Repository Issue Activity (beta)

#505 mat_transpose_f32_row2col2d在非方阵输入下结果不完整

#495 Require example of FP8 GEMM using CUTLASS CUTE

#500 Require TMA + MMA WS (sm 120) example for CuTe

#50 🌤🌤 CONTRIBUTE 🎉🎉

#428 Skills

#493 Require example of TMA (Async proxy) + WS (ld.matrix + mma.async) for SM120

#416 [Proposal] Upstream SM90 WGMMA HGEMM TN?

#413 reduce算子16位下的精度问题

#422 are you accepting cute dsl python kernels?

#411 关于ptx指令时使用uint32_t RA[4]的含义与调试？

#414 SGEMM benchmark may hit CUBLAS_STATUS_ALLOC_FAILED due to cublas handle leak in kernels/sgemm/sgemm_cublas.cu

#409 Potential OOB shared memory read in online safe softmax (MD block reduction)

#405 Reduce kernel maybe need some modifications in certain areas to make it more efficient.

#397 请教一个online softmax 实现的问题

#398 请问什么时候支持B200?

#399 sgemm中bank conflict计算问题

#394 A100 and H100 GPU Performance

#396 wmma smem 数据搬运问题

#392 softmax只能单block内规约

#393 hgemm在h20性能上问题

#381 elementwise 算子问题

#377 Typo at kernels/nvidia-nsight/README.md

#379 sigmoid.cu文件修改Ss Ks之后，kernel输出与pytorch输出不能对齐

#386 Why does single-stage runs faster than multi-stage for triton_fused_softmax.py?

#390 Why can't the fused_layer_norm support > 64KB BLOCK_SIZE