Releases: ROCm/hipTensor
hipTensor 2.2.0 for ROCm 7.2.0
Added
- Added software-managed plan cache support.
- Added
hiptensorHandleWritePlanCacheToFileto write the plan cache of a hipTensor handle to a file. - Added
hiptensorHandleReadPlanCacheFromFileto read a plan cache from a file into a hipTensor handle. - Added
simple_contraction_plan_cacheto demonstrate plan cache usages. - Added
plan_cache_testto test the plan cache across various tensor ranks. - Added C API headers to enable compatibility with C programs.
- Added a CMake function to allow projects to query architecture support.
- Added an option to configure the memory layout for tests and benchmarks.
Changed
- Updated C++ standard from C++17 to C++20.
- Include files
hiptensor/hiptensor.hppandhiptensor/hiptensor_types.hppare now deprecated. Usehiptensor/hiptensor.handhiptensor/hiptensor_types.hinstead. - Converted include guards from #ifndef/#define/#endif to #pragma once.
Resolved issues
- Removed large tensor sizes causing problem in benchmarks.
hipTensor 2.0.0 for ROCm 7.1.1
hipTensor code for ROCm 7.1.1 did not change. The library was rebuilt for the updated ROCm 7.1.1 stack.
hipTensor 2.0.0 for ROCm 7.1.0
hipTensor code for ROCm 7.1.0 did not change. The library was rebuilt for the updated ROCm 7.1.0 stack.
hipTensor 2.0.0 for ROCm 7.0.2
hipTensor code for ROCm 7.0.2 did not change. The library was rebuilt for the updated ROCm 7.0.2 stack.
hipTensor 2.0.0 for ROCm 7.0.1
hipTensor code for ROCm 7.0.1 did not change. The library was rebuilt for the updated ROCm 7.0.1 stack.
hipTensor 2.0.0 for ROCm 7.0.0
Added
- Added element-wise binary operation support.
- Added element-wise trinary operation support.
- Added support for new GPU target gfx950.
- Added dynamic unary and binary operator support for element-wise operations and permutation.
- Added a CMake check for
f8datatype availability. - Added
hiptensorDestroyOperationDescriptorto free all resources related to the provided descriptor. - Added
hiptensorOperationDescriptorSetAttributeto set attribute of ahiptensorOperationDescriptor_tobject. - Added
hiptensorOperationDescriptorGetAttributeto retrieve an attribute of the providedhiptensorOperationDescriptor_tobject. - Added
hiptensorCreatePlanPreferenceto allocate thehiptensorPlanPreference_tand enabled users to limit the applicable kernels for a given plan or operation. - Added
hiptensorDestroyPlanPreferenceto free all resources related to the provided preference. - Added
hiptensorPlanPreferenceSetAttributeto set attribute of ahiptensorPlanPreference_tobject. - Added
hiptensorPlanGetAttributeto retrieve information about an already-created plan. - Added
hiptensorEstimateWorkspaceSizeto determine the required workspaceSize for the given operation. - Added
hiptensorCreatePlanto allocate ahiptensorPlan_tobject, select an appropriate kernel for a given operation and prepare a plan that encodes the execution. - Added
hiptensorDestroyPlanto free all resources related to the provided plan.
Changed
- Removed architecture support for gfx940 and gfx941.
- Generalized opaque buffer now for any descriptor.
- Replaced
hipDataTypewithhiptensorDataType_tfor all supported types, for example,HIP_R_32FtoHIPTENSOR_R_32F. - Replaced
hiptensorComputeType_twithhiptensorComputeDescriptor_tfor all supported types. - Replaced
hiptensorInitTensorDescriptorwithhiptensorCreateTensorDescriptor. - Changed handle type and API usage from
*handletohandle. - Replaced
hiptensorContractionDescriptor_twithhipTensorOperationDescriptor_t. - Replaced
hiptensorInitContractionDescriptorwithhiptensorCreateContraction. - Replaced
hiptensorContractionFind_twithhiptensorPlanPreference_t. - Replaced
hiptensorInitContractionFindwithhiptensorCreatePlanPreference. - Replaced
hiptensorContractionGetWorkspaceSizewithhiptensorEstimateWorkspaceSize. - Replaced
HIPTENSOR_WORKSPACE_RECOMMENDEDwithHIPTENSOR_WORKSPACE_DEFAULT. - Replaced
hiptensorContractionPlan_twithhiptensorPlan_t. - Replaced
hiptensorInitContractionPlanwithhiptensorCreatePlan. - Replaced
hiptensorContractionwithhiptensorContract. - Replaced
hiptensorPermutationwithhiptensorPermute. - Replaced
hiptensorReductionwithhiptensorReduce. - Replaced
hiptensorElementwiseBinarywithhiptensorElementwiseBinaryExecute. - Replaced
hiptensorElementwiseTrinarywithhiptensorElementwiseTrinaryExecute. - Removed function
hiptensorReductionGetWorkspaceSize.
hipTensor 1.5.0 for ROCm 6.4.4
hipTensor code for ROCm 6.4.4 did not change. The library was rebuilt for the updated ROCm 6.4.4 stack.
hipTensor 1.5.0 for ROCm 6.4.3
hipTensor code for ROCm 6.4.3 did not change. The library was rebuilt for the updated ROCm 6.4.3 stack.
hipTensor 1.5.0 for ROCm 6.4.2
hipTensor code for ROCm 6.4.2 did not change. The library was rebuilt for the updated ROCm 6.4.2 stack.
hipTensor 1.5.0 for ROCm 6.4.1
hipTensor code for ROCm 6.4.1 did not change. The library was rebuilt for the updated ROCm 6.4.1 stack.