Patchwork * doc/install.texi: Document availability of cloog-0.16

login
register
mail settings
Submitter Jack Howarth
Date Jan. 9, 2011, 12:27 a.m.
Message ID <20110109002733.GA1326@bromo.med.uc.edu>
Download mbox | patch
Permalink /patch/77995/
State New
Headers show

Comments

Jack Howarth - Jan. 9, 2011, 12:27 a.m.
On Sat, Jan 08, 2011 at 06:09:43PM -0500, Tobias Grosser wrote:
> On 01/08/2011 05:52 PM, Jack Howarth wrote:
>> On Sat, Jan 08, 2011 at 03:35:49PM -0500, Jack Howarth wrote:
>>> On Sat, Jan 08, 2011 at 10:44:54AM -0500, Tobias Grosser wrote:
>>>> On 01/08/2011 07:10 AM, Gerald Pfeifer wrote:
>>>>> On Thu, 6 Jan 2011, Tobias Grosser wrote:
>>>>>> thanks for the comments. I just added information to that install.texi
>>>>>> file, which made it really hard to read. I now tried to restructure
>>>>>> the whole comment and hope it became easier to read.
>>>>>
>>>>> Definitely, thanks a lot. Now I managed to parse this just fine. ;-)
>>>>>
>>>>> Mind the long lines and break them, otherwise this looks good.
>>>>
>>>> OK. Thanks.
>>>>
>>>> I will commit this as soon as the related CLooG packages are uploaded.
>>>>
>>>> Cheers
>>>> Tobi
>>>
>>> Tobi,
>>>     I am confused by the configure code that sets CFLAGS to -O1 when
>>> gcc 4.2.x is in use. Reverting this change on x86_64-apple-darwin10
>>> still allows cloog 0.16.1 to build on x86_64-apple-darwin10 without
>>> any regressions in the cloog testsuite. This is built against the
>>> gcc 4.2.1 compiler in Xcode 3.2.5. I also find that building against
>>> gcc-4.2 and g++-4.2 from Xcode 3.1.4 under powerpc-apple-darwin9
>>> produces no cloog testsuite failures. Perhaps the cloog/isl/configure
>>> test could be enhanced to not set -O1 if the gcc 4.2.x is "Apple Inc. build"
>>> since Apple uses a very forked branch of gcc 4.2.1...
>>>
>>> gcc -v
>>> Using built-in specs.
>>> Target: i686-apple-darwin10
>>> Configured with: /var/tmp/gcc/gcc-5664~89/src/configure --disable-checking --enable-werror --prefix=/usr --mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --program-transform-name=/^[cg][^.-]*$/s/$/-4.2/ --with-slibdir=/usr/lib --build=i686-apple-darwin10 --program-prefix=i686-apple-darwin10- --host=x86_64-apple-darwin10 --target=i686-apple-darwin10 --with-gxx-include-dir=/include/c++/4.2.1
>>> Thread model: posix
>>> gcc version 4.2.1 (Apple Inc. build 5664)
>>>
>>> I'll test current gcc trunk built with -enable-cloog-backend=isl against
>>> a -O3 isl build to check for regressions in graphite.
>>>             Jack
>>
>> Tobi,
>>    A build of current gcc trunk with -enable-cloog-backend=isl against cloog 0.16.1 built
>> entirely at -O3 shows no regressions in the graphite.exp tests on x86_64-apple-darwin10.
>> So it seems unnecessary to resort to -O1 in the isl build within cloog when Apple's
>> gcc 4.2.1 is in use.
>>               Jack
>
> Hi Jack,
>
> thanks for testing. Even if I am not sure that the graphite test suite  
> is large enough to find all subtile bugs, but in case of Apple's gcc  
> 4.2.1 it is probably save to assume it is not a gcc 4.2.1 any more. Do  
> you think you could provide a patch for Cloog that implements your  
> suggestions, such that Sven can include it in the next Cloog release.
>
> Thanks a lot
> Tobi

Tobi,
   The simplest fix would be...


so that the test is short-circuited on Apple's compiler builds. Applying this change
directly to isl/configure, I've confirmed that this properly applies -O1 for a GNU gcc
4.2 compiler and not for Apple's compilers.
             Jack
Tobias Grosser - Jan. 9, 2011, 12:34 a.m.
On 01/08/2011 07:27 PM, Jack Howarth wrote:
> On Sat, Jan 08, 2011 at 06:09:43PM -0500, Tobias Grosser wrote:
>> On 01/08/2011 05:52 PM, Jack Howarth wrote:
>>> On Sat, Jan 08, 2011 at 03:35:49PM -0500, Jack Howarth wrote:
>>>> On Sat, Jan 08, 2011 at 10:44:54AM -0500, Tobias Grosser wrote:
>>>>> On 01/08/2011 07:10 AM, Gerald Pfeifer wrote:
>>>>>> On Thu, 6 Jan 2011, Tobias Grosser wrote:
>>>>>>> thanks for the comments. I just added information to that install.texi
>>>>>>> file, which made it really hard to read. I now tried to restructure
>>>>>>> the whole comment and hope it became easier to read.
>>>>>>
>>>>>> Definitely, thanks a lot. Now I managed to parse this just fine. ;-)
>>>>>>
>>>>>> Mind the long lines and break them, otherwise this looks good.
>>>>>
>>>>> OK. Thanks.
>>>>>
>>>>> I will commit this as soon as the related CLooG packages are uploaded.
>>>>>
>>>>> Cheers
>>>>> Tobi
>>>>
>>>> Tobi,
>>>>      I am confused by the configure code that sets CFLAGS to -O1 when
>>>> gcc 4.2.x is in use. Reverting this change on x86_64-apple-darwin10
>>>> still allows cloog 0.16.1 to build on x86_64-apple-darwin10 without
>>>> any regressions in the cloog testsuite. This is built against the
>>>> gcc 4.2.1 compiler in Xcode 3.2.5. I also find that building against
>>>> gcc-4.2 and g++-4.2 from Xcode 3.1.4 under powerpc-apple-darwin9
>>>> produces no cloog testsuite failures. Perhaps the cloog/isl/configure
>>>> test could be enhanced to not set -O1 if the gcc 4.2.x is "Apple Inc. build"
>>>> since Apple uses a very forked branch of gcc 4.2.1...
>>>>
>>>> gcc -v
>>>> Using built-in specs.
>>>> Target: i686-apple-darwin10
>>>> Configured with: /var/tmp/gcc/gcc-5664~89/src/configure --disable-checking --enable-werror --prefix=/usr --mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --program-transform-name=/^[cg][^.-]*$/s/$/-4.2/ --with-slibdir=/usr/lib --build=i686-apple-darwin10 --program-prefix=i686-apple-darwin10- --host=x86_64-apple-darwin10 --target=i686-apple-darwin10 --with-gxx-include-dir=/include/c++/4.2.1
>>>> Thread model: posix
>>>> gcc version 4.2.1 (Apple Inc. build 5664)
>>>>
>>>> I'll test current gcc trunk built with -enable-cloog-backend=isl against
>>>> a -O3 isl build to check for regressions in graphite.
>>>>              Jack
>>>
>>> Tobi,
>>>     A build of current gcc trunk with -enable-cloog-backend=isl against cloog 0.16.1 built
>>> entirely at -O3 shows no regressions in the graphite.exp tests on x86_64-apple-darwin10.
>>> So it seems unnecessary to resort to -O1 in the isl build within cloog when Apple's
>>> gcc 4.2.1 is in use.
>>>                Jack
>>
>> Hi Jack,
>>
>> thanks for testing. Even if I am not sure that the graphite test suite
>> is large enough to find all subtile bugs, but in case of Apple's gcc
>> 4.2.1 it is probably save to assume it is not a gcc 4.2.1 any more. Do
>> you think you could provide a patch for Cloog that implements your
>> suggestions, such that Sven can include it in the next Cloog release.
>>
>> Thanks a lot
>> Tobi
>
> Tobi,
>     The simplest fix would be...
>
> --- isl/m4/ax_cc_maxopt.m4.org	2011-01-08 18:12:56.000000000 -0500
> +++ isl/m4/ax_cc_maxopt.m4	2011-01-08 19:20:53.000000000 -0500
> @@ -151,7 +151,7 @@
>        AX_GCC_ARCHFLAG($acx_maxopt_portable)
>
>        # drop to -O1 for gcc 4.2
> -     $CC --version |
> +     $CC --version | grep -v 'Apple Inc' |
>   	sed -e 's/.* \(@<:@0-9@:>@@<:@0-9@:>@*\)\.\(@<:@0-9@:>@@<:@0-9@:>@*\).*/\1 \2/' |
>   	(read major minor
>   	    if test $major -eq 4 -a $minor -eq 2; then
>
> so that the test is short-circuited on Apple's compiler builds. Applying this change
> directly to isl/configure, I've confirmed that this properly applies -O1 for a GNU gcc
> 4.2 compiler and not for Apple's compilers.
>               Jack

Thanks for this quick patch. Let's see what Sven says.

Tobi
Jack Howarth - Jan. 9, 2011, 1:22 a.m.
On Sat, Jan 08, 2011 at 07:34:08PM -0500, Tobias Grosser wrote:
...
>
> Thanks for this quick patch. Let's see what Sven says.
>
> Tobi

Tobi,
   I also tested a -O3 cloog-isl build on i386-apple-darwin10 using the
Xcode 3.2.5 compilers and there are no regressions in the cloog testsuite
at -m32 as well.
   On a different issue, do you think that the warnings...

libtool: compile:  gcc -DPACKAGE_NAME=\"\" -DPACKAGE_TARNAME=\"\" -DPACKAGE_VERSION=\"\" -DPACKAGE_STRING=\"\" -DPACKAGE_BUGREPORT=\"\" -DPACKAGE_URL=\"\" -DPACKAGE=\"cloog\" -DVERSION=\"0.16.1\" -DSTDC_HEADERS=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_SYS_STAT_H=1 -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_MEMORY_H=1 -DHAVE_STRINGS_H=1 -DHAVE_INTTYPES_H=1 -DHAVE_STDINT_H=1 -DHAVE_UNISTD_H=1 -DHAVE_DLFCN_H=1 -DLT_OBJDIR=\".libs/\" -DSTDC_HEADERS=1 -DHAVE_STRTOL=1 -DHAVE_SYS_RESOURCE_H=1 "-DCLOOG_RUSAGE=/**/" -DHAVE_DECL_MP_GET_MEMORY_FUNCTIONS=1 -DCLOOG_INT_GMP=1 -I. -I./include -I./include -I./isl/include -Iisl/include -I/sw_i386/include -I/sw_i386/include -Wall -O3 -fomit-frame-pointer -malign-double -fstrict-aliasing -ffast-math -MT libcloog_isl_la-loop.lo -MD -MP -MF .deps/libcloog_isl_la-loop.Tpo -c source/loop.c  -fno-common -DPIC -o .libs/libcloog_isl_la-loop.o
source/loop.c: In function 'cloog_loop_sort':
source/loop.c:1059: warning: 'now' may be used uninitialized in this function
source/loop.c: In function 'cloog_loop_nest':
source/loop.c:1120: warning: 'now' may be used uninitialized in this function
source/loop.c: In function 'cloog_loop_generate_backtrack':
source/loop.c:1470: warning: 'now2' may be used uninitialized in this function
source/loop.c:1470: warning: 'now' may be used uninitialized in this function
source/loop.c: In function 'cloog_loop_generate_general':
source/loop.c:1627: warning: 'now' may be used uninitialized in this function
source/loop.c:1627: warning: 'now2' may be used uninitialized in this function
source/loop.c: In function 'cloog_loop_generate_scalar':
source/loop.c:1755: warning: 'now' may be used uninitialized in this function

are harmless or should now and now2 be initialized?
             Jack

Patch

--- isl/m4/ax_cc_maxopt.m4.org	2011-01-08 18:12:56.000000000 -0500
+++ isl/m4/ax_cc_maxopt.m4	2011-01-08 19:20:53.000000000 -0500
@@ -151,7 +151,7 @@ 
      AX_GCC_ARCHFLAG($acx_maxopt_portable)
 
      # drop to -O1 for gcc 4.2
-     $CC --version |
+     $CC --version | grep -v 'Apple Inc' |
 	sed -e 's/.* \(@<:@0-9@:>@@<:@0-9@:>@*\)\.\(@<:@0-9@:>@@<:@0-9@:>@*\).*/\1 \2/' |
 	(read major minor
 	    if test $major -eq 4 -a $minor -eq 2; then