2 Pull requests merged by 2 people

• Fix SM90 beta=1 hang and stream-K launch errors

  #2172 merged Mar 13, 2025

• Blockwise Improvement and Programmatic Dependent Launch

  #2161 merged Mar 10, 2025

5 Pull requests opened by 5 people

• Update mma_atom.hpp

  #2159 opened Mar 9, 2025

• fix: the bug of example_simt_canonical

  #2160 opened Mar 10, 2025

• Fix sm100 gemm wrong static constexpr that breaks compilation on Windows

  #2167 opened Mar 13, 2025

• Fix CUTE_DEVICE for cast_smem_ptr_to_unit

  #2171 opened Mar 13, 2025

• [Doc]Fix typo of cute document. "test"-->"text"

  #2174 opened Mar 14, 2025

6 Issues closed by 6 people

• [QST] `wgmma_sm90.cu` tutorial pipelining

  #2173 closed Mar 14, 2025

• [QST] Permute in K mode for consistent LDSM results

  #2140 closed Mar 14, 2025

• [DOC] Possible typos in fundamental_types.md document

  #2091 closed Mar 14, 2025

• [BUG] "Got cutlass error: Error Internal at: " even though compilation is successful

  #2157 closed Mar 13, 2025

• [BUG] Unable to run CUTLASS example 65_distributed_gemm

  #2097 closed Mar 13, 2025

• [QST] Quickstart guide for 5080.

  #2165 closed Mar 12, 2025

8 Issues opened by 7 people

• [BUG] Segmentation fault with high N, C, K values in Conv2dFprop

  #2175 opened Mar 14, 2025

• [BUG] Cute gemm fails to dispatch to batched outer product

  #2170 opened Mar 13, 2025

• [QST]How can I implement W4A16 gemm kernel with cute api?

  #2168 opened Mar 13, 2025

• [QST] Adding new parameter to Conv2dFprop in Python

  #2166 opened Mar 12, 2025

• [QST] Variable size Gemm that can be Cuda graphed

  #2164 opened Mar 12, 2025

• [QST] if(runtime_value) (cute::copy or cute::gemm) Generates NaNs

  #2163 opened Mar 12, 2025

• [BUG] Possible race condition in sm90_gemm_array_tma_warpspecialized_cooperative

  #2162 opened Mar 11, 2025

• [DOC] CUTLASS INT4 GEMM: Missing SM89 Dispatch Configuration for L40S/4090

  #2158 opened Mar 8, 2025

30 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [QST] Why tma_load.get_slice(0) here always need 0?

  #1929 commented on Mar 8, 2025 • 0 new comments

• [QST] Getting a template error trying to use cutlass's depthwise 2D convolution with pytorch

  #2077 commented on Mar 8, 2025 • 0 new comments

• [QST] in implicit gemm conv, why does not support split-k when group !=1 ?

  #2049 commented on Mar 9, 2025 • 0 new comments

• [QST] bfloat16 x int8 GEMM

  #1936 commented on Mar 11, 2025 • 0 new comments

• [BUG] Unaligned access in test/unit/gemm/threadblock/batched_gemv.cu

  #2003 commented on Mar 11, 2025 • 0 new comments

• [BUG] Tmem tiled copy with non power-of-2 size fails to compile

  #2094 commented on Mar 12, 2025 • 0 new comments

• [QST] How to apply StreamK to hopper warp specialized GEMM

  #2075 commented on Mar 12, 2025 • 0 new comments

• [QST] Adding a flag in Tensor Ref Class

  #2080 commented on Mar 12, 2025 • 0 new comments

• [QST]The Persistent Tile Scheduler in CUTLASS?

  #1685 commented on Mar 12, 2025 • 0 new comments

• [QST] How to define a new custom kernel

  #1930 commented on Mar 12, 2025 • 0 new comments

• [QST] Global variable inside conv2d kernel

  #1987 commented on Mar 12, 2025 • 0 new comments

• [QST] Question about example 69

  #2096 commented on Mar 13, 2025 • 0 new comments

• [QST] Why does GenerateSM90_TensorOp_16b_WGMMA_alignx_gemm not generate C.element = DataType.void?

  #2144 commented on Mar 13, 2025 • 0 new comments

• [QST]From index into a coordinate (or coordniate into a index), it has two different implementations, how should one distinguish and understand the scenarios for their use?

  #2128 commented on Mar 13, 2025 • 0 new comments

• [QST]How Does TMA Work in CUTLASS for Writing from Shared Memory to Global Memory?

  #2008 commented on Mar 13, 2025 • 0 new comments

• [BUG] wmma should be enabled w/ clang.

  #2006 commented on Mar 13, 2025 • 0 new comments

• [QST]Behavior of TMA Store and Wait Mechanism in CUTLASS

  #2002 commented on Mar 13, 2025 • 0 new comments

• [BUG] Funcionality TensorOp 80+ s8 * s8 + s32 => {s32, s8} not working

  #1981 commented on Mar 13, 2025 • 0 new comments

• [QST] Is there a Cutlass GEMM example to read inputs with custom padding?

  #1922 commented on Mar 13, 2025 • 0 new comments

• [BUG] Logic issue in nondeterministic reduction mode of Stream-K tile scheduler.

  #2027 commented on Mar 13, 2025 • 0 new comments

• [QST] Are there plans to add specialisations for Sm90?

  #1123 commented on Mar 14, 2025 • 0 new comments

• [BUG] calling cast_smem_ptr_to_uint(device fn) from make_gmma_desc(host device fn) is not allowed

  #1997 commented on Mar 14, 2025 • 0 new comments

• [QST]Is the Key Difference Between mbarrier and barrier Their Handling of Producer-Consumer Count?

  #1999 commented on Mar 14, 2025 • 0 new comments

• [QST]How to Handle Synchronization with Different Thread Counts for Producer and Consumer in CUTLASS?

  #1998 commented on Mar 14, 2025 • 0 new comments

• [BUG]Is tcgen05.fence supported by Cutlass-3.8.0 ?

  #2098 commented on Mar 14, 2025 • 0 new comments

• [QST] Permutation layout for contiguous stores

  #2127 commented on Mar 14, 2025 • 0 new comments

• [QST] Question about UniversalFMA

  #2101 commented on Mar 14, 2025 • 0 new comments

• [QST] FP8 with row-wise scaling on Ada-Lovelace

  #1937 commented on Mar 14, 2025 • 0 new comments

• Use Compile-Time Constexpr If

  #2035 commented on Mar 9, 2025 • 0 new comments

• Allow cluster sizes across m,n,k to be reported in cutlass profiler

  #2078 commented on Mar 14, 2025 • 0 new comments