ROCm aotriton issues - Githubissues

ROCm / aotriton

Ahead of Time (AOT) Triton Math Library

MIT License

42 stars 15 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Misc changes and performance tuning for 0.8b release

#57 xinyazhang closed 1 week ago
0
[Queued PR] Port fixes from 0.7.2b

#56 xinyazhang closed 1 week ago
1
Restore the support of causal=True and seqlen_q != seqlen_k

#55 xinyazhang closed 1 week ago
1
[Documentation]: large bf16 inputs leads to nan

#54 xinyazhang opened 3 weeks ago
0
Add versioning support in multiple levels.

#53 xinyazhang closed 2 weeks ago
0
libaotriton_v2.so: Fix 'argument list too long" error

#52 prarit closed 1 month ago
1
Add docker based package builder and switch to system compiler

#51 xinyazhang closed 2 weeks ago
8
Kernel Storage V2

#50 xinyazhang closed 2 weeks ago
7
GQA Support

#49 xinyazhang closed 3 weeks ago
0
Code Clean Up

#48 xinyazhang closed 1 month ago
0
[Documentation]: Forward kernel returns NaN when inputs are irregular (including causal) and sm_scale is 0

#47 xinyazhang opened 1 month ago
1
Merge improvements of 0.7.1b release into main

#46 xinyazhang closed 1 month ago
1
FA Kernel Update for Accuracy and Performance

#45 xinyazhang closed 2 months ago
0
Ignore colon suffixes in gcnArchName

#44 xinyazhang closed 2 months ago
0
Fix numerical error by applying qk_scale at inner loop instead of outer loop

#43 xinyazhang closed 2 months ago
2
Add cmake option AOTRITON_NAME_SUFFIX to resolve name conflicts

#42 xinyazhang closed 2 months ago
1
Add PyTorch compatibility matrix to README.md

#41 xinyazhang closed 2 months ago
0
Support hipGraph usage in PyTorch

#40 xinyazhang closed 3 months ago
3
Improve Backward Performance and Navi31 Support

#39 xinyazhang closed 3 months ago
35
[Documentation]: The overall tuning idea of aotriton

#38 hubotao1 opened 4 months ago
6
[Perf] Is it possible that the kernels wrapped with AOT have the similar performance comparing with the original ones?

#37 xinji1 opened 4 months ago
1
Switch to upstream Triton compiler, and related changes

#36 xinyazhang closed 4 months ago
3
[Issue]: build fail: ModuleNotFoundError: No module named 'triton._C.libtriton.triton'

#35 minzhezhou closed 5 months ago
1
[Issues]: The Gap between AOT and JIT Triton on Flash Attention kernel

#34 jinsong-mao opened 5 months ago
0
Install .so if AOTRITON_NO_SHARED is OFF

#33 jithunnair-amd closed 5 months ago
0
[Issue]: failed to run the tune_flash.py

#32 jinsong-mao opened 5 months ago
13
Add varlen support to AOTriton's Flash Attention

#31 xinyazhang closed 5 months ago
3
How to run benchmark tests[Issue]:

#30 jinsong-mao closed 5 months ago
8
Refactor the build system

#29 xinyazhang closed 6 months ago
1
[Issue]: Unable to build, Unknown CMake command "pybind11_add_module"

#28 RandUser123sa closed 1 week ago
10
[Feature]: CDNA1 Support

#27 IMbackK opened 6 months ago
0
Adding mutex.h for TE pytorch extension compilation

#26 wangye805 closed 6 months ago
0
the atten_bwd_dk_dv is bad in performance on mi300x

#25 jinsong-mao closed 5 months ago
1
[mGPU] Run hipModuleLoadDataEx for each GPU device.

#24 xinyazhang closed 6 months ago
0
Resolve cmake conflicts when adding aotriton into TE via add_subdirectory

#23 wangye805 closed 6 months ago
0
Add FP32 and Bias to fulfill the functionalities required by `torch.nn.attention.SDPBackend.EFFICIENT_ATTENTION`

#22 xinyazhang closed 7 months ago
0
[Question] Autotune kernel based on `third_party/triton`

#21 xinji1 closed 7 months ago
2
<vector> is required regardless of AOTRITON_USE_ZSTD

#20 xinyazhang closed 7 months ago
1
Add new triton kernel debug_fill_dropout_rng

#19 xinyazhang closed 7 months ago
0
[Issue]: Pytorch fails to compile locally due to aotriton failing to build the hsaco objects

#18 Zakhrov closed 1 month ago
9
[Feature]: Fix the mandatory boundary_check when loading bias tensor

#17 xinyazhang opened 7 months ago
0
[Feature]: Memory Efficient Flash Attention for gfx1100 (7900xtx)

#16 supernovae closed 3 months ago
31
[Feature]: C++ version `mk_aotensor`

#15 xinji1 closed 7 months ago
2
Add matrix bias to forward/backward kernel

#14 xinyazhang closed 7 months ago
1
Switch Tuning database to SQLite3 for Incremental Tuning

#13 xinyazhang closed 7 months ago
0
[Documentation]: Shall we modify the configurations in `v2python` for the other kernels?

#12 xinji1 opened 8 months ago
5
[Issue]: Release Tags

#11 trixirt closed 8 months ago
2
Fix the performance regression introduced during support of irregular shapes.

#10 xinyazhang closed 8 months ago
0
Update README.md

#9 groenenboomj closed 8 months ago
1
Add strides to all input tensors

#8 xinyazhang closed 9 months ago
0