User-visible Changes

Version 0.93

Warning

Version 0.93 makes some changes to the PyCUDA programming interface. In all cases where documented features were changed, the old usage continues to work, but results in a warning. It is recommended that you update your code to remove the warning.

• OpenGL interoperability in pycuda.gl.
• Document pycuda.gpuarray.GPUArray.__len__(). Change its definition to match numpy.
• Add pycuda.gpuarray.GPUArray.bind_to_texref_ext().
• Let pycuda.gpuarray.GPUArray operators deal with generic data types, including type promotion.
• Add pycuda.gpuarray.take().
• Fix thread handling by making internal context stack thread-local.
• Add pycuda.reduction.ReductionKernel.
• Add pycuda.gpuarray.sum(), pycuda.gpuarray.dot(), pycuda.gpuarray.subset_dot().
• Synchronous and asynchronous memory transfers are now separate from each other, the latter having an _async suffix. The now-synchronous forms still take a pycuda.driver.Stream argument, but this practice is deprecated and prints a warning.
• pycuda.gpuarray.GPUArray no longer has an associated pycuda.driver.Stream. Asynchronous GPUArray transfers are now separate from synchronous ones and have an _async suffix.
• Support for features added in CUDA 2.2.
• pycuda.driver.SourceModule has been moved to pycuda.compiler.SourceModule. It is still available by the old name, but will print a warning about the impending deprecation.
• pycuda.driver.Device.get_attribute() with a pycuda.driver.device_attribute attr can now be spelled dev.attr, with no further namespace detours. (Suggested by Ian Cullinan) Likewise for pycuda.driver.Function.get_attribute()
• pycuda.driver.Function.registers, pycuda.driver.Function.lmem, and pycuda.driver.Function.smem have been deprecated in favor of the mechanism above. See pycuda.driver.Function.num_regs for more.
• Add PyCUDA version query mechanism, see pycuda.VERSION.

Version 0.92

Note

If you’re upgrading from prior versions, you may delete the directory $HOME/.pycuda-compiler-cache to recover now-unused disk space.

Note

During this release time frame, I had the honor of giving a talk on PyCUDA for a class that a group around Nicolas Pinto was teaching at MIT. If you’re interested, the slides for it are available.

• Make pycuda.tools.DeviceMemoryPool official functionality, after numerous improvements. Add pycuda.tools.PageLockedMemoryPool for pagelocked memory, too.
• Properly deal with automatic cleanup in the face of several contexts.
• Fix compilation on Python 2.4.
• Fix 3D arrays. (Nicolas Pinto)
• Improve error message when nvcc is not found.
• Automatically run Python GC before throwing out-of-memory errors.
• Allow explicit release of memory using pycuda.driver.DeviceAllocation.free(), pycuda.driver.HostAllocation.free(), pycuda.driver.Array.free(), pycuda.tools.PooledDeviceAllocation.free(), pycuda.tools.PooledHostAllocation.free().
• Make configure switch ./configure.py --cuda-trace to enable API tracing.
• Add a documentation chapter and examples on Metaprogramming with PyCUDA.
• Add pycuda.gpuarray.empty_like() and pycuda.gpuarray.zeros_like().
• Add and document pycuda.gpuarray.GPUArray.mem_size in anticipation of stride/pitch support in pycuda.gpuarray.GPUArray.
• Merge Jozef Vesely’s MD5-based RNG.
• Document pycuda.driver.from_device() and pycuda.driver.from_device_like().
• Add pycuda.elementwise.ElementwiseKernel.
• Various documentation improvements. (many of them from Nicholas Tung)
• Move PyCUDA’s compiler cache to the system temporary directory, rather than the users home directory.

Version 0.91

• Add support for compiling on CUDA 1.1. Added version query pycuda.driver.get_version(). Updated documentation to show 2.0-only functionality.
• Support for Windows and MacOS X, in addition to Linux. (Gert Wohlgemuth, Cosmin Stejerean, Znah on the Nvidia forums, and David Gadling)
• Support more arithmetic operators on pycuda.gpuarray.GPUArray. (Gert Wohlgemuth)
• Add pycuda.gpuarray.arange(). (Gert Wohlgemuth)
• Add pycuda.curandom. (Gert Wohlgemuth)
• Add pycuda.cumath. (Gert Wohlgemuth)
• Add pycuda.autoinit.
• Add pycuda.tools.
• Add pycuda.tools.DeviceData and pycuda.tools.OccupancyRecord.
• pycuda.gpuarray.GPUArray parallelizes properly on GTX200-generation devices.
• Make pycuda.driver.Function resource usage available to the program. (See, e.g. pycuda.driver.Function.registers.)
• Cache kernels compiled by pycuda.compiler.SourceModule. (Tom Annau)
• Allow for faster, prepared kernel invocation. See pycuda.driver.Function.prepare().
• Added memory pools, at pycuda.tools.DeviceMemoryPool as experimental, undocumented functionality. For some workloads, this can cure the slowness of pycuda.driver.mem_alloc().
• Fix the memset family of functions.
• Improve Error Reporting.
• Add order parameter to pycuda.driver.matrix_to_array() and pycuda.driver.make_multichannel_2d_array().

Acknowledgments

• Gert Wohlgemuth ported PyCUDA to MacOS X and contributed large parts of pycuda.gpuarray.GPUArray.
• Alexander Mordvintsev contributed fixes for Windows XP.
• Cosmin Stejerean provided multiple patches for PyCUDA’s build system.
• Tom Annau contributed an alternative SourceModule compiler cache as well as Windows build insight.
• Nicholas Tung improved PyCUDA’s documentation.
• Jozef Vesely contributed a massively improved random number generator derived from the RSA Data Security, Inc. MD5 Message Digest Algorithm.
• Chris Heuser provided a test cases for multi-threaded PyCUDA.
• The reduction templating is based on code by Mark Harris at Nvidia.
• Andrew Wagner provided a test case and contributed the port of the convolution example. The original convolution code is based on an example provided by Nvidia.
• Hendrik Riedmann contributed the matrix transpose and list selection examples.
• Peter Berrington contributed a working example for CUDA-OpenGL interoperability.

Licensing

PyCUDA is licensed to you under the MIT/X Consortium license:

Copyright (c) 2009 Andreas Klöckner and Contributors.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Frequently Asked Questions

The FAQ is now maintained collaboratively in the PyCUDA Wiki.