October 2006 - Darwin-dev - lists.apple.com

Re: Port from Win to OSX with Inline functions
by seebs＠plethora.net 10 Oct '06

10 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com In message <4A08BBCE-BFE7-46A7-8BDD-A4A7323E011F(a)thatsawesomesoftware.com>, Chr is Edgington writes: >I thought "inline" was a compiler issue - not a linker issue? It can affect both; there's some very weird magic to handle the case where: 1. A function can't ALWAYS be inlined, so a definition needs to be provided. but 2. There is more than one module that may provide that definition. -s _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Port from Win to OSX with Inline functions
by Chris Edgington 10 Oct '06

10 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com I thought "inline" was a compiler issue - not a linker issue? Definitely the prototype is inline. Here is the sample I created: ------ macbook:~/dev/inline chris$ cat macinline1.c #include <stdio.h> #define INLINER inline INLINER void macfn1(void); INLINER void macfn1(void) { printf("macfn1-1-->\n"); return; } void macfn2(void) { printf("macfn2-->\n"); macfn1(); return; } ------ ------ ------ macbook:~/dev/inline chris$ make gcc -o macinline macmain.o macinline1.o macinline2.o /usr/bin/ld: multiple definitions of symbol _macfn1 macinline1.o definition of _macfn1 in section (__TEXT,__text) macinline2.o definition of _macfn1 in section (__TEXT,__text) collect2: ld returned 1 exit status make: *** [macinline] Error 1 Since, by definition: 1) An inline function is short 2) Its definition must be available to anyone who calls it Later, -Chris _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… 2) By default - an inlined function will also have a regular non- inlined copy in the object file. So - the short-term answer is to redefine __INLINE as "extern inline __attribute((always_inline))" and to build with optimizations enabled. Ultimately, however, I will likely need to change the prorotypes and function definitions to be static - so they work whether or not they are inlined. Actually, the only thing inline means is that it's OK if there are multiple definitions of a function and the linker shouldn't complain. I believe is that your problem is that your prototypes aren't inline -- you MUST fix at least that -- otherwise, the non-inlined prototype will take precidence over the inline definition. Here is the assembly code produced by the compiler with optimization turned on: macbook:~/dev/inline chris$ cat macinline1.s .cstring LC0: .ascii "macfn1-1-->\0" .text .globl _macfn1 _macfn1: pushl %ebp movl %esp, %ebp pushl %ebx subl $20, %esp call ___i686.get_pc_thunk.bx "L00000000001$pb": leal LC0-"L00000000001$pb"(%ebx), %eax movl %eax, (%esp) call L_puts$stub addl $20, %esp popl %ebx popl %ebp ret .cstring LC1: .ascii "macfn2-->\0" .text .globl _macfn2 _macfn2: pushl %ebp movl %esp, %ebp pushl %ebx subl $20, %esp call ___i686.get_pc_thunk.bx "L00000000002$pb": leal LC1-"L00000000002$pb"(%ebx), %eax movl %eax, (%esp) call L_puts$stub leal LC0-"L00000000002$pb"(%ebx), %eax movl %eax, (%esp) call L_puts$stub addl $20, %esp popl %ebx popl %ebp ret .section __IMPORT,__jump_table,symbol_stubs,self_modifying_code +pure_instructions,5 L_puts$stub: .indirect_symbol _puts hlt ; hlt ; hlt ; hlt ; hlt .subsections_via_symbols .section __TEXT,__textcoal_nt,coalesced,pure_instructions .weak_definition ___i686.get_pc_thunk.bx .private_extern ___i686.get_pc_thunk.bx ___i686.get_pc_thunk.bx: movl (%esp), %ebx ret As you can see - the function call is inlined - there is no call to macfn1 - the printf has been inlined. However, the compiler is still embedding a ".globl _macfn1" version of the function. When I build with a copy of this .C file included in the build (with macfn2 renamed to macfn3) - the linker err's out with this: Honestly - that is what I would expect? If a global function is defined in multiple .o files, how is the linker supposed to know that those functions are all the same - certainly not by the name? I always define inline functions in their header (with no separate declaration). If the inline keyword is in the declaration, there is no need to declare them static -- in fact this will cause code bloat if they are referenced from multiple files and not inlined. Simply using inline will cause the linker to keep just one implementation, but for this to happen, all declarations must be inline. I do not see this happening. In fact, I see nothing in the generated assembly that would clue the linker into the fact that the function was inlined - and again, this is not surprising to me - since inline is a compile-time operation, not a link-time operation. In fact, here is a statement from the gcc inline doc pointer from Colin ... "When an inline function is not static, then the compiler must assume that there may be calls from other source files; since a global symbol can be defined only once in any program, the function must not be defined in the other source files, so the calls therein cannot be integrated. Therefore, a non-static inline function is always compiled on its own in the usual fashion." This email sent to site_archiver(a)lists.apple.com

1 0

Re: Port from Win to OSX with Inline functions
by Colin Hirsch 10 Oct '06

10 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mutt/1.5.9i > >>1) No inlining at all happens unless optimization is enabled - > >>regardless of the always_inline attribute. > > > >Yup, I filed a bug on that and it was closed as "works ad designed" > >-- seem the compiler folks, consider inlining an "optimization". > > In my experience, the always_inline attribute always inlines as long > as the function is inlineable, regardless of optimization level. > However, declaring a function as inline without the always_inline > attribute will cause it to not be inlined when building unoptimized > code. The GCC documentation has a quite comprehensive page on inlining: http://gcc.gnu.org/onlinedocs/gcc-4.0.3/gcc/Inline.html Regards, Colin _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Re: can I get maven onto my Mac?
by Shawn Erickson 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=T6PeS9pnfCNIIAa1VeVTr9jwE9TgxTm2zS6yskhc90LmPs6B+bY1j0+BUjFvespDiYzqfBKUz8OckNCwRgYaYXaqJPBehxaFOzoRfXIh7FMw7LjgAQgp/DKWaRoX19BPhUgGzis6OeN2brbdTTgk7CV3bCKrKgCktYOGB2yT90E= On 10/9/06, Paul Archibald <pxl(a)sonic.net> wrote: I doubt that is a correct path to the JDK ... Java on Mac OS X doesn't live at that location and Apple is the only supported provider for Java on the Mac. In general you should set JAVA_HOME to "/Library/Java/Home" if you want to use the default version of Java for the system (currently defaulted to Java 5 on an up-to-date Mac OS X 10.4 system). If you want to use a specific version of Java you would use a path like the following /System/Library/Frameworks/JavaVM.framework/Versions/<version number>/Home where <version number> is 1.3.1, 1.4.2, 1.5.0, 1.6.0, etc. depending on the versions of Java found on the system. -Shawn _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… 209-148-127-8:~ paularchibald$ export JAVA_HOME=usr/java/jdk1.5.0_02 This email sent to site_archiver(a)lists.apple.com

1 0

Re: can I get maven onto my Mac?
by Markus Hitter 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Am 09.10.2006 um 23:41 schrieb Paul Archibald: A typical step would be to create this directory: sudo mkdir /usr/local Most likely you want an absolute path here: export JAVA_HOME=/usr/java/jdk1.5.0_02 Does this directory exist? Well, you'll have to find it, then: sudo find / -name mvn (ignore the errors) Markus - - - - - - - - - - - - - - - - - - - Dipl. Ing. Markus Hitter http://www.jump-ing.de/ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… (I don't have a "local"). export JAVA_HOME=usr/java/jdk1.5.0_02 -bash: mvn: command not found Judging by your mistakes, you either want to learn ho to install common Unix software on Mac OS X or you want to have a look at something like Fink or Darwinports. This email sent to site_archiver(a)lists.apple.com

1 0

Re: can I get maven onto my Mac?
by Paul Archibald 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Port from Win to OSX with Inline functions
by Eric Albert 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Oct 9, 2006, at 7:34 AM, Steve Sisak wrote: -Eric _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… At 10:03 AM -0400 10/9/06, Chris Edgington wrote: FYI ... found a short-term solution ... and some interesting things ... 1) No inlining at all happens unless optimization is enabled - regardless of the always_inline attribute. Yup, I filed a bug on that and it was closed as "works ad designed" -- seem the compiler folks, consider inlining an "optimization". In my experience, the always_inline attribute always inlines as long as the function is inlineable, regardless of optimization level. However, declaring a function as inline without the always_inline attribute will cause it to not be inlined when building unoptimized code. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Port from Win to OSX with Inline functions
by Steve Sisak 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com FYI ... found a short-term solution ... and some interesting things ... ProcessPixel->CopyBitRange->__rlwimi Since, by definition: 1) An inline function is short 2) Its definition must be available to anyone who calls it HTH, -Steve _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… At 10:03 AM -0400 10/9/06, Chris Edgington wrote: 1) No inlining at all happens unless optimization is enabled - regardless of the always_inline attribute. Yup, I filed a bug on that and it was closed as "works ad designed" -- seem the compiler folks, consider inlining an "optimization". In my case I had some C++ inline functions for pixel manipulation that called an inline utility function that evaluated to a compiler intrinsic. With optimizations off, this evaluated to 5 levels of function calls instead of a single instruction -- since this code is called 3 times per pixel, performance was degraded to the point it was no longer usable. I eventually had to resort to #define instead of inline for performance-critical code -- throwing away the type safety an inline function provides. Despite the compiler folks objections, there really does need to be an inline_dammmit attribute. 2) By default - an inlined function will also have a regular non-inlined copy in the object file. So - the short-term answer is to redefine __INLINE as "extern inline __attribute((always_inline))" and to build with optimizations enabled. Ultimately, however, I will likely need to change the prorotypes and function definitions to be static - so they work whether or not they are inlined. Actually, the only thing inline means is that it's OK if there are multiple definitions of a function and the linker shouldn't complain. I believe is that your problem is that your prototypes aren't inline -- you MUST fix at least that -- otherwise, the non-inlined prototype will take precidence over the inline definition. I always define inline functions in their header (with no separate declaration). If the inline keyword is in the declaration, there is no need to declare them static -- in fact this will cause code bloat if they are referenced from multiple files and not inlined. Simply using inline will cause the linker to keep just one implementation, but for this to happen, all declarations must be inline. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Port from Win to OSX with Inline functions
by Chris Edgington 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com FYI ... found a short-term solution ... and some interesting things ... -Chris On Oct 6, 2006, at 4:48 PM, Chris Edgington wrote: Anybody have any thoughts on how to resolve this? Thanks, -Chris _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… 1) No inlining at all happens unless optimization is enabled - regardless of the always_inline attribute. 2) By default - an inlined function will also have a regular non- inlined copy in the object file. So - the short-term answer is to redefine __INLINE as "extern inline __attribute((always_inline))" and to build with optimizations enabled. Ultimately, however, I will likely need to change the prorotypes and function definitions to be static - so they work whether or not they are inlined. I'm porting a large application from Windows to OSX. I'm trying to minimize changes to my clients codebase. One area that is giving me problems is inline functions. All over the place the client has functions in header files marked as inline (this is a performance- critical app). These header files get included by multiple .C files. When building with gcc on OSX - what ends up happening is that the compiler puts a copy of each inlined function in each .o file, then when I goto link I get linker errors because of multiple definitions of the same symbol. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Re: Port from Win to OSX with Inline functions
by David Leimbach 09 Oct '06

09 Oct '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=p2jftIKakTthd9YmRFfReotie27C2nbUG4/q7SRlqI20Yd7iaIXWjAR21WIsPW2t8StopQrTq45rsusRiCpE8siMgpmD+rU/6WhWL/DzujzKu8bava13+X+PPoYIlxn3ncsrt9nIJb+EAVQNLjnHwoZ89E7WfYJvS5JBELrhMq8= Am 07.10.2006 um 01:46 schrieb David Leimbach: > Inline is a C99 keyword and still only a "hint" to the compiler. > > To my knowledge, putting "inline" in a header file doesn't make it's > symbols static. And inline functions aren't allowed to also be static. > Is it a fact ? Actually I misread. It was that inline functions may not contain static objects per section 6.7.4.3. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… On 10/8/06, Nat! <nat(a)mulle-kybernetik.com> wrote: This email sent to site_archiver(a)lists.apple.com

1 0