NVIDIA / cutlass Public

Discussions

• 1

  🙏

  "Too many predicates" error from `PredicatedTileAccessIteratorPredicates`

  masahi asked Feb 22, 2022 in Q&A · Answered

  

  

  7
• 1

  💬

  [2.8] What is new?

  hwu36 started Dec 20, 2021 in General

  

  

  2
• 1

  🙏

  How to correctly use conv2d split-k parallel

  masahi asked Jan 24, 2022 in Q&A · Answered

  

  

  

  

  15
• 1

  🙏

  Tuning wgrad kernels with split-k, in practice

  masahi asked Jan 17, 2022 in Q&A · Answered

  

  

  

  

  17
• 3

  💬

  TVM+CUTLASS MLSys'22 paper

  hwu36 started Jan 18, 2022 in General

  

  

  0
• 1

  🙏

  Why this cutlass_tensorop_s1688gemm_f16_64x128_64x2_tn_align4 GEMM has blocks filled with zeros?

  dsilvavinicius asked Oct 4, 2021 in Q&A · Unanswered

  

  

  9
• 1

  🙏

  3xTF32 GEMM example slower than SIMT?

  masahi asked Dec 22, 2021 in Q&A · Answered

  

  

  2
• 2

  🙏

  Epilogue with mutiple sources

  masahi asked Oct 18, 2021 in Q&A · Answered

  

  

  

  

  30
• 5

  💬

  CUTLASS is integrated into TVM

  hwu36 started Oct 29, 2021 in General

  

  

  2
• 2

  🙏

  How to compare CUTLASS with CUBLAS

  puddingfjz asked Nov 26, 2021 in Q&A · Answered

  

  

  

  

  3
• 2

  💬

  CUDA 11.3 significantly improved the performance of CUTLASS

  hwu36 started Apr 18, 2021 in General

  

  

  

  

  8
• 10

  💬

  [2.8] 3xTF32: FP32 accuracy with 2x Performance

  hwu36 started Nov 9, 2021 in General

  

  

  1
• 1

  🙏

  Profiling Conv2d kernels by piggy-backing on Gemm profiler

  masahi asked Nov 8, 2021 in Q&A · Answered

  

  

  

  

  3
• 2

  💬

  GTC 2021 Nov Talk

  hwu36 started Nov 7, 2021 in General

  

  

  0
• 3

  🙏

  Allowing "source" (bias tensor) and output tensor to have different data type

  masahi asked Nov 1, 2021 in Q&A · Answered

  

  

  2
• 1

  💬

  GTC 2021 Nov Braindate

  hwu36 started Nov 1, 2021 in General

  

  

  0
• 1

  🙏

  What is a good layout for problem with dimensions N = image size (512 * 512 or 1024 * 1024, for example), M = 64, K = 4?

  dsilvavinicius asked Sep 28, 2021 in Q&A · Answered

  

  

  4
• 3

  💬

  [2.7] What Is New

  hwu36 started Sep 25, 2021 in General

  

  

  0
• 0

  🙏

  Whether deadlock is possible when using Gemm with SplitKSerial=true ?

  uiemUI asked Sep 12, 2021 in Q&A · Unanswered

  

  

  

  

  3
• 1

  🙏

  cutlass_profiler build with WMMA API fails?

  navdeepkk asked Jul 6, 2021 in Q&A · Unanswered

  

  

  

  

  4
• 4

  💬

  [2.6] What is New

  hwu36 started Sep 7, 2021 in General

  

  

  0
• 1

  🙏

  Are swizzling functions for reducing shared memory bank conflicts same for different architectures?

  navdeepkk asked Aug 7, 2021 in Q&A · Unanswered

  

  

  3
• 1

  💬

  cuda 11.4 is out.

  hwu36 started Jul 1, 2021 in General

  

  

  4
• 1

  🙏

  Shared memory bank conflict problem

  FindDefinition asked Jun 17, 2021 in Q&A · Unanswered

  

  

  5
• 1

  🙏

  What is "align" CL argument in cutlass_profiler?

  navdeepkk asked Apr 6, 2021 in Q&A · Unanswered

  

  

  5