|
| View previous topic :: View next topic |
| Author |
Message |
SarahA
Joined: 29 Aug 2006 Posts: 16
|
Posted: Wed May 12, 2010 9:44 am Post subject: mcmodel medium linking for x86_64/EMT64 |
|
|
Has anyone seen this error linking a large-host memory application? This is with PGI 10.4.0 I can't find this entrypoint in the library set either.
| Code: | -g -r8 -D_BIG_ENDIAN -Mextend -Mcuda=cc13,keepptx -fastsse -mcmodel=medium -Minform,inform -Minfo=all,ftn,intensity -o slv_5d
chiralcuda.o: In function `mopr5_eo_':
/home/users/saraha/PreBenchmark/Src/./chiralcuda.F:141: undefined reference to `pgf90_dev_auto_alloc_i8'
/home/users/saraha/PreBenchmark/Src/./chiralcuda.F:287: undefined reference to `pgf90_dev_auto_dealloc_i8'
cg_overlap_5d.o: In function `mopr5_5d_eo_':
/home/users/saraha/PreBenchmark/Src/./cg_overlap_5d.F:323: undefined reference to `pgf90_dev_auto_alloc_i8'
/home/users/saraha/PreBenchmark/Src/./cg_overlap_5d.F:323: undefined reference to `pgf90_dev_auto_alloc_i8'
/home/users/saraha/PreBenchmark/Src/./cg_overlap_5d.F:336: undefined reference to `pgf90_dev_auto_dealloc_i8'
/home/users/saraha/PreBenchmark/Src/./cg_overlap_5d.F:336: undefined reference to `pgf90_dev_auto_dealloc_i8 |
|
|
| Back to top |
|
 |
SarahA
Joined: 29 Aug 2006 Posts: 16
|
Posted: Thu May 13, 2010 3:26 pm Post subject: undefined reference to `pgf90_dev_auto_dealloc_i8' |
|
|
I guess no one has run into this. Here is a simple reproducer using PGI fortran 10.4.0
nvcc -c bigtest_kernel.cu
pgfortran -Mcuda -mcmodel=medium bigtest_kernel.o bigtest.F
I get: | Code: | pgfortran -Mcuda -mcmodel=medium bigtest_kernel.o bigtest.F
bigtest.F:
bigtest.o: In function `MAIN':
/home/users/saraha/./bigtest.F:11: undefined reference to `pgf90_dev_auto_alloc_i8'
/home/users/saraha/./bigtest.F:11: undefined reference to `pgf90_dev_auto_dealloc_i8' |
where bigtest_kernel.cu is just | Code: | | extern "C" {__global__ void ackernel_( float *arg ) {} } |
and bigtest.F is | Code: | integer, parameter:: M=400
integer, parameter:: N=2097152
real, device, dimension(1000000) :: a_dev
common /biggy/ a(N,M)
interface
subroutine ackernel( arg )
use iso_c_binding
real, device, dimension(100000) :: arg
end subroutine ackernel
end interface
call ackernel<<<1,100>>>(a_dev)
print *, a(1,1)
stop
end
| [/quote] |
|
| Back to top |
|
 |
mkcolg
Joined: 30 Jun 2004 Posts: 4996 Location: The Portland Group Inc.
|
Posted: Thu May 13, 2010 5:18 pm Post subject: |
|
|
Hi SarahA,
Using the medium memory model with CUDA Fortran isn't supported yet. It's in our feature queue, but had been a lower priority item. I've asked our developers to increase it's priority so hopefully we have have it available soon.
Thanks!
Mat |
|
| Back to top |
|
 |
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2002 phpBB Group
|