April 2013 - petsc-dev - mailman3.cels.anl.gov

usage of string.h
by Satish Balay 25 Apr '13

25 Apr '13

I see the usage of string.h is removed from petsc sources. https://bitbucket.org/petsc/petsc/commits/b083e2b6e0be6f731fe9e787c55cf4647… https://bitbucket.org/petsc/petsc/commits/573b0fb4b27458e353e9b121e629dc5c6… The remove of strings.h from str.c is added back in https://bitbucket.org/petsc/petsc/commits/3964eb8815a373d98cd54ebd45324cfa2… Now I see a breakage in windows64 build with strstr() [usded in str.c] - fixable by adding back string.h Is this the correct fix? Whats the intention of remove these includes? thanks, Satish

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Chetan Jhurani 25 Apr '13

25 Apr '13

> -----Original Message----- > From: Satish Balay [mailto:[email protected]] > Sent: Thursday, April 25, 2013 12:35 PM > To: Chetan Jhurani > Cc: 'For users of the development version of PETSc' > Subject: Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h > > BTW: Which version of compiler gives error? What exact error? VS 2012. The first file with error is this. src\sys\objects\mpinit.c(261) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier src\sys\objects\mpinit.c(279) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier src\sys\objects\mpinit.c(287) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier src\sys\objects\mpinit.c(329) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier src\sys\objects\mpinit.c(408) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier src\sys\objects\mpinit.c(449) : error C2065: 'MPI_UNSIGNED_LONG_LONG' : undeclared identifier Essentially, all files that use MPIU_SIZE_T see it as MPI_UNSIGNED_LONG_LONG and give this error. Chetan > Satish > > On Thu, 25 Apr 2013, Chetan Jhurani wrote: > > > Hi, > > > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > > #define MPIU_SIZE_T MPI_UNSIGNED > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG > > #else > > #error "Unknown size for size_t! Send us a bugreport at petsc-maint(a)mcs.anl.gov" > > #endif > > > > In case this is important, I'm on windows, no mpi and can > > send more details if needed. I'm using a slightly older > > version of petsc, but I see that the next branch does not > > contain the required definition in mpiuni.h > > > > > https://bitbucket.org/petsc/petsc/src/d93c087c991ba51c9078189a5f0b6f06a5177… > =next > > > > Chetan > > > >

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Jed Brown 25 Apr '13

25 Apr '13

Satish Balay <balay(a)mcs.anl.gov> writes: > It won't get used in PETSc unless PETSC_SIZEOF_LONG_LONG is defined > [which is only for 64bit compilers that support 'long long' anyway. Yes, that sounds fine to me.

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Satish Balay 25 Apr '13

25 Apr '13

On Thu, 25 Apr 2013, Jed Brown wrote: > Chetan Jhurani <chetan.jhurani(a)gmail.com> writes: > > > Hi, > > > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > > #define MPIU_SIZE_T MPI_UNSIGNED > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG > > mpiuni/mpi.h also has this unguarded statement. Do all Windows > compilers now provide 'long long'? > > #define MPI_LONG_LONG_INT sizeof(long long) Hm - perhaps its ok to have this unguarded and add: diff --git a/include/mpiuni/mpi.h b/include/mpiuni/mpi.h index 5adfe26..a6a4f97 100644 --- a/include/mpiuni/mpi.h +++ b/include/mpiuni/mpi.h @@ -168,6 +168,7 @@ typedef int MPI_Offset; #define MPI_UNSIGNED sizeof(unsigned) #define MPI_UNSIGNED_CHAR sizeof(unsigned char) #define MPI_UNSIGNED_LONG sizeof(unsigned long) +#define MPI_UNSIGNED_LONG_LONG sizeof(unsigned long long) #define MPI_COMPLEX 2*sizeof(float) #define MPI_C_COMPLEX 2*sizeof(float) #define MPI_C_DOUBLE_COMPLEX 2*sizeof(double) It won't get used in PETSc unless PETSC_SIZEOF_LONG_LONG is defined [which is only for 64bit compilers that support 'long long' anyway. satish > > We could make mpiuni define MPIU_SIZE_T itself instead of jumping > through hoops to find the matching type. > > Chetan, what's your perspective about using stdint.h? >

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Satish Balay 25 Apr '13

25 Apr '13

BTW: Which version of compiler gives error? What exact error? Satish On Thu, 25 Apr 2013, Chetan Jhurani wrote: > Hi, > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > #define MPIU_SIZE_T MPI_UNSIGNED > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG > #else > #error "Unknown size for size_t! Send us a bugreport at petsc-maint(a)mcs.anl.gov" > #endif > > In case this is important, I'm on windows, no mpi and can > send more details if needed. I'm using a slightly older > version of petsc, but I see that the next branch does not > contain the required definition in mpiuni.h > > https://bitbucket.org/petsc/petsc/src/d93c087c991ba51c9078189a5f0b6f06a5177… > > Chetan > >

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Satish Balay 25 Apr '13

25 Apr '13

On Thu, 25 Apr 2013, Chetan Jhurani wrote: > > -----Original Message----- > > From: Jed Brown [mailto:[email protected]] On Behalf Of Jed Brown > > Sent: Thursday, April 25, 2013 11:09 AM > > To: Chetan Jhurani; 'For users of the development version of PETSc' > > Subject: Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h > > > > Chetan Jhurani <chetan.jhurani(a)gmail.com> writes: > > > > > Hi, > > > > > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > > > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > > > > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > > > #define MPIU_SIZE_T MPI_UNSIGNED > > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG > > > > mpiuni/mpi.h also has this unguarded statement. Do all Windows > > compilers now provide 'long long'? > > > > #define MPI_LONG_LONG_INT sizeof(long long) > > > > We could make mpiuni define MPIU_SIZE_T itself instead of jumping > > through hoops to find the matching type. > > > > Chetan, what's your perspective about using stdint.h? > > Info about long long: > > Visual Studio 2005 supports long long (sizeof = 8) > even when compiling in 32 bit mode, so I assume all > later ones do. For 64bit ints petscsys.h uses __int64 which is supported by VC6 aswell. So perhaps this should be used. Satish > > Trilinos 11.0+ uses long long for 64-bit indices. I have tested > it with two compilers (2005/2012) and I'm guessing others > have used other versions. > > About stdint.h: > > I see that it is there with Visual Studio 2012 on > my machine, but not with 2005. Below is more info > on stdint and MS VS. I never had a need to use > it so can't say more than this. > > http://stackoverflow.com/questions/126279/c99-stdint-h-header-and-ms-visual… > > Chetan > > >

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Jed Brown 25 Apr '13

25 Apr '13

Chetan Jhurani <chetan.jhurani(a)gmail.com> writes: > Info about long long: > > Visual Studio 2005 supports long long (sizeof = 8) > even when compiling in 32 bit mode, so I assume all > later ones do. > > Trilinos 11.0+ uses long long for 64-bit indices. I have tested > it with two compilers (2005/2012) and I'm guessing others > have used other versions. We use 'long long' or '__int64', but I don't have any data about whether the '__int64' option is still needed. It sounds like maybe not. Trilinos is C++ where as PETSc needs only a C compiler, and what is "portable enough" for them does not necessarily carry over. > About stdint.h: > > I see that it is there with Visual Studio 2012 on > my machine, but not with 2005. Below is more info > on stdint and MS VS. I never had a need to use > it so can't say more than this. > > http://stackoverflow.com/questions/126279/c99-stdint-h-header-and-ms-visual… Yeah, we've had this discussion before.

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Chetan Jhurani 25 Apr '13

25 Apr '13

> -----Original Message----- > From: Jed Brown [mailto:[email protected]] On Behalf Of Jed Brown > Sent: Thursday, April 25, 2013 11:09 AM > To: Chetan Jhurani; 'For users of the development version of PETSc' > Subject: Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h > > Chetan Jhurani <chetan.jhurani(a)gmail.com> writes: > > > Hi, > > > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > > #define MPIU_SIZE_T MPI_UNSIGNED > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG > > mpiuni/mpi.h also has this unguarded statement. Do all Windows > compilers now provide 'long long'? > > #define MPI_LONG_LONG_INT sizeof(long long) > > We could make mpiuni define MPIU_SIZE_T itself instead of jumping > through hoops to find the matching type. > > Chetan, what's your perspective about using stdint.h? Info about long long: Visual Studio 2005 supports long long (sizeof = 8) even when compiling in 32 bit mode, so I assume all later ones do. Trilinos 11.0+ uses long long for 64-bit indices. I have tested it with two compilers (2005/2012) and I'm guessing others have used other versions. About stdint.h: I see that it is there with Visual Studio 2012 on my machine, but not with 2005. Below is more info on stdint and MS VS. I never had a need to use it so can't say more than this. http://stackoverflow.com/questions/126279/c99-stdint-h-header-and-ms-visual… Chetan

1 0

Re: [petsc-dev] MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Jed Brown 25 Apr '13

25 Apr '13

Chetan Jhurani <chetan.jhurani(a)gmail.com> writes: > Hi, > > mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads > to a problem in defining MPIU_SIZE_T here in petscsys.h. > > #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) > #define MPIU_SIZE_T MPI_UNSIGNED > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) > #define MPIU_SIZE_T MPI_UNSIGNED_LONG > #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) > #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG mpiuni/mpi.h also has this unguarded statement. Do all Windows compilers now provide 'long long'? #define MPI_LONG_LONG_INT sizeof(long long) We could make mpiuni define MPIU_SIZE_T itself instead of jumping through hoops to find the matching type. Chetan, what's your perspective about using stdint.h?

1 0

MPI_UNSIGNED_LONG_LONG in mpiuni.h
by Chetan Jhurani 25 Apr '13

25 Apr '13

Hi, mpiuni.h does not define MPI_UNSIGNED_LONG_LONG, which leads to a problem in defining MPIU_SIZE_T here in petscsys.h. #if (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_INT) #define MPIU_SIZE_T MPI_UNSIGNED #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG) #define MPIU_SIZE_T MPI_UNSIGNED_LONG #elif (PETSC_SIZEOF_SIZE_T) == (PETSC_SIZEOF_LONG_LONG) #define MPIU_SIZE_T MPI_UNSIGNED_LONG_LONG #else #error "Unknown size for size_t! Send us a bugreport at petsc-maint(a)mcs.anl.gov" #endif In case this is important, I'm on windows, no mpi and can send more details if needed. I'm using a slightly older version of petsc, but I see that the next branch does not contain the required definition in mpiuni.h https://bitbucket.org/petsc/petsc/src/d93c087c991ba51c9078189a5f0b6f06a5177… Chetan

1 0