Comments on: Multithreaded simple data type access and atomic variables http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables Sun, 05 Feb 2012 21:17:46 +0000 hourly 1 http://wordpress.org/?v= By: Alexander Sandler http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-28429 Alexander Sandler Sun, 05 Feb 2012 20:39:53 +0000 http://www.alexandersandler.net/?page_id=770#comment-28429 <a href='#comment-27528' rel="nofollow">@yt</a> Yes, your implementation uses i386 assembler. Whenever you try to do something like this on some other platform, it probably will not compile.<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('28429','Alexander Sandler'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('28429','Alexander Sandler','<a href=\'#comment-27528\' rel=\"nofollow\">@yt<\/a>\r\nYes, your implementation uses i386 assembler. Whenever you try to do something like this on some other platform, it probably will not compile.'); return false;">Quote</a></div> @yt
Yes, your implementation uses i386 assembler. Whenever you try to do something like this on some other platform, it probably will not compile.
Reply – Quote
]]> By: yt http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-27528 yt Sun, 16 Oct 2011 07:39:27 +0000 http://www.alexandersandler.net/?page_id=770#comment-27528 Thanks Alex,I has found we can implement these function by gcc asm,like __sync_fetch_and_add,I write this: typedef atomic_t int; static inline atomic_t fetch_and_add(atomic_t *value,atomic_t key) { asm volatile("lock;addl %1,%0" :"+m"(*value) :"ir"(key) :"memory"); return *value; } but it's limited for intel x86,reference this in detail: http://lxr.linux.no/linux+v2.6.26.4/include/asm-x86/atomic_32.h<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('27528','yt'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('27528','yt','Thanks Alex,I has found we can implement these function by gcc asm,like __sync_fetch_and_add,I write this:\r\ntypedef atomic_t int;\r\nstatic inline atomic_t fetch_and_add(atomic_t *value,atomic_t key)\r\n{\r\n asm volatile(\"lock;addl %1,%0\"\r\n :\"+m\"(*value)\r\n :\"ir\"(key)\r\n :\"memory\");\r\n return *value;\r\n}\r\nbut it\'s limited for intel x86,reference this in detail:\r\nhttp:\/\/lxr.linux.no\/linux+v2.6.26.4\/include\/asm-x86\/atomic_32.h'); return false;">Quote</a></div> Thanks Alex,I has found we can implement these function by gcc asm,like __sync_fetch_and_add,I write this:
typedef atomic_t int;
static inline atomic_t fetch_and_add(atomic_t *value,atomic_t key)
{
asm volatile(“lock;addl %1,%0″
:”+m”(*value)
:”ir”(key)
:”memory”);
return *value;
}
but it’s limited for intel x86,reference this in detail:
http://lxr.linux.no/linux+v2.6.26.4/include/asm-x86/atomic_32.h
Reply – Quote
]]> By: Alain Spineux http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-25382 Alain Spineux Fri, 08 Apr 2011 11:08:12 +0000 http://www.alexandersandler.net/?page_id=770#comment-25382 This and the two before are great articles, useful and pleasant to read. Thanks<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('25382','Alain Spineux'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('25382','Alain Spineux','This and the two before are great articles, useful and pleasant to read.\r\nThanks'); return false;">Quote</a></div> This and the two before are great articles, useful and pleasant to read.
Thanks
Reply – Quote
]]> By: 簡單而有趣的雙執行緒同步問題 « Hitripod http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24758 簡單而有趣的雙執行緒同步問題 « Hitripod Tue, 11 Jan 2011 17:28:01 +0000 http://www.alexandersandler.net/?page_id=770#comment-24758 [...] 今天看到一系列有趣的文章：1, 2, 3 [...]
Reply – Quote
]]> By: Zamir http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24450 Zamir Tue, 02 Nov 2010 15:48:15 +0000 http://www.alexandersandler.net/?page_id=770#comment-24450 Alex, This is a wonderful piece. Thank you for helping us understand atomic operations in C!<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24450','Zamir'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24450','Zamir','Alex,\r\nThis is a wonderful piece. Thank you for helping us understand atomic operations in C!'); return false;">Quote</a></div> Alex,
This is a wonderful piece. Thank you for helping us understand atomic operations in C!
Reply – Quote
]]> By: Mohammed Shahid http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24357 Mohammed Shahid Sat, 09 Oct 2010 18:45:39 +0000 http://www.alexandersandler.net/?page_id=770#comment-24357 Wouldn't __sync_fetch_and_sub correspond to fetch-and-subtract rather than fetch-and-substitute ?<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24357','Mohammed Shahid'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24357','Mohammed Shahid','Wouldn\'t __sync_fetch_and_sub correspond to fetch-and-subtract rather than fetch-and-substitute ?'); return false;">Quote</a></div> Wouldn’t __sync_fetch_and_sub correspond to fetch-and-subtract rather than fetch-and-substitute ?
Reply – Quote
]]> By: Michael Cole http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24296 Michael Cole Wed, 22 Sep 2010 15:59:02 +0000 http://www.alexandersandler.net/?page_id=770#comment-24296 Exceptionally beneficial bless you, It looks like your trusty readers would most likely want far more writing of this nature carry on the excellent work.<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24296','Michael Cole'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24296','Michael Cole','Exceptionally beneficial bless you, It looks like your trusty readers would most likely want far more writing of this nature carry on the excellent work.'); return false;">Quote</a></div> Exceptionally beneficial bless you, It looks like your trusty readers would most likely want far more writing of this nature carry on the excellent work.
Reply – Quote
]]> By: Alexander Sandler http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24255 Alexander Sandler Mon, 13 Sep 2010 13:19:22 +0000 http://www.alexandersandler.net/?page_id=770#comment-24255 <a href='#comment-24164' rel="nofollow">@Manju</a> 1. __sync_fetch_and_add(pts, 0) is the way! 2. The question of interrupt safety is irrelevant here because all these functions translate into single processor instruction - interrupts land between instructions, not in the middle of an instruction. 3. There are couple of solutions for this. Most of them are not pretty. First there's a __sync_lock_test_and_set(). gcc documentation says that one should be careful when using this function because it is not implemented on all gcc platforms. Should be safe on Intel though. Another option is to do assignments before you use the variable atomically and then to do relative adjustments only (those you can do atomically). This is not very convenient, but this is all gcc got. Hope it helps.<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24255','Alexander Sandler'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24255','Alexander Sandler','<a href=\'#comment-24164\' rel=\"nofollow\">@Manju<\/a>\r\n1. __sync_fetch_and_add(pts, 0) is the way!\r\n2. The question of interrupt safety is irrelevant here because all these functions translate into single processor instruction - interrupts land between instructions, not in the middle of an instruction.\r\n3. There are couple of solutions for this. Most of them are not pretty. First there\'s a __sync_lock_test_and_set(). gcc documentation says that one should be careful when using this function because it is not implemented on all gcc platforms. Should be safe on Intel though.\r\nAnother option is to do assignments before you use the variable atomically and then to do relative adjustments only (those you can do atomically). This is not very convenient, but this is all gcc got.\r\nHope it helps.'); return false;">Quote</a></div> @Manju
1. __sync_fetch_and_add(pts, 0) is the way!
2. The question of interrupt safety is irrelevant here because all these functions translate into single processor instruction – interrupts land between instructions, not in the middle of an instruction.
3. There are couple of solutions for this. Most of them are not pretty. First there’s a __sync_lock_test_and_set(). gcc documentation says that one should be careful when using this function because it is not implemented on all gcc platforms. Should be safe on Intel though.
Another option is to do assignments before you use the variable atomically and then to do relative adjustments only (those you can do atomically). This is not very convenient, but this is all gcc got.
Hope it helps.
Reply – Quote
]]> By: Manju http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24164 Manju Sun, 29 Aug 2010 04:43:40 +0000 http://www.alexandersandler.net/?page_id=770#comment-24164 One more 3. whats the equivalent of atomic_set ? I see in atomic.h that atomic_set is defined as #define atomic_read(v) ((v)->counter) So a normal assignment statement should be safe here ?<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24164','Manju'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24164','Manju','One more\r\n3. whats the equivalent of atomic_set ?\r\n I see in atomic.h that atomic_set is defined as\r\n#define atomic_read(v) ((v)->counter)\r\nSo a normal assignment statement should be safe here ?'); return false;">Quote</a></div> One more
3. whats the equivalent of atomic_set ?
I see in atomic.h that atomic_set is defined as
#define atomic_read(v) ((v)->counter)
So a normal assignment statement should be safe here ?
Reply – Quote
]]> By: Manju http://www.alexonlinux.com/multithreaded-simple-data-type-access-and-atomic-variables/comment-page-1#comment-24162 Manju Sun, 29 Aug 2010 04:30:41 +0000 http://www.alexandersandler.net/?page_id=770#comment-24162 Hello Alex, Thanks for informative article and interesting conversations following that. I have two queries. 1. I need to do a automic_read which of the __sync shd I use. __sync_fetch_and_add(pts, 0) be fine ? or is there any other specific function for read 2. Are sync* function interrupt safe ? i.e. will gcc take care of disabling interrupts while performing these operations ? Thanks<div class="comment-remix-meta"><a href="#" class="replyto" onclick="replyto('24162','Manju'); return false;">Reply</a> - <a href="#" class="quote" onclick="quote('24162','Manju','Hello Alex,\r\nThanks for informative article and interesting conversations following that.\r\nI have two queries.\r\n1. I need to do a automic_read which of the __sync shd I use.\r\n __sync_fetch_and_add(pts, 0) be fine ? or is there any other specific function for read\r\n\r\n2. Are sync* function interrupt safe ? i.e. will gcc take care of disabling interrupts while performing these operations ?\r\n\r\nThanks'); return false;">Quote</a></div> Hello Alex,
Thanks for informative article and interesting conversations following that.
I have two queries.
1. I need to do a automic_read which of the __sync shd I use.
__sync_fetch_and_add(pts, 0) be fine ? or is there any other specific function for read

2. Are sync* function interrupt safe ? i.e. will gcc take care of disabling interrupts while performing these operations ?

Thanks

Reply – Quote
]]>