leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

v3.9 update (#2203)

Browse Source 
* v3.9 update

* voidD

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-04-02 12:11:18 -07:00
committed by GitHub
parent 62750a2b75
commit 6f4921858b
129 changed files with 7719 additions and 2036 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
CHANGELOG.md
View File

@@ -17,14 +17,16 @@
* Enhancement and new support of block-wise and group-wise GEMM for Hopper and Blackwell architectures:
- Enhancement of [blockwise GEMM](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling.cu) for Hopper architecture.
- Enhancement of [groupwise GEMM](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_groupwise_scaling.cu) for Hopper architecture.
- Support for [grouped GEMM with blockwise scaling](./examples/68_hopper_fp8_warp_specialized_grouped_gemm_with_blockwise_scaling/) for Hopper architecture.
- Support for [grouped GEMM with blockwise and groupwise scaling](./examples/68_hopper_fp8_warp_specialized_grouped_gemm_with_blockwise_scaling/) for Hopper architecture.
- Support for [blockwise GEMM](./examples/81_blackwell_gemm_blockwise/81_blackwell_gemm_blockwise.cu) for Blackwell architecture.
- Support for [groupwise GEMM](./examples/81_blackwell_gemm_blockwise/81_blackwell_gemm_groupwise.cu) for Blackwell architecture.
* Added support for enhanced kernel performance search in CUTLASS:
- Support for [grouped GEMM with blockwise](./examples/81_blackwell_gemm_blockwise/81_blackwell_grouped_gemm_blockwise.cu) and [groupwise scaling](./examples/81_blackwell_gemm_blockwise/81_blackwell_grouped_gemm_groupwise.cu) for Blackwell architecture.
* Added support for enhanced kernel performance search (auto-tuning) in CUTLASS profiler:
- Sorting performance results by GFLOPs/second: Users can now sort the final performance report based on GFLOPs/second, making it easier to identify the most efficient kernels.
- Exhaustive search for best kernel performance in GFLOPs/second: The profiler now searches for the best-performing kernel across a range of problem sizes, swizzle sizes, rasterization orders, and dynamic cluster configurations to maximize performance.
- Performance search under a fixed GEMM shape: Enables exhaustive tuning within a fixed GEMM shape, exploring various kernel parameters to find the best configuration.
- More detailed introductions and examples to leverage this feature can be found in [profiler.md](./media/docs/profiler.md#exhaustive-search-mode-and-top-k-output-ranking-according-to-performance-in-gflopss).
- More detailed introductions and examples to leverage this feature can be found in [profiler.md](./media/docs/cpp/profiler.md#exhaustive-search-mode-and-top-k-output-ranking-according-to-performance-in-gflopss).
* Support `void` as the D element in sm100 kernel epilogues.
## [3.8.0](https://github.com/NVIDIA/cutlass/releases/tag/v3.8.0) (2025-01-25)
@@ -40,7 +42,7 @@
- [Pipelines that implement Blackwell specific synchronization](./include/cutlass/pipeline/sm100_pipeline.hpp).
- [Cluster launch control API supporting preferred and fallback cluster shapes](./include/cutlass/cluster_launch.hpp).
- Data types including NVFP4, MXFP4, MXFP6, and MXFP8 and all their supported element and scale factor types.
- Tile schedulers using [Blackwell's Cluster Launch Control (CLC) feature](./media/docs/blackwell_cluster_launch_control.md) to implement dynamic persistence scheduling for [GEMMs](./include/cutlass/gemm/kernel/sm100_tile_scheduler.hpp), and [stream-K](./include/cutlass/gemm/kernel/sm100_tile_scheduler_stream_k.hpp).
- Tile schedulers using [Blackwell's Cluster Launch Control (CLC) feature](./media/docs/cpp/blackwell_cluster_launch_control.md) to implement dynamic persistence scheduling for [GEMMs](./include/cutlass/gemm/kernel/sm100_tile_scheduler.hpp), and [stream-K](./include/cutlass/gemm/kernel/sm100_tile_scheduler_stream_k.hpp).
- Extensions to testbeds and reference check code for unit tests and CUTLASS profiler.
* Full support for Blackwell SM100 kernels in CUTLASS 3.x API:
- [Blackwell specific kernel layers](./include/cutlass/gemm/kernel/sm100_gemm_tma_warpspecialized.hpp) that
@@ -78,11 +80,11 @@
- A set of new [Hopper grouped GEMM kernels](./examples/69_hopper_mixed_dtype_grouped_gemm/) that support mixed A and B datatypes.
- A new [Hopper FP8 GEMM with groupwise scaling](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_groupwise_scaling.cu).
* Documentation updates:
- [Quickstart - instantiating a Blackwell block-scaled GEMM](./media/docs/quickstart.md#instantiating-a-blackwell-gemm-kernel).
- Detailed [Blackwell block-scaled GEMM functionality documentation](./media/docs/blackwell_functionality.md)
- A new [functionality documentation](./media/docs/functionality.md) specifically for 3.x API comprehensively documenting all supported kernel types, data types, kernel features, minimum CUDA tookit support etc for 3.x supported architectures.
- [Quickstart - instantiating a Blackwell block-scaled GEMM](./media/docs/cpp/quickstart.md#instantiating-a-blackwell-gemm-kernel).
- Detailed [Blackwell block-scaled GEMM functionality documentation](./media/docs/cpp/blackwell_functionality.md)
- A new [functionality documentation](./media/docs/cpp/functionality.md) specifically for 3.x API comprehensively documenting all supported kernel types, data types, kernel features, minimum CUDA tookit support etc for 3.x supported architectures.
- Updates to [compatibility](./README.md#compatibility) section regarding supported compilers, operating systems, CUDA Toolkits, Hardware Architectures, and [Target Architecture](./README.md#Target-Architecture).
- Updates to [profiler documentation](./media/docs/profiler.md) for testing mixed input GEMM kernels on Hopper.
- Updates to [profiler documentation](./media/docs/cpp/profiler.md) for testing mixed input GEMM kernels on Hopper.
## [3.7.0](https://github.com/NVIDIA/cutlass/releases/tag/v3.7.0) (2025-01-11)
- [Hopper blockwise scaling FP8 GEMM](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling.cu) uses 2D scaling tensor, assigning one value per threadblock. This allows a finer-grained scaling to be applied for each output tile per gemm-k iteration. The operands and scaling tensors are loaded from global memory to shared memory using TMA and cp_async, respectively. The scaling is applied inside the mainloop. Details with figures are [here](https://github.com/NVIDIA/cutlass/pull/1932#issue-2645398439).
@@ -95,7 +97,7 @@
+ Fix `cute::SM80_CP_ASYNC_CACHEALWAYS`, `cute::SM80_CP_ASYNC_CACHEGLOBAL`, `cute::SM80_CP_ASYNC_CACHEALWAYS_ZFILL`, `cute::SM80_CP_ASYNC_CACHEGLOBAL_ZFILL` to avoid implicitly selecting `ZFILL` behavior on predication.
+ Remove `cute::copy_vec<T>` in favor of `cute::copy_aligned` and `cute::copy(AutoVectorizingCopyWithAssumedAlignment<NumBits>,...)`.
+ A refactor of default epilogue struct `DefaultEpilogue` [API](./include/cutlass/epilogue/collective/default_epilogue.hpp) to avoid reading non-void `ElementC` value for `ElementC = void` kernel.
- New CUTLASS profiler flags: `profiling-duration`, `min-iterations`, and `kernels-file` documented in [profiler.md](./media/docs/profiler.md#cutlass-profiler).
- New CUTLASS profiler flags: `profiling-duration`, `min-iterations`, and `kernels-file` documented in [profiler.md](./media/docs/cpp/profiler.md#cutlass-profiler).
- Various improvements and fixes from the community and CUTLASS team. Thanks to everyone who submitted PRs!
- Optimal code generation with CUDA toolkit versions 12.6.
@@ -109,12 +111,12 @@
- A refactor to the CUTLASS 3.x convolution `kernel::ConvUniversal` [API](./include/cutlass/conv/kernel/sm90_implicit_gemm_tma_warpspecialized.hpp) to bring it in line with `gemm::GemmUniversal`. Now the 3.x convolution API is no longer considered as a beta API.
- [An improved mixed input GEMM](./examples/55_hopper_mixed_dtype_gemm/README.md) and a [lookup table implementation](./examples/55_hopper_mixed_dtype_gemm/55_hopper_int4_fp8_gemm.cu) for `INT4`x`FP8` scale-only mode.
- [EVT nodes for Top-K selection and softmax](./include/cutlass/epilogue/fusion/sm90_visitor_topk_softmax.hpp) and [GEMM example using those](./examples/61_hopper_gemm_with_topk_and_softmax/61_hopper_gemm_with_topk_and_softmax.cu).
- [Programmatic Dependent Launch](./include/cutlass/arch/grid_dependency_control.h) (PDL) that leverages a new Hopper feature to speedup two back-to-back kernels, and its corresponding [documentations](./media/docs/dependent_kernel_launch.md).
- [A new debugging tool, synclog](./include/cutlass/arch/synclog.hpp), for dumping out all synchronization events from within a kernel to a file. Please see [synclog documentation](./media/docs/utilities.md#debugging-asynchronous-kernels-with-cutlasss-built-in-synclog-tool) for details.
- [Programmatic Dependent Launch](./include/cutlass/arch/grid_dependency_control.h) (PDL) that leverages a new Hopper feature to speedup two back-to-back kernels, and its corresponding [documentations](./media/docs/cpp/dependent_kernel_launch.md).
- [A new debugging tool, synclog](./include/cutlass/arch/synclog.hpp), for dumping out all synchronization events from within a kernel to a file. Please see [synclog documentation](./media/docs/cpp/utilities.md#debugging-asynchronous-kernels-with-cutlasss-built-in-synclog-tool) for details.
- A new TMA-enabled [epilogue](./include/cutlass/epilogue/collective/sm90_epilogue_array_tma_warpspecialized.hpp) for grouped GEMM that brings significant performance improvement, as well as its EVT support.
- A SIMT-enabled pointer-array [epilogue](./include/cutlass/epilogue/collective/sm70_epilogue_vectorized_array.hpp).
- A new [Ping-Pong kernel schedule for Grouped GEMM](./include/cutlass/gemm/kernel/sm90_gemm_array_tma_warpspecialized_pingpong.hpp) and some other optimizations.
- [A new instantiation strategy for CUTLASS profiler kernels](./python/cutlass_library/sm90_shapes.py) along with [improved documentation for instantiation level in CUTLASS profiler](./media/docs/profiler.md#instantiating-more-kernels-with-hopper).
- [A new instantiation strategy for CUTLASS profiler kernels](./python/cutlass_library/sm90_shapes.py) along with [improved documentation for instantiation level in CUTLASS profiler](./media/docs/cpp/profiler.md#instantiating-more-kernels-with-hopper).
- A new hardware support for comparisons and computations of [`cutlass::bfloat16_t`](./include/cutlass/bfloat16.h)
- Fixed use of isnan on Windows for [`half_t`](./test/unit/core/functional.cu).
- Various improvements and fixes from the community and CUTLASS team. Thanks to everyone who submitted PRs!
@@ -124,7 +126,7 @@
- [Minimal SM90 WGMMA + TMA GEMM example in 100 lines of code](./examples/cute/tutorial/wgmma_sm90.cu)
- [Exposure of L2 `cache_hint`s in TMA copy atoms](./include/cute/arch/copy_sm90_tma.hpp#L48)
- Exposure of raster order and tile swizzle extent in [CUTLASS library profiler](./media/docs/profiler.md#GEMM), and
- Exposure of raster order and tile swizzle extent in [CUTLASS library profiler](./media/docs/cpp/profiler.md#GEMM), and
[example 48](./examples/48_hopper_warp_specialized_gemm/48_hopper_warp_specialized_gemm.cu).
- [TMA store based and EVT supported epilogues](./include/cutlass/epilogue/collective/sm90_epilogue_array_tma_warpspecialized.hpp) for [Hopper pointer array batched kernels](./test/unit/gemm/device/sm90_gemm_f16_f16_f16_tensor_op_f32_ptr_array.cu).
- A new [`GemmSparseUniversal` API for CUTLASS 2.x Ampere kernels](./include/cutlass/gemm/device/gemm_sparse_universal.h) to enable serial and parallel split-k for sparse tensor cores and new tiny tile sizes to better support LLM inferrence:
@@ -137,7 +139,7 @@
- Support for residual add (beta != 0) in convolution kernels.
- A new convolution [epilogue](./examples/16_ampere_tensorop_conv2dfprop/ampere_tensorop_conv2dfprop.cu#L269) for CUTLASS 2.x to support non-packed NHWC output.
- A refactor of [include files throughout CUTLASS core directories](./include/cutlass/gemm/collective/collective_mma_decl.hpp) to reduce circular dependencies and [tests to guard against them](./test/self_contained_includes/CMakeLists.txt).
- [A guide for setting up VSCode to work well with CUTLASS](./media/docs/ide_setup.md) and [expanded code style guide](./media/docs/programming_guidelines.md).
- [A guide for setting up VSCode to work well with CUTLASS](./media/docs/cpp/ide_setup.md) and [expanded code style guide](./media/docs/cpp/programming_guidelines.md).
- Better support for MSVC as a host compiler.
- Many performance optimizations, improvements, and bug fixes including fixes for FlashAttention-2.
- Optimal code generation with CUDA toolkit versions 12.4 and 12.5u1.
@@ -145,7 +147,7 @@
## [3.5.0](https://github.com/NVIDIA/cutlass/releases/tag/v3.5.0) (2024-04-09)
- Implicit GEMM Convolutions targeting Hopper SM90A via WGMMA + [TMA im2col](./include/cute/atom/copy_traits_sm90_im2col.hpp)
+ Native implementation in CUTLASS 3.x using CuTe, mirroring the [same design hierarchy as that of GEMMs](./media/docs/gemm_api_3x.md).
+ Native implementation in CUTLASS 3.x using CuTe, mirroring the [same design hierarchy as that of GEMMs](./media/docs/cpp/gemm_api_3x.md).
+ Support for 1D, 2D, and 3D convolutions in a [rank-agnostic fashion](./include/cutlass/conv/convnd_problem_shape.hpp).
+ Support for [Fprop](./test/unit/conv/device_3x/fprop/sm90_conv3d_fprop_implicit_gemm_s8_s8_s32_tensorop_s32.cu), [Dgrad](./test/unit/conv/device_3x/dgrad/sm90_conv2d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu), and [Wgrad](./test/unit/conv/device_3x/wgrad/sm90_conv1d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu) algorithms
+ [CUTLASS profiler support](./python/cutlass_library/conv3x_emitter.py) for 2D and 3D convolutions implemented via the 3.x API.
@@ -157,7 +159,7 @@
- 32x and 16x tile sizes are added to CUTLASS 2.x to improve the performance of narrow-tall and wide-short matrices.
+ [Ampere FP16 TN](./test/unit/gemm/device/gemm_f16t_f16n_f16t_tensor_op_f32_sm80.cu) and [NT](./test/unit/gemm/device/gemm_f16n_f16t_f16t_tensor_op_f32_sm80.cu#L227-L301), [Ampere INT8 TN](./test/unit/gemm/device/gemm_s8t_s8n_s8t_tensor_op_s32_sm80.cu#L392-L1342), [Ampere INT4 TN](./test/unit/gemm/device/gemm_s4t_s4n_s4t_tensor_op_s32_sm80.cu#L372-L934).
+ [Turing FP16 TN](./test/unit/gemm/device/gemm_f16t_f16n_f16t_tensor_op_f32_sm75.cu#L55-L394), [Turing INT8 TN](./test/unit/gemm/device/gemm_s8t_s8n_s8t_tensor_op_s32_sm75.cu#L166-L537), [Turing INT4 TN](./test/unit/gemm/device/gemm_s4t_s4n_s4t_tensor_op_s32_sm75.cu#L310-L564).
- Updates to CuTe documentation for [`cute::Tensor<>`](./media/docs/cute/03_tensor.md), [MMA atoms](./media/docs/cute/0t_mma_atom.md), and an overhauled [CuTe GEMM tutorial series](./examples/cute/tutorial).
- Updates to CuTe documentation for [`cute::Tensor<>`](./media/docs/cpp/cute/03_tensor.md), [MMA atoms](./media/docs/cpp/cute/0t_mma_atom.md), and an overhauled [CuTe GEMM tutorial series](./examples/cute/tutorial).
- Extensions to CuTe to support [L2 prefetching](./include/cute/algorithm/prefetch.hpp) and [TMA store+reductions](./include/cute/arch/copy_sm90_tma.hpp#L1337).
- Remove C++11 requirement on a few CUTLASS 2.x API header files. All CUTLASS files now require C++17.
- Fixes to greatly reduce build warnings.
@@ -176,7 +178,7 @@
* Beta release of [Group-GEMM](./examples/57_hopper_grouped_gemm) utilizing TMA and WGMMA (requires CUDA 12.3 or above).
* [Ampere Sparse GEMM](./examples/15_ampere_sparse_tensorop_gemm/ampere_sparse_tensorop_gemm_with_visitor.cu) supports Epilogue Visitor Tree (EVT) now.
* NamedBarriers usability improvement and list of [ReservedNamedBarriers](./include/cutlass/arch/barrier.h) has been officially released.
* Improved [CuTe documentation](./media/docs/cute/) including improved clarity and depth of [Quickstart](./media/docs/cute/00_quickstart.md), [CuTe Layout](./media/docs/cute/01_layout.md), and [CuTe Layout Algebra](./media/docs/cute/02_layout_algebra.md). Associated code comments, post-conditions, and details in [CuTe Core Unit Tests](./test/unit/cute/core/) also improved.
* Improved [CuTe documentation](./media/docs/cpp/cute/) including improved clarity and depth of [Quickstart](./media/docs/cute/00_quickstart.md), [CuTe Layout](./media/docs/cpp/cute/01_layout.md), and [CuTe Layout Algebra](./media/docs/cpp/cute/02_layout_algebra.md). Associated code comments, post-conditions, and details in [CuTe Core Unit Tests](./test/unit/cute/core/) also improved.
## [3.3](https://github.com/NVIDIA/cutlass/releases/tag/v3.3.0) (2023-10-31)
* [Mixed-input Hopper GEMMs](./examples/55_hopper_mixed_dtype_gemm) support covering 16-bit x 8-bit input operand types.
@@ -227,7 +229,7 @@
* Epilogue builders. Similar to mainloop builders (see [example 49](./examples/49_hopper_gemm_with_collective_builder/49_collective_builder.cu)), epilogue builders aim to generate the best-possible epilogue while exposing incremental opt-ins for greater customization.
* Profiler support for overriding kernel and epilogue builder auto schedules for 3.x API kernels, allowing specific policies to be run in the CUTLASS profiler.
* Performance optimizations for the [*warp-specialized persistent ping-pong*](./include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_pingpong.hpp) kernel.
* Changes to the [GEMM API 3.x](./media/docs/gemm_api_3x.md), involving the host-facing arguments and the underlying `Params` structs.
* Changes to the [GEMM API 3.x](./media/docs/cpp/gemm_api_3x.md), involving the host-facing arguments and the underlying `Params` structs.
* [FMHA Backward Pass](./examples/41_fused_multi_head_attention/fused_multi_head_attention_backward.cu) from Meta xFormers.
* [Streamk GEMM with Broadcast](./examples/47_ampere_gemm_universal_streamk/ampere_gemm_universal_streamk_broadcast.cu) enables epilogue broadcast with StreamK GEMM.
* [Batched B2B GEMM](./examples/13_two_tensor_op_fusion) now can run multiple Back-to-Back GEMM with the same problem size in parallel.
@@ -239,10 +241,10 @@
* Updates and bugfixes from the community (thanks!)
## [3.0.0](https://github.com/NVIDIA/cutlass/releases/tag/v3.0.0) (2023-01-23)
* [CuTe](./media/docs/cute/00_quickstart.md), a [new core library and backend](./include/cute) for CUTLASS 3.0 that defines a single Layout vocabulary type and an associated algebra of layouts for a much more expressive and composable abstraction for tensors, sets of parallel agents, and operations by said agents on tensors.
* [A new conceptual operation hierarchy](./media/docs/cutlass_3x_design.md) that replaces the architecture-centric hierarchy of CUTLASS 2.x and [documentation for CUTLASS 3.0's GEMM API changes](./media/docs/gemm_api_3x.md).
* Strict API backwards compatibility that exposes both 2.x and 3.x API kernels through the same [`device::GemmUniversalAdapter`](./include/cutlass/gemm/device/gemm_universal_adapter.h) and [`kernel::GemmUniversal`](./include/cutlass/gemm/kernel/gemm_universal.hpp) types, allowing users to include both APIs in the same translation units. More information can be found in the [3.x backwards compatibility section](./media/docs/cutlass_3x_backwards_compatibility.md).
* Updates to [Functionality](./media/docs/functionality.md) which directs users on which kernels are supported via CUTLASS-2 and CUTLASS-3.
* [CuTe](./media/docs/cpp/cute/00_quickstart.md), a [new core library and backend](./include/cute) for CUTLASS 3.0 that defines a single Layout vocabulary type and an associated algebra of layouts for a much more expressive and composable abstraction for tensors, sets of parallel agents, and operations by said agents on tensors.
* [A new conceptual operation hierarchy](./media/docs/cpp/cutlass_3x_design.md) that replaces the architecture-centric hierarchy of CUTLASS 2.x and [documentation for CUTLASS 3.0's GEMM API changes](./media/docs/cpp/gemm_api_3x.md).
* Strict API backwards compatibility that exposes both 2.x and 3.x API kernels through the same [`device::GemmUniversalAdapter`](./include/cutlass/gemm/device/gemm_universal_adapter.h) and [`kernel::GemmUniversal`](./include/cutlass/gemm/kernel/gemm_universal.hpp) types, allowing users to include both APIs in the same translation units. More information can be found in the [3.x backwards compatibility section](./media/docs/cpp/cutlass_3x_backwards_compatibility.md).
* Updates to [Functionality](./media/docs/cpp/functionality.md) which directs users on which kernels are supported via CUTLASS-2 and CUTLASS-3.
* Updates to [Compatibility](./README.md#compatibility) Section regarding supported compilers, operating systems, CUDA Toolkits, Hardware Architectures and [Target Architecture](./README.md#Target-Architecture).
* New warp-specialized GEMM [kernel schedules](./include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized.hpp) and [mainloops](./include/cutlass/gemm/collective/sm90_mma_tma_gmma_ss_warpspecialized.hpp) targeting Hopper architecture that achieve great performance with TMA, WGMMA, and threadblock clusters.
* Extensions to CUTLASS profiler to support threadblock cluster shapes in library and profiler tile configurations.
@@ -420,7 +422,7 @@
* Global memory iterators supporting Fprop, Dgrad, and Wgrad
* `MmaMultistage` for implicit GEMM convolution for NVIDIA Ampere architecture
* `MmaPipeline` for implicit GEMM convolution for NVIDIA Volta and Turing architectures
* [Documentation](./media/docs/implicit_gemm_convolution.md) describing Implicit GEMM Convolution algorithm and implementation
* [Documentation](./media/docs/cpp/implicit_gemm_convolution.md) describing Implicit GEMM Convolution algorithm and implementation
## [2.3.0](https://github.com/NVIDIA/cutlass/releases/tag/v2.3.0) (2020-09-23)
* [NVIDIA Ampere Architecture features](https://devblogs.nvidia.com/nvidia-ampere-architecture-in-depth/)
@@ -434,7 +436,7 @@
* NVIDIA Ampere GPU Architecture examples and documentation:
* [Tensor Float 32](./examples/14_ampere_tf32_tensorop_gemm/ampere_tf32_tensorop_gemm.cu) and
* [Sparse Tensor Cores](./examples/15_ampere_sparse_tensorop_gemm/ampere_sparse_tensorop_gemm.cu)
* Documentation added on CUTLASS [efficient row-major epilogue](./media/docs/gemm_api.md#efficient-epilogue)
* Documentation added on CUTLASS [efficient row-major epilogue](./media/docs/cpp/gemm_api.md#efficient-epilogue)
## [2.2.0](https://github.com/NVIDIA/cutlass/releases/tag/v2.2.0) (2020-06-08)
* [NVIDIA Ampere Architecture features](https://devblogs.nvidia.com/nvidia-ampere-architecture-in-depth/)
@@ -454,7 +456,7 @@
* Disabled F16C by default for compatibility - enable on cmake command line with `-DCUTLASS_ENABLE_F16C=ON`
## [2.1.0](https://github.com/NVIDIA/cutlass/releases/tag/v2.1.0) (2020-04-06)
* BLAS-style host-side API added to [CUTLASS Library](./media/docs/quickstart.md#cutlass-library)
* BLAS-style host-side API added to [CUTLASS Library](./media/docs/cpp/quickstart.md#cutlass-library)
* API to launch compiled kernel instances for GEMM and planar complex GEMM
* Planar Complex GEMM kernels targeting Volta and Turing Tensor Cores
* Computes complex matrix products on matrices stored as disjoint real and imaginary parts
@@ -468,10 +470,10 @@
* Encapsulated functionality embodying modern C++11 programming techniques
* Optimized containers and data types for efficient, generic, portable device code
* Updates to:
* [Quick start guide](./media/docs/quickstart.md)
* [Quick start guide](./media/docs/cpp/quickstart.md)
* [Documentation](./README.md#documentation)
* [Utilities](./media/docs/utilities.md)
* [CUTLASS Profiler](./media/docs/profiler.md)
* [Utilities](./media/docs/cpp/utilities.md)
* [CUTLASS Profiler](./media/docs/cpp/profiler.md)
* Native Turing Tensor Cores
* Efficient GEMM kernels targeting Turing Tensor Cores
* Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands