- mpiwg-large-counts - mailman3.cels.anl.gov

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Rolf Rabenseifner 24 Oct '19

24 Oct '19

I was reading web pages about segmented address space. And they show two possibilities (which may go only into the past): - flattened address space - segmented with some high order bits refering to the Segment and low order bits to address within the segment. If I understood correctly, then the major problem behind MPI_Aint_diff and MPI_Aint_add was the following: (my example is with 8 bit MPI_Aint) addr1 01111111 = 127 (signed int) = 127 (unsigned int) addr2 10000001 = -127 (signed int) = 129 (unsigned int) Internally the addreses are viewed by the hardware and OS as unsigned. MPI_Aint is interpreting the same bits as signed int. addr2-addr1 = 129 -127 = 2 (as unsigned int) but in a real application code with "-" operator: = -127 -127 = -254 --> signed int Overflow because 8 bit can express only -128 .. +127 --> detected or automatically corrected with +256 --> -254+256 = 2 The segmented problem was solved very early through the wording on "sequential storage" (unchanged since MPI-1.1 !!! :-) in the section "Correct use of addresses". What I see, that before changing anything, the mpiwg-large-counts WG should really understand MPI-3.1. For me, it looked like that there was some misunderstanding of the concept that absolute and relative addresses and number of bytes that can be stored in MPI_Aint. Best regards Rolf ----- Original Message ----- > From: "Jeff Squyres" <jsquyres(a)cisco.com> > To: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de> > Cc: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org> > Sent: Thursday, October 24, 2019 3:28:47 PM > Subject: Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements > On Oct 24, 2019, at 2:19 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: >> >>> Does MPI support segmented address spaces or not? >> >> When I understand correctly then MPI is and was always prepared for segmented >> address spaces as long as all integer types are signed integers and the >> segmentation is done through some higher order bits. > > > As I was pondering your replies from yesterday and thinking about how to > respond, this is exactly the same point that I got to. > > I think what I have been failing to elucidate is that this is not how modern > segmented address spaces work. It's not sufficient to have a few bits to > specify which segment to use. Instead, you typically need three things: > > 1. The type of memory segment being described (which may be just a few bits). > In reality, this "type" may correlate to which API you need to use to access > this memory (e.g., main RAM, RDMA, accelerator memory, persistent memory, other > IO memory, ...etc.). > 2. A context of some kind (usually more than a small number of bits -- perhaps > 32 or 64 bits). > 3. A pointer (usually 32 or 64 bits). > > For example, if I'm accessing IO memory of some flavor, I probably need some > kind of context/handle that points to a descriptor with information about my > data segment on the IO peripheral in question (e.g., a NIC), and then a pointer > for the specific place that I want in that specific memory. I use this > context/handle + pointer with some underlying API to get/set that data. > > I think that's the disconnect here: the prior thinking about MPI_Aint may have > assumed "it's either main RAM or it's XYZ memory, and the rest of the bits are > enough to uniquely describe the address." But that doesn't seem sufficient any > more to describe the evolution of complex, segmented memory systems (and be > able to use their supporting APIs). > > Meaning: perhaps a pure integer MPI_Aint may have been sufficient in the past. > But I'm not sure that's useful any more. > > -- > Jeff Squyres > jsquyres(a)cisco.com -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Jeff Squyres (jsquyres) 24 Oct '19

24 Oct '19

On Oct 24, 2019, at 2:19 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: > >> Does MPI support segmented address spaces or not? > > When I understand correctly then MPI is and was always prepared for segmented address spaces as long as all integer types are signed integers and the segmentation is done through some higher order bits. As I was pondering your replies from yesterday and thinking about how to respond, this is exactly the same point that I got to. I think what I have been failing to elucidate is that this is not how modern segmented address spaces work. It's not sufficient to have a few bits to specify which segment to use. Instead, you typically need three things: 1. The type of memory segment being described (which may be just a few bits). In reality, this "type" may correlate to which API you need to use to access this memory (e.g., main RAM, RDMA, accelerator memory, persistent memory, other IO memory, ...etc.). 2. A context of some kind (usually more than a small number of bits -- perhaps 32 or 64 bits). 3. A pointer (usually 32 or 64 bits). For example, if I'm accessing IO memory of some flavor, I probably need some kind of context/handle that points to a descriptor with information about my data segment on the IO peripheral in question (e.g., a NIC), and then a pointer for the specific place that I want in that specific memory. I use this context/handle + pointer with some underlying API to get/set that data. I think that's the disconnect here: the prior thinking about MPI_Aint may have assumed "it's either main RAM or it's XYZ memory, and the rest of the bits are enough to uniquely describe the address." But that doesn't seem sufficient any more to describe the evolution of complex, segmented memory systems (and be able to use their supporting APIs). Meaning: perhaps a pure integer MPI_Aint may have been sufficient in the past. But I'm not sure that's useful any more. -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Rolf Rabenseifner 24 Oct '19

24 Oct '19

Dear Jeff and all, > Does MPI support segmented address spaces or not? When I understand correctly then MPI is and was always prepared for segmented address spaces as long as all integer types are signed integers and the segmentation is done through some higher order bits. Best regards Rolf ----- Rolf Rabenseifner via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org> wrote: > See below > > ----- Original Message ----- > > From: "Jeff Squyres" <jsquyres(a)cisco.com> > > To: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de> > > Cc: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org> > > Sent: Wednesday, October 23, 2019 5:12:33 PM > > Subject: Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements > > > On Oct 23, 2019, at 10:44 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: > >> > >>> typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; > >>> > >>> Is that incorrect? > >> > >> I expect, that it is not correct. My apologies for that. > >> > >> In MPI-1.0 - MPI-2.1 all address calculation had to be done with - and + > >> operators. > >> Only for the case, that somebody switches integer overflow detection on, > >> we added MPI_Aint_add and MPI_Aint_diff. > > > > I'm confused with how to reconcile that with MPI-3.1 4.1.12 "Correct Use of > > Addresses", p115:13-15: > > > > "Also, in machines with a segmented address space, addresses are not unique and > > address arithmetic has some peculiar properties. Thus, the use of addresses, > > that is, displacements relative to the start address MPI_BOTTOM, has to be > > restricted." > > > > Doesn't this explicitly state that segmented address spaces are supported? > > Yes, some higher bits may be used as a flag. > If two absolute addresses are within the same array or structure (i.e. variable), > i.e., within the save sequential storage, then a minus operator must still work. > > > >> We asked ourself, whether we want to deprecate the use of + and - operators, > >> i.e., to add this to Table 2.1 on page 18 and the forum decided "no" per straw > >> vote. > >> We definitely never removed the operators + and - from the list of > >> valid operations for MPI_Aint. > > > > I'm not sure how to reconcile your statements with MPI-3.1 2.5.6 "Absolute > > Addresses and Relative Address Displacements" p16:39-43: > > > > "For retrieving absolute addresses or any calculation with absolute addresses, > > one should use the routines and functions provided in Section 4.1.5. Section > > 4.1.12 provides additional rules for the correct use of absolute addresses. For > > expressions with relative displacements or other usage without absolute > > addresses, intrinsic operators (e.g., +, -, *) can be used." > > > > If MPI_Aint is supposed to be used for absolute addresses, this tells me that > > MPI_Aint_add/diff must be used for all mathematical operations. > > The following definitions apply in understanding how to implement an > ISO International Standard and other normative ISO deliverables (TS, PAS, IWA). > > - "shall" indicates a requirement. > - "should" indicates a recommendation. > > > ... any calculation with absolute addresses, > > one ***should*** use the routines and functions provided in Section 4.1.5 > > ... indicates a recommendation, not a must > > > > >> MPI-3.1 on page 17 lines 16-18 clearly tells that any int, MPI:Aint, and > >> MPI_Offset value > >> can be assigned to a MPI_Count variable, i.e., all 4 must be integers of some > >> byte-size. > > > > I think it actually says something subtly different than that. It says: > > > > "The size of the MPI_Count type is determined by the MPI implementation with the > > restriction that it must be minimally capable of encoding any value that may be > > stored in a variable of type int, MPI_Aint, or MPI_Offset in C and of type > > INTEGER, INTEGER (KIND=MPI_ADDRESS_KIND), or INTEGER (KIND=MPI_OFFSET_KIND) in > > Fortran." > > > > Meaning: an MPI_Count must be big enough to hold an MPI_Aint. It does not say > > that you can assign one to the other. In my hypothetical "typedef struct" > > example, this means that MPI_Count would need to be a 128 bit value. > > > > ---- > > > > All this being said, what all this discussion probably means is that there are > > discrepancies in the standard that should be fixed and/or clarified. But I > > think the discussion starts with a fundamental question: > > > > Does MPI support segmented address spaces or not? > > > > -- > > Jeff Squyres > > jsquyres(a)cisco.com > > -- > Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . > High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . > University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . > Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . > Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) . > _______________________________________________ > mpiwg-large-counts mailing list > mpiwg-large-counts(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Rolf Rabenseifner 23 Oct '19

23 Oct '19

See below ----- Original Message ----- > From: "Jeff Squyres" <jsquyres(a)cisco.com> > To: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de> > Cc: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org> > Sent: Wednesday, October 23, 2019 5:12:33 PM > Subject: Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements > On Oct 23, 2019, at 10:44 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: >> >>> typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; >>> >>> Is that incorrect? >> >> I expect, that it is not correct. My apologies for that. >> >> In MPI-1.0 - MPI-2.1 all address calculation had to be done with - and + >> operators. >> Only for the case, that somebody switches integer overflow detection on, >> we added MPI_Aint_add and MPI_Aint_diff. > > I'm confused with how to reconcile that with MPI-3.1 4.1.12 "Correct Use of > Addresses", p115:13-15: > > "Also, in machines with a segmented address space, addresses are not unique and > address arithmetic has some peculiar properties. Thus, the use of addresses, > that is, displacements relative to the start address MPI_BOTTOM, has to be > restricted." > > Doesn't this explicitly state that segmented address spaces are supported? Yes, some higher bits may be used as a flag. If two absolute addresses are within the same array or structure (i.e. variable), i.e., within the save sequential storage, then a minus operator must still work. >> We asked ourself, whether we want to deprecate the use of + and - operators, >> i.e., to add this to Table 2.1 on page 18 and the forum decided "no" per straw >> vote. >> We definitely never removed the operators + and - from the list of >> valid operations for MPI_Aint. > > I'm not sure how to reconcile your statements with MPI-3.1 2.5.6 "Absolute > Addresses and Relative Address Displacements" p16:39-43: > > "For retrieving absolute addresses or any calculation with absolute addresses, > one should use the routines and functions provided in Section 4.1.5. Section > 4.1.12 provides additional rules for the correct use of absolute addresses. For > expressions with relative displacements or other usage without absolute > addresses, intrinsic operators (e.g., +, -, *) can be used." > > If MPI_Aint is supposed to be used for absolute addresses, this tells me that > MPI_Aint_add/diff must be used for all mathematical operations. The following definitions apply in understanding how to implement an ISO International Standard and other normative ISO deliverables (TS, PAS, IWA). - "shall" indicates a requirement. - "should" indicates a recommendation. > ... any calculation with absolute addresses, > one ***should*** use the routines and functions provided in Section 4.1.5 ... indicates a recommendation, not a must > >> MPI-3.1 on page 17 lines 16-18 clearly tells that any int, MPI:Aint, and >> MPI_Offset value >> can be assigned to a MPI_Count variable, i.e., all 4 must be integers of some >> byte-size. > > I think it actually says something subtly different than that. It says: > > "The size of the MPI_Count type is determined by the MPI implementation with the > restriction that it must be minimally capable of encoding any value that may be > stored in a variable of type int, MPI_Aint, or MPI_Offset in C and of type > INTEGER, INTEGER (KIND=MPI_ADDRESS_KIND), or INTEGER (KIND=MPI_OFFSET_KIND) in > Fortran." > > Meaning: an MPI_Count must be big enough to hold an MPI_Aint. It does not say > that you can assign one to the other. In my hypothetical "typedef struct" > example, this means that MPI_Count would need to be a 128 bit value. > > ---- > > All this being said, what all this discussion probably means is that there are > discrepancies in the standard that should be fixed and/or clarified. But I > think the discussion starts with a fundamental question: > > Does MPI support segmented address spaces or not? > > -- > Jeff Squyres > jsquyres(a)cisco.com -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Jeff Squyres (jsquyres) 23 Oct '19

23 Oct '19

On Oct 23, 2019, at 10:44 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de> wrote: > >> typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; >> >> Is that incorrect? > > I expect, that it is not correct. My apologies for that. > > In MPI-1.0 - MPI-2.1 all address calculation had to be done with - and + operators. > Only for the case, that somebody switches integer overflow detection on, > we added MPI_Aint_add and MPI_Aint_diff. I'm confused with how to reconcile that with MPI-3.1 4.1.3 "Correct Use of Addresses", p115:13-15: "Also, in machines with a segmented address space, addresses are not unique and address arithmetic has some peculiar properties. Thus, the use of addresses, that is, displacements relative to the start address MPI_BOTTOM, has to be restricted." Doesn't this explicitly state that segmented address spaces are supported? > We asked ourself, whether we want to deprecate the use of + and - operators, > i.e., to add this to Table 2.1 on page 18 and the forum decided "no" per straw vote. > We definitely never removed the operators + and - from the list of > valid operations for MPI_Aint. I'm not sure how to reconcile your statements with MPI-3.1 2.5.6 "Absolute Addresses and Relative Address Displacements" p16:39-43: "For retrieving absolute addresses or any calculation with absolute addresses, one should use the routines and functions provided in Section 4.1.5. Section 4.1.12 provides additional rules for the correct use of absolute addresses. For expressions with relative displacements or other usage without absolute addresses, intrinsic operators (e.g., +, -, *) can be used." If MPI_Aint is supposed to be used for absolute addresses, this tells me that MPI_Aint_add/diff must be used for all mathematical operations. > MPI-3.1 on page 17 lines 16-18 clearly tells that any int, MPI:Aint, and MPI_Offset value > can be assigned to a MPI_Count variable, i.e., all 4 must be integers of some byte-size. I think it actually says something subtly different than that. It says: "The size of the MPI_Count type is determined by the MPI implementation with the restriction that it must be minimally capable of encoding any value that may be stored in a variable of type int, MPI_Aint, or MPI_Offset in C and of type INTEGER, INTEGER (KIND=MPI_ADDRESS_KIND), or INTEGER (KIND=MPI_OFFSET_KIND) in Fortran." Meaning: an MPI_Count must be big enough to hold an MPI_Aint. It does not say that you can assign one to the other. In my hypothetical "typedef struct" example, this means that MPI_Count would need to be a 128 bit value. ---- All this being said, what all this discussion probably means is that there are discrepancies in the standard that should be fixed and/or clarified. But I think the discussion starts with a fundamental question: Does MPI support segmented address spaces or not? -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [Mpiwg-large-counts] Pythonizing + BigCount: virtual meeting on Nov 13?
by Jeff Hammond 23 Oct '19

23 Oct '19

Come on, Squyres. You know we have to call it ExaCounts in order for our publicly funded colleagues to take any interest in this. Jeff Sent from my iPhone > On Oct 23, 2019, at 10:11 AM, Jeff Squyres (jsquyres) via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org> wrote: > > Martin S -- > > The LargeCount / BigCount / WhopperCount / GinormousCount / UseYouFavoriteWordCount WG would like to reserve the 2-hour virtual meeting on Wed Nov 13. This is the Wed before SC, and I'm sure it will be incredibly well attended! ;-) > > We'd like to cover two things: > > 1. The state of LaTeX Pythonizing > (Recall: the Pythonization -- yes, that's totally a word; you must use it in a sentence with a colleague today -- of the MPI LaTeX is a separate effort than the Embiggening -- yes, that's also a word -- of "count" parameters. Specifically: we're doing a comparable-to-MPI-3.1 Pythonization of the MPI bindings in LaTeX without Embiggening or otherwise changing any parameters. Pythonization opens the door to multiple interesting things in the future, including Embiggening "count" parameters). > > 2. The state of BigCount / Embiggening > There was a reading and much discussion at the last Forum meeting about Embiggening. We'd like to get some Forum feedback before the next Forum meeting in December. > > -- > Jeff Squyres > jsquyres(a)cisco.com > > _______________________________________________ > mpiwg-large-counts mailing list > mpiwg-large-counts(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Rolf Rabenseifner 23 Oct '19

23 Oct '19

Dear Jeff, > typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; > > Is that incorrect? I expect, that it is not correct. My apologies for that. In MPI-1.0 - MPI-2.1 all address calculation had to be done with - and + operators. Only for the case, that somebody switches integer overflow detection on, we added MPI_Aint_add and MPI_Aint_diff. We asked ourself, whether we want to deprecate the use of + and - operators, i.e., to add this to Table 2.1 on page 18 and the forum decided "no" per straw vote. We definitely never removed the operators + and - from the list of valid operations for MPI_Aint. MPI-3.1 on page 17 lines 16-18 clearly tells that any int, MPI:Aint, and MPI_Offset value can be assigned to a MPI_Count variable, i.e., all 4 must be integers of some byte-size. Best regards Rolf ----- Original Message ----- > From: "Jeff Squyres" <jsquyres(a)cisco.com> > To: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de>, "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org> > Sent: Wednesday, October 23, 2019 4:19:53 PM > Subject: Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements > On Oct 9, 2019, at 1:44 PM, Rolf Rabenseifner via mpiwg-large-counts > <mpiwg-large-counts(a)lists.mpi-forum.org> wrote: > >> I never said that that a number is a displacement. >> MPI_Aint is clearly used for >> - addresses relative to a buffer begin, >> - absolute addresses (returned by MPI_GET_ADDRESS) >> - number of bytes. >> >> MPI-3.1 says >> 2.5.6 Absolute Addresses and Relative Address Displacements >> Some MPI procedures use address arguments that represent an absolute address in >> the calling >> program, or relative displacement arguments that represent differences of two >> absolute >> addresses. The datatype of such arguments is MPI_Aint in C and INTEGER (KIND= >> MPI_ADDRESS_KIND) in Fortran. > > I thought the whole point of MPI_Aint was to support segmented address spaces. > Indeed, the next sentence in MPI-3.1 states: > > "These types must have the same width and encode address values in the same > manner such that address values in one language may be passed directly to > another language without conversion." > > This says to me that MPI_Aint can be encoded, which I take to mean the > following: > > 1. This is exactly the reason why we have MPI_AINT_ADD / MPI_AINT_DIFF. > 2. Presuming I have an MPI_ADDRESS_KIND value in Fortran that represents a 128 > bit integer, the following is valid in C: > > typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; > > Is that incorrect? > > -- > Jeff Squyres > jsquyres(a)cisco.com -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .

1 0

Re: [Mpiwg-large-counts] Large Count - the principles for counts, sizes, and byte and nonbyte displacements
by Jeff Squyres (jsquyres) 23 Oct '19

23 Oct '19

On Oct 9, 2019, at 1:44 PM, Rolf Rabenseifner via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org> wrote: > I never said that that a number is a displacement. > MPI_Aint is clearly used for > - addresses relative to a buffer begin, > - absolute addresses (returned by MPI_GET_ADDRESS) > - number of bytes. > > MPI-3.1 says > 2.5.6 Absolute Addresses and Relative Address Displacements > Some MPI procedures use address arguments that represent an absolute address in the calling > program, or relative displacement arguments that represent differences of two absolute > addresses. The datatype of such arguments is MPI_Aint in C and INTEGER (KIND= > MPI_ADDRESS_KIND) in Fortran. I thought the whole point of MPI_Aint was to support segmented address spaces. Indeed, the next sentence in MPI-3.1 states: "These types must have the same width and encode address values in the same manner such that address values in one language may be passed directly to another language without conversion." This says to me that MPI_Aint can be encoded, which I take to mean the following: 1. This is exactly the reason why we have MPI_AINT_ADD / MPI_AINT_DIFF. 2. Presuming I have an MPI_ADDRESS_KIND value in Fortran that represents a 128 bit integer, the following is valid in C: typedef struct { uint64_t context; uint64_t pointer; } MPI_Aint; Is that incorrect? -- Jeff Squyres jsquyres(a)cisco.com

1 0

Pythonizing + BigCount: virtual meeting on Nov 13?
by Jeff Squyres (jsquyres) 23 Oct '19

23 Oct '19

Martin S -- The LargeCount / BigCount / WhopperCount / GinormousCount / UseYouFavoriteWordCount WG would like to reserve the 2-hour virtual meeting on Wed Nov 13. This is the Wed before SC, and I'm sure it will be incredibly well attended! ;-) We'd like to cover two things: 1. The state of LaTeX Pythonizing (Recall: the Pythonization -- yes, that's totally a word; you must use it in a sentence with a colleague today -- of the MPI LaTeX is a separate effort than the Embiggening -- yes, that's also a word -- of "count" parameters. Specifically: we're doing a comparable-to-MPI-3.1 Pythonization of the MPI bindings in LaTeX without Embiggening or otherwise changing any parameters. Pythonization opens the door to multiple interesting things in the future, including Embiggening "count" parameters). 2. The state of BigCount / Embiggening There was a reading and much discussion at the last Forum meeting about Embiggening. We'd like to get some Forum feedback before the next Forum meeting in December. -- Jeff Squyres jsquyres(a)cisco.com

1 0

Re: [Mpiwg-large-counts] A perfect trick in C with similar API as with Fortran overloading - instead of the many new _l functions in C
by Jeff Squyres (jsquyres) 21 Oct '19

21 Oct '19

I agree with Hammy. I think we all agree that adding a bunch of new _x and/or _l symbols sucks. But it's unfortunately the least evil solution. :-\ C11 was our best hope to avoid that (from the user perspective at least), but that didn't work out. [Image result for sad panda needs a hug gif] On Oct 21, 2019, at 2:43 PM, Jeff Hammond <jeff.science(a)gmail.com<mailto:[email protected]>> wrote: If users want such a preprocessor solution, they should get it from a third-party implementation. I do not think it belongs in the MPI standard. Jeff On Oct 21, 2019, at 11:40 AM, Rolf Rabenseifner <rabenseifner(a)hlrs.de<mailto:[email protected]>> wrote: Dear all, Yes, I understand. And providing the new API in a new module mpi_x.h, but with same function names MPI_Send ... is then also an evil. Especially because the two MPI_Send(..,int count,..) from mpi.h and MPI_Send(..,MPI_Count count,..) from mpi_x.h cannot be together within the same lib because C does not allow to define an ABI-function-Name that is not identical to the API-function-name. Therefore next proposal: We do what we planned, but can additionally provide an mpi_l.h header file containing only a set of aliases #define MPI_Send(...) MPI_Send_l(...) ... User who changed somehow all their 4 byte integer in into 8 byte integer can then directly use the new routines without changing all the MPI function calls. As far as I understand, wouldn't this add-on proposal fall into the evil category, or would it? Best regards Rolf ----- Original Message ----- From: "Jeff Squyres" <jsquyres(a)cisco.com<mailto:[email protected]>> To: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> Cc: "Rolf Rabenseifner" <rabenseifner(a)hlrs.de<mailto:[email protected]>>, "Jeff Hammond" <jeff.science(a)gmail.com<mailto:[email protected]>> Sent: Monday, October 21, 2019 3:41:53 PM Subject: Re: [Mpiwg-large-counts] A perfect trick in C with similar API as with Fortran overloading - instead of the many new _l functions in C I agree: breaking ABI in this way is truly evil. Specifically: compile your app with implementation version X. Then run your app with same implementation, but version Y (this is not uncommon for commercial ISV MPI apps). Things fail in weird, non-obvious ways (or worse, they *don't* fail, but silently give you wrong results). Speaking as someone who writes code for a living, changing the signature of an existing, well-established function is evil. Changing it in ways that a compiler may not even warn you about (in forward- and backward-compatility scenarios) is TRULY evil. On Oct 21, 2019, at 9:35 AM, Jeff Hammond via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: ABI is very much a user problem. We are not breaking ABI for this. What is being done now is five years in the making. I’m not sure why you didn’t comment until recently. Jeff Sent from my iPhone On Oct 20, 2019, at 10:49 PM, Rolf Rabenseifner via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: Dear Tony and all, the group has labored over each area... 1 sided has had the least attention so far... All of MPI is to be made 64 bit clean not parts. My proposal ia for whole MPI. Only for few routines we need some additional _x versions. Partial answer to your earliest question in the note: Changing the meaning of existing APIs was disallowed some time ago. Certainly, for in arguments of count , replacing int with MPI_Count is a partial solution. But it changes the APIs ... The goal is backward compatibility. And this one is provided. In C, a program written for the assumption of big count and compiled accidentally with an MPI-3 compliant MPI will silently build and fail at runtime ... rounding of integers ... Let's recommend to the users of the long coutnt feature that they should at a check of the version. In C, in can be tested at compile time with cpp. and changing the AbI violates the ABI for tools and such — two possible reasons why not chosen ... but we will have to go refresh on all the reasons . Not a user problem ... I am sure you will get extensive feedback on these questions from the whole group ? :-) And also by the plenum. Best regards Rolf ----- Tony Skjellum <skjellum(a)gmail.com<mailto:[email protected]>> wrote: Rolf, the group has labored over each area... 1 sided has had the least attention so far... All of MPI is to be made 64 bit clean not parts. Partial answer to your earliest question in the note: Changing the meaning of existing APIs was disallowed some time ago. Certainly, for in arguments of count , replacing int with MPI_Count is a partial solution. But it changes the APIs ... In C, a program written for the assumption of big count and compiled accidentally with an MPI-3 compliant MPI will silently build and fail at runtime ... rounding of integers ... and changing the AbI violates the ABI for tools and such — two possible reasons why not chosen ... but we will have to go refresh on all the reasons . I am sure you will get extensive feedback on these questions from the whole group ? :-) Tony Anthony Skjellum, PhD 205-807-4968 On Oct 20, 2019, at 2:25 PM, Rolf Rabenseifner via mpiwg-large-counts <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: Dear all, a participant in my latest course asked me, whether we not substituting in by MPI_Count in all "int xxxxx" input arguments. The compiler automatically would cast to MPI_Count which may be a fast instruction. Then we need the new _x or _l functions only for those rare functions with MPI_Count as output or array argument. Was this ever discussed? Second question: Does the large Count roup plan, only to use MPI_Count for data, i.e., only in communication, I/O, and derived type routines? Is it correct, that we keep the number of - request handles in request arrays, - ranks an size in/of communicator and group handles, will stay int? How do we handle disp in 1-sided communication? With this approach, it seems that explicit MPI_Count versions of a subroutine is needed only for MPI_[I][NEIGHBOR_]{ALLTOALL|ALLGATHER}{V|W} MPI_[I]{GATHER|SCATTER}V MPI_[I]REDUCE_SCATTER (18 routines) MPI_WIN_SHARED_QUERY --> *disp (1 Routine) MPI_[UN]PACK MPI_PACK_SIZE _x/_l version should have MPI_Aint (same as for MPI_...PACK..._external) (3 routines) MPI_GET_COUNT MPI_GET_ELEMENTS (has already an _X version) MPI_TYPE_SIZE (has already an _X version) Which solution exists for MPI_CONVERSION_FN_NULL? All other 100 routines with "int count" and "int size" seems to be solved with directly changing it into "MPI_Count count" and "MPI_Count size". Best regards Rolf ----- Original Message ----- From: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> To: "mpiwg-large-counts" <mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]>> Cc: "Jeff Squyres" <jsquyres(a)cisco.com<mailto:[email protected]>> Sent: Saturday, October 19, 2019 10:26:23 PM Subject: [Mpiwg-large-counts] Pythonizing instructions As discussed in our last meeting, I've written a first take at instructions for Chapter Authors for how to write Pythonized functions. These instructions are not yet intended for a wide audience -- they are intended solely for the working group. Let's iterate on them here in the WG and get them to a good enough state before sharing them widely: https://github.com/mpiwg-large-count/large-count-issues/wiki/pythonizing Puri/Tony/Dan: please try to use these instructions to Pythonize a few bindings. The PR for the Pythonization is https://github.com/mpiwg-large-count/mpi-standard/pull/2. All even-numbered chapters are done, and a small number of odd-numbered chapters. Grep for `funcdef` in chapter *.tex files to find a chapter that hasn't been Pythonized yet (funcdef is the LaTeX macro for hard-coded LaTeX macros). Send your feedback here and/or we'll discuss on Wednesday. -- Jeff Squyres jsquyres(a)cisco.com _______________________________________________ mpiwg-large-counts mailing list mpiwg-large-counts(a)lists.mpi-forum.org https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de<mailto:[email protected]> . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner<http://www.hlrs.de/people/rabenseifner> . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) . _______________________________________________ mpiwg-large-counts mailing list mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de<mailto:[email protected]> . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner<http://www.hlrs.de/people/rabenseifner> . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) . _______________________________________________ mpiwg-large-counts mailing list mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts _______________________________________________ mpiwg-large-counts mailing list mpiwg-large-counts(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-large-counts -- Jeff Squyres jsquyres(a)cisco.com<mailto:[email protected]> -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de<mailto:[email protected]> . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner<http://www.hlrs.de/people/rabenseifner> . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) . -- Jeff Squyres jsquyres(a)cisco.com<mailto:[email protected]>

1 0