[BUG]strange tv layout for LdMatrix16x16x8bOp
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 10.5k
- Forks
- 2.1k
- Avg merge
- 3d 11h
- Merged PRs (30d)
- 7
Description
When building atom copy ldmatrix for 8b element, the layout doesn't seem normal. I expect it to be (32, 16): (16, 1). Since Nvidia haven't released a cuteDSL FP8 Gemm example for blackwell geforce, is it possibly a bug?
a_dtype = cutlass.Int8
atom_copy_ldmatrix_A = cute.make_copy_atom(
cute.nvgpu.warp.LdMatrix16x16x8bOp(2),
a_dtype,
)
print(f"copy_atom_op: LdMatrix16x16x8bOp")
print(f"DEBUG: atom_copy_ldmatrix_A: \n {atom_copy_ldmatrix_A}")
--------------------------------
# Output
copy_atom_op: LdMatrix16x16x8bOp
DEBUG: atom_copy_ldmatrix_A:
Copy Atom
ThrID: 32:1
TV Layout Src: ((2,2,4,2),16):((16,128,32,256),1)
TV Layout Dst: ((4,8),(1,2,2,4)):((32,1),(1,16,8,128))
Value type: i8
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the shown Python reproduction and the cute.nvgpu.warp.LdMatrix16x16x8bOp entry point. Inspect how the copy atom constructs its source and destination layouts, then compare the printed result with the expected (32, 16):(16, 1) layout. Done means determining whether the output is an implementation bug or an incorrect expectation, with a focused reproduction or fix identified.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp, python
- Domain
- hpc
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100