[Llvm-bgq-discuss] mathcalls.h and ppu_intrinsics.h

[Jozsef Bakosi]

Another solution is

#undef __powerpc__
#define __x86_64__
#define R123_USE_MULHILO64_MULHI_INTRIN 0
#define R123_USE_GNU_UINT128 1
#include <Random123/philox.h> or #include <Random123/threefry.h>
#define __powerpc__
#undef __x86_64__

Which has the advantage that the Random123 source does not need to be modified,
i.e., the #include <ppu_intrinsics.h> does not need to be taken out from their
Random123/philox.h and Random123/threefry.h.

Jozsef

On 01.25.2017 15:26, Jozsef Bakosi wrote:
> Oh, and of course I had to take out the #include <ppu_intrinsics.h> from their
> Random123/philox.h.
> 
> J
> 
> On 01.25.2017 15:23, Jozsef Bakosi wrote:
> > Hi Hal,
> > 
> > When I commented out the inclusion of ppu_intrinsics.h, I got:
> > 
> > <...>/include/Random123/philox.h:198:40: error: use of undeclared identifier '__mulhdu'
> > _mulhilo_cuda_intrin_tpl(64, uint64_t, R123_MULHILO64_MULHI_INTRIN)
> > 
> > Thus it is __mulhdu, so I guess ppu_intrinsics.h would pull in the assembly
> > intrinsics for the mulhdu (Multiply High Double Word Unsigned) instruction.
> > Also, as you can see, this is coming out of the Random123 random number
> > generator library:
> > http://www.thesalmons.org/john/random123/releases/latest/docs/index.html.
> > 
> > Based on their documentation I was able to work around the problem by overriding
> > a couple of their pre-processor symbols before including their header:
> > 
> > #define R123_USE_MULHILO64_MULHI_INTRIN 0
> > #define R123_USE_GNU_UINT128 1
> > 
> > #include <Random123/philox.h>
> > 
> > This turns off looking for the intrinsics that would be enabled via
> > ppu_intrinsics.h and falls back to a non-intrinsics version.
> > 
> > Though this compiles fine, I have not been able to test this solution for
> > correctness or performance yet.
> > 
> > I guess a ppu_intrinsics.h that is compatible with mathcalls.h would allow me to
> > use the intrinsics, but I'm okay with this solution for now.
> > 
> > Jozsef
> > -- 
> > Jozsef Bakosi
> > Computational Physics and Algorithms (CCS-2)
> > MS D413, Los Alamos National Laboratory
> > Los Alamos, NM 87545, USA
> > email: jbakosi at lanl.gov
> > phone: 505-665-0950
> > fax: 505-665-4972
> > 
> > On 01.24.2017 18:05, Hal Finkel wrote:
> > > Hi Jozsef,
> > > 
> > > Interesting. Do you know what is being used out of ppu_intrinsics.h? That is
> > > really a GCC-internal header file. Maybe we need a Clang equivalent, or
> > > maybe this is easy to work around?
> > > 
> > >  -Hal
> > > 
> > > 
> > > On 01/24/2017 03:02 PM, Jozsef Bakosi wrote:
> > > > Hi folks,
> > > > 
> > > > I'm trying to compile code including ppu_intrinsics.h as well as math.h and they
> > > > seem to be incompatible with each other. Here is the error I get:
> > > > 
> > > > =============================================
> > > > In file included from /soft/compilers/bgclang/r284961-stable/libc++/include/c++/v1/unordered_map:369:
> > > > In file included from /soft/compilers/bgclang/r284961-stable/libc++/include/c++/v1/__hash_table:19:
> > > > In file included from /soft/compilers/bgclang/r284961-stable/libc++/include/c++/v1/cmath:305:
> > > > In file included from /soft/compilers/bgclang/r284961-stable/libc++/include/c++/v1/math.h:301:
> > > > In file included from /bgsys/drivers/V1R2M2/ppc64/gnu-linux-4.7.2/powerpc64-bgq-linux/sys-include/math.h:70:
> > > > /bgsys/drivers/V1R2M2/ppc64/gnu-linux-4.7.2/powerpc64-bgq-linux/sys-include/bits/mathcalls.h:182:1: error: exception specification in declaration does not match previous
> > > >        declaration
> > > > __MATHCALLX (fabs,, (_Mdouble_ __x), (__const__));
> > > > ^
> > > > /bgsys/drivers/V1R2M2/ppc64/gnu-linux-4.7.2/powerpc64-bgq-linux/sys-include/math.h:59:3: note: expanded from macro '__MATHCALLX'
> > > >    __MATHDECLX (_Mdouble_,function,suffix, args, attrib)
> > > >    ^
> > > > /bgsys/drivers/V1R2M2/ppc64/gnu-linux-4.7.2/powerpc64-bgq-linux/sys-include/math.h:62:22: note: expanded from macro '__MATHDECLX'
> > > >    __MATHDECL_1(type, __CONCAT(__,function),suffix, args) __attribute__ (attrib)
> > > >                       ^
> > > > /bgsys/drivers/V1R2M2/ppc64/gnu-linux-4.7.2/powerpc64-bgq-linux/sys-include/sys/cdefs.h:89:23: note: expanded from macro '__CONCAT'
> > > > #define __CONCAT(x,y)   x ## y
> > > >                          ^
> > > > <scratch space>:360:1: note: expanded from here
> > > > __fabs
> > > > ^
> > > > /usr/lib/gcc/ppc64-redhat-linux/4.4.4/include/ppu_intrinsics.h:411:16: note: expanded from macro '__fabs'
> > > > #define __fabs __ppu_fabs
> > > >                 ^
> > > > /usr/lib/gcc/ppc64-redhat-linux/4.4.4/include/ppu_intrinsics.h:416:1: note: previous declaration is here
> > > > __fabs(double x)
> > > > ^
> > > > /usr/lib/gcc/ppc64-redhat-linux/4.4.4/include/ppu_intrinsics.h:411:16: note: expanded from macro '__fabs'
> > > > #define __fabs __ppu_fabs
> > > >                 ^
> > > > =============================================
> > > > 
> > > > Does anyone have an idea of a workaround?
> > > > 
> > > > Thanks,
> > > > Jozsef
> > > 
> > > -- 
> > > Hal Finkel
> > > Lead, Compiler Technology and Programming Languages
> > > Leadership Computing Facility
> > > Argonne National Laboratory