The accelerate-cuda package

[Tags:bsd3, library]

This library implements a backend for the Accelerate language instrumented for parallel execution on CUDA-capable NVIDIA GPUs. For further information, refer to the main Accelerate package:

To use this backend you will need:

1. A CUDA-enabled NVIDIA GPU with, for full functionality, compute capability 1.3 or greater. See the table on Wikipedia for supported GPUs:

2. The CUDA SDK, available from the NVIDIA Developer Zone:

See the Haddock documentation for additional information related to using this backend.

Compile modules that use the CUDA backend with the -threaded flag.


Dependencies accelerate (==0.15.1.*), array (>=0.3), base (>=4.7 && <4.9), binary (>=0.7), bytestring (>=0.9), cryptohash (>=0.7), cuda (>=0.7), directory (>=1.0), fclabels (>=2.0), filepath (>=1.0), hashable (>=1.1), hashtables (>=1.0.1), language-c-quote (>=0.4.4 && <0.9 || >0.10.1), mainland-pretty (>=0.2), mtl (>=2.0), old-time (>=1.0), pretty (>=1.0), process (>=1.0), SafeSemaphore (>=0.9), srcloc (>=0.2), template-haskell (>=2.2), text (>=0.11), transformers (>=0.2), unix (>=2.4), unordered-containers (>=0.1.4), Win32 (>=2.2.1) [details]
License BSD3
Author Manuel M T Chakravarty, Robert Clifton-Everest, Gabriele Keller, Sean Lee, Trevor L. McDonell
Maintainer Trevor L. McDonell <>
Stability Experimental
Category Compilers/Interpreters, Concurrency, Data, Parallelism
Home page
Bug tracker
Source repository this: git clone -b release/0.15(tag
Uploaded Tue Jan 19 21:47:45 UTC 2016 by TrevorMcDonell
Distributions NixOS:
Downloads 3606 total (35 in the last 30 days)
0 []
Status Docs not available [build log]
All reported builds failed as of 2016-01-19 [all 1 reports]


  • Data
    • Array
      • Accelerate
        • Data.Array.Accelerate.CUDA
          • Data.Array.Accelerate.CUDA.Foreign


debugEnable tracing message flags. These are read from the command-line arguments, which is convenient but may cause problems interacting with the user program, so is not enabled by default. The available options: * -ddump-cc: print the generated code, kernel table management information, nvcc compiler warnings, and thread & resource statistics * -ddebug-cc: compile code with debugging symbols, suitable for 'cuda-gdb' * -ddump-exec: print each kernel name as it is invoked * -ddump-gc: print memory management information * -dverbose: other, uncategorised messages * -fflush-cache: delete the persistent kernel cache DisabledAutomatic
bounds-checksEnable bounds checkingEnabledAutomatic
unsafe-checksEnable bounds checking in unsafe operationsDisabledAutomatic
internal-checksEnable internal consistency checksDisabledAutomatic

Use -f <flag> to enable a flag, or -f -<flag> to disable that flag. More info


Maintainer's Corner

For package maintainers and hackage trustees