-f-cuda not working

[rzil]

I'm trying to install accelerate-examples on my MacBook Pro using this
cabal install -f-cuda accelerate-examples

For some reason, however, it is still trying to install the cuda package which doesn't work because my graphics card doesn't support CUDA.

[tmcdonell]

hm, can you cabal install -f-cuda accelerate-examples --dry-run and show the list. I am guessing that cabal isn't applying the -f-cuda bit to all packages, and something else is still trying to pull it in still; maybe accelerate-fft or gloss-[raster-]accelerate.

[rzil]

cabal install -f-cuda accelerate-examples --dry-run
Resolving dependencies...
In order, the following would be installed (use -v for more details):
cuda-0.6.5.1
accelerate-cuda-0.15.0.0
cufft-0.1.1.0 (latest: 0.1.2.0)
accelerate-fft-0.15.0.0
gloss-accelerate-1.8.15.0
gloss-raster-accelerate-1.8.15.0
accelerate-examples-0.15.0.0

On 28 Feb 2015, at 2:53 am, Trevor L. McDonell notifications@github.com wrote:

hm, can you cabal install -f-cuda accelerate-examples --dry-run and show the list. I am guessing that cabal isn't applying the -f-cuda bit to all packages, and something else is still trying to pull it in still; maybe accelerate-fft or gloss-[raster-]accelerate.

—
Reply to this email directly or view it on GitHub #243 (comment).

[tmcdonell]

Try the following?

cabal install accelerate-fft -f-cuda
cabal install gloss-accelerate -f-cuda
cabal install gloss-raster-accelerate -f-cuda

[rzil]

That worked as far as installing it is concerned. I also had to brew install llvm and create some symlinks but yeah, it installed. Only problem is that the example programs don't work!

[tmcdonell]

If you're running under the interpreter, I am guessing that the example programs do actually work, but they just take longer to compute than you were willing to wait (;

accelerate-crystal --benchmark

May be informative...

The interpreter is only meant to be a reference implementation for the semantics of the language, it is not designed for speed. For that, you'll either need to install the CUDA backend (which doesn't work on your machine) or one of the (incomplete) CPU backends.

[rzil]

Oh I see. I was hoping to use OpenCL but now I can see that it is not available for some reason.

[tmcdonell]

Yes, the OpenCL backend has languished. If you know OpenCL, then we could use your expertise... (;

Moving forward, we could support AMD GPUs via the LLVM backend.

[mchakravarty]

Moving forward, we could support AMD GPUs via the LLVM backend.

Makes the most sense to me, especially give SPIR-V: https://www.khronos.org/vulkan

[rzil]

I checked out the source for accelerate-opencl and I am updating it. I’m stuck at the following error. What happened to these objects Let2 and PairArrays?

Data/Array/Accelerate/OpenCL/CodeGen.hs:91:9:
Not in scope: data constructor ‘Let2’
Perhaps you meant ‘Let’ (imported from Data.Array.Accelerate.AST)

Data/Array/Accelerate/OpenCL/CodeGen.hs:95:9:
Not in scope: data constructor ‘PairArrays’

Ruben

On 4 Mar 2015, at 1:53 pm, Manuel M T Chakravarty notifications@github.com wrote:

Moving forward, we could support AMD GPUs via the LLVM backend.

Makes the most sense to me, especially give SPIR-V: https://www.khronos.org/vulkan https://www.khronos.org/vulkan
—
Reply to this email directly or view it on GitHub #243 (comment).

[tmcdonell]

Ah, that was replaced by Atuple and Aprj quite a while ago... (298a3f2)
So getting that backend up to speed could take some effort.