Skip to content

Releases: seeder-research/uMagNUS

uMagNUS (v2.3.0)

Choose a tag to compare

@xfong xfong released this 22 Aug 14:03

Update release with the following changes:

  • Fixed issue with make process
  • Fixed a bug with initialization in the GUI
  • Updated the reduce kernels to squeeze out some speed

uMagNUS (v2.2.2)

Choose a tag to compare

@xfong xfong released this 12 May 05:33

Improvements to kernel launches:

  • Sizes of work-groups and total number of work-items are determined based on GPU parameters
  • Sizes of work-groups for reduction kernels are reduced to improve GPU occupancy

uMagNUS (v2.2.1)

Choose a tag to compare

@xfong xfong released this 14 Apr 05:06

Feature updates:

  • added RemoveCustomEnergies()
  • Dockerfile added for portable deployment in HPC

uMagNUS (v2.2.0)

Choose a tag to compare

@xfong xfong released this 09 Apr 16:02

Feature release:

  • User may define custom torque terms (similar to custom field terms) to be included in the LLG calculation using AddTorqueTerm()
  • Added CustomQuantity() that takes user-defined *data.Slice as input to create a Quantity
  • Added CustomTorque() that takes a Quantity as a user-defined torque term in the LLG calculation

uMagNUS (v2.1.2)

Choose a tag to compare

@xfong xfong released this 07 Apr 09:18

Minor update to add documentation.
Binaries in this release were built using Golang 1.17.8

uMagNUS (v2.1.1)

Choose a tag to compare

@xfong xfong released this 30 Mar 13:03

Updated build process

  • added a stub library for the OpenCL API base on the C headers released by Khronos Group (v2022.01.04). Build of Golang bindings to OpenCL will use the included stubs but at runtime, uMagNUS will search for the OpenCL library installed in the system
  • Tested binaries were built using Golang 1.17.7 on several platforms:
    • msys2 on Windows 10 and Windows 11 (32-bit and 64-bit binaries were tested to run on Nvidia MX150 and Intel UHD 630)
    • CentOS 7 (used devtoolset-9 for build, tested on Nvidia K80 using 470.xx driver series with known errors for 32-bit binary)
    • Ubuntu 18.04 (32-bit and 64-bit binaries tested to run on Intel HD 630)
    • Ubuntu 20.04 (32-bit and 64-bit binaries tested to run on Nvidia P100, P2000 and RTX 2080 SUPER with 510.xx driver series)

uMagNUS (v2.1.0)

Choose a tag to compare

@xfong xfong released this 28 Mar 10:27

Several significant changes to the build process were implemented:

  • uMagNUS-clCompiler and uMagNUS-kernelLoader were created to help developers compile and link their custom OpenCL kernels using the OpenCL platform compiler to check for compile and link errors
  • ocl2go and ocl2go64 in opencl and opencl64 packages have been merged
  • Makefiles were updated to simplify the build process for users
  • Offline compilation of kernels into umagnus and umagnus64 libraries are enabled
  • During initialization, opencl and opencl64 packages search for the umagnus and umagnus64 libraries, respectively, in the directories defined in the user environment variables (LD_LIBRARY_PATH in Linux, PATH in windows). If found, the precompiled kernels are used to run uMagNUS and uMagNUS64. If not found, the kernels will be compiled from the source code hard coded into uMagNUS and
    uMagNUS64 when they were built

uMagNUS (v2.0.1)

Choose a tag to compare

@xfong xfong released this 02 Mar 04:52

Urgent bug fix:

  • Fixes an issue where uMagNUS (32-bit and 64-bit binaries) is unable to query device and platform information in Windows

uMagNUS (v2.0.0)

Choose a tag to compare

@xfong xfong released this 28 Feb 12:47

Full feature release

New features:

  • uMagNUS (32-bit backend) and uMagNUS64 (64-bit backend) binaries
  • ESDIRK implicit solvers for initial value problems

Fixes:

  • Removed MTGP psudo-RNG
  • Updated VkFFT to v1.2.21
  • GPU selection via command line argument
    • Comma separated list to select GPUs used for execution

uMagNUS (v1.3.1)

Choose a tag to compare

@xfong xfong released this 15 Dec 02:43

Patch release with the following fixes:

  • fixed nodemagspins
  • increased number of cells in rk4temperature to reduce variance in error