triton-lang/triton issues and pull requests

#4736 - Racecheck Bug when tl.min used with tl.sum

Issue - State: open - Opened by thumbe3 2 months ago - 5 comments

#4734 - Refactor compiler specializations to consider backend

Pull Request - State: closed - Opened by giuseros 2 months ago - 7 comments

#4728 - Segmentation fault occurred in debug mode

Issue - State: closed - Opened by kuaizilh 2 months ago - 1 comment

#4727 - Assertion error from linear layouts

Issue - State: open - Opened by peterbell10 2 months ago - 6 comments

#4722 - Refactor the C code template in `third_party/nvidia/backend/driver.py`

Pull Request - State: closed - Opened by sfzhu93 2 months ago - 5 comments

#4721 - [SWP] When num_stages = 2, do not pipeline indirect loads

Pull Request - State: closed - Opened by manman-ren 2 months ago - 4 comments

#4717 - atomic_add slows down attention backwards due to layout conversions

Issue - State: closed - Opened by bertmaher 2 months ago - 8 comments

#4716 - [AMD] Add buffer support

Pull Request - State: open - Opened by giuseros 2 months ago - 17 comments

#4698 - [Frontend] [BC breaking] Always follow C semantics on %

Pull Request - State: open - Opened by lezcano 2 months ago - 4 comments

#4689 - [SWP] add unit test for two performance warnings

Pull Request - State: closed - Opened by sfzhu93 2 months ago - 4 comments

#4686 - [tools/triton-tensor-layout] Allow parsing ttgir files with triton_nvidia_gpu ops

Pull Request - State: closed - Opened by bertmaher 2 months ago

#4684 - The ptx code do not use the cp.async future

Issue - State: closed - Opened by l1351868270 2 months ago - 2 comments

#4683 - [Frontend] Add TRITON_FRONT_END_DEBUGGING for easier debugging of frontend errors

Pull Request - State: closed - Opened by peterbell10 2 months ago - 1 comment

#4674 - [PROTON][Experimental] Initialize instruction sampling support for NVIDIA GPUs

Pull Request - State: closed - Opened by Jokeren 2 months ago - 1 comment

#4672 - [Bug]Why the ptr type of tt.atomic_rmw don't allow TT_TensorPtr?

Issue - State: open - Opened by tfruan2000 2 months ago - 6 comments

#4669 - Don't set target machine in LLVM pass builder when using LLVM IR level plugins

Pull Request - State: open - Opened by CRobeck 2 months ago - 1 comment

#4668 - [RUNTIME] Add a global metadata hook

Pull Request - State: open - Opened by Jokeren 2 months ago

#4667 - [Frontend] Improve the error when doing tensor[uint32] * -3

Pull Request - State: closed - Opened by lezcano 2 months ago

#4666 - [AMD] Bring back hoist Q out of the loop FA optimization

Pull Request - State: open - Opened by oplavsic 2 months ago - 2 comments

#4665 - [AMD] Default to stream-pipeline-v2

Pull Request - State: closed - Opened by sjw36 2 months ago - 1 comment

#4664 - 3-d tl.dot for sparse matmul failed to compile when lowering to LLVM-IR

Issue - State: open - Opened by yetiansh 2 months ago

#4663 - set warp synchronous true for reducing RankedTensorType without layout

Pull Request - State: closed - Opened by quintinwang5 2 months ago

#4662 - [FRONTEND] Adding unroll loops count to tl.range for scf for

Pull Request - State: open - Opened by plotfi 2 months ago - 2 comments

#4661 - Workaround for MLIR_ENABLE_DUMP being ignored

Pull Request - State: closed - Opened by kapilsh 2 months ago

#4660 - [frontend] Warn on usage of fp8e4b15 on Hopper rather than error

Pull Request - State: closed - Opened by lezcano 2 months ago

#4659 - [Pipeliner] Properly fail instead of assert if cannot predicate op

Pull Request - State: closed - Opened by sjw36 2 months ago

#4658 - [NFC] Use `const auto&` instead of `auto` in `ConvertLayoutOpToLLVM.cpp` and `RewriteTensorPointer.cpp`

Pull Request - State: closed - Opened by anmyachev 2 months ago

#4657 - [NFC] Use `const auto&` instead of `auto` in `bin/triton-tensor-layout.cpp` file

Pull Request - State: closed - Opened by anmyachev 2 months ago

#4656 - [NFC] Remove dead code for `hex==True` branch in `getFormatSubstr` function

Pull Request - State: closed - Opened by anmyachev 2 months ago - 1 comment

#4655 - Pass the target machine to the LLVM pass builder

Pull Request - State: closed - Opened by giuseros 2 months ago

#4654 - [BACKEND] Reduce shared memory usage when pipelining multiple TMA stores

Pull Request - State: closed - Opened by ThomasRaoux 2 months ago

#4653 - [BACKEND] Relax layout supported by SplitOp

Pull Request - State: closed - Opened by ThomasRaoux 2 months ago

#4652 - LLVM ERROR: mma16816 data type not supported when invoking `tl.dot` with dequantized tensor

Issue - State: open - Opened by shadowpa0327 2 months ago - 1 comment

#4651 - [AMD] Disable LDS optimization for tensors of rank=1

Pull Request - State: closed - Opened by sjw36 2 months ago - 6 comments

#4650 - [AMD] Enable dynamic peeling for stream-pipeliner

Pull Request - State: closed - Opened by sjw36 2 months ago

#4649 - Fix test_reduce1d test with numpy-2.0

Pull Request - State: closed - Opened by Retribution98 2 months ago

#4648 - [Tutorial] Add device side tensormap update to persistent matmul tutorial

Pull Request - State: closed - Opened by peterbell10 2 months ago

#4647 - [PROTON] Improve user experience on the CUPTI backend

Pull Request - State: closed - Opened by Jokeren 2 months ago

#4646 - [Frontend] Warn on implicit casting in the condition in tl.where

Pull Request - State: closed - Opened by lezcano 2 months ago - 1 comment

#4645 - [BACKEND] Add a loop unroller pass

Pull Request - State: open - Opened by htyu 2 months ago - 7 comments

#4644 - [TUTORIALS] fix improper type checks

Pull Request - State: closed - Opened by htyu 2 months ago

#4643 - [TEST] Use `device` fixture for `assert_helper.py` and `print_helper.py`

Pull Request - State: closed - Opened by anmyachev 2 months ago - 1 comment

#4642 - Build fails for Grace Hopper system

Issue - State: closed - Opened by wahabk 2 months ago - 8 comments

#4641 - Use `device` fixture for `runtime/test_cache.py` and `runtime/test_launch.py`

Pull Request - State: closed - Opened by anmyachev 2 months ago

#4640 - [BACKEND] Update llvm to llvm/llvm-project@c08c6a71cfc5

Pull Request - State: closed - Opened by htyu 2 months ago

#4639 - [BACKEND] Update llvm to llvm/llvm-project@c08c6a71cfc5

Pull Request - State: closed - Opened by htyu 2 months ago - 8 comments

#4638 - Enable MaskedLoad and pointer canonicalization pass

Pull Request - State: open - Opened by giuseros 2 months ago - 6 comments

#4637 - Build LLVMAarch64CodeGen if CMAKE_OSX_ARCHITECTURES is arm64.

Pull Request - State: closed - Opened by jlebar 2 months ago

#4636 - [doc] Elaborate param axis for reduce ops

Pull Request - State: closed - Opened by ByronHsu 2 months ago

#4635 - [DOC] Add pybind11 to build-time dependencies

Pull Request - State: closed - Opened by Jokeren 2 months ago

#4634 - [TEST] Majority of `test_bitwise_op` cases don't generate any artifacts

Issue - State: open - Opened by minjang 2 months ago

#4633 - [Frontend][Backend] Add device-side tma descriptor update API

Pull Request - State: closed - Opened by peterbell10 2 months ago - 3 comments

#4632 - [BUILD] Prepare for future CUDA updates using more flexible configurations

Pull Request - State: closed - Opened by Jokeren 3 months ago

#4631 - Revert "[AMD] Disable block merging to avoid block argument explosion (#4176)"

Pull Request - State: closed - Opened by joviliast 3 months ago - 4 comments

#4630 - Verify the `tt.dot` operation thru the dot verification interface of the dialect which defines the C layout.

Pull Request - State: closed - Opened by chengjunlu 3 months ago

#4628 - How can I compile and use the triton-opt tool or triton-llvm-opt tool separately?

Issue - State: open - Opened by Mark3123 3 months ago - 1 comment

#4627 - [BACKEND] Linear Layout with `stmatrix` part 1: support `stmatrix` for `convert_layout` ops

Pull Request - State: closed - Opened by Jokeren 3 months ago - 2 comments

#4626 - Add the option to configure FTZ

Pull Request - State: open - Opened by cchan 3 months ago

#4625 - [BACKEND] Update llvm to llvm/llvm-project@56152fa377

Pull Request - State: closed - Opened by antiagainst 3 months ago

#4624 - Revert "[BACKEND] Update llvm to llvm/llvm-project@f142f8afe2 (#4619)"

Pull Request - State: closed - Opened by antiagainst 3 months ago - 1 comment

#4622 - [NFC]simplified spmd ops to llvm pass.

Pull Request - State: closed - Opened by linuxlonelyeagle 3 months ago - 5 comments

#4621 - [Frontend] Recursively include triton.language modules in triton_key()

Pull Request - State: closed - Opened by peterbell10 3 months ago

#4620 - [BACKEND] Update llvm to llvm/llvm-project@f142f8afe2

Pull Request - State: closed - Opened by antiagainst 3 months ago

#4619 - [BACKEND] Update llvm to llvm/llvm-project@f142f8afe2

Pull Request - State: closed - Opened by antiagainst 3 months ago - 1 comment

#4618 - [SWP] Attempt to move all scheduling logic to a scheduling pass

Pull Request - State: closed - Opened by manman-ren 3 months ago

#4617 - Allow fallback to legacy layout for DotOp layout

Pull Request - State: closed - Opened by whitneywhtsang 3 months ago - 1 comment

#4616 - Remove `test_xpu_backend.py`

Pull Request - State: closed - Opened by whitneywhtsang 3 months ago

#4615 - [NFC]simplify SPMDOpToLLVM pattern.

Pull Request - State: closed - Opened by linuxlonelyeagle 3 months ago

#4614 - Update maintainers' list

Pull Request - State: closed - Opened by ThomasRaoux 3 months ago

#4613 - [Frontend] [BC breaking] Implement PyTorch/JAX/NumPy 2.0 typecast semantics for scalars

Pull Request - State: open - Opened by lezcano 3 months ago - 3 comments

#4612 - Create Membar_update.c++

Pull Request - State: closed - Opened by 1proprogrammerchant 3 months ago

#4611 - Create allocation_improvement.cpp

Pull Request - State: closed - Opened by 1proprogrammerchant 3 months ago

#4610 - update

Pull Request - State: closed - Opened by 1proprogrammerchant 3 months ago

#4609 - when I use cmake/make to compile the Triton source code, I encountered the error "cannot find -lNVGPUIR: No such file or directory."

Issue - State: open - Opened by Mark3123 3 months ago

#4608 - Move `PTXAsmFormatTest` into `third_party/nvidia/unittest` folder

Pull Request - State: closed - Opened by anmyachev 3 months ago - 1 comment

#4607 - [BACKEND] Remove barrier invalidate op

Pull Request - State: closed - Opened by ThomasRaoux 3 months ago

#4606 - Fix mantissa width for fp64

Pull Request - State: closed - Opened by int3 3 months ago

#4605 - Don't eagerly import all backends from tl.extra

Pull Request - State: closed - Opened by int3 3 months ago - 2 comments

#4604 - [frontend] Improve error messages about the support of fp8 types

Pull Request - State: closed - Opened by lezcano 3 months ago - 6 comments

#4603 - 3xTF32 precision issues

Issue - State: closed - Opened by alexsamardzic 3 months ago - 7 comments

#4602 - About the neccessity of num_warps

Issue - State: closed - Opened by Edenzzzz 3 months ago - 3 comments

#4601 - [BACKEND] Update LLVM to llvm/llvm-project@56152fa37730

Pull Request - State: closed - Opened by joviliast 3 months ago - 2 comments

#4600 - [BACKEND] Update LLVM to lvm/llvm-project@441b672bbdc6

Pull Request - State: closed - Opened by joviliast 3 months ago

#4599 - [testing] Complete test_const with the rest of the failing tests.

Pull Request - State: closed - Opened by lezcano 3 months ago - 1 comment

#4598 - [RFC] To support the lowering of nested layout "#slice->#dot->#mma"

Pull Request - State: closed - Opened by chengjunlu 3 months ago - 1 comment

#4597 - [backend][nvidia] Fix shift exponent is too large ASAN error.

Pull Request - State: closed - Opened by chsigg 3 months ago

#4596 - [backend][fp8] Float8E4M3FNUZ -> Float8E4M3FN for NVIDIA PTX

Pull Request - State: closed - Opened by chsigg 3 months ago

#4595 - Windows 11 cannot install Triton

Issue - State: open - Opened by moonthr 3 months ago - 2 comments

#4594 - [WIP] [AMD] Emit AMD specific intrinsics for dot

Pull Request - State: open - Opened by binarman 3 months ago - 1 comment

#4593 - [AMD] Reorder convert_layout after local_dealloc

Pull Request - State: closed - Opened by jungpark-mlir 3 months ago - 1 comment

#4592 - [SWP] add LoopScheduling pass and move scheduling heuristics from SWP to the new pass

Pull Request - State: open - Opened by manman-ren 3 months ago

#4591 - fused_moe SIGSEGV

Issue - State: open - Opened by nivibilla 3 months ago

#4590 - fused_moe

Issue - State: closed - Opened by nivibilla 3 months ago

#4589 - [frontend] added overflow checks in `debug` mode

Pull Request - State: closed - Opened by ptillet 3 months ago

#4588 - [CODEGEN] Support CUDA 12.6

Pull Request - State: closed - Opened by Jokeren 3 months ago - 3 comments

#4587 - [FRONTEND] cleaner handling of DEBUG=1

Pull Request - State: closed - Opened by ptillet 3 months ago - 1 comment

#4586 - [Tutorial] Add parameter shape check in 01 Vector Addition

Pull Request - State: closed - Opened by Qi-Zhan 3 months ago - 1 comment

#4585 - The number of generated TTIR files

Issue - State: closed - Opened by LHW-CLOUD 3 months ago - 1 comment

#4584 - [SWP] move schedulePrologueAndEpilogue to be before lowering

Pull Request - State: closed - Opened by manman-ren 3 months ago - 4 comments

#4583 - Cannot build triton

Issue - State: open - Opened by fireblade2534 3 months ago - 3 comments

GitHub / triton-lang/triton issues and pull requests