Release Notes 2010 and Earlier

From Jacket Wiki

Jump to: navigation, search
Back to Release Notes

Contents

Version 1.6

  • Requires CUDA 3.1 drivers according to the following list:
  • Users are not required to install the CUDA toolkit. Jacket 1.6 was built with CUDA 3.1. On Linux and Mac, if you have another CUDA Toolkit installed, you might need to use LD_LIBRARY_PATH to give precedence to Jacket's CUDA libraries (see Jacket Installation).

12/14/2010 - v1.6.1

11/15/2010 - v1.6

The major goals of this release were introducing the Statistics Library and improving the stability of the memory subsystem to avoid "Out of memory" and "GPU failure" error messages. Build 9695.

Version 1.5

  • Requires CUDA 3.1 drivers according to the following list:
  • Users are not required to install the CUDA toolkit. Jacket 1.5 was built with CUDA 3.1. In Linux and Mac, Jacket's CUDA libraries must be given precedence (via LD_LIBRARY_PATH) relative to existing CUDA toolkit installations.

10/18/2010 - v1.5.1

  • Build 8873

Additions:

  • The following new functions are available:
    • IMFILTER - 2-D filtering of multidimensional images
    • RGB2GRAY - Convert RGB image or colormap to grayscale
    • MEDFILT2 - 2-D median filtering
    • EDGE - Find edges in grayscale image
    • INTERSECT - Find set intersection of two vectors
    • SORTKEYS - Sorts a column matrix based on input keys
    • PLANEROT - Givens plane rotation

Changes:

  • FFT and IFFT support bigger inputs
  • BWDIST is faster
  • SUM, MIN, MAX improvements:
  • GFOR expanded support for iterator arithmetic (e.g. A(:,i+2), A(:,ceil(i+.4)*2))

09/16/2010 - v1.5

  • Build 8250

Additions:

Changes:

Known Issues:

  • FFT may fail on systems with multiple GPUs of different compute capabilities. This issue will be fixed in CUDA 3.2 from NVIDIA.
  • Sporadic Out-of-memory errors may be encountered on GPUs with 256MB or smaller space.
  • FIND is not supported for floating-point matrices/vectors having greater than 2^24 elements.
  • fft_example and blas_example run into GPU failures on CUDA 1.0 cards when run multiple times.
  • conv_example, interactive_ocean_example, shaded_ocean_example run into "Internal CUFFT errors" when run on the non-default GPU on machines with multiple GPUs. This will be resolved with the CUDA 3.2 release.
  • There are known issues specific to graphics on OSX. This does not effect the Jacket core on the OSX platform. We hope to resolve these soon.

Version 1.4

07/27/2010 - v1.4.1

  • Build 6737
  • Requires CUDA 3.1 drivers
  • Users are not required to install the CUDA toolkit. Jacket 1.4.1 was built with CUDA 3.1. In Linux and Mac, Jacket's CUDA libraries must be given precedence (via LD_LIBRARY_PATH) relative to existing CUDA toolkit installations.

Additions:

  • Jacket support for 64-bit MAC OSX (versions 10.6.3 and higher)
  • GPROFILE beta version of GPU profiler for MATLAB (console version)
  • GLAUNCH beta version of inline MATLAB CUDA kernel construction (write CUDA code within M files)
  • GFOR capabilities enabled for CUMSUM and CUMPROD

Changes:

  • MTIMES performance improved for smaller matrices
  • Jacket SDK no longer locks mex files by default, allowing recompiles without restarting MATLAB
  • JIT inside GFOR bug fix
  • Running FFT and clearing all data multiple times no longer produces Jacket Errors
  • Minor fixes to JacketDemo
  • Graphics Library - Middle mouse button now correctly recognized under Windows
  • INTERP2 fixed occurrence of erroneous extrap values (NaN) on some non-square input/output combinations

Known Issues:

  • Jacket DLA not yet available for Mac OSX 64-bit
  • CUMPROD gives incorrect values for certain sizes on Fermi-based GPUs

07/12/2010 - v1.4

  • Build 6121
  • Requires CUDA 3.1 drivers
  • Users are not required to install the CUDA toolkit. Jacket 1.4 was built with CUDA 3.1. In Linux and Mac, Jacket's CUDA libraries must be given precedence (via LD_LIBRARY_PATH) relative to existing CUDA toolkit installations.

Additions:

  • Added support for the NVIDIA Fermi architecture (GTX400 and Tesla C2000 series)
  • Dramatically improved the performance of Jacket's JIT (Just-In-Time) compilation technology
    • Operations involving random scalar constants do not incur a recompile
    • Removed dependencies on MINGW and NVCC
  • Logical indexing now supported for SUBSREF and SUBSASGN, e.g. B = A(A > x)
  • MTIMES supports mixed types, no longer uses CUBLAS, and achieves better performance than CUBLAS
  • SUM, MIN, MAX, ANY, ALL now supported over any number of columns, rows, or dimensions
  • MIN, MAX indexed output now supported for complex single and complex double inputs
  • SUM, MIN, MAX over columns is greatly accelerated; vectors accelerated too
  • FIND performance improvements
  • CONVN, BLKDIAG, DOT performance improvements
  • CUMSUM now supported for matrices also
  • SORT, CONVN now supported in double-precision
  • HESS(A) and [P,H] = HESS(A) now supported (see Jacket DLA)
  • LEGENDRE now supported
  • Expanded GFOR support for:
  • PCG now supported, this is a system solver that uses the Preconditioned Conjugate Gradient Method for dense matrices
  • Image Processing Library now available. Direct access to the NVIDIA Performance Primitives (NPP) enabling new image processing functionality such as ERODE and DILATE.

Changes:

  • Memory subsystem is now more stable and incurs less fragmentation resulting in fewer "Out of memory" errors
  • MLDIVIDE fixed behavior in case of singular inputs
  • FFT no longer gives incorrect values or CUFFT errors for certain sizes
  • HIST now uses the same binning method and therefore, similar results as MATLAB
  • SQRT on negative real numbers correctly outputs complex data
  • SORT on rows more consistent with MATLAB behavior
  • GRADIENT fixed erroneous behavior in some cases
  • INTERP1 no longer segfaults and is partially supported

Known Issues:

  • FULL results in a GPU failure on certain systems
  • BSXFUN may perform slower in certain situations
  • FFT fails for sizes less than 32 elements on some single-precision cards

Version 1.3

04/06/2010 - v1.3

  • Build 3889
  • Requires 190.xx or higher Drivers
  • This release packages CUDA 2.3 toolkit; users are not required to install the CUDA toolkit. Existing toolkits are fine as long as Jacket's CUDA toolkit is given precedence.

Additions:

  • Added a new PCA example to demonstrate Jacket's linear algebra functionality available with JacketDLA and based on CULA (www.culatools.com)

Linear algebra additions:

Changes:

  • GFOR support is now present for:
  • Enhanced performance for:
  • GDOUBLE support for Linear Algebra functions:
  • Fixed issue where using GRAND and GRANDN repeatedly would cause loss of randomness after a point.
  • CONV2 (separable convolution)
  • GINFO prints out more detailed information about System Configuration that now also includes CUDA driver version and License features.
  • ALL and ANY support GSINGLE and GDOUBLE complex inputs.
  • CROSS supported with GDOUBLE inputs.
  • The Windows installer has been revamped, with options to manage licenses and download license files automatically, as well as download of CUDA driver for your system.

Known Issues:

  • Although Jacket does not contain out-of-the-box support for SuSE Linux Enterprise Server, it is possible to run Jacket with SLES with a few modifications. Please contact support@accelereyes.com for more information on Jacket Support for SLES.
  • Some linear algebra operations, particularly MLDIVIDE, DET, and INV, have variations in precision.
  • For certain cases (primarily Linux 64 systems), CONV2 returns a GPU failure or incorrect values.
  • KRON is not supported for use in a GFOR loop yet, but does not throw an explicit error message.
  • SQRT of a negative real matrix does not produce the a complex number correctly and does not display and error message

Version 1.2

01/04/2010 - v1.2.2

  • Build 3170
  • Requires CUDA 2.3 Drivers and Toolkit
  • This is a development release; please note the following issues (specific to this release) before installing:
    • This development version requires Linux users to set LD_LIBRARY_PATH in the following order: <PATH TO JACKET>/cula/lib:<PATH TO CUDA>/lib (Replace lib by lib64 if using 64-bit Linux)
    • This version of Jacket has an issue with the memory subsystem that will be fixed in the next version. This may manifest itself in the GPU issuing out of memory errors.

Additions:

Changes:

  • GONES, GZEROS, GEYE can now return double-precision on double-precision cards. This is possible by setting a flag as follows: gpu_entry(10,true) to enable default-gdouble, and gpu_entry(10,false) to disable it.
  • GONES, GZEROS, GEYE, GRAND and GRANDN (without arguments) now return a 1x1 number.
  • SQRT and POWER now supported for complex inputs
  • Optimizations made to subscripted referencing and assignments
  • The divide operation now preserves class: if A is a GDOUBLE, 1/A has the same class
  • Fixed issue with incorrect cached kernel being used in multiplication
  • Fixed issue where MLDIVIDE failed with CPU arguments
  • Fixed failures with complex GDOUBLE to complex GSINGLE conversion
  • Fixed failures with MOD, BLKDIAG for some inputs
  • Fixed a failure for CAT(1,A,B) for 2D inputs (order of 1000x500).

Known issues:

  • In addition to LD_LIBRARY_PATH, MacOSX users need to set $DYLD_LIBRARY_PATH to the following:<PATH TO JACKET>/cula/lib/ in environment.plist.
  • If you are running Jacket in Linux on 'screen', note that problems may occur with commands such as GINFO and GACTIVATE. This is because Jacket sets the environment variable KMP_DUPLICATE_LIB_OK to TRUE to enable the SVD algorithm to run.
  • BSXFUN, COV, STD, VAR, CORRCOEF currently give incorrect results for some data (Please see release notes for 1.2). This is slated to be fixed soon.

Versions from 2009 and earlier

For older versions, visit Release Notes 2009 and Earlier.

Retrieved from "http://wiki.accelereyes.com/wiki/index.php/Release_Notes_2010_and_Earlier"
Views
Personal tools