leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

CUTLASS 3.7 (#2045)

Browse Source 
* CUTLASS 3.7

* clean up changelog

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-01-18 09:53:07 -05:00
committed by GitHub
co-authored by yuzhai Haicheng Wu
parent 902dff3663
commit b78588d163
2030 changed files with 8947 additions and 3475 deletions
Download Patch File Download Diff File Expand all files Collapse all files
media/docs/code_organization.md
+1 -1
View File
@@ -238,7 +238,7 @@ of tests run may vary over time as more are added.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/cutlass_3x_backwards_compatibility.md
+1 -1
View File
@@ -442,7 +442,7 @@ as the mappings are not always bijective.
# Copyright
Copyright (c) 2023 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/doxygen_mainpage.md
+1 -1
View File
@@ -38,7 +38,7 @@ has a variety of examples.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/efficient_gemm.md
+1 -1
View File
@@ -257,7 +257,7 @@ targeting NVIDIA GPUs.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/functionality.md
+1 -1
View File
@@ -281,7 +281,7 @@ CUDA exposes warp-level matrix operations in the CUDA C++ WMMA API. The CUDA C++
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/fundamental_types.md
+1 -1
View File
@@ -345,7 +345,7 @@ support on current and future NVIDIA GPUs.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/gemm_api.md
+1 -1
View File
@@ -542,7 +542,7 @@ to inline PTX.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/gemm_api_3x.md
+2 -1
View File
@@ -128,6 +128,7 @@ using CollectiveMainloop = typename cutlass::gemm::collective::CollectiveBuilder
// Step 2: Specify the collective layer epilogue type
using CollectiveEpilogue = cutlass::epilogue::collective::DefaultEpilogue<
ElementC,
cutlass::gemm::TagToStrideC_t<LayoutC>,
cutlass::gemm::TagToStrideC_t<LayoutC>,
cutlass::epilogue::thread::LinearCombination<ElementC, 1, ElementAccumulator, ElementAccumulator>>;
@@ -673,7 +674,7 @@ please refer to CuTe's tutorial, e.g., the sections on
# Copyright
Copyright (c) 2023 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/implicit_gemm_convolution.md
+1 -1
View File
@@ -761,7 +761,7 @@ Convolution can also be run by the CUTLASS Profiler.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/layout.md
+1 -1
View File
@@ -272,7 +272,7 @@ Permuted Shared Memory Layouts:
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/pipeline.md
+1 -1
View File
@@ -179,7 +179,7 @@ for more details.
# Copyright
Copyright (c) 2023 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/profiler.md
+11 -3
View File
@@ -159,9 +159,17 @@ Profiling:
capacity of the last-level cache.
--profiling-iterations=<iterations> Number of iterations to profile each kernel. If zero, kernels
are launched up to the profiling duration.
are launched up to the profiling duration. If non-zero, this
overrides `profiling-duration` and `min-iterations`.
--warmup-iterations=<iterations> Number of iterations to execute each kernel prior to profiling.
--profiling-duration=<duration> Time to spend profiling each kernel (ms). Overriden by
`profiling-iterations` when `profiling-iterations` != 0.
Note that `min-iterations` must also be satisfied.
--min-iterations=<iterations> Minimum number of iterations to spend profiling each kernel, even if
`profiling-duration` has been met.
--warmup-iterations=<iterations> Number of iterations to execute each kernel prior to profiling (default: 10).
--sleep-duration=<duration> Number of ms to sleep between profiling periods (ms).
@@ -624,7 +632,7 @@ reference_device: Passed
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/programming_guidelines.md
+1 -1
View File
@@ -1161,7 +1161,7 @@ However, A is certainly M major if interpreted as a matrix.
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/quickstart.md
+1 -1
View File
@@ -655,7 +655,7 @@ $ cmake .. -DCUTLASS_NVCC_ARCHS='70;75;80' -DCUTLASS_LIBRARY_KERNELS=tensorop*s*
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/terminology.md
+1 -1
View File
@@ -82,7 +82,7 @@ replaced by [MMA and Copy atoms from CuTe](/media/docs/cute/0t_mma_atom.md).
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/tile_iterator_concept.md
+1 -1
View File
@@ -473,7 +473,7 @@ struct WriteableReadableRandomAccessContiguousTileIteratorConcept {
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```
media/docs/utilities.md
+1 -1
View File
@@ -434,7 +434,7 @@ Please note that `synclog` is an experimental feature, and its functionality is
# Copyright
Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2017 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
```