perlmulticore/perlmulticore.pod

=head1 NAME

The Perl Multicore Specification and Implementation

=head1 SYNOPSIS

  #include "perlmultiore.h"

  // in your XS function:

  perlinterp_release ();
  do_the_C_thing ();
  perlinterp_acquire ();

=head1 DESCRIPTION

This header file implements a simple mechanism for XS modules to allow
re-use of the perl interpreter for other threads while doing some lengthy
operation, such as cryptography, SQL queries, disk I/O and so on.

The design goals for this mechanism were to be simple to use, very
efficient when not needed, low code and data size overhead and broad
applicability.

The newest version of this document can be found at
L<http://perlmulticore.schmorp.de/>.

The newest version of the header file itself, can be downloaded from
L<http://perlmulticore.schmorp.de/perlmulticore.h>.

=head1 HOW DO I USE THIS IN MY MODULES?

The usage is very simple - you include this header file in your XS module. Then, before you
do your lengthy operation, you release the perl interpreter:

   perlinterp_release ();

And when you are done with your computation, you acquire it again:

   perlinterp_acquire ();

And that's it. This doesn't load any modules and consists of only a few
machine instructions when no module to take advantage of it is loaded.

Here is a simple example, an C<flock> wrapper implemented in XS. Unlike
perl's built-in C<flock>, it allows other threads (for example, those
provided by L<Coro>) to execute, instead of blocking the whole perl
interpreter. For the sake of this example, it requires a file descriptor
instead of a handle.

   #include "perlmulticore.h" // this header file

   // and in the XS portion
   int flock (int fd, int operation)
           CODE:
           perlinterp_release ();
           RETVAL = flock (fd, operation);
           perlinterp_acquire ();
           OUTPUT:
           RETVAL

Another example would be to modify L<DBD::mysql> to allow other
threads to execute while executing SQL queries. One way to do this
is find all C<mysql_st_internal_execute> and similar calls (such as
C<mysql_st_internal_execute41>), and adorn them with release/acquire
calls:

   {
     perlinterp_release ();
     imp_sth->row_num= mysql_st_internal_execute(sth, ...);
     perlinterp_acquire ();
   }

=head2 HOW ABOUT NOT-SO LONG WORK?

Sometimes you don't know how long your code will take - in a compression
library for example, compressing a few hundred Kilobyte of data can take
a while, while 50 Bytes will compress so fast that even attempting to do
something else could be more costly than just doing it.

This is a very hard problem to solve. The best you can do at the moment is
to release the perl interpreter only when you think the work to be done
justifies the expense.

As a rule of thumb, if you expect to need more than a few thousand cycles,
you should release the interpreter, else you shouldn't. When in doubt,
release.

For example, in a compression library, you might want to do this:

   if (bytes_to_be_compressed > 2000) perlinterp_release ();
   do_compress (...);
   if (bytes_to_be_compressed > 2000) perlinterp_acquire ();

Make sure the if conditions are exactly the same and don't change, so you
always call acquire when you release, and vice versa.

When you don't have a handy indicator, you might still do something
useful. For example, if you do some file locking with C<fcntl> and you
expect the lock to be available immediately in most cases, you could try
with C<F_SETLK> (which doesn't wait), and only release/wait/acquire when
the lock couldn't be set:

   int res = fcntl (fd, F_SETLK, &flock);

   if (res)
     {
       // error, assume lock is held by another process and do it the slow way
       perlinterp_release ();
       res = fcntl (fd, F_SETLKW, &flock);
       perlinterp_acquire ();
     }

=head1 THE HARD AND FAST RULES

As with everything, there are a number of rules to follow.

=over 4

=item I<Never> touch any perl data structures after calling C<perlinterp_release>.

Possibly the most important rule of them all, anything perl is
completely off-limits after C<perlinterp_release>, until you call
C<perlinterp_acquire>, after which you can access perl stuff again.

That includes anything in the perl interpreter that you didn't prove to be
safe, and didn't prove to be safe in older and future versions of perl:
global variables, local perl scalars, even if you are sure nobody accesses
them and you only try to "read" their value, and so on.

If you need to access perl things, do it before releasing the
interpreter with C<perlinterp_release>, or after acquiring it again with
C<perlinterp_acquire>.

=item I<Always> call C<perlinterp_release> and C<perlinterp_acquire> in pairs.

For each C<perlinterp_release> call there must be a C<perlinterp_acquire>
call. They don't have to be in the same function, and you can have
multiple calls to them, as long as every C<perlinterp_release> call is
followed by exactly one C<perlinterp_acquire> call.

For example., this would be fine:

   perlinterp_release ();

   if (!function_that_fails_with_0_return_value ())
     {
       perlinterp_acquire ();
       croak ("error");
       // croak doesn't return
     }

   perlinterp_acquire ();
   // do other stuff

=item I<Never> nest calls to C<perlinterp_release> and C<perlinterp_acquire>.

That simply means that after calling C<perlinterp_release>, you must
call C<perlinterp_acquire> before calling C<perlinterp_release>
again. Likewise, after C<perlinterp_acquire>, you can call
C<perlinterp_release> but not another C<perlinterp_acquire>.

=item I<Always> call C<perlinterp_release> first.

Also simple: you I<must not> call C<perlinterp_acquire> without having
called C<perlinterp_release> before.

=item I<Never> underestimate threads.

While it's easy to add parallel execution ability to your XS module, it
doesn't mean it is safe. After you release the perl interpreter, it's
perfectly possible that it will call your XS function in another thread,
even while your original function still executes. In other words: your C
code must be thread safe, and if you use any library, that library must be
thread-safe, too.

Always assume that the code between C<perlinterp_release> and
C<perlinterp_acquire> is executed in parallel on multiple CPUs at the same
time. If your code can't cope with that, you could consider using a mutex
to only allow one such execution, which is still better than blocking
everybody else from doing anything:

   static pthread_mutex_t my_mutex = PTHREAD_MUTEX_INITIALIZER;

   perlinterp_release ();
   pthread_mutex_lock (&my_mutex);
   do_your_non_thread_safe_thing ();
   pthread_mutex_unlock (&my_mutex);
   perlinterp_acquire ();

=item I<Don't> get confused by having to release first.

In many real world scenarios, you acquire a resource, do something, then
release it again. Don't let this confuse you, with this, you already own
the resource (the perl interpreter) so you have to I<release> first, and
I<acquire> it again later, not the other way around.

=back


=head1 DESIGN PRINCIPLES

This section discusses how the design goals were reached (you be the
judge), how it is implemented, and what overheads this implies.

=over 4

=item Simple to Use

All you have to do is identify the place in your existing code where you
stop touching perl stuff, do your actual work, and start touching perl
stuff again.

Then slap C<perlinterp_release ()> and C<perlinterp_acquire ()> around the
actual work code.

You have to include F<perlmulticore.h> and distribute it with your XS
code, but all these things border on the trivial.

=item Very Efficient

The definition for C<perlinterp_release> and C<perlinterp_release> is very
short:

   #define perlinterp_release() perl_multicore_api->pmapi_release ()
   #define perlinterp_acquire() perl_multicore_api->pmapi_acquire ()

Both are macros that read a pointer from memory (perl_multicore_api),
dereference a function pointer stored at that place, and call the
function, which takes no arguments and returns nothing.

The first call to C<perlinterp_release> will check for the presence
of any supporting module, and if none is loaded, will create a dummy
implementation where both C<pmapi_release> and C<pmapi_acquire> execute
this function:

  static void perl_multicore_nop (void) { }

So in the case of no magical module being loaded, all calls except the
first are two memory accesses and a predictable function call of an empty
function.

Of course, the overhead is much higher when these functions actually
implement anything useful, but you always get what you pay for.

With L<Coro::Multicore>, every release/acquire involves two pthread
switches, two coro thread switches, a bunch of syscalls, and sometimes
interacting with the event loop.

A dedicated thread pool such as the one L<IO::AIO> uses could reduce
these overheads, and would also reduce the dependencies (L<AnyEvent> is a
smaller and more portable dependency than L<Coro>), but it would require a
lot more work on the side of the module author wanting to support it than
this solution.

=item Low Code and Data Size Overhead

On a 64 bit system, F<perlmulticore.h> uses exactly C<8> octets (one
pointer) of your data segment, to store the C<perl_multicore_api>
pointer. In addition it creates a C<16> octet perl string to store the
function pointers in, and stores it in a hash provided by perl for this
purpose.

This is pretty much the equivalent of executing this code:

   $existing_hash{perl_multicore_api} = "123456781234567812345678";

And that's it, which is, as I think, indeed very little.

As for code size, on my amd64 system, every call to C<perlinterp_release>
or C<perlinterp_acquire> results in a variation of the following 9-10
octet sequence:

   150>   mov    0x200f23(%rip),%rax  # <perl_multicore_api>
   157>   callq  *0x8(%rax)

The biggest part if the initialisation code, which consists of 11 lines of
typical XS code. On my system, all the code in F<perlmulticore.h> compiles
to less than 160 octets of read-only data.

=item Broad Applicability

While there are alternative ways to achieve the goal of parallel execution
with threads that might be more efficient, this mechanism was chosen
because it is very simple to retrofit existing modules with it, and it

The design goals for this mechanism were to be simple to use, very
efficient when not needed, low code and data size overhead and broad
applicability.

=back


=head1 DISABLING PERL MULTICORE AT COMPILE TIME

You can disable the complete perl multicore API by defining the
symbol C<PERL_MULTICORE_DISABLE> to C<1> (e.g. by specifying
F<-DPERL_MULTICORE_DISABLE> as compiler argument).

This will leave no traces of the API in the compiled code, suitable
"empty" C<perl_release> and C<perl_acquire> definitions will be provided.

This could be added to perl's C<CPPFLAGS> when configuring perl on
platforms that do not support threading at all for example.


=head1 AUTHOR

   Marc A. Lehmann <perlmulticore@schmorp.de>
   http://perlmulticore.schmorp.de/

=head1 LICENSE

The F<perlmulticore.h> header file is put into the public
domain. Where this is legally not possible, or at your
option, it can be licensed under creativecommons CC0
license: L<https://creativecommons.org/publicdomain/zero/1.0/>.

Revision:	1.1
Committed:	Thu Jul 2 22:39:56 2015 UTC (9 years, 3 months ago) by root
Branch:	MAIN
Log Message:	* empty log message *
#	Content
1	=head1 NAME
2
3	The Perl Multicore Specification and Implementation
4
5	=head1 SYNOPSIS
6
7	#include "perlmultiore.h"
8
9	// in your XS function:
10
11	perlinterp_release ();
12	do_the_C_thing ();
13	perlinterp_acquire ();
14
15	=head1 DESCRIPTION
16
17	This header file implements a simple mechanism for XS modules to allow
18	re-use of the perl interpreter for other threads while doing some lengthy
19	operation, such as cryptography, SQL queries, disk I/O and so on.
20
21	The design goals for this mechanism were to be simple to use, very
22	efficient when not needed, low code and data size overhead and broad
23	applicability.
24
25	The newest version of this document can be found at
26	L<http://perlmulticore.schmorp.de/>.
27
28	The newest version of the header file itself, can be downloaded from
29	L<http://perlmulticore.schmorp.de/perlmulticore.h>.
30
31	=head1 HOW DO I USE THIS IN MY MODULES?
32
33	The usage is very simple - you include this header file in your XS module. Then, before you
34	do your lengthy operation, you release the perl interpreter:
35
36	perlinterp_release ();
37
38	And when you are done with your computation, you acquire it again:
39
40	perlinterp_acquire ();
41
42	And that's it. This doesn't load any modules and consists of only a few
43	machine instructions when no module to take advantage of it is loaded.
44
45	Here is a simple example, an C<flock> wrapper implemented in XS. Unlike
46	perl's built-in C<flock>, it allows other threads (for example, those
47	provided by L<Coro>) to execute, instead of blocking the whole perl
48	interpreter. For the sake of this example, it requires a file descriptor
49	instead of a handle.
50
51	#include "perlmulticore.h" // this header file
52
53	// and in the XS portion
54	int flock (int fd, int operation)
55	CODE:
56	perlinterp_release ();
57	RETVAL = flock (fd, operation);
58	perlinterp_acquire ();
59	OUTPUT:
60	RETVAL
61
62	Another example would be to modify L<DBD::mysql> to allow other
63	threads to execute while executing SQL queries. One way to do this
64	is find all C<mysql_st_internal_execute> and similar calls (such as
65	C<mysql_st_internal_execute41>), and adorn them with release/acquire
66	calls:
67
68	{
69	perlinterp_release ();
70	imp_sth->row_num= mysql_st_internal_execute(sth, ...);
71	perlinterp_acquire ();
72	}
73
74	=head2 HOW ABOUT NOT-SO LONG WORK?
75
76	Sometimes you don't know how long your code will take - in a compression
77	library for example, compressing a few hundred Kilobyte of data can take
78	a while, while 50 Bytes will compress so fast that even attempting to do
79	something else could be more costly than just doing it.
80
81	This is a very hard problem to solve. The best you can do at the moment is
82	to release the perl interpreter only when you think the work to be done
83	justifies the expense.
84
85	As a rule of thumb, if you expect to need more than a few thousand cycles,
86	you should release the interpreter, else you shouldn't. When in doubt,
87	release.
88
89	For example, in a compression library, you might want to do this:
90
91	if (bytes_to_be_compressed > 2000) perlinterp_release ();
92	do_compress (...);
93	if (bytes_to_be_compressed > 2000) perlinterp_acquire ();
94
95	Make sure the if conditions are exactly the same and don't change, so you
96	always call acquire when you release, and vice versa.
97
98	When you don't have a handy indicator, you might still do something
99	useful. For example, if you do some file locking with C<fcntl> and you
100	expect the lock to be available immediately in most cases, you could try
101	with C<F_SETLK> (which doesn't wait), and only release/wait/acquire when
102	the lock couldn't be set:
103
104	int res = fcntl (fd, F_SETLK, &flock);
105
106	if (res)
107	{
108	// error, assume lock is held by another process and do it the slow way
109	perlinterp_release ();
110	res = fcntl (fd, F_SETLKW, &flock);
111	perlinterp_acquire ();
112	}
113
114	=head1 THE HARD AND FAST RULES
115
116	As with everything, there are a number of rules to follow.
117
118	=over 4
119
120	=item I<Never> touch any perl data structures after calling C<perlinterp_release>.
121
122	Possibly the most important rule of them all, anything perl is
123	completely off-limits after C<perlinterp_release>, until you call
124	C<perlinterp_acquire>, after which you can access perl stuff again.
125
126	That includes anything in the perl interpreter that you didn't prove to be
127	safe, and didn't prove to be safe in older and future versions of perl:
128	global variables, local perl scalars, even if you are sure nobody accesses
129	them and you only try to "read" their value, and so on.
130
131	If you need to access perl things, do it before releasing the
132	interpreter with C<perlinterp_release>, or after acquiring it again with
133	C<perlinterp_acquire>.
134
135	=item I<Always> call C<perlinterp_release> and C<perlinterp_acquire> in pairs.
136
137	For each C<perlinterp_release> call there must be a C<perlinterp_acquire>
138	call. They don't have to be in the same function, and you can have
139	multiple calls to them, as long as every C<perlinterp_release> call is
140	followed by exactly one C<perlinterp_acquire> call.
141
142	For example., this would be fine:
143
144	perlinterp_release ();
145
146	if (!function_that_fails_with_0_return_value ())
147	{
148	perlinterp_acquire ();
149	croak ("error");
150	// croak doesn't return
151	}
152
153	perlinterp_acquire ();
154	// do other stuff
155
156	=item I<Never> nest calls to C<perlinterp_release> and C<perlinterp_acquire>.
157
158	That simply means that after calling C<perlinterp_release>, you must
159	call C<perlinterp_acquire> before calling C<perlinterp_release>
160	again. Likewise, after C<perlinterp_acquire>, you can call
161	C<perlinterp_release> but not another C<perlinterp_acquire>.
162
163	=item I<Always> call C<perlinterp_release> first.
164
165	Also simple: you I<must not> call C<perlinterp_acquire> without having
166	called C<perlinterp_release> before.
167
168	=item I<Never> underestimate threads.
169
170	While it's easy to add parallel execution ability to your XS module, it
171	doesn't mean it is safe. After you release the perl interpreter, it's
172	perfectly possible that it will call your XS function in another thread,
173	even while your original function still executes. In other words: your C
174	code must be thread safe, and if you use any library, that library must be
175	thread-safe, too.
176
177	Always assume that the code between C<perlinterp_release> and
178	C<perlinterp_acquire> is executed in parallel on multiple CPUs at the same
179	time. If your code can't cope with that, you could consider using a mutex
180	to only allow one such execution, which is still better than blocking
181	everybody else from doing anything:
182
183	static pthread_mutex_t my_mutex = PTHREAD_MUTEX_INITIALIZER;
184
185	perlinterp_release ();
186	pthread_mutex_lock (&my_mutex);
187	do_your_non_thread_safe_thing ();
188	pthread_mutex_unlock (&my_mutex);
189	perlinterp_acquire ();
190
191	=item I<Don't> get confused by having to release first.
192
193	In many real world scenarios, you acquire a resource, do something, then
194	release it again. Don't let this confuse you, with this, you already own
195	the resource (the perl interpreter) so you have to I<release> first, and
196	I<acquire> it again later, not the other way around.
197
198	=back
199
200
201	=head1 DESIGN PRINCIPLES
202
203	This section discusses how the design goals were reached (you be the
204	judge), how it is implemented, and what overheads this implies.
205
206	=over 4
207
208	=item Simple to Use
209
210	All you have to do is identify the place in your existing code where you
211	stop touching perl stuff, do your actual work, and start touching perl
212	stuff again.
213
214	Then slap C<perlinterp_release ()> and C<perlinterp_acquire ()> around the
215	actual work code.
216
217	You have to include F<perlmulticore.h> and distribute it with your XS
218	code, but all these things border on the trivial.
219
220	=item Very Efficient
221
222	The definition for C<perlinterp_release> and C<perlinterp_release> is very
223	short:
224
225	#define perlinterp_release() perl_multicore_api->pmapi_release ()
226	#define perlinterp_acquire() perl_multicore_api->pmapi_acquire ()
227
228	Both are macros that read a pointer from memory (perl_multicore_api),
229	dereference a function pointer stored at that place, and call the
230	function, which takes no arguments and returns nothing.
231
232	The first call to C<perlinterp_release> will check for the presence
233	of any supporting module, and if none is loaded, will create a dummy
234	implementation where both C<pmapi_release> and C<pmapi_acquire> execute
235	this function:
236
237	static void perl_multicore_nop (void) { }
238
239	So in the case of no magical module being loaded, all calls except the
240	first are two memory accesses and a predictable function call of an empty
241	function.
242
243	Of course, the overhead is much higher when these functions actually
244	implement anything useful, but you always get what you pay for.
245
246	With L<Coro::Multicore>, every release/acquire involves two pthread
247	switches, two coro thread switches, a bunch of syscalls, and sometimes
248	interacting with the event loop.
249
250	A dedicated thread pool such as the one L<IO::AIO> uses could reduce
251	these overheads, and would also reduce the dependencies (L<AnyEvent> is a
252	smaller and more portable dependency than L<Coro>), but it would require a
253	lot more work on the side of the module author wanting to support it than
254	this solution.
255
256	=item Low Code and Data Size Overhead
257
258	On a 64 bit system, F<perlmulticore.h> uses exactly C<8> octets (one
259	pointer) of your data segment, to store the C<perl_multicore_api>
260	pointer. In addition it creates a C<16> octet perl string to store the
261	function pointers in, and stores it in a hash provided by perl for this
262	purpose.
263
264	This is pretty much the equivalent of executing this code:
265
266	$existing_hash{perl_multicore_api} = "123456781234567812345678";
267
268	And that's it, which is, as I think, indeed very little.
269
270	As for code size, on my amd64 system, every call to C<perlinterp_release>
271	or C<perlinterp_acquire> results in a variation of the following 9-10
272	octet sequence:
273
274	150> mov 0x200f23(%rip),%rax # <perl_multicore_api>
275	157> callq *0x8(%rax)
276
277	The biggest part if the initialisation code, which consists of 11 lines of
278	typical XS code. On my system, all the code in F<perlmulticore.h> compiles
279	to less than 160 octets of read-only data.
280
281	=item Broad Applicability
282
283	While there are alternative ways to achieve the goal of parallel execution
284	with threads that might be more efficient, this mechanism was chosen
285	because it is very simple to retrofit existing modules with it, and it
286
287	The design goals for this mechanism were to be simple to use, very
288	efficient when not needed, low code and data size overhead and broad
289	applicability.
290
291	=back
292
293
294	=head1 DISABLING PERL MULTICORE AT COMPILE TIME
295
296	You can disable the complete perl multicore API by defining the
297	symbol C<PERL_MULTICORE_DISABLE> to C<1> (e.g. by specifying
298	F<-DPERL_MULTICORE_DISABLE> as compiler argument).
299
300	This will leave no traces of the API in the compiled code, suitable
301	"empty" C<perl_release> and C<perl_acquire> definitions will be provided.
302
303	This could be added to perl's C<CPPFLAGS> when configuring perl on
304	platforms that do not support threading at all for example.
305
306
307	=head1 AUTHOR
308
309	Marc A. Lehmann <perlmulticore@schmorp.de>
310	http://perlmulticore.schmorp.de/
311
312	=head1 LICENSE
313
314	The F<perlmulticore.h> header file is put into the public
315	domain. Where this is legally not possible, or at your
316	option, it can be licensed under creativecommons CC0
317	license: L<https://creativecommons.org/publicdomain/zero/1.0/>.
318