NVIDIA Fuser issues - Githubissues

NVIDIA / Fuser

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

Other

271 stars 53 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

[WIP][DO NOT REVIEW] Enable slice in vectorization analysis

#3457 jjsjann123 opened 36 minutes ago
1
Support 2D inner reduction scheduler with autotuning

#3456 rdspring1 opened 23 hours ago
1
Invalid indexing path when resize is used with a residual path

#3455 naoyam opened 23 hours ago
5
[WIP] (Yet another) indexing war for resize

#3454 naoyam opened 1 day ago
3
Restructure baseline benchmarks

#3453 Priya2698 opened 1 day ago
1
Return a bool indicating if all nodes are visited

#3452 naoyam closed 1 day ago
2
Skip sequence parallel test when D=1

#3451 cowanmeg closed 1 day ago
1
Cleanup `currentLoadStage`

#3450 zasdfgbnm closed 1 day ago
1
Add `ParallelDimensionMap::getNumThreadsEachBlock`

#3449 zasdfgbnm closed 1 day ago
1
Split `insertMBarrierWaitBeforeFirstRead`

#3448 zasdfgbnm closed 1 day ago
2
Stride `MatmulOp` according to set allocation domain

#3447 Priya2698 opened 2 days ago
0
Use mbarrier for WAR for circular buffering

#3446 zasdfgbnm opened 2 days ago
3
Fix stride-order based allocation domain computation when output has reduction axis

#3445 Priya2698 opened 2 days ago
2
Unshard tensor sizes before binding.

#3444 wujingyue opened 2 days ago
0
Fix gdimz in LaunchParams

#3443 rdspring1 closed 2 days ago
1
kill `onlyOneSerialForLoopOnStack`

#3442 zasdfgbnm closed 2 days ago
3
Remove `getStageDepthFor`, `getPrefetchDistanceFor`, `circularBufferDepth`, `circularBufferPrefetchDistance`

#3441 zasdfgbnm closed 2 days ago
1
[WIP] Enable translation of Hopper matmuls

#3440 jacobhinkle opened 2 days ago
0
Patch vectorization on permuted inputs for PadOp

#3439 jjsjann123 closed 1 day ago
10
fix race in async copy

#3438 liqiangxl opened 2 days ago
10
Variable renaming in AllocationInserter

#3437 zasdfgbnm closed 2 days ago
2
Avoid canScheduleCompileTime for dynamic shape check

#3436 jacobhinkle closed 1 day ago
4
Add `ir_utils::createRangeLoop`

#3435 zasdfgbnm closed 3 days ago
1
Minor cleanup in `CloneTmaCircularBufferLoopAndInsertSync`

#3434 zasdfgbnm closed 2 days ago
2
Add interface to directly work with CircularBufferOptions

#3433 zasdfgbnm closed 3 days ago
1
Fix the setting of `gdimx` in 2d and 3d inner reduction heuristics

#3432 rdspring1 closed 1 day ago
1
Add ReductionParams to python frontend

#3431 rdspring1 closed 2 days ago
3
Allocate index for all circular buffer stages

#3430 zasdfgbnm closed 2 days ago
3
Fix comment double buffer -> circular buffer

#3429 zasdfgbnm closed 3 days ago
2
Race reported between Write access and Read access in fusion using async copy

#3428 liqiangxl opened 3 days ago
2
thread predicate is missing in async copy

#3427 liqiangxl opened 4 days ago
0
Minor cleanups and improvements

#3426 wujingyue closed 3 days ago
1
[WIP] Resize scheduler

#3425 naoyam opened 5 days ago
0
Remove unnecessary propagation that's already done by the ops API.

#3424 wujingyue closed 1 day ago
4
use bdimy = 1 to WAR smem race

#3423 liqiangxl closed 1 day ago
8
Remove an unused variable

#3422 wujingyue closed 3 days ago
1
Extend isResharding to allow DID loop split.

#3421 wujingyue opened 5 days ago
8
Print spaces for readability.

#3420 wujingyue closed 3 days ago
2
Dynamic shape host latency is slow

#3419 jacobhinkle closed 1 day ago
2
Add `unsafeBind` to `LaunchParams` to allow modification.

#3418 rdspring1 closed 5 days ago
2
Remove unnecessary cast

#3417 cowanmeg closed 5 days ago
1
Allow inlining past loop broadcasts

#3416 jacobhinkle opened 6 days ago
7
set maxrregcount in outer reduction heuristic

#3415 liqiangxl closed 3 days ago
3
Only check actually used IDs in predicate elimination

#3414 jacobhinkle opened 6 days ago
18
avoid calc reg count from block size in getMaxRegCount

#3413 liqiangxl closed 6 days ago
2
Fix a typo

#3412 wujingyue closed 6 days ago
1
Add support for stmatrix in the unit test HopperMatmulTest/HSH_NT_128BSwizzle

#3411 protonu opened 6 days ago
1
Preserve additional IDs when mutating TensorDomains in OptOutMutator

#3410 jacobhinkle closed 5 days ago
2
Additional IDs are lost when mutating a TensorDomain

#3409 jacobhinkle closed 5 days ago
0
Clean-up: reduction producer IterDomains aren't mapped anyway.

#3408 wujingyue closed 6 days ago
1