NVIDIA / Fuser

A Fusion Code Generator for NVIDIA GPUs (commonly known as "nvFuser")

NVIDIA/Fuser Issues

`RuntimeError: Stride mismatch with contiguity info.`
Closed a month ago7
Flipping declaration order of matmul inputs results in scheduling error
Updated a month ago
group norm segmented into pointwise + persistent + pointwise
Updated a month ago9
Check whether JAX exists for `pytest_ops.py` tests and give warning instead of failing if JAX is not installed
Updated a month ago
Segfault on FusionDefinition().define_scalar()
Updated a month ago
Almost Exact graph does not map non-size-one broadcast domains
Closed a month ago3
RuntimeError: producer->getMemoryType() == MemoryType::Global
Updated a month ago1
matmul expreval cannot handle output stride order
Updated a month ago
`allocation order inference` should be a hint rather than a requirement.
Updated a month ago
failure in lightning-thunder ci
Closed a month ago3
Confusing extra epilogue casting loop between lds and stg
Closed a month ago2
Data-driven heuristics
Updated a month ago
Factor out validation from Expr subclass constructors
Updated a month ago
Merging IterDomains requires that their iteration types match.
Updated a month ago7
Tracking outer reduction CPP benchmark regressions
Updated a month ago2
`getExprsBetween` is not getting expressions between vals
Updated a month ago5
Invalid input domain extent: 0
Closed a month ago1
CUDA NVRTC compile error in `fd.ops.add`
Closed a month ago1
[obsolete issues]
Closed a month ago
Task formalism in our IR
Updated a month ago4
DIDx aware ATen evaluation for matmul and linear.
Updated a month ago6
pad propagation and replay issues.
Updated a month ago3
Squeezed IterDomain ?S536{1} must concretize to IterType::Broadcast but found ?S536{1}.
Closed a month ago7
Write a sharded transformer block in nvFuser API.
Updated 2 months ago21
Implement User Scheduling in Python Frontend
Updated 2 months ago2
Investigate a new SDPA IR node
Updated 2 months ago5
[Python Benchmarks] Add a mode to selectively run a few inputs for functionality check
Updated 2 months ago
CpAsync with shared memory predicate is not permitted
Updated 2 months ago
Opportunistically encourage uniform data path for some integer scalars
Updated 2 months ago3
Invalid loop promotion due to propagation of incomplete promotion
Closed 2 months ago
NoOp scheduler claims zero-dimensional non-trivial fusions
Closed 2 months ago1
IdModel: Self mapping
Updated 2 months ago
IdModel buildAllGraphs() fails in presegment pass
Updated 2 months ago7
Failure to find the right loop promotion with rfactor domains
Closed 2 months ago1
CatOp is not preceded by PadOp
Closed 2 months ago3
fp8 recipe support with full reduction
Updated 2 months ago
allocation order inference creating issue in multidevice_matmul tests
Updated 2 months ago
Cannot define an `MmaOp` where batch dimension is `Broadcast`
Updated 2 months ago1
Benchmark failure in lit-gpt falcon-7b model
Updated 2 months ago1
bank conflicts due to L1 cache
Updated 2 months ago
Allow reductions in matmul epilogues
Updated 2 months ago3
Five distributed tests are failing on one GPU.
Updated 2 months ago1
allocation domain propagation to support reshape operations
Updated 2 months ago
Initial top-down presegmentation phase
Updated 2 months ago2
Reduction scheduler does not handle allocation domain properly and trigger assert when reduction output has specified allocation domain
Updated 2 months ago21
add a inner_outer_reduction scheduler
Updated 2 months ago
Support different cp.async and load stages for operands A and B
Updated 2 months ago
What is the benefit of using nanosleep in grid_sync?
Closed 3 months ago2
allocation order propagation for matmul/linear
Updated 3 months ago4
Add Double-Buffer TMA Support for Pointwise and Normalization Kernels
Updated 3 months ago