like a normal uint32_t/uint64_t.

Another use for these functions is to manipulate floating point values
directly.

Silly example: toggle the sign bit of a float.

   /* On gcc-4.7 on amd64, */
   /* this results in a single add instruction to toggle the bit, and 4 extra */
   /* instructions to move the float value to an integer register and back. */

   x = ecb_binary32_to_float (ecb_float_to_binary32 (x) ^ 0x80000000U)

=item float  ecb_binary16_to_float  (uint16_t x) [-UECB_NO_LIBM]

=item float  ecb_binary32_to_float  (uint32_t x) [-UECB_NO_LIBM]

=item double ecb_binary32_to_double (uint64_t x) [-UECB_NO_LIBM]

The reverse operation of the previous function - takes the bit
representation of an IEEE binary16, binary32 or binary64 number and
converts it to the native C<float> or C<double> format.

This function should work even when the native floating point format isn't
IEEE compliant, of course at a speed and code size penalty, and of course
also within reasonable limits (it tries to convert normals and denormals,
and might be lucky for infinities, and with extraordinary luck, also for
negative zero).

On all modern platforms (where C<ECB_STDFP> is true), the compiler should
be able to optimise away this function completely.

=back

=head2 ARITHMETIC

=over 4

=item x = ecb_mod (m, n)

Returns C<m> modulo C<n>, which is the same as the positive remainder
of the division operation between C<m> and C<n>, using floored
division. Unlike the C remainder operator C<%>, this function ensures that
the return value is always positive and that the two numbers I<m> and
I<m' = m + i * n> result in the same value modulo I<n> - in other words,
C<ecb_mod> implements the mathematical modulo operation, which is missing
in the language.

C<n> must be strictly positive (i.e. C<< >= 1 >>), while C<m> must be
negatable, that is, both C<m> and C<-m> must be representable in its
type (this typically excludes the minimum signed integer value, the same
limitation as for C</> and C<%> in C).

Current GCC versions compile this into an efficient branchless sequence on
almost all CPUs.

For example, when you want to rotate forward through the members of an
array for increasing C<m> (which might be negative), then you should use
C<ecb_mod>, as the C<%> operator might give either negative results, or
change direction for negative values:

   for (m = -100; m <= 100; ++m)
     int elem = myarray [ecb_mod (m, ecb_array_length (myarray))];

=item x = ecb_div_rd (val, div)

=item x = ecb_div_ru (val, div)

Returns C<val> divided by C<div> rounded down or up, respectively.
C<val> and C<div> must have integer types and C<div> must be strictly
positive. Note that these functions are implemented with macros in C
and with function templates in C++.

=back

=head2 UTILITY

=over 4

=item element_count = ecb_array_length (name)

Returns the number of elements in the array C<name>. For example:

  int primes[] = { 2, 3, 5, 7, 11 };
  int sum = 0;

  for (i = 0; i < ecb_array_length (primes); i++)
    sum += primes [i];

=back

=head2 SYMBOLS GOVERNING COMPILATION OF ECB.H ITSELF

These symbols need to be defined before including F<ecb.h> the first time.

=over 4

=item ECB_NO_THREADS

If F<ecb.h> is never used from multiple threads, then this symbol can
be defined, in which case memory fences (and similar constructs) are
completely removed, leading to more efficient code and fewer dependencies.

Setting this symbol to a true value implies C<ECB_NO_SMP>.

=item ECB_NO_SMP

The weaker version of C<ECB_NO_THREADS> - if F<ecb.h> is used from
multiple threads, but never concurrently (e.g. if the system the program
runs on has only a single CPU with a single core, no hyperthreading and so
on), then this symbol can be defined, leading to more efficient code and
fewer dependencies.

=item ECB_NO_LIBM

When defined to C<1>, do not export any functions that might introduce
dependencies on the math library (usually called F<-lm>) - these are
marked with [-UECB_NO_LIBM].

=back


Revision:	1.62
Committed:	Wed Feb 18 20:29:27 2015 UTC (9 years, 4 months ago) by root
Branch:	MAIN
Changes since 1.61:	+51 -5 lines
Log Message:	* empty log message *