8 Pull requests merged by 7 people

• [DOC] Add more exposition to composition example

  #2536 merged Aug 12, 2025

• Fix typo in smem_allocator.py

  #2517 merged Aug 11, 2025

• fix typo

  #2529 merged Aug 11, 2025

• NIT: Grammar

  #2537 merged Aug 11, 2025

• Made documentation updates in batched_gemm.cu

  #2538 merged Aug 11, 2025

• Fix a copy error in the SM70 main loop

  #2540 merged Aug 11, 2025

• Support both CUDA 12 and 13 cccl header locations

  #2543 merged Aug 11, 2025

• Fix incorrect K dim in CuTe MMA Atom doc

  #2544 merged Aug 11, 2025

10 Pull requests opened by 10 people

• Fix typo in cute.nvgpu.warpgroup.mma doc

  #2548 opened Aug 7, 2025

• Make swizzle in pycute work

  #2553 opened Aug 7, 2025

• Liberate runtime check in example 13

  #2554 opened Aug 7, 2025

• ex77 backwards GQA

  #2556 opened Aug 7, 2025

• Add missing CUDA_ARCH guard for `__nanosleep` in example

  #2558 opened Aug 11, 2025

• fix a typo.

  #2561 opened Aug 12, 2025

• Add movmatrix support

  #2562 opened Aug 12, 2025

• Fix typo in tv layout stride in elementwise_add

  #2564 opened Aug 12, 2025

• Fix default cluster callback values to 1 to avoid profiler failure when these values are not set in command line.

  #2565 opened Aug 13, 2025

• Fix arch guards in a few examples

  #2567 opened Aug 13, 2025

6 Issues closed by 4 people

• [QST] [CuTeDSL] What functions should be in `cute.jit` vs `cute.kernel`

  #2546 closed Aug 8, 2025

• [QST] Question about logical_divide (only on older version of cutlass)

  #2545 closed Aug 7, 2025

• [BUG] Kernel Launch GridDim was miscalculated in examples/cute/tutorial/blackwell/04_mma_tma_2sm_sm100.cu

  #2494 closed Aug 7, 2025

• Why not use sync after loading from TMEM to RMEM in example 02_mma_tma_sm100.cu

  #2525 closed Aug 7, 2025

• [QST] Deadlock in producer consumer loop

  #2404 closed Aug 7, 2025

• [QST] [CuTeDSL] Branching: ValueError: unable to convert ... in type <class 'list'> to Numeric

  #2531 closed Aug 6, 2025

7 Issues opened by 6 people

• [QST]How to optimize the tensorrt-llm's mixed gemm

  #2566 opened Aug 13, 2025

• [QST] Why are my registers being loaded from global memory?

  #2563 opened Aug 12, 2025

• [QST][Cute-DSL] How to return more than 1 value from llvm.inline_asm

  #2560 opened Aug 11, 2025

• [BUG] `88_hopper_fmha_fp8` example fails to compile on some CUDA archs

  #2559 opened Aug 11, 2025

• [BUG] [CuTeDSL] LLVM ERROR: pthread_create failed: Resource temporarily unavailable

  #2551 opened Aug 7, 2025

• Missing PYPI releases

  #2549 opened Aug 7, 2025

• [BUG] Grouped GEMM with GroupScaling Fails with TMA Error on Hopper When a Problem has K=0

  #2547 opened Aug 7, 2025

21 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] [CuTeDSL] operand #0 does not dominate this use

  #2532 commented on Aug 6, 2025 • 0 new comments

• [QST] how blackwell gemm deal with Not 16-byte aligned data?

  #2491 commented on Aug 7, 2025 • 0 new comments

• [QST]Will K-Major Blockwise Scale Config be supported on the Hopper architecture?

  #2464 commented on Aug 7, 2025 • 0 new comments

• [QST]Can I use list of tensor as input in 68_hopper_fp8_warp_specialized_grouped_gemm_with_blockwise_scaling

  #2423 commented on Aug 7, 2025 • 0 new comments

• [QST] b2b gemm fp32

  #2415 commented on Aug 7, 2025 • 0 new comments

• [BUG] Issue with TMA-REDUCE in Split-K GEMM for Inputs with Negative Values on H100

  #2535 commented on Aug 7, 2025 • 0 new comments

• [DOC]构建 cutlass_profiler

  #2226 commented on Aug 7, 2025 • 0 new comments

• [QST] Adding new parameter to Conv2dFprop in Python

  #2166 commented on Aug 7, 2025 • 0 new comments

• [QST] How to correctly vectorize copy with cast?

  #2440 commented on Aug 8, 2025 • 0 new comments

• [FEA] [Cute-DSL] Passing flags to ptxas

  #2486 commented on Aug 9, 2025 • 0 new comments

• [QST]how to use 68_hopper_fp8_warp_specialized_grouped_gemm_with_blockwise_scaling

  #2412 commented on Aug 12, 2025 • 0 new comments

• [QST] CUTLASS GEMV with very large matrices [50000, 50000]

  #2409 commented on Aug 12, 2025 • 0 new comments

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

  #2164 commented on Aug 12, 2025 • 0 new comments

• [QST] How to implement a fused mixed precision matrix multiplication such as w4a4 + w16a16?

  #2058 commented on Aug 12, 2025 • 0 new comments

• [FEA] Blackwell support for python libraries.

  #2237 commented on Aug 12, 2025 • 0 new comments

• [BUG] The `print_layout` function failed to format the output correctly.

  #2496 commented on Aug 13, 2025 • 0 new comments

• [QST]What's wrong with my usage of cutlass profiler???

  #2458 commented on Aug 13, 2025 • 0 new comments

• [QST] incomplete type "cutlass::gemm::device::DefaultGemmConfiguration<MMAOp, SmArch, ElementInputA, ElementInputB, ElementOutput, ElementAccumulator>"

  #2461 commented on Aug 13, 2025 • 0 new comments

• Remove duplicate function calls

  #1584 commented on Aug 13, 2025 • 0 new comments

• bwl1289/fix/cmake-build-fixes

  #2305 commented on Aug 12, 2025 • 0 new comments

• Make unittest compilation faster

  #2402 commented on Aug 8, 2025 • 0 new comments