atomicAdd for floating point operations how to specify sm_20 arch

Accelerated Computing CUDA CUDA Programming and Performance
1

hi

I have read that to use atomicAdd for floating point operations in gpu with compute capability 2.x, we have to specify sm_20 arch.
Someone please aid me to do this. I am using VS 2008, win 7 64 bit.
I was using “simpleAtomicIntrinsics” example from sdk . It works if I use atomics for integers.
When I give floating point numbers the compiler gives the following error

///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
1>------ Build started: Project: simpleAtomicIntrinsics, Configuration: Debug x64 ------

1>Compiling with CUDA Build Rule…

1>“C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v3.2\bin\nvcc.exe” -gencode=arch=compute_11,code="sm_11,compute_11" -gencode=arch=compute_20,code="sm_20,compute_20" --machine 64 -ccbin “C:\Program Files (x86)\Microsoft Visual Studio 9.0\VC\bin” -Xcompiler “/EHsc /W3 /nologo /Od /Zi /MTd " -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v3.2/include” -I"./" -I"…/…/common/inc" -I"…/…/…/shared/inc" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v3.2\include" -maxrregcount=32 --compile -o “x64\Debug/simpleAtomicIntrinsics.cu.obj” simpleAtomicIntrinsics.cu

1>simpleAtomicIntrinsics.cu

c:/ProgramData/NVIDIA Corporation/NVIDIA GPU Computing SDK 3.2/C/src/simpleAtomicIntrinsics/simpleAtomicIntrinsics.cu(86): error: no instance of overloaded function “atomicAdd” matches the argument list

1> argument types are: (float *, float)

////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

someone please help me to fix this.

2

Each time a description of the change, there is always plenty to talk about around the world. They are not exempt. A statement usually is: "i have to change almost immediately versions, so do not expect me to date continue to attack …

3

Can you post the code you are trying to use? It sounds like you are feeding atomicAdd the wrong arguments.

4

Sorry to bump an old topic, but i’ve got the same problem here.
I added “-arch=sm_20” to Project > Properties > CUDA C/C++ > Command line > Additional Options, but it won’t work.
Is there anything else ?

5

In VC …property pages …go to CUDA Runtime API/GPU…and select the device archi…

6

Okay ! You’re sure it’s the problem ? Else I’ll be totally stuck tuesday at work :s
Thank you

7

Well sadly, it doesn’t work. I don’t have “CUDA Runtime API/GPU” but only “CUDA C/C++” in property pages, and if i go to Device > Code generation “compute_20,sm_20” doesn’t work.

The weird thing is, if I try “-arch=sm_11” it does tell me

error : no instance of overloaded function "atomicAdd" matches the argument list

argument types are: (float *, float)

which is fine,

but with “-arch=sm_20” I only have “error : identifier “atomicAdd” is undefined” …

What’s wrong : ( ?

EDIT : Okay, on VS2010 :

And don’t put “compute_20,sm_20;%CodeGeneration;” as I did !

8

Well sadly, it doesn’t work. I don’t have “CUDA Runtime API/GPU” but only “CUDA C/C++” in property pages, and if i go to Device > Code generation “compute_20,sm_20” doesn’t work.

The weird thing is, if I try “-arch=sm_11” it does tell me

error : no instance of overloaded function "atomicAdd" matches the argument list

argument types are: (float *, float)

which is fine,

but with “-arch=sm_20” I only have “error : identifier “atomicAdd” is undefined” …

What’s wrong : ( ?

EDIT : Okay, on VS2010 :

And don’t put “compute_20,sm_20;%CodeGeneration;” as I did !