hg: ppc-aix-port/jdk7u/jdk: Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds
Changeset: 5c3a27b86fd1 Author: spoole <spoole@linux.vnet.ibm.com> Date: 2012-07-12 13:50 +0100 URL: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/5c3a27b86fd1 Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds ! src/solaris/bin/java_md_solinux.c
Hi Steve, this change breaks the Linux build, because USE_PTHREADS is defined in /usr/work/openjdk/nb/fdbg/linuxx86_64/jdk7u/jdk/make/common/Defs-linux.gmk, but it is not really exported in the CFLAGS/CPPFLAGS as this is done for AIX. So there are two possibilities here: we could either reintroduce the "superflous" __linux__ or we could export USE_PTHREADS in the CFLAGS/CPPFLAGS in the same way as this is done for AIX. Although we wanted to do as few changes as possible in shared code, I'd vote for the second option, because this is really a feature we want to use (PTHREADS) and it should be defined once in the makefile for the corresponding platform. Going that way would simplify further ports like for example HPUX. What do you think? Do you want me to do the change or do you have a Linux build yourself? Regards, Volker On Thu, Jul 12, 2012 at 2:50 PM, <spoole@linux.vnet.ibm.com> wrote:
Changeset: 5c3a27b86fd1 Author: spoole <spoole@linux.vnet.ibm.com> Date: 2012-07-12 13:50 +0100 URL: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/5c3a27b86fd1
Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds
! src/solaris/bin/java_md_solinux.c
Hi Volker, I also prefer the second option. I think in the Linux case, one would need only to export USE_PTHREADS in CPPFLAGS because, confusingly, it looks like CPPFLAGS is used for different things on Linux compared to AIX. On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus'). On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options). So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON. I'm still grappling with Steve's instructions on building this stuff (whilst he suns himself in the glorious British summer weather [1] for a bit), so if you're able to quickly verify (or refute) what I'm saying above, feel free to push a suitable change up to the project. Regards, Neil [1] http://goo.gl/qQIJq On Mon, 2012-07-16 at 10:15 +0200, Volker Simonis wrote:
Hi Steve,
this change breaks the Linux build, because USE_PTHREADS is defined in /usr/work/openjdk/nb/fdbg/linuxx86_64/jdk7u/jdk/make/common/Defs-linux.gmk, but it is not really exported in the CFLAGS/CPPFLAGS as this is done for AIX.
So there are two possibilities here: we could either reintroduce the "superflous" __linux__ or we could export USE_PTHREADS in the CFLAGS/CPPFLAGS in the same way as this is done for AIX. Although we wanted to do as few changes as possible in shared code, I'd vote for the second option, because this is really a feature we want to use (PTHREADS) and it should be defined once in the makefile for the corresponding platform. Going that way would simplify further ports like for example HPUX.
What do you think? Do you want me to do the change or do you have a Linux build yourself?
Regards, Volker
On Thu, Jul 12, 2012 at 2:50 PM, <spoole@linux.vnet.ibm.com> wrote:
Changeset: 5c3a27b86fd1 Author: spoole <spoole@linux.vnet.ibm.com> Date: 2012-07-12 13:50 +0100 URL: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/5c3a27b86fd1
Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds
! src/solaris/bin/java_md_solinux.c
-- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
Hi Neil, yes you're right - just had to export USE_PTHREADS through CPPFLAGS on Linux. I've just pushed the fix: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/35172a51cc76 Regards, Volker PS: weather is similar in Germany:) On Mon, Jul 16, 2012 at 4:28 PM, Neil Richards <neil.richards@ngmr.net> wrote:
Hi Volker, I also prefer the second option.
I think in the Linux case, one would need only to export USE_PTHREADS in CPPFLAGS because, confusingly, it looks like CPPFLAGS is used for different things on Linux compared to AIX.
On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus').
On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options).
So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON.
I'm still grappling with Steve's instructions on building this stuff (whilst he suns himself in the glorious British summer weather [1] for a bit), so if you're able to quickly verify (or refute) what I'm saying above, feel free to push a suitable change up to the project.
Regards, Neil
On Mon, 2012-07-16 at 10:15 +0200, Volker Simonis wrote:
Hi Steve,
this change breaks the Linux build, because USE_PTHREADS is defined in /usr/work/openjdk/nb/fdbg/linuxx86_64/jdk7u/jdk/make/common/Defs-linux.gmk, but it is not really exported in the CFLAGS/CPPFLAGS as this is done for AIX.
So there are two possibilities here: we could either reintroduce the "superflous" __linux__ or we could export USE_PTHREADS in the CFLAGS/CPPFLAGS in the same way as this is done for AIX. Although we wanted to do as few changes as possible in shared code, I'd vote for the second option, because this is really a feature we want to use (PTHREADS) and it should be defined once in the makefile for the corresponding platform. Going that way would simplify further ports like for example HPUX.
What do you think? Do you want me to do the change or do you have a Linux build yourself?
Regards, Volker
On Thu, Jul 12, 2012 at 2:50 PM, <spoole@linux.vnet.ibm.com> wrote:
Changeset: 5c3a27b86fd1 Author: spoole <spoole@linux.vnet.ibm.com> Date: 2012-07-12 13:50 +0100 URL: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/5c3a27b86fd1
Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds
! src/solaris/bin/java_md_solinux.c
-- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
Hi Neil,
On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus').
On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options).
So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON.
Are you sure this is right? I could not find any documentation for that, e.g. on the xlC manuals. Reason I ask is I always thought both CXXFLAGS and CPPFLAGS originally came from the GNU automake toolchain, and that what you describe as "the Linux way" is just the correct way to use those variables - but that many people use CPPFLAGS as "C++-Flags" because that is the first thing which comes to mind. I may be wrong. I think it's mostly all convention anyway, the only tools I think do something with those variables are autoconf/automake. When working on the AIX port of the SAP JVM, we tried to keep the GNU meaning of CPPFLAGS intact: CPPFLAGS for -D../-I.., CXXFLAGS/CFLAGS for languange specific settings. However, this has the tendency to detoriate, so you probably will find both meanings in the makefiles. Kind Regards, Thomas Stuefe SAP Germany
I'm still grappling with Steve's instructions on building this stuff (whilst he suns himself in the glorious British summer weather [1] for a bit), so if you're able to quickly verify (or refute) what I'm saying above, feel free to push a suitable change up to the project.
Regards, Neil
On Mon, 2012-07-16 at 10:15 +0200, Volker Simonis wrote:
Hi Steve,
this change breaks the Linux build, because USE_PTHREADS is defined in /usr/work/openjdk/nb/fdbg/linuxx86_64/jdk7u/jdk/make/common/Defs-linux.gmk, but it is not really exported in the CFLAGS/CPPFLAGS as this is done for AIX.
So there are two possibilities here: we could either reintroduce the "superflous" __linux__ or we could export USE_PTHREADS in the CFLAGS/CPPFLAGS in the same way as this is done for AIX. Although we wanted to do as few changes as possible in shared code, I'd vote for the second option, because this is really a feature we want to use (PTHREADS) and it should be defined once in the makefile for the corresponding platform. Going that way would simplify further ports like for example HPUX.
What do you think? Do you want me to do the change or do you have a Linux build yourself?
Regards, Volker
On Thu, Jul 12, 2012 at 2:50 PM, <spoole@linux.vnet.ibm.com> wrote:
Changeset: 5c3a27b86fd1 Author: spoole <spoole@linux.vnet.ibm.com> Date: 2012-07-12 13:50 +0100 URL: http://hg.openjdk.java.net/ppc-aix-port/jdk7u/jdk/rev/5c3a27b86fd1
Updated using pthreads in java_md_solinux.c to just rely on USE_PTHREADS. Removed the superflous __linux__ as USE_PTHREADS is already explictly turned on for linux builds
! src/solaris/bin/java_md_solinux.c
-- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
On Tue, 2012-07-17 at 08:21 +0200, Thomas Stüfe wrote:
Hi Neil,
On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus').
On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options).
So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON.
Are you sure this is right? I could not find any documentation for that, e.g. on the xlC manuals.
Reason I ask is I always thought both CXXFLAGS and CPPFLAGS originally came from the GNU automake toolchain, and that what you describe as "the Linux way" is just the correct way to use those variables - but that many people use CPPFLAGS as "C++-Flags" because that is the first thing which comes to mind.
I may be wrong. I think it's mostly all convention anyway, the only tools I think do something with those variables are autoconf/automake.
When working on the AIX port of the SAP JVM, we tried to keep the GNU meaning of CPPFLAGS intact: CPPFLAGS for -D../-I.., CXXFLAGS/CFLAGS for languange specific settings. However, this has the tendency to detoriate, so you probably will find both meanings in the makefiles.
Kind Regards, Thomas Stuefe SAP Germany
Hi Thomas, To be honest, I'm more certain on the Linux side of this story than I am on the AIX side. :-) When writing the reply, I did find some references via Google that seemed to confirm the AIX story I was spinning. Naturally, as I didn't take a note of them at that point, trying to track those references down again has proven elusive, thus casting doubt on their validity. Obviously, the use of these variables is controlled by the make environment, rather than the compiler, as it is the (potentially implicit) rules in the make system that use the variables (in whatever combination) to give command line options to the compiler. So, if/as it's GNU make being used on both systems, I suppose it makes sense its behaviour re. these variables should also be (made to be) similar. If this is the case, I'm hoping I'll see duplication of (-DUSE_PTHREADS) options when compiling C++ code (with Steve's change) on AIX. If I see that, I'll have confirmation as to what correction to Steve's change is necessary to bring the use of these variables into line (with the GNU make / autoconf / automake) norm. If that option isn't duplicated in the invocation of the C++ compiler on AIX, it may suggest there are other factors in play (eg. the implicit rule for C++ compilation is being overridden somewhere). In that case, I'll have to dig deeper to figure out what's going on. Regards, Neil -- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
I think the build (at least the HotSpot build) doesn't use implicit rules at all. I have no reference for this, but we recently found out that we can considerably speed up the Windows build if we disable implicit make rules, so this implicates that they are not really used. Regards, Volker On Wed, Jul 18, 2012 at 1:58 PM, Neil Richards <neil.richards@ngmr.net> wrote:
On Tue, 2012-07-17 at 08:21 +0200, Thomas Stüfe wrote:
Hi Neil,
On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus').
On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options).
So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON.
Are you sure this is right? I could not find any documentation for that, e.g. on the xlC manuals.
Reason I ask is I always thought both CXXFLAGS and CPPFLAGS originally came from the GNU automake toolchain, and that what you describe as "the Linux way" is just the correct way to use those variables - but that many people use CPPFLAGS as "C++-Flags" because that is the first thing which comes to mind.
I may be wrong. I think it's mostly all convention anyway, the only tools I think do something with those variables are autoconf/automake.
When working on the AIX port of the SAP JVM, we tried to keep the GNU meaning of CPPFLAGS intact: CPPFLAGS for -D../-I.., CXXFLAGS/CFLAGS for languange specific settings. However, this has the tendency to detoriate, so you probably will find both meanings in the makefiles.
Kind Regards, Thomas Stuefe SAP Germany
Hi Thomas, To be honest, I'm more certain on the Linux side of this story than I am on the AIX side. :-)
When writing the reply, I did find some references via Google that seemed to confirm the AIX story I was spinning. Naturally, as I didn't take a note of them at that point, trying to track those references down again has proven elusive, thus casting doubt on their validity.
Obviously, the use of these variables is controlled by the make environment, rather than the compiler, as it is the (potentially implicit) rules in the make system that use the variables (in whatever combination) to give command line options to the compiler.
So, if/as it's GNU make being used on both systems, I suppose it makes sense its behaviour re. these variables should also be (made to be) similar.
If this is the case, I'm hoping I'll see duplication of (-DUSE_PTHREADS) options when compiling C++ code (with Steve's change) on AIX. If I see that, I'll have confirmation as to what correction to Steve's change is necessary to bring the use of these variables into line (with the GNU make / autoconf / automake) norm.
If that option isn't duplicated in the invocation of the C++ compiler on AIX, it may suggest there are other factors in play (eg. the implicit rule for C++ compilation is being overridden somewhere). In that case, I'll have to dig deeper to figure out what's going on.
Regards, Neil
-- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
Hi, Neil, Volker, I also looked and did not find any proof that CPPFLAGS would *not* be synonymous for C++ on AIX. So, Neil may be right with his AIX claim. The "hardest" definition I could find (beside wikipedia) is the GNU make manual: http://www.gnu.org/software/make/manual/make.html#Implicit-Variables I think in the end it is just our decision - as Volker said, the make does not seem to use implicit rules but always spells them out. It would be my preference to stick to the GNU standard, but I am pretty sure it is already used both ways in the OpenJDK make files. Regards, Thomas On Wed, Jul 18, 2012 at 4:19 PM, Volker Simonis <volker.simonis@gmail.com> wrote:
I think the build (at least the HotSpot build) doesn't use implicit rules at all. I have no reference for this, but we recently found out that we can considerably speed up the Windows build if we disable implicit make rules, so this implicates that they are not really used.
Regards, Volker
On Wed, Jul 18, 2012 at 1:58 PM, Neil Richards <neil.richards@ngmr.net> wrote:
On Tue, 2012-07-17 at 08:21 +0200, Thomas Stüfe wrote:
Hi Neil,
On AIX, CPPFLAGS is used for invocations of the C++ compiler (the 'PP' standing for 'Plus-Plus').
On Linux, CPPFLAGS is for the C Pre-Processor, and so is used for invocations of both the C and C++ compilers. (Linux has CXXFLAGS for C++-specific options).
So, on Linux, adding to CPPFLAGS_COMMON should cater for both scenarios, I believe, whilst on AIX, one needs to add to both CPPFLAGS_COMMON and CFLAGS_COMMON.
Are you sure this is right? I could not find any documentation for that, e.g. on the xlC manuals.
Reason I ask is I always thought both CXXFLAGS and CPPFLAGS originally came from the GNU automake toolchain, and that what you describe as "the Linux way" is just the correct way to use those variables - but that many people use CPPFLAGS as "C++-Flags" because that is the first thing which comes to mind.
I may be wrong. I think it's mostly all convention anyway, the only tools I think do something with those variables are autoconf/automake.
When working on the AIX port of the SAP JVM, we tried to keep the GNU meaning of CPPFLAGS intact: CPPFLAGS for -D../-I.., CXXFLAGS/CFLAGS for languange specific settings. However, this has the tendency to detoriate, so you probably will find both meanings in the makefiles.
Kind Regards, Thomas Stuefe SAP Germany
Hi Thomas, To be honest, I'm more certain on the Linux side of this story than I am on the AIX side. :-)
When writing the reply, I did find some references via Google that seemed to confirm the AIX story I was spinning. Naturally, as I didn't take a note of them at that point, trying to track those references down again has proven elusive, thus casting doubt on their validity.
Obviously, the use of these variables is controlled by the make environment, rather than the compiler, as it is the (potentially implicit) rules in the make system that use the variables (in whatever combination) to give command line options to the compiler.
So, if/as it's GNU make being used on both systems, I suppose it makes sense its behaviour re. these variables should also be (made to be) similar.
If this is the case, I'm hoping I'll see duplication of (-DUSE_PTHREADS) options when compiling C++ code (with Steve's change) on AIX. If I see that, I'll have confirmation as to what correction to Steve's change is necessary to bring the use of these variables into line (with the GNU make / autoconf / automake) norm.
If that option isn't duplicated in the invocation of the C++ compiler on AIX, it may suggest there are other factors in play (eg. the implicit rule for C++ compilation is being overridden somewhere). In that case, I'll have to dig deeper to figure out what's going on.
Regards, Neil
-- Unless stated above: IBM email: neil_richards at uk.ibm.com IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
participants (4)
-
Neil Richards
-
spoole＠linux.vnet.ibm.com
-
Thomas Stüfe
-
Volker Simonis