NVIDIA / NVIDIA/cutlass

[BUG] BF16 * BF16 GEMM with pingpong schedule and non-zero beta hangs

Open
#2,152 4 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

? - Needs Triage bug inactive-30d inactive-90d
Dominant language
C++
Stars
10.5k
Forks
2.1k
Avg merge
3d 11h
Merged PRs (30d)
7

Description

cmake

cmake -DCMAKE_BUILD_TYPE:STRING=Release -DCUTLASS_NVCC_ARCHS:STRING=90a -DCUTLASS_NVCC_KEEP:STRING=OFF -DCUTLASS_ENABLE_F16C:STRING=ON -DCUTLASS_LIBRARY_KERNELS:STRING=cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_bf16_bf16_128x128x64*tnn_align8 -DCUTLASS_LIBRARY_INSTANTIATION_LEVEL:STRING=max -DCUTLASS_LIBRARY_IGNORE_KERNELS:STRING=gemm_grouped*,gemm_planar*,scl,sclzr -DCUTLASS_ENABLE_CUBLAS:STRING=ON -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=TRUE -DCMAKE_C_COMPILER:FILEPATH=/usr/bin/gcc -DCMAKE_CXX_COMPILER:FILEPATH=/usr/bin/g++ --no-warn-unused-cli -S/home/manish_magic_dev/repos/cutlass/cutlass_tree_1/cutlass -B/home/manish_magic_dev/repos/cutlass/cutlass_tree_1/build

Profiler run command

beta=1, 2x2x1 cluster hangs
./tools/profiler/cutlass_profiler --kernels=cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_bf16_bf16_128x128x64_2x2x1_0_tnn_align8_warpspecialized_pingpong_epi_tma --beta=1
beta=0, 2x2x1 cluster does not hang
./tools/profiler/cutlass_profiler --kernels=cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_bf16_bf16_128x128x64_2x2x1_0_tnn_align8_warpspecialized_pingpong_epi_tma --beta=0
beta=1, 1x2x1 cluster hangs hangs
./tools/profiler/cutlass_profiler --kernels=cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_bf16_bf16_128x128x64_1x2x1_0_tnn_alig
n8_warpspecialized_pingpong_epi_tma --beta=1
beta=1, 1x1x1 cluster does not hang
./tools/profiler/cutlass_profiler --kernels=cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_bf16_bf16_128x128x64_1x1x1_0_tnn_align8_warpspecialized_pingpong_epi_tma --beta=1

Conclusions: Pingpong schedule with, non-zero beta, non-unity cluster shape (TMA-multicast) hangs.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the cutlass_profiler entry point and reproduce the named BF16 GEMM kernel using the supplied beta and cluster-shape commands. Compare beta=1 against beta=0 and compare the 2x2x1 and 1x1x1 configurations to isolate the hang. Done means the pingpong schedule completes for non-zero beta with non-unity cluster shapes.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
hpc, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.