Skip to content

MIOpen v1.6.0
Compare
Choose a tag to compare
@daniellowell daniellowell released this 19 Nov 03:16
· 2080 commits to master since this release
1.6.0
ffedda8
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Notes:

  • Training in fp16 (half precision) including mixed-precision is now fully supported
  • Batch Normalization in fp16 (half precision) including mixed-precision are now available
  • Performance improvements for 3x3 and 1x1 single-precision convolutions
  • Layer fusions for BatchNorm+Activation are now available
  • Layer fusions with convolutions now support varying strides and padding configurations

Changes:

  • rocBLAS is now used as the default BLAS library for the HIP backend (minimum version 14.3.0)
  • Fixed various bugs in convolution kernels
  • Fixed issues with bad references in layer fusion
  • Fixed gfx803 assembily issues
  • Added support fp16 Winograd convolutions
  • Added support for fp16 pooling
  • Improved error reporting for convolutions and layer fusions
  • Improved documentation

Known Issues:

  • RNNs do not support fp16
  • OpenCL backend does not have full fp16 support
  • Layer fusions for convolution 1x1 fp16 are not supported
Assets 6
Loading