Skip to content

Latest commit

 

History

History
401 lines (259 loc) · 8.87 KB

CHANGELOG.md

File metadata and controls

401 lines (259 loc) · 8.87 KB
Raw

Changelog for HIPIFY

Documentation for HIPIFY is available at https://rocmdocs.amd.com/projects/HIPIFY/en/latest/.

HIPIFY for ROCm 6.3.1

Added

  • CUDA 12.6.2 support
  • cuDNN 9.5.1 support
  • LLVM 19.1.3 support
  • Full hipBLAS 64-bit APIs support
  • Full rocBLAS 64-bit APIs support

Resolved issues

  • Added missing support for device intrinsics and built-ins: __all_sync, __any_sync, __ballot_sync, __activemask, __match_any_sync, __match_all_sync, __shfl_sync, __shfl_up_sync, __shfl_down_sync, and __shfl_xor_sync

HIPIFY for ROCm 6.3.0

Added

  • CUDA 12.6.1 support
  • cuDNN 9.5.0 support
  • LLVM 19.1.1 support
  • rocBLAS 64-bit APIs support
  • Initial support for direct hipification of cuDNN into MIOpen under the --roc option
  • Initial support for direct hipification of cuRAND into rocRAND under the --roc option
  • [#1650] Added a filtering ability for the supplementary hipification scripts

Resolved issues

  • Correct roc header files support

Known issues

  • [#1617] Support for fp8 data types

HIPIFY for ROCm 6.2.4

Added

  • cuDNN 9.3.0 support

Resolved issues

  • Removed some post HIP 6.2 APIs from support
  • Added hipification support for HIP functions hipSetValidDevices, hipMemcpy2DArrayToArray, hipMemcpyAtoA, hipMemcpyAtoD, hipMemcpyAtoA, hipMemcpyAtoHAsync, and hipMemcpyHtoAAsync
  • Fixed an issue with Skipped some replacements when hipification didn't occur at all

HIPIFY for ROCm 6.2.1

Added

  • CUDA 12.5.1 support
  • cuDNN 9.2.1 support
  • LLVM 18.1.8 support
  • hipBLAS 64-bit APIs support
  • Support for Math Constants math_constants.h

HIPIFY for ROCm 6.2.0

Added

  • CUDA 12.4.1 support
  • cuDNN 9.1.1 support
  • LLVM 18.1.6 support
  • Full hipBLASLt support

Resolved issues

  • Apply reinterpret_cast for an explicit conversion between pointer-to-function and pointer-to-object; affected functions: hipFuncGetAttributes, hipFuncSetAttribute, hipFuncSetCacheConfig, hipFuncSetSharedMemConfig, hipLaunchKernel, and hipLaunchCooperativeKernel

HIPIFY for ROCm 6.1.2

Added

  • cuDNN 9.0.0 support
  • LLVM 18.1.2 support
  • New options:
    • --clang-resource-directory to specify the clang resource path - the path to the parent folder for the include folder that contains __clang_cuda_runtime_wrapper.h and other header files used during the hipification process

Resolved issues

  • Clang resource files used during hipification are now searchable and also can be specified by the --clang-resource-directory option

HIPIFY for ROCm 6.1.0

Added

  • CUDA 12.3.2 support
  • cuDNN 8.9.7 support
  • LLVM 17.0.6 support
  • Full hipSOLVER support
  • Full rocSPARSE support
  • New options:
    • --amap to hipify as much as possible, ignoring --default-preprocessor behavior

Resolved issues

  • Code blocks skipped by the Preprocessor are not hipified anymore under the --default-preprocessor option

HIPIFY for ROCm 6.0.2

Resolved issues

  • Use the new locations of header files of some HIP and ROCm libraries (hipRAND, hipFFT, rocSOLVER)

HIPIFY for ROCm 6.0.0

Added

  • CUDA 12.2.2 support
  • cuDNN 8.9.5 support
  • LLVM 17.0.3 support
  • Improved support for Windows and Visual Studio 2019 and 2022
  • More rocSPARSE support
  • ABI changes are shown in the 'C' ('Changed') column for CUDA, HIP, and ROC API

Known issues

  • [#837] Added a new function to call transformation type "additional non-const arg"
  • [#1014] Added a new function to call transformation type "replace argument with a const"

HIPIFY for ROCm 5.7.0

Added

  • CUDA 12.2.0 support
  • cuDNN 8.9.2 support
  • LLVM 16.0.6 support
  • Initial rocSPARSE support
  • Initial CUDA2ROC documentation generation for rocBLAS, rocSPARSE, and MIOpen:
    • In separate files: hipify-clang --md --doc-format=full --doc-roc=separate
    • In one file: hipify-clang --md --doc-format=full --doc-roc=joint
  • New options:
    • --use-hip-data-types (Use hipDataType instead of hipblasDatatype_t or rocblas_datatype)
    • --doc-roc=\<value\> (ROC documentation generation: skip (default), separate, and joint; the --md or --csv option must be included)

Known issues

  • [#822] Added a new function to call transformation type "additional const by value arg"
  • [#830] Added a new function to call transformation type "move arg from place X to place Y"

HIPIFY for ROCm 5.6.0

Added

  • CUDA 12.1.0 support
  • cuDNN 8.8.1 support
  • LLVM 16.0.0 support
  • New options:
    • --default-preprocessor (synonymous with --skip-excluded-preprocessor-conditional-blocks)
    • --no-undocumented-features
    • --no-warnings-on-undocumented-features
    • --versions

Resolved issues

  • Accessing half2 struct members (undocumented feature)
  • Retargeted INSTALL to the bin subfolder

HIPIFY for ROCm 5.5.0

Added

  • Partial CUDA 12.0.0 support
  • cuDNN 8.7.0 support
  • Initial MIOpem support
  • cuBLAS 64-bit API (CUDA 12.0) initial support
  • rocBLAS and MIOpen synthetic tests
  • LLVM 15.0.7 support

Changed

  • Synthetic unit tests for cuBLAS2rocBLAS and cuDNN2MIOpen

HIPIFY for ROCm 5.4.1

Added

  • CUDA 11.8.0 support
  • cuDNN 8.6.0 support
  • Device types support
  • LLVM 15.0.4 support

Resolved issues

  • Removed RPATH definitions (Linux)

HIPIFY for ROCm 5.4.0

Added

  • hipRTC support
  • Error handling API support
  • hipBLAS synthetic tests
  • LLVM 15.0.0 support

HIPIFY for ROCm 5.3.0

Added

  • CUDA 11.7.0 support
  • cuDNN 8.4.1 support
  • CUB initial support
  • More synthetic tests
  • New options:
    • --hip-kernel-execution-syntax

Changed

  • LLVM 3.8.0 is out of support
  • HIPIFY-specific options support in unit testing

Resolved issues

  • Patches for LLVM 14.0.x (Windows only)
  • Add GNUInstallDirs for CMake on Linux

HIPIFY for ROCm 5.2.0

Added

  • CUDA 11.6.0 support
  • cuDNN 8.3.3 support
  • LLVM 14.0.0 support

HIPIFY for ROCm 5.1.0

Added

  • CUDA 11.5.1 support
  • cuDNN 8.3.2 support

Resolved issues

  • Hipification of cuOccupancyMaxPotentialBlockSize and cuOccupancyMaxPotentialBlockSizeWithFlags

HIPIFY for ROCm 5.0.0

Added

  • CUDA 11.4.2 support
  • cuDNN 8.3.2 support
  • Initial hipRTC support
  • GNU C/C++ 11.2 support
  • Visual Studio 2022 support
  • LLVM 13.0.0 support
  • New options:
    • --experimental
    • --cuda-kernel-execution-syntax

Changed

  • Support for different formats of locally generated documentation
  • Experimentally supported APIs

Resolved issues

  • Packaging for Debian and RPM Linux distributions
  • Undo argument typecasting for four driver API functions (cuStreamWaitValue32, cuStreamWaitValue64, cuStreamWriteValue32, and cuStreamWriteValue64) because the arguments in the corresponding HIP functions are now uint

HIPIFY for ROCm 4.5.0

Added

  • cuDNN 8.2.4 support
  • Initial graph API support
  • GNU C/C++ 11.1 support
  • LLVM 12.0.1 support

Changed

  • Synthetic unit tests
  • -std=c++14 by default

Resolved issues

  • Abandoned HIP_DYNAMIC_SHARED

HIPIFY for ROCm 4.3.0

Added

  • CUDA 11.3.0 support
  • cuDNN 8.2.0 support
  • LLVM 12.0.0 support

Resolved issues

  • Added missing type casting arguments for cuStreamWaitValue32(64) and cuStreamWriteValue32(64)

HIPIFY for ROCm 4.2.0

Added

  • CUDA 11.2.2 support
  • cuDNN 8.1.1 support
  • Initial device API support
  • LLVM 11.1.0 support
  • New options:
    • --doc-format=<value>, with full (default), strict, and compact options

Changed

  • Tests on kernel launch syntax

HIPIFY for ROCm 4.1.0

Added

  • CUDA 11.2.0 support
  • Stream-ordered memory API support
  • cuDNN 8.1.1 support
  • LLVM 11.0.1 support

Changed

  • Initial support for API versioning

Resolved issues

  • Patches for LLVM 10.0.x and 11.0.0 (Windows and Linux)

HIPIFY for ROCm 4.0.0

Added

  • LLVM 11.0.0 support

HIPIFY for ROCm 3.10.0

Changed

  • Revised CUDA and HIP API and data type versioning
  • Revised and removed deprecated CUDA and HIP APIs and data types

HIPIFY for ROCm 3.9.0

Added

  • CUDA 11.0.1 support

  • CUDA2HIP documentation generation in Markdown and CSV formats

  • Versioning support for CUDA and HIP APIs

  • New options:

    • --md (generate Markdown documentation)
    • --csv (generate CSV documentation)

    Changed

  • Improved hipify-perl generation

HIPIFY for ROCm 3.8.0

Added

  • cuDNN 8.0.2 support
  • compile_commands.json support (-p <build-path>)

Changed

  • Improved hipify-perl generation

HIPIFY for ROCm 3.7.0

Added

  • CUDA 11.0.0 support
  • Linux packaging
  • LLVM 10.0.1 support

HIPIFY for ROCm 3.6.0

Added

  • deprecated flag for all corresponding CUDA and HIP APIs

Changed

  • Added warning for all deprecated APIs

HIPIFY for ROCm 3.5.0

Added

  • CUDA 10.2.0 support
  • cuDNN 7.6.5 support
  • LLVM 10.0.0 support

Changed

  • hipify-clang and clang options separator (--) support