Re: [AMBER-Developers] AMBER11 bugfix #19 is broken

From: David A Case <>
Date: Wed, 23 Nov 2011 09:31:11 -0500

On Wed, Nov 23, 2011, Mark Williamson wrote:

> mpicc -ip -O3 -no-prec-div -xHost -DMPICH_IGNORE_CXX_SEEK
> -DMPICH_IGNORE_CXX_SEEK -I/usr/local/Cluster-Apps/cuda/3.2/cuda/include
> -IB40C -IB40C/KernelCommon
> -I/usr/local/Cluster-Apps.nehalem/mvapich2/1.6/ofa/include -c gpu.cpp
> gpu.cpp(445): error: class "_gpuContext" has no member "totalMemory"
> if (gpu->totalMemory < 2 * 1024 * 1024 * 1024)
> ^
> gpu.cpp(445): warning #61: integer operation result is out of range
> if (gpu->totalMemory < 2 * 1024 * 1024 * 1024)

Appears that Jason left out commits 4152f12336d9 and d6dc39655857e when
preparing bugfix.19. (To be fair, the latter was fixed after the bugfix
was prepared.) Looks(?) like somehow Jason had an inconsistent version of
gputypes.h in the working directory where he prepared bugfix.19.]

> Seeing that the cruise control framework is now in place,

I think all we need is a cuda cruise control build for amber11-with-patches,
which would have caught this problem. [We currently only test the serial
version of amber11-with-patches.]

I suggest just updating bugfix.19 in place, but we could make a bugfix.20 if
people think it is cleaner. I really hope we can limp home until Amber12,
when things get cleaned up and we no longer have enormous feature enhancements
disguised as bugfixes.


AMBER-Developers mailing list
Received on Wed Nov 23 2011 - 07:00:02 PST
Custom Search