AMD has announced the release of ROCm 6.3.0, the latest version of its advanced generative AI and large language model server. The update features the rocJPEG library, a high-performance JPEG decoding SDK designed for AMD GPUs. The ROCm Compute Profiler and ROCm Systems Profiler have undergone a renaming to align with their updated focus within the ROCm software stack. The SHARK AI toolkit for high-speed inferencing and serving has been introduced, featuring support for the Stable Diffusion XL model on AMD Instinct:tm:MI300 devices.

ROCm 6.3.0 introduces support for PyTorch 2.4, enabling the Transformer Engine to optionally choose a flexible and optimized Attention solution for AMD GPUs. The hipStreamLegacy API has been integrated, serving as the equivalent to NVIDIA cudaStreamLegacy. The ROCm Offline Installer Creator 6.3 now includes a feature that allows for the uninstallation of the previous version of ROCm on a non-connected target system prior to the installation of a new version.

ROCm 6.3.0 Release

ROCm 6.3.0 release notes

The release notes provide a summary of notable changes since the previous ROCm release.

• Release highlights

• Operating system and hardware support changes

• ROCm components versioning

• Detailed component changes

• ROCm known issues

• ROCm resolved issues

• ROCm upcoming changes

If you’re using Radeon PRO or Radeon GPUs in a workstation setting with a display connected, continue to use ROCm 6.2.3. See the [Use ROCm on Radeon GPUs](https://rocm.docs.amd.com/projects/radeon/en/latest/index.html) documentation to verify compatibility and system requirements.

Release highlights

The following are notable new features and improvements in ROCm 6.3.0. For changes to individual components, see
Detailed component changes.

rocJPEG added

ROCm 6.3.0 introduces the rocJPEG library to the ROCm software stack. rocJPEG is a high performance
JPEG decode SDK for AMD GPUs. For more information, see the  rocJPEG documentation.

ROCm Compute Profiler and ROCm Systems Profiler

These ROCm components have been renamed to reflect their new direction as part of the ROCm software
stack.

• ROCm Compute Profiler, formerly Omniperf. For more information, see the  ROCm Compute Profiler
  documentation and
  https://github.com/ROCm/rocprofiler-compute on GitHub.

• ROCm Systems Profiler, formerly Omnitrace. For more information, see the  ROCm Systems Profiler
  documentation and
  https://github.com/ROCm/rocprofiler-systems on GitHub.
  For future compatibility, the Omnitrace project is available at  https://github.com/ROCm/omnitrace.
  See the  Omnitrace documentation.

  Update any references to the old binary names `omniperf` and `omnitrace` to
  ensure compatibility with the new `rocprof-compute` and `rocprof-sys-*` binaries.
  This might include updating environment variables, commands, and paths as
  needed to avoid disruptions to your profiling or tracing workflows.
  
  See [ROCm Compute Profiler 3.0.0](#rocm-compute-profiler-3-0-0) and [ROCm Systems
  Profiler 0.1.0](#rocm-systems-profiler-0-1-0).
  

SHARK AI toolkit for high-speed inferencing and serving introduced

SHARK is an open-source toolkit for high-performance serving of popular generative AI and large
language models. In its initial release, SHARK contains the  Shortfin high-performance serving
engine, which is the SHARK inferencing
library that includes example server applications for popular models.

This initial release includes support for serving the Stable Diffusion XL model on AMD Instinct
MI300 devices using ROCm. See the SHARK  release
page on GitHub to get started.

PyTorch 2.4 support added

ROCm 6.3.0 adds support for PyTorch 2.4. See the  Compatibility
matrixfor the complete list of PyTorch versions tested for compatibility with ROCm.

Flash Attention kernels in Triton and Composable Kernel (CK) added to Transformer Engine

Composable Kernel-based and Triton-based Flash Attention kernels have been integrated into
Transformer Engine via the ROCm Composable Kernel and AOTriton libraries. The
Transformer Engine can now optionally select a flexible and optimized Attention
solution for AMD GPUs. For more information, see  Fused Attention Backends on
ROCmon GitHub.

HIP compatibility

HIP now includes the hipStreamLegacy API. It's equivalent to NVIDIA cudaStreamLegacy. For more
information, see  Global enum and
definesin the HIP runtime API documentation.

Unload active amdgpu-dkms module without a system reboot

On Instinct MI200 and MI300 systems, you can now unload the active amdgpu-dkms modules, and reinstall
and reload newer modules without a system reboot. If the new dkms package includes newer firmware
components, the driver will first reset the device and then load newer firmware components.

ROCm Offline Installer Creator updates

The ROCm Offline Installer Creator 6.3 introduces a new feature to uninstall the previous version of
ROCm on the non-connected target system before installing a new version. This feature is only supported
on the Ubuntu distribution. See the  ROCm Offline Installer
Creatordocumentation for more information.

OpenCL ICD loader separated from ROCm

The OpenCL ICD loader is no longer delivered as part of ROCm, and must be installed separately
as part of the  ROCm installation
process. For Ubuntu and RHEL
installations, the required package is installed as part of the setup described in
Prerequisites.
In other supported Linux distributions like SUSE, the required package must be installed in separate steps, which are included in the installation instructions.

Because the OpenCL path is now separate from the ROCm installation for versioned and multi-version
installations, you must manually define the LD_LIBRARY_PATH to point to the ROCm
installation library as described in the  Post-installation
instructions.
If the LD_LIBRARY_PATH is not set as needed for versioned or multi-version installations, OpenCL
applications like clinfo will fail to run and return an error.

ROCT Thunk Interface integrated into ROCr runtime

The ROCT Thunk Interface package is now integrated into the ROCr runtime. As a result, the ROCT package
is no longer included as a separate package in the ROCm software stack.

ROCm documentation updates

ROCm documentation continues to be updated to provide clearer and more comprehensive guidance for a
wider variety of user needs and use cases.

• Documentation for Tensile is now available. Tensile is a library that creates
  benchmark-driven backend implementations for GEMMs, serving primarily as a
  backend component of rocBLAS. See the  Tensile
  documentation.

• New documentation has been added to explain the advantages of enabling the IOMMU in passthrough
  mode for Instinct accelerators and Radeon GPUs. See  Input-Output Memory Management
  Unit.

• The HIP documentation has been updated and includes the following new topics:

  • What is HIP?
  • HIP environment variables
  • Initializationand  error handling
  • Hardware features
  • Call stack
  • External resource interoperability

• The following HIP documentation topics have been updated:

  • HIP FAQ
  • Deprecated APIs
  • Performance guidelines

• The following HIP documentation topics have been reorganized to improve usability:

  • HIP documentation landing page
  • HIP runtime API reference topics
  • Programming guide

Operating system and hardware support changes

ROCm 6.3.0 adds support for the following operating system and kernel versions:

• Ubuntu 24.04.2 (kernel: 6.8 [GA], 6.11 [HWE])
• Ubuntu 22.04.5 (kernel: 5.15 [GA], 6.8 [HWE])
• RHEL 9.5 (kernel: 5.14.0)
• Oracle Linux 8.10 (kernel: 5.15.0)

See installation instructions at  ROCm installation for
Linux.

ROCm 6.3.0 marks the end of support (EoS) for:

• Ubuntu 24.04.1
• Ubuntu 22.04.4
• RHEL 9.3
• RHEL 8.9
• Oracle Linux 8.9

Hardware support remains unchanged in this release.

See the  Compatibility matrix for more information about operating system and hardware compatibility.

ROCm components

The following table lists the versions of ROCm components for ROCm 6.3.0, including any version
changes from 6.2.4 to 6.3.0. Click the component's updated version to go to a list of its changes.
Click {fab}github to go to the component's source code on GitHub.

Category	Group	Name	Version
Libraries	Machine learning and computer vision	Composable Kernel	1.1.0
		MIGraphX	2.11.0
		MIOpen	3.2.0 ⇒  3.3.0
		MIVisionX	3.0.0 ⇒  3.1.0
		rocAL	2.0.0 ⇒  2.1.0
		rocDecode	0.6.0 ⇒  0.8.0
		rocJPEG	0.6.0
		rocPyDecode	0.1.0 ⇒  0.2.0
		RPP	1.8.0 ⇒  1.9.1
	Communication	RCCL	2.20.5 ⇒  2.21.5
	Math	hipBLAS	2.2.0 ⇒  2.3.0
		hipBLASLt	0.8.0 ⇒  0.10.0
		hipFFT	1.0.16 ⇒  1.0.17
		hipfort	0.4.0 ⇒  0.5.0
		hipRAND	2.11.1 ⇒  2.11.0  *
		hipSOLVER	2.2.0 ⇒  2.3.0
		hipSPARSE	3.1.1 ⇒  3.1.2
		hipSPARSELt	0.2.1 ⇒  0.2.2
		rocALUTION	3.2.0 ⇒  3.2.1
		rocBLAS	4.2.4 ⇒  4.3.0
		rocFFT	1.0.30 ⇒  1.0.31
		rocRAND	3.1.1 ⇒  3.2.0
		rocSOLVER	3.26.2 ⇒  3.27.0
		rocSPARSE	3.2.1 ⇒  3.3.0
		rocWMMA	1.5.0 ⇒  1.6.0
		Tensile	4.41.0 ⇒  4.42.0
	Primitives	hipCUB	3.2.1 ⇒  3.3.0
		hipTensor	1.3.0 ⇒  1.4.0
		rocPRIM	3.2.2 ⇒  3.3.0
		rocThrust	3.1.1 ⇒  3.3.0
Tools	System management	AMD SMI	24.6.3 ⇒  24.7.1
		ROCm Data Center Tool	0.3.0 ⇒  0.3.0
		rocminfo	1.0.0
		ROCm SMI	7.3.0 ⇒  7.4.0
		ROCmValidationSuite	1.0.0 ⇒  1.1.0
			Performance	ROCm Bandwidth Test	1.4.0
				ROCm Compute Profiler	2.0.1 ⇒  3.0.0
ROCm Systems Profiler	1.11.2 ⇒  0.1.0
ROCProfiler	2.0.0 ⇒  2.0.0
ROCprofiler-SDK	0.4.0 ⇒  0.5.0
ROCTracer	4.1.0
	Development	HIPIFY	18.0.0 ⇒  18.0.0
		ROCdbgapi	0.76.0 ⇒  0.77.0
		ROCm CMake	0.14.0
		ROCm Debugger (ROCgdb)	14.2 ⇒  15.2
		ROCr Debug Agent	2.0.3
Compilers		HIPCC	1.1.1
		llvm-project	18.0.0 ⇒  18.0.0
Runtimes		HIP	6.2.4 ⇒  6.3.0
		ROCr Runtime	1.14.0

Detailed component changes

The following sections describe key changes to ROCm components.

AMD SMI (24.7.1)

Added

• Support for amd-smi metric --ecc & amd-smi metric --ecc-blocks on Guest VMs.

• Support for GPU metrics 1.6 to amdsmi_get_gpu_metrics_info()

• New violation status outputs and APIs: amdsmi_status_t amdsmi_get_violation_status(), amd-smi metric --throttle, and amd-smi monitor --violation. This feature is only available on MI300+ ASICs

• Ability to view XCP (Graphics Compute Partition) activity within amd-smi metric --usage. Partition-specific features are only available on MI300+ ASICs

• Added LC_PERF_OTHER_END_RECOVERY CLI output to amd-smi metric --pcie and updated amdsmi_get_pcie_info() to include this value. This feature is only available on MI300+ ASICs

• Ability to retrieve a set of GPUs that are nearest to a given device at a specific link type level

  • Added amdsmi_get_link_topology_nearest() function to amd-smi C and Python Libraries.

• More supported utilization count types to amdsmi_get_utilization_count()

• amd-smi set -L/--clk-limit ... command. This is equivalent to rocm-smi's --extremum command which sets sclk's or mclk's soft minimum or soft maximum clock frequency.

• Unittest functionality to test amdsmi API calls in Python

• GPU memory overdrive percentage to amd-smi metric -o

  • Added amdsmi_get_gpu_mem_overdrive_level() function to AMD SMI C and Python Libraries.

• Ability to retrieve connection type and P2P capabilities between two GPUs

  • Added amdsmi_topo_get_p2p_status() function to amd-smi C and Python Libraries.
  • Added retrieving P2P link capabilities to CLI amd-smi topology.

• New amdsmi_kfd_info_t type and added information under amd-smi list

• Subsystem device ID to amd-smi static --asic. There are no underlying changes to amdsmi_get_gpu_asic_info.

• Target_Graphics_Version to amd-smi static --asic and amdsmi_get_gpu_asic_info().

Changed

• Updated BDF commands to use KFD SYSFS for BDF: amdsmi_get_gpu_device_bdf(). This change aligns BDF output with ROCm SMI.

• Moved Python tests directory path install location.

  • /opt/<rocm-path>/share/amd_smi/pytest/.. to /opt/<rocm-path>/share/amd_smi/tests/python_unittest/..
  • Removed PyTest dependency. Python testing now depends on the unittest framework only.

• Changed the power parameter in amdsmi_get_energy_count() to energy_accumulator.

  • Changes propagate forwards into the Python interface as well. Backwards compatibility is maintained.

• Updated Partition APIs and struct information and added partition_id to amd-smi static --partition.

  • As part of an overhaul to partition information, some partition information will be made available in the amdsmi_accelerator_partition_profile_t.
  • This struct will be filled out by a new API, amdsmi_get_gpu_accelerator_partition_profile().
  • Future data from these APIs will eventually be added to amd-smi partition.

Removed

• amd-smi reset --compute-partition and ... --memory-partition and associated APIs

  • This change is part of the partition redesign. Reset functionality will be reintroduced in a later update.
  • Associated APIs include amdsmi_reset_gpu_compute_partition() and amdsmi_reset_gpu_memory_partition()

• Usage of _validate_positive is removed in parser and replaced with _positive_int and _not_negative_int as appropriate.

  • This will allow 0 to be a valid input for several options in setting CPUs where appropriate (for example, as a mode or NBIOID).

Optimized

• Adjusted ordering of gpu_metrics calls to ensure that pcie_bw values remain stable in amd-smi metric & amd-smi monitor.
  • With this change additional padding was added to PCIE_BW amd-smi monitor --pcie

Known issues

• See  AMD SMI manual build issue.

Resolved issues

• Improved Offline install process and lowered dependency for PyYAML.

• Fixed CPX not showing total number of logical GPUs.

• Fixed incorrect implementation of the Python API amdsmi_get_gpu_metrics_header_info().

• amdsmitst TestGpuMetricsRead now prints metric in correct units.

Upcoming changes

• Python API for amdsmi_get_energy_count() will deprecate the power field in a future ROCm release and use energy_accumulator field instead.

• New memory and compute partition APIs will be added in a future ROCm release.

  • These APIs will be updated to fully populate the CLI and allowing compute (accelerator) partitions to be set by profile ID.

  • One API will be provided, to reset both memory and compute (accelerator).

  • The following APIs will remain:

    amdsmi_status_t
    amdsmi_set_gpu_compute_partition(amdsmi_processor_handle processor_handle,
                                      amdsmi_compute_partition_type_t compute_partition);
    amdsmi_status_t
    amdsmi_get_gpu_compute_partition(amdsmi_processor_handle processor_handle,
                                      char *compute_partition, uint32_t len);
    amdsmi_status_t
    amdsmi_get_gpu_memory_partition(amdsmi_processor_handle processor_handle,
    
                                      char *memory_partition, uint32_t len);
    amdsmi_status_t
    amdsmi_set_gpu_memory_partition(amdsmi_processor_handle processor_handle,
                                      amdsmi_memory_partition_type_t memory_partition);

• amd-smi set --compute-partition "SPX/DPX/CPX..." will no longer be supported in a future ROCm release.

  • This is due to aligning with Host setups and providing more robust partition information through the APIs outlined above. Furthermore, new APIs which will be available on both BM/Host can set by profile ID.

• Added a preliminary amd-smi partition command.

  • The new partition command can display GPU information, including memory and accelerator partition information.
  • The command will be at full functionality once additional partition information from amdsmi_get_gpu_accelerator_partition_profile() has been implemented.

See the full [AMD SMI changelog](https://github.com/ROCm/amdsmi/blob/rocm-6.3.x/CHANGELOG.md) for more details and examples.

HIP (6.3.0)

Added

• New HIP APIs:
  • hipGraphExecGetFlags returns the flags on executable graph.
  • hipGraphNodeSetParams updates the parameters of a created node.
  • hipGraphExecNodeSetParams updates the parameters of a created node on an executable graph.
  • hipDrvGraphMemcpyNodeGetParams gets a memcpy node's parameters.
  • hipDrvGraphMemcpyNodeSetParams sets a memcpy node's parameters.
  • hipDrvGraphAddMemFreeNode creates a memory free node and adds it to a graph.
  • hipDrvGraphExecMemcpyNodeSetParams sets the parameters for a memcpy node in the given graphExec.
  • hipDrvGraphExecMemsetNodeSetParams sets the parameters for a memset node in the given graphExec.
  • hipExtHostAlloc preserves the functionality of hipHostMalloc.

Changed

• Un-deprecated HIP APIs:
  • hipHostAlloc
  • hipFreeHost

Optimized

• Disabled CPU wait in device synchronize to avoid idle time in applications such as Hugging Face models and PyTorch.
• Optimized multi-threaded dispatches to improve performance.
• Limited the software batch size to control the number of command submissions for runtime to handle efficiently.
• Optimizes HSA callback performance when a large number of events are recorded by multiple threads and submitted to multiple GPUs.

Resolved issues

• Soft hang in runtime wait event when run TensorFlow.
• Memory leak in the API hipGraphInstantiate when kernel is launched using hipExtLaunchKernelGGL with event.
• Memory leak when the API hipGraphAddMemAllocNode is called.
• The _sync() version of crosslane builtins such as shfl_sync(),
  __all_sync() and __any_sync(), continue to be hidden behind the
  preprocessor macro HIP_ENABLE_WARP_SYNC_BUILTINS, and will be enabled
  unconditionally in the next ROCm release.

Upcoming changes

• Deprecated HIP APIs:
  • hipHostMalloc to be replaced by hipExtHostAlloc.
  • hipHostFree to be replaced by hipFreeHost.

hipBLAS (2.3.0)

Added

• Level 3 functions have an additional ILP64 API for both C and Fortran (_64 name suffix) with int64_t function arguments

Changed

• amdclang is used as the default compiler instead of g++.
• Added a dependency on the hipblas-common package.

hipBLASLt (0.10.0)

Added

• Support for the V2 CPP extension API for backward compatibility
• Support for data type INT8 in with INT8 out
• Support for data type FP32/FP64 for gfx110x
• Extension API hipblaslt_ext::matmulIsTuned
• Output atol and rtol for hipblaslt-bench validation
• Output the bench command for the hipblaslt CPP ext API path if HIPBLASLT_LOG_MASK=32 is set
• Support odd sizes for FP8/BF8 GEMM

Changed

• Reorganized and added more sample code.
• Added a dependency with the hipblas-common package and removed the dependency with the hipblas package.

Optimized

• Support fused kernel for HIPBLASLT_MATMUL_DESC_AMAX_D_POINTER for the FP8/BF8 data type
• Improved the library loading time.
• Improved the overall performance of the first returned solution.

Upcoming changes

• The V1 CPP extension API will be deprecated in a future release of hipBLASLt.

hipCUB (3.3.0)

Added

• Support for large indices in hipcub::DeviceSegmentedReduce::* has been added, with the exception
  of DeviceSegmentedReduce::Arg*. Although rocPRIM's backend provides support for all reduce
  variants, CUB does not support large indices in DeviceSegmentedReduce::Arg*. For this reason,
  large index support is not available for hipcub::DeviceSegmentedReduce::Arg*.

Changed

• Changed the default value of rmake.py -a to default_gpus. This is equivalent to gfx906:xnack-,gfx1030,gfx1100,gfx1101,gfx1102.
• The NVIDIA backend now requires CUB, Thrust, and libcu++ 2.3.2.

Resolved issues

• Fixed an issue in rmake.py where the list storing cmake options would contain individual characters instead of a full string of options.
• Fixed an issue where config.hpp was not included in all hipCUB headers, resulting in build errors.

hipFFT (1.0.17)

Changed

• The AMD backend is now compiled using amdclang++ instead of hipcc. The NVIDIA CUDA backend still uses hipcc-nvcc.
• CLI11 replaces Boost Program Options as the command line parser for clients.
• Building with the address sanitizer option sets xnack+ for the relevant GPU architectures.

hipfort (0.5.0)

Added

• Added ROC-TX to the hipfort interfaces.

Changed

• Updated the hipSOLVER bindings.

HIPIFY (18.0.0)

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
• Added a filtering ability for the supplementary hipification scripts

Resolved issues

• Correct roc header files support

Known issues

• Support for fp8 data types

hipRAND (2.11.0*)

Changed

• Updated the default value for the -a argument from rmake.py to gfx906:xnack-,gfx1030,gfx1100,gfx1101,gfx1102.

Known issues

• In ROCm 6.3.0, the hipRAND package version is incorrectly set to 2.11.0. In ROCm
  6.2.4, the hipRAND package version was 2.11.1. The hipRAND version number will be corrected in a
  future ROCm release.

Resolved issues

• Fixed an issue in rmake.py where the list storing the CMake options would contain individual characters instead of a full string of options.

hipSOLVER (2.3.0)

Added

• Auxiliary functions:
  • hipsolverSetDeterministicMode, hipsolverGetDeterministicMode
• Compatibility-only functions:
  • potrf
    • hipsolverDnXpotrf_bufferSize
    • hipsolverDnXpotrf
  • potrs
    • hipsolverDnXpotrs
  • geqrf
    • hipsolverDnXgeqrf_bufferSize
    • hipsolverDnXgeqrf

Changed

• Binaries in debug builds no longer have a -d suffix.
• Changed rocSPARSE and SuiteSparse to be runtime dependencies by default. The BUILD_WITH_SPARSE CMake option can still be used
  to convert them into build-time dependencies (now off by default).
• The --no-sparse option for the install script now only affects the hipSOLVER clients and their dependency on hipSPARSE. Use the
  BUILD_HIPSPARSE_TESTS CMake option to enable tests for the hipsolverSp API (on by default).

Upcoming changes

• The Fortran bindings provided in hipsolver_module.f90 have been deprecated.
  The Fortran bindings provided by the hipfort project are recommended instead.

hipSPARSE (3.1.2)

Added

• Added an alpha version of the hipsparse-bench executable to facilitate comparing NVIDIA CUDA cuSPARSE and rocSPARSE backends.

Changed

• Changed the default compiler from hipcc to amdclang in the install script and CMake files.
• Improved the user documentation.

Resolved issues

• Fixed the gfortran dependency for the Azure Linux operating system.

Known issues

• In hipsparseSpSM_solve(), the external buffer is passed as a parameter. This does not match the NVIDIA CUDA cuSPARSE API. This extra external buffer parameter will be removed in a future release. For now, this extra parameter can be ignored and nullptr passed as it is unused internally by hipsparseSpSM_solve().

hipSPARSELt (0.2.2)

Added

• Support for a new data type combination: INT8 inputs, BF16 output, and INT32 Matrix Core accumulation
• Support for row-major memory order (HIPSPARSE_ORDER_ROW)

Changed

• Changed the default compiler to amdclang++.

Upcoming changes

• hipsparseLtDatatype_t is deprecated and will be removed in the next major release of ROCm. hipDataType should be used instead.

hipTensor (1.4.0)

Added

• Added support for tensor reduction, including APIs, CPU reference, unit tests, and documentation

Changed

• ASAN builds only support xnack+ targets.
• ASAN builds use -mcmodel=large to accommodate library sizes greater than 2GB.
• Updated the permute backend to accommodate changes to element-wise operations.
• Updated the actor-critic implementation.
• Various documentation formatting updates.

Optimized

• Split kernel instances to improve build times.

Resolved issues

• Fixed a bug in randomized tensor input data generation.
• Fixed the default strides calculation to be in column-major order.
• Fixed a small memory leak by properly destroying HIP event objects in tests.
• Default strides calculations now follow column-major convention.

llvm-project (18.0.0)

Resolved issues

• Fixed an issue where the compiler would incorrectly compile a program that used the __shfl(var, srcLane, width) function when one of the parameters to the function is undefined along some path
  to the function. See  issue #3499 on GitHub.

MIOpen (3.3.0)

Added

• [RNN] LSTM forward pass
• [Mha] Mask is added for forward pass
• [GLU] Gated Linear Unit (this is an experimental feature)
• [PReLU] Implemented PReLU backward pass (this is an experimental feature)

Optimized

• MI300 TunaNet Update: CK forward pass and WRW Solvers updated

Resolved issues

• Fixed unset stream when calling hipMemsetAsync.
• Fixed a memory leak issue caused by an incorrect transpose in find 2.0. See PR  #3285 on GitHub.
• Fixed a memcopy data race by replacing hipMemcpy with hipMemcpyWithStream.

MIVisionX (3.1.0)

Changed

• rocDecode is no longer installed by the setup script.
• The rocDecode dependency has been removed from the package installation.

Known issues

• See  MIVisionX memory access fault in Canny edge detection.
• Package installation requires the manual installation of OpenCV.
• Installation on CentOS/RedHat/SLES requires the manual installation of the FFMPEG Dev package.
• Hardware decode requires installation with --usecase=graphics in addition to --usecase=rocm.

Upcoming changes

• Optimized audio augmentations support for VX_RPP

RCCL (2.21.5)

Added

• MSCCL++ integration for specific contexts
• Performance collection to rccl_replayer
• Tuner Plugin example for Instinct MI300
• Tuning table for a large number of nodes
• Support for amdclang++
• New Rome model

Changed

• Compatibility with NCCL 2.21.5
• Increased channel count for MI300X multi-node
• Enabled MSCCL for single-process multi-threaded contexts
• Enabled CPX mode for MI300X
• Enabled tracing with rocprof
• Improved version reporting
• Enabled GDRDMA for Linux kernel 6.4.0+

Resolved issues

• Fixed an issue where, on systems running Linux kernel 6.8.0 such as Ubuntu 24.04, Direct Memory
  Access (DMA) transfers between the GPU and NIC were disabled, impacting multi-node RCCL
  performance. See  issue #3772 on GitHub.
• Fixed model matching with PXN enable

Known issues

• MSCCL is temporarily disabled for AllGather collectives.
  • This can impact in-place messages (< 2 MB) with ~2x latency.
  • Older RCCL versions are not impacted.
  • This issue will be addressed in a future ROCm release.
• Unit tests do not exit gracefully when running on a single GPU.
  • This issue will be addressed in a future ROCm release.

rocAL (2.1.0)

Added

• rocAL Pybind support for package installation has been added. To use the rocAL python module, set the PYTHONPATH: export PYTHONPATH=/opt/rocm/lib:$PYTHONPATH
• Last batch policy, pad last batch, stick to shard, and shard size support have been added for the coco, caffe, caffe2, mxnet, tf, and cifar10 image readers.

Changed

• rocDecode is no longer installed by the setup script.
• The rocDecode dependency has been removed from the package installation.

Optimized

• CTest has been updated.

Resolved issues

• Test failures have been fixed.

Known issues

• The package installation requires the manual installation of TurboJPEG and RapidJSON.
• CentOS/RedHat/SLES requires the manual installation of the FFMPEG Dev package.
• Hardware decode requires installation with --usecase=graphics in addition to --usecase=rocm.

Upcoming changes

• Optimized audio augmentations support.

rocALUTION (3.2.1)

Changed

• The default compiler has been changed from hipcc to amdclang in the installation script and cmake files.
• Changed the address sanitizer build targets. Now only gfx908:xnack+, gfx90a:xnack+, gfx940:xnack+, gfx941:xnack+, and gfx942:xnack+ are built with BUILD_ADDRESS_SANITIZER=ON.

Resolved issues

• Fixed hang in RS-AMG for Navi on some specific matrix sparsity patterns.
• Fixed wrong results in Apply on multi-GPU setups.

rocBLAS (4.3.0)

Added

• Level 3 and EX functions have an additional ILP64 API for both C and Fortran (_64 name suffix) with int64_t function arguments

Changed

• amdclang is used as the default compiler instead of hipcc
• Internal performance scripts use AMD SMI instead of the deprecated ROCm SMI

Optimized

• Improved performance of Level 2 gbmv
• Improved performance of Level 2 gemv for float and double precisions for problem sizes (TransA == N && m==n && m % 128 == 0) measured on a gfx942 GPU

Resolved issues

• Fixed the stbsv_strided_batched_64 Fortran binding

Upcoming changes

• rocblas_Xgemm_kernel_name APIs are deprecated

ROCdbgapi (0.77.0)

Added

• Support for setting precise ALU exception reporting

rocDecode (0.8.0)

Changed

• Clang is now the default CXX compiler.
• The new minimum supported version of va-api is 1.16.
• New build and runtime options have been added to the rocDecode-setup.py setup script.

Removed

• Make tests have been removed. CTEST is now used for both Make tests and package tests.
• mesa-amdgpu-dri-drivers has been removed as a dependency on RHEL and SLES.

Resolved issues

• Fixed a bug in the size of output streams in the videoDecodeBatch sample.

rocFFT (1.0.31)

Added

• rocfft-test now includes a --smoketest option.

• Implemented experimental APIs to allow computing FFTs on data
  distributed across multiple MPI ranks. These APIs can be enabled with the
  ROCFFT_MPI_ENABLE CMake option. This option defaults to OFF.

  When ROCFFT_MPI_ENABLE is ON:

  • rocfft_plan_description_set_comm can be called to provide an
    MPI communicator to a plan description, which can then be passed
    to rocfft_plan_create. Each rank calls
    rocfft_field_add_brick to specify the layout of data bricks on
    that rank.

  • An MPI library with ROCm acceleration enabled is required at
    build time and at runtime.

Changed

• Compilation uses amdclang++ instead of hipcc.
• CLI11 replaces Boost Program Options as the command line parser for clients and samples.
• Building with the address sanitizer option sets xnack+ on relevant GPU
  architectures and address-sanitizer support is added to runtime-compiled
  kernels.

ROCgdb (15.2)

Added

• Support for precise ALU exception reporting for supported architectures. Precise ALU exceptions reporting is controlled with the following commands:
  • set amdgpu precise-alu-exceptions
  • show amdgpu precise-alu-exceptions

Changed

• The sysroot or solib-search-path settings can now be used to locate files containing GPU code objects when opening a core dump. This allows opening GPU code objects on systems different from the one where the core dump was generated.

Resolved issues

• Fixed possible hangs when opening some AMDGPU core dumps in ROCgdb.
• Addressed cases where the roccoremerge utility improperly handled LOAD segment copy from the host core dump to the combined core dump.

ROCm Compute Profiler (3.0.0)

Changed

• Renamed to ROCm Compute Profiler from Omniperf.
  • New package name: rocprofiler-compute
  • New repository:  https://github.com/ROCm/rocprofiler-compute
  • New binary name: rocprof-compute

Known issues

• See  ROCm Compute Profiler post-upgrade.

• See  ROCm Compute Profiler CTest failure in CI.

ROCm Data Center Tool (0.3.0)

Added

• RVS integration
• Real time logging for diagnostic command
• --version command
• XGMI_TOTAL_READ_KB and XGMI_TOTAL_WRITE_KB monitoring metrics

Known issues

• See  ROCm Data Center Tool incorrect RHEL9 package version.

ROCm SMI (7.4.0)

Added

• Added rsmi_dev_memory_partition_capabilities_get which returns driver memory partition capablities.Driver now has the ability to report what the user can set memory partition modes to. User can now see available
  memory partition modes upon an invalid argument return from memory partition mode set (rsmi_dev_memory_partition_set).

• Support for GPU metrics 1.6 to rsmi_dev_gpu_metrics_info_get(). Updated
  rsmi_dev_gpu_metrics_info_get() and structure rsmi_gpu_metrics_t to include new fields for
  PVIOL / TVIOL, XCP (Graphics Compute Partitions) stats, and pcie_lc_perf_other_end_recovery.

• Ability to view raw GPU metrics using rocm-smi --showmetrics.

Changed

• Added back in C++ tests for memorypartition_read_write

• Updated rsmi_dev_memory_partition_set to not return until a successful restart of AMD GPU Driver.

• All APIs now have the ability to catch driver reporting invalid arguments.

Removals

• Removed --resetcomputepartition, and --resetmemorypartition options and associated APIs.
  • This change is part of the partition feature redesign.
  • The related APIs rsmi_dev_compute_partition_reset() and rsmi_dev_memory_partition_reset().

Resolved issues

• Fixed rsmi_dev_target_graphics_version_get, rocm-smi --showhw, and rocm-smi --showprod not displaying properly for MI2x or Navi 3x ASICs.

Upcoming changes

• C++ tests for memorypartition_read_write are to be re-enabled in a future ROCm release.

See the full [ROCm SMI changelog](https://github.com/ROCm/rocm_smi_lib/blob/rocm-6.3.x/CHANGELOG.md) for more details and examples.

ROCm Systems Profiler (0.1.0)

Changed

• Renamed to ROCm Systems Profiler from Omnitrace.
  • New package name: rocprofiler-systems
  • New repository:  https://github.com/ROCm/rocprofiler-systems
  • Reset the version to 0.1.0
  • New binary prefix: rocprof-sys-*

Known issues

• See  ROCm Systems Profiler post-upgrade.

ROCm Validation Suite (1.1.0)

Added

• Support for hipBLASLT blas library and option to select blas library in conf file.

Changed

• Babel parameters made runtime configurable.

Known issues

• See  ROCm Validation Suite needs specified configuration file.

rocPRIM (3.3.0)

Added

• The --test smoke option has been added to rtest.py. When rtest.py is called with this option it runs a subset of tests such that the total test time is 5 minutes. Use python3 ./rtest.py --test smoke or python3 ./rtest.py -t smoke to run the smoke test.
• The --seed option has been added to run_benchmarks.py. The --seed option specifies a seed for the generation of random inputs. When the option is omitted, the default behavior is to use a random seed for each benchmark measurement.
• Added configuration autotuning to device partition (rocprim::partition, rocprim::partition_two_way, and rocprim::partition_three_way), to device select (rocprim::select, rocprim::unique, and rocprim::unique_by_key), and to device reduce by key (rocprim::reduce_by_key) to improve performance on selected architectures.
• Added rocprim::uninitialized_array to provide uninitialized storage in local memory for user-defined types.
• Added large segment support for rocprim:segmented_reduce.
• Added a parallel nth_element device function similar to std::nth_element. nth_element places elements that are smaller than the nth element before the nth element, and elements that are bigger than the nth element after the nth element.
• Added deterministic (bitwise reproducible) algorithm variants rocprim::deterministic_inclusive_scan, rocprim::deterministic_exclusive_scan, rocprim::deterministic_inclusive_scan_by_key, rocprim::deterministic_exclusive_scan_by_key, and rocprim::deterministic_reduce_by_key. These provide run-to-run stable results with non-associative operators such as float operations, at the cost of reduced performance.
• Added a parallel partial_sort and partial_sort_copy device functions similar to std::partial_sort and std::partial_sort_copy. partial_sort and partial_sort_copy arrange elements such that the elements are in the same order as a sorted list up to and including the middle index.

Changed

• Changed the default value of rmake.py -a to default_gpus. This is equivalent to gfx906:xnack-,gfx1030,gfx1100,gfx1101,gfx1102.
• Modified the input size in device adjacent difference benchmarks. Observed performance with these benchmarks might be different.
• Changed the default seed for device_benchmark_segmented_reduce.

Removed

• rocprim::thread_load() and rocprim::thread_store() have been deprecated. Use dereference() instead.

Resolved issues

• Fixed an issue in rmake.py where the list storing cmake options would contain individual characters instead of a full string of options.
• Resolved an issue in rtest.py where it crashed if the build folder was created without release or debug subdirectories.
• Resolved an issue with rtest.py on Windows where passing an absolute path to --install_dir caused a FileNotFound error.
• rocPRIM functions are no longer forcefully inlined on Windows. This significantly reduces the build time of debug builds.
• block_load, block_store, block_shuffle, block_exchange, and warp_exchange now use placement new instead of copy assignment (operator=) when writing to local memory. This fixes the behavior of custom types with non-trivial copy assignments.
• Fixed a bug in the generation of input data for benchmarks, which caused incorrect performance to be reported in specific cases. It may affect the reported performance for one-byte types (uint8_t and int8_t) and instantiations of custom_type. Specifically, device binary search, device histogram, device merge and warp sort are affected.
• Fixed a bug for rocprim::merge_path_search where using unsigned offsets would produce incorrect results.
• Fixed a bug for rocprim::thread_load and rocprim::thread_store where float and double were not cast to the correct type, resulting in incorrect results.
• Resolved an issue where tests were failing when they were compiled with -D_GLIBCXX_ASSERTIONS=ON.
• Resolved an issue where algorithms that used an internal serial merge routine caused a memory access fault that resulted in potential performance drops when using block sort, device merge sort (block merge), device merge, device partial sort, and device sort (merge sort).
• Fixed memory leaks in unit tests due to missing calls to hipFree() and the incorrect use of hipGraphs.
• Fixed an issue where certain inputs to block_sort_merge(), device_merge_sort_merge_path(), device_merge(), and warp_sort_stable() caused an assertion error during the call to serial_merge().

ROCProfiler (2.0.0)

Added

• JSON output plugin for rocprofv2. The JSON file matches Google Trace Format making it easy to load on Perfetto, Chrome tracing, or Speedscope. For Speedscope, use --disable-json-data-flows option as speedscope doesn't work with data flows.
• --no-serialization flag to disable kernel serialization when rocprofv2 is in counter collection mode. This allows rocprofv2 to avoid deadlock when profiling certain programs in counter collection mode.
• FP64_ACTIVE and ENGINE_ACTIVE metrics to AMD Instinct MI300 accelerator
• New HIP APIs with struct defined inside union.
• Early checks to confirm the eligibility of ELF file in ATT plugin
• Support for kernel name filtering in rocprofv2
• Barrier bit to read and stop packets

Changed

• Extended lifetime for proxy queues
• Setting the trace-start option for rocprof to off now disables kernel tracing
• libpciaccess-dev functions now load with dlopen
• PcieAccessApi* api and void* libpciaccess_handle are now initialized to nullptr

Removed

• Obsolete BSD and GPL licenses
• libsystemd-dev from CMakeLists.txt

Optimized

• ROCProfiler Performance improved to reduce profiling time for large workloads of counter collection

Resolved issues

• Bandwidth measurement in AMD Instinct MI300 accelerator
• Perfetto plugin issue of roctx trace not getting displayed
• --help for counter collection
• Signal management issues in queue.cpp
• Perfetto tracks for multi-GPU
• Perfetto plugin usage with rocsys
• Incorrect number of columns in the output CSV files for counter collection and kernel tracing
• The ROCProfiler hang issue when running kernel trace, thread trace, or counter collection on Iree benchmark for AMD Instinct MI300 accelerator
• Build errors thrown during parsing of unions
• The system hang caused while running --kernel-trace with Perfetto for certain applications
• Missing profiler records issue caused while running --trace-period
• The hang issue of ProfilerAPITest of runFeatureTests on AMD Instinct MI300 accelerator
• Segmentation fault on Navi32

ROCprofiler-SDK (0.5.0)

Added

• Start and end timestamp columns to the counter collection csv output
• Check to force tools to initialize context id with zero
• Support to specify hardware counters for collection using rocprofv3 as rocprofv3 --pmc [COUNTER [COUNTER ...]]

Changed

• --marker-trace option for rocprofv3 now supports the legacy ROC-TX library libroctx64.so when the application is linked against the new library librocprofiler-sdk-roctx.so
• Replaced deprecated hipHostMalloc and hipHostFree functions with hipExtHostAlloc and hipFreeHost for ROCm versions starting 6.3
• Updated rocprofv3 --help options
• Changed naming of "agent profiling" to a more descriptive "device counting service". To convert existing tool or user code to the new name, use the following sed:

  find . -type f -exec sed -i 's/rocprofiler_agent_profile_callback_t/rocprofiler_device_counting_service_callback_t/g; s/rocprofiler_configure_agent_profile_counting_service/rocprofiler_configure_device_counting_service/g; s/agent_profile.h/device_counting_service.h/g; s/rocprofiler_sample_agent_profile_counting_service/rocprofiler_sample_device_counting_service/g' {} +
  
• Changed naming of "dispatch profiling service" to a more descriptive "dispatch counting service". To convert existing tool or user code to the new names, the following sed can be used:

  -type f -exec sed -i -e 's/dispatch_profile_counting_service/dispatch_counting_service/g' -e 's/dispatch_profile.h/dispatch_counting_service.h/g' -e 's/rocprofiler_profile_counting_dispatch_callback_t/rocprofiler_dispatch_counting_service_callback_t/g' -e 's/rocprofiler_profile_counting_dispatch_data_t/rocprofiler_dispatch_counting_service_data_t/g'  -e 's/rocprofiler_profile_counting_dispatch_record_t/rocprofiler_dispatch_counting_service_record_t/g' {} +
  
• FETCH_SIZE metric on gfx94x now uses TCC_BUBBLE for 128B reads
• PMC dispatch-based counter collection serialization is now per-device instead of being global across all devices

Removed

• gfx8 metric definitions
• rocprofv3 installation from sbin directory

Resolved issues

• Introduced subdirectory creation when rocprofv3 --output-file used to specify a folder path
• Fixed misaligned stores (undefined behavior) for buffer records
• Fixed crash when only scratch reporting is enabled
• Fixed MeanOccupancy metrics
• Fixed aborted-application validation test to properly check for hipExtHostAlloc command
• Fixed implicit reduction of SQ and GRBM metrics
• Fixed support for derived counters in reduce operation
• Bug fixed in max-in-reduce operation
• Introduced fix to handle a range of values for select() dimension in expressions parser
• Fixed Navi3x kernel tracing issues by setting the conditional aql::set_profiler_active_on_queue only when counter collection is registered

rocPyDecode (0.2.0)

Added

• RGB and YUV pytorch tensors
• Python distribution wheel (.whl)
• Multiple usecase samples

Changed

• Clang replaces hipcc as the default CXX compiler.

Removed

• Make tests have been removed. CTEST is now used for both Make tests and package tests.

Optimized

• Setup script - build and runtime install options
• Prerequisite installation helper Python scripts
• Same GPU memory viewed as pytorch tensor

Resolved issues

• Fixed setup issues.

rocRAND (3.2.0)

Added

• Added host generator for MT19937
• Support for rocrand_generate_poisson in hipGraphs
• Added engine, distribution, mode, throughput_gigabytes_per_second, and lambda columns for the csv format in
  benchmark_rocrand_host_api and benchmark_rocrand_device_api. To see these new columns, set --benchmark_format=csv
  or --benchmark_out_format=csv --benchmark_out="outName.csv".

Changed

• Updated the default value for the -a argument from rmake.py to gfx906:xnack-,gfx1030,gfx1100,gfx1101,gfx1102.
• rocrand_discrete for MTGP32, LFSR113 and ThreeFry generators now uses the alias method, which is faster than binary search in CDF.

Resolved issues

• Fixed an issue in rmake.py where the list storing the CMake options would contain individual characters instead of a full string of options.

rocSOLVER (3.27.0)

Added

• 64-bit APIs for existing functions:
  • LACGV_64
  • LARF_64
  • LARFG_64
  • GEQR2_64 (with batched and strided_batched versions)
  • GEQRF_64 (with batched and strided_batched versions)
  • POTF2_64 (with batched and strided_batched versions)
  • POTRF_64 (with batched and strided_batched versions)
  • POTRS_64 (with batched and strided_batched versions)

Changed

• The rocSPARSE library is now an optional dependency at runtime. If rocSPARSE
  is not available, rocSOLVER's sparse refactorization and solvers functions
  will return rocblas_status_not_implemented.

Optimized

• Improved the performance of LARFG, LARF, and downstream functions such as GEQR2 and GEQRF on wave64 architectures
• Improved the performance of BDSQR and GESVD
• Improved the performance of STEDC and divide and conquer Eigensolvers

Resolved issues

• Fixed a memory allocation issue in SYEVJ that could cause failures on clients that manage their own memory.
• Fixed a synchronizarion issue with SYEVJ that could led to a convergence failure for large matrices.
• Fixed a convergence issue in STEIN stemming from numerical orthogonality of the initial choice of eigenvectors.
• Fixed a synchronization issue in STEIN.

Known issues

• A known issue in STEBZ can lead to errors in routines based on bisection to compute eigenvalues for symmetric/Hermitian matrices (for example, SYEVX/HEEVX and SYGVX/HEGVX), as well as singular values (for example, BDSVDX and GESVDX).

rocSPARSE (3.3.0)

Added

• rocsparse_create_extract_descr, rocsparse_destroy_extract_descr, rocsparse_extract_buffer_size, rocsparse_extract_nnz, and rocsparse_extract APIs to allow extraction of the upper or lower part of sparse CSR or CSC matrices.

Changed

• Change the default compiler from hipcc to amdclang in install script and CMake files.
• Change address sanitizer build targets so that only gfx908:xnack+, gfx90a:xnack+, gfx940:xnack+, gfx941:xnack+, and gfx942:xnack+ are built when BUILD_ADDRESS_SANITIZER=ON is configured.

Optimized

• Improved user documentation

Resolved issues

• Fixed the csrmm merge path algorithm so that diagonal is clamped to the correct range.
• Fixed a race condition in bsrgemm that could on rare occasions cause incorrect results.
• Fixed an issue in hyb2csr where the CSR row pointer array was not being properly filled when n=0, coo_nnz=0, or ell_nnz=0.
• Fixed scaling in rocsparse_Xhybmv when only performing y=beta*y, for example, where alpha==0 in y=alpha*Ax+beta*y.
• Fixed rocsparse_Xgemmi failures when the y grid dimension is too large. This occurred when n >= 65536.
• Fixed the gfortran dependency for the Azure Linux operating system.

rocThrust (3.2.0)

Added

• Merged changes from upstream CCCL/thrust 2.3.2
  • Only the NVIDIA backend uses tuple and pair types from libcu++, other backends continue to use the original Thrust implementations and hence do not require libcu++ (CCCL) as a dependency.
• Added the thrust::hip::par_det execution policy to enable bitwise reproducibility on algorithms that are not bitwise reproducible by default.

Changed

• Changed the default value of rmake.py -a to default_gpus. This is equivalent to gfx906:xnack-,gfx1030,gfx1100,gfx1101,gfx1102.
• Enabled the upstream (thrust) test suite for execution by default. It can be disabled by using the -DENABLE_UPSTREAM_TESTS=OFF cmake option.

Resolved issues

• Fixed an issue in rmake.py where the list storing cmake options would contain individual characters instead of a full string of options.
• Fixed the HIP backend not passing TestCopyIfNonTrivial from the upstream (thrust) test suite.
• Fixed tests failing when compiled with -D_GLIBCXX_ASSERTIONS=ON.

rocWMMA (1.6.0)

Added

• Added OCP F8/BF8 datatype support

Changed

• Optimized some aos<->soa transforms with half-rotation offsets
• Refactored the rocBLAS reference entry point for validation and benchmarking
• ROCWMMA_* preprocessor configurations are now all assigned values
• Updated the default architecture targets for ASAN builds
• Updated the actor-critic implementation

Resolved issues

• Fixed a bug in F64 validation due to faulty typecasting
• Fixed a bug causing runtime compilation errors with hipRTC
• Various documentation updates and fixes

RPP (1.9.1)

Added

• RPP Glitch and RPP Pixelate have been added to the HOST and HIP backend.
• The following audio support was added to the HIP backend:
  • Resample
  • Pre-emphasis filter
  • Down-mixing
  • To Decibels
  • Non-silent region

Changed

• Test prerequisites have been updated.
• AMD advanced build flag.

Removed

• Older versions of TurboJPEG have been removed.

Optimized

• Updated the test suite.

Resolved issues

• macOS build
• RPP Test Suite: augmentations fix
• Copy: bugfix for NCDHW layout
• MIVisionX compatibility fix: Resample and pre-emphasis filter

Known issues

• Package installation only supports the HIP backend.

Upcoming changes

• Optimized audio augmentations

Tensile (4.42.0)

Added

• Testing and documentation for MasterSolutionLibrary.ArchitectureIndexMap and remapSolutionIndicesStartingFrom
• Functions for writing master file
• tPrint and reconcile printing options
• Python unit test coverage report
• Factor embed library logic into function and test
• clang++ as cxx compiler option for Windows
• Logic to cope with different compilers
  -toFile function to include generateManifest and moved to utilities
• Profiling CI job
• Support for amdclang and use defaults
• Architecture management functions in TensileCreateLibrary
• TensileCreateLibrary CLI reference docs
• New documentation for sphinx prototype and build out skeleton
• Contributor and developer guide
• Prediction model for optimal number of Stream-K tiles to run
  • Two-tile algorithm with Stream-K after DP
  • Atomic two-tile Stream-K and clean-up tuning parameters
  • Using glob to find logic files in TensileCreateLibrary
  • Function to confirm supported compiler rather than raw logic

Changed

• Improved rocBLAS build output by allowing warning suppression, ignoring developer warnings, displaying progress bar and quiet printing
• Reordered extensions for Windows in which function
• updated amdclang++ and asm directories
• Updated duplicate marking tests with mocks
• Restored print ordering
• Print option
• Bumped rocm-docs-core from 1.2.0 to 1.5.0 in /docs/sphinx
• Refactored kernel duplicate matching
• Refactored generateLogicDataAndSolutions
• Restricted XCC mapping to gfx942
• Refactored argument parsing in TensileCreateLibrary
• Disabled failing rhel9 tests
• Changed line length to 100 characters for formatting
• Changed YAML operations to use C libyaml backend
• Improved warning text
• Updated clang support for Windows
• Updated supportedCompiler function
• Clang support on Windows to require use of conditional choices and defaults
• Refactored sanity check in TensileCreateLibrary
• Moved client config logic from TensileCreateLibrary main into createClientConfig
• Updated verifyManifest in TensileCreateLibrary
• Updated RTD configs
• Cleaned up CMake to avoid redundant work during client builds
• Updated Stream-K debug settings

Removed

• Deprecated flag from CI profiling job
• Diagnostic print
• Globals from prepAsm
• Deprecated package-library option
• Duplicate which function and minor cleanup

Optimized

• To optimize the performance of Stream-K kernels:
  • Introduced analytical grid size prediction model
  • Remapped XCC-based workgroup

Resolved issues

• Fixed stream-K XCC configs for gfx942
• Updated WMMA capability command for ISA 10+
• Fixed progress bar character encoding error on Windows
• Fixed solution redundancy removal
• Fixed tuning imports for pyyaml
• Fixed printing of ASM capabilities for ROCm versions prior to 6.3
• Fixed code objects by filtering kernels with build errors and unprocessed kernels
• Fixed fully qualified std::get in contraction solutions
• Fixed add -v flag and change system invocation
• Used conditional imports for new dependencies to fix yaml CSafe load and dump import and rich terminal print import
• Fixed comments on scalarStaticDivideAndRemainder

ROCm known issues

ROCm known issues are noted on {fab}github  GitHub. For known
issues related to individual components, review the  Detailed component changes.

Instinct MI300X reports incorrect raw GPU timestamps

On MI300X accelerators, the command processor firmware reports incorrect raw GPU timestamps. This
issue is under investigation and will be addressed in a future release. See  GitHub issue #4079.

Instinct MI300 series: backward weights convolution performance issue

A performance issue affects certain tensor shapes during backward weights convolution when using
FP16 or FP32 data types on Instinct MI300 series accelerators. This issue will be addressed in a future ROCm release.
See  GitHub issue #4080.

To mitigate the issue during model training, set the following environment variables:

export MIOPEN_FIND_MODE=3
export MIOPEN_FIND_ENFORCE=3

These settings enable auto-tuning on the first occurrence of a new tensor shape. The tuning results
are stored in the user database, eliminating the need for repeated tuning when the same shape is
encountered in subsequent runs. See the
MIOpensection in the workload optimization guide to learn more about the MIOpen auto-tuning capabilities.

TransferBench package not functional

TransferBench packages included in the ROCm 6.3.0 release are not compiled properly and are not
functional for most GPU targets, except for gfx906. Full functionality will be available
in a future ROCm release. See  GitHub issue #4081.

TransferBench is a utility for benchmarking simultaneous transfers between user-specified devices
(CPUs or GPUs). See the documentation at  TransferBench
documentation. If you
want to use TransferBench, access the properly compiled packages at
https://github.com/ROCm/TransferBench/releases.

ROCm Compute Profiler post-upgrade

In ROCm 6.3.0, the omniperf package is now named rocprofiler-compute. As a result, running apt install omniperf will fail to locate the package.
Instead, use apt install rocprofiler-compute. See  ROCm Compute Profiler 3.0.0.

When upgrading from ROCm 6.2 to 6.3, any existing /opt/rocm-6.2/../omniperf folders are not
automatically removed. To clean up these folders, manually uninstall Omniperf using apt remove omniperf.
See  GitHub issue #4082.

ROCm Systems Profiler post-upgrade

In ROCm 6.3.0, the omnitrace package is now named rocprofiler-systems. As a result, running apt install omnitrace will fail to locate the package.
Instead, use apt install rocprofiler-systems. See  ROCm Systems Profiler 0.1.0.

When upgrading from ROCm 6.2 to 6.3, any existing /opt/rocm-6.2/../omnitrace folders are not
automatically removed. To clean up these folders, manually uninstall Omnitrace using apt remove omnitrace.
See  GitHub issue #4083.

Stale file due to OpenCL ICD loader deprecation

When upgrading from ROCm 6.2.x to ROCm 6.3.0, the  removal of the rocm-icd-loader
package leaves a stale file in the old rocm-6.2.x
directory. This has no functional impact. As a workaround, manually uninstall the
rocm-icd-loader package to remove the stale file. This issue will be addressed in a future ROCm
release. See  GitHub issue #4084.

ROCm Compute Profiler CTest failure in CI

When running ROCm Compute Profiler's (rocprof-compute) CTest in the Azure CI environment, the
rocprof-compute execution test fails. This issue is due to an outdated test file that was not renamed
(omniperf to rocprof-compute), and due to the ROCM_PATH environment variable not being set in
the Azure CI environment, causing the tool to be unable to extract chip information as expected.
This issue will be addressed in a future ROCm release. See  GitHub issue #4085.

MIVisionX memory access fault in Canny edge detection

Canny edge detection kernels might access out-of-bounds memory locations while
computing gradient intensities on edge pixels. This issue is isolated to
Canny-specific use cases on Instinct MI300 series accelerators. This issue is
resolved in the  MIVisionX develop branchand will be part of a future ROCm release. See  GitHub issue #4086.

Transformer Engine test_distributed_fused_attn aborts with fatal Python error

The test_distributed_fused_attn Pytest case for JAX in  Transformer Engine
for ROCm fails with a fatal Python
error under certain conditions. The root cause is unrelated Transformer Engine
but due to some issue within XLA. This XLA issue is under investigation and
will be addressed in a future release. See  GitHub issue #4087.

AMD SMI manual build issue

Manual builds of AMD SMI fail due to a broken link in its build configuration.
This affects past AMD SMI releases as well. The fix is underway and will be
applied to all branches at  https://github.com/ROCm/amdsmi.
See  GitHub issue #4088.

ROCm Data Center Tool incorrect RHEL9 package version

In previous versions of ROCm Data Center Tool (RDC) included with ROCm 6.2 for RHEL9, RDC's version
number was incorrectly set to 1.0.0. ROCm 6.3 includes RDC with the correct version number.
See  GitHub issue #4089.

If you're using RHEL9, you must first uninstall the existing ROCm 6.2 RDC 1.0.0 package with `sudo yum
remove rdc` before upgrading to the ROCm 6.3 RDC package `sudo yum install rdc`.

ROCm Validation Suite needs specified configuration file

ROCm Validation Suite might fail for certain platforms if executed without the -c option and
specifying the configuration file. See  RVS command line
optionsfor more information. This issue will be addressed in a future release.
See  GitHub issue #4090.

ROCm resolved issues

The following are previously known issues resolved in this release. For resolved issues related to
individual components, review the  Detailed component changes.

Bandwidth limitation in gang and non-gang modes on Instinct MI300A

Fixed an issue where expected target peak non-gang performance (~60 GB/s) and target peak gang
performance (~90 GB/s) were not achieved. Previously, both gang and non-gang performance were
observed to be limited at 45 GB/s. See  issue #3496 on
GitHub.

ROCm upcoming changes

The following changes to the ROCm software stack are anticipated for future releases.

AMDGPU wavefront size compiler macro deprecation

The __AMDGCN_WAVEFRONT_SIZE__ macro is deprecated and support will be removed in an upcoming
release. It is recommended that any use of this macro be removed. For more information, see  AMDGPU
support.

HIPCC Perl scripts deprecation

The HIPCC Perl scripts (hipcc.pl and hipconfig.pl) will be removed in an upcoming release.

Release ROCm 6.3.0 Release · ROCm/ROCm