flagos-ai/FlagTree

View on GitHub

FlagTree is a unified compiler supporting multiple AI chip backends for custom Deep Learning operations, which is forked from triton-lang/triton.

Stars: 353
Forks: 152
Open beginner issues: 2
Indexed issues: 60
Avg merge: 2d 22h
Merged PRs (30d): 92

Dominant language: Python
License: MIT
Last GitHub push: Sep 20, 2026
Latest indexed: Sep 16, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

55 open issues indexed

[Roadmap][DRAFT] FlagTree on AMD Radeon — 26Q4 / 27Q1 Open

Difficulty 5/5 Over a week Newbie friendliness 18/100

flagos-ai/FlagTree#1256 · 1 comment · Opened Sep 21, 2026
[BUG][xpu] TRITONXPU_COMPARE_FUSION=1 时 cmpf+masked store 在 tile≥4096 编译崩溃（triton_xpu.cmpf result 类型被判成 ui32） Open

bug flagos2.2-rc2 P1

flagos-ai/FlagTree#1249 · 1 comment · 1 assignee · Opened Sep 21, 2026
[BUG][xpu] tl.exp2 被降级成 exp（静默错结果·比值 1/ln2），且 block≥1024 时 triton_xpu.unpack 编译崩溃 Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1248 · 2 comments · Opened Sep 21, 2026
[BUG][Ascend] Lost stores: masked store skips 2xfp32 vector groups nondeterministically Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1245 · 1 comment · Opened Sep 20, 2026
[BUG][Ascend] TritonToLinalgIncubated hits llvm_unreachable at TritonToLinalgPass.cpp:440 on a kernel with a dead block containing an early return Open

Difficulty 4/5 3-5 days Newbie friendliness 58/100

flagos-ai/FlagTree#1244 · Opened Sep 20, 2026
[BUG][Ascend]TritonToLinalgIncubated: unconditional cast<RankedTensorType> fails on a non-ranked-tensor type Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1243 · Opened Sep 20, 2026
[BUG][Ascend] TritonToLinalgIncubated leaves a live memref→tensor materialization and aborts with a dangling OpOperand assertion Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1242 · Opened Sep 20, 2026
[BUG][Ascend] TritonToLinalgIncubated asserts "Cannot destroy a value that still has uses!" (BlockOperand) on valid Triton kernel with nested opaque control flow Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

flagos-ai/FlagTree#1241 · Opened Sep 20, 2026
[BUG][Ascend] PlanMemory pass crashes with "buffer memory has been released" Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1237 · Opened Sep 20, 2026
[BUG][iluvatar] make_ttgir crash in TritonILUVATARGPUSmeLoad: tt.expand_dims layout conflict with SME mask layout Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1235 · Opened Sep 20, 2026
[BUG][nvidia] 0.7.0rc2 TLE import dlopens libflagcx.so linked against libcudart.so.12 — breaks 'import flag_gems' on CUDA 13 Open

bug flagos2.2-rc2 nvidia P1 tle

Difficulty 4/5 3-5 days Newbie friendliness 52/100

flagos-ai/FlagTree#1234 · 1 assignee · Opened Sep 20, 2026
[BUG][enflame] 0.7.0rc2 emits 'enable_i64=true' into --convert-gpu-to-gcu — rejected by the tops1.9.10 GCU toolkit ('no such option enable_i64') Open

bug enflame flagos2.2-rc2 P1

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1233 · 1 assignee · Opened Sep 20, 2026
[BUG][metax] 0.7.0rc2 ICE on tl.dot with BLOCK_SIZE_M=8 — MACAMmaEncodingAttr assertion 'tn and tk not meet conditon' (0.6.1 compiles fine) Open

bug flagos2.2-rc2 metax P0

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1232 · 1 assignee · Opened Sep 20, 2026
[HCU/gfx936] fp64 tl.dot silently falls back to FMA — please add upstream support for the fp64 matrix unit (v_mmac_16x16x4_f64) Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

flagos-ai/FlagTree#1231 · Opened Sep 20, 2026
[BUILD] Ascend wheel does not declare cann-shmem, so pip install succeeds and import fails later Open

Difficulty 4/5 3-5 days Newbie friendliness 64/100

flagos-ai/FlagTree#1229 · Opened Sep 20, 2026
[BUG][iluvatar] output corrupted when a redundant runtime branch wraps a load inside the loop Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1228 · 1 comment · Opened Sep 19, 2026
[BUG][iluvatar] tl.assume inside a dead `if` branch makes the kernel drop its final atomic_add Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1227 · Opened Sep 19, 2026
[iluvatar3.6] flagtree 0.7.0rc2+iluvatar3.6 的 libtriton.so 依赖 GLIBC_2.38，在官方 Ubuntu22.04(glibc2.35)镜像中无法导入 Open

bug flagos2.2-rc2 P0

flagos-ai/FlagTree#1225 · 1 assignee · Opened Sep 19, 2026
[BUG][mthreads]软件流水线在循环内含运行时条件规约/store 块时静默算错（num_stages=3 触发，改 1 恢复） Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

flagos-ai/FlagTree#1223 · Opened Sep 18, 2026
[BUG][mthreads] 同一 kernel、同一输入，仅改变编译选项 num_stages 与 num_warps得到不同结果 Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1222 · Opened Sep 18, 2026
[BUG][mthreads][ILUVATAR] 前端解析崩溃：嵌套 @triton.jit helper 接收调用方局部 tl.constexpr 时触发 region isolation（'tt.splat' op using value defined outside the region） Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1221 · 1 comment · Opened Sep 18, 2026
[Ascend] _topk_topp_kernel fails to compile — ConvertTritonIRToLinalgIR PassManager::run failed Open

bug flagos2.2-rc2 P1

flagos-ai/FlagTree#1220 · 1 comment · 1 assignee · Opened Sep 18, 2026
[BUG][xpu] do_bench (event 模式) estimate_ms==0 时除零（昆仑芯 Event.elapsed_time 恒 0） Open

bug flagos2.2-rc2 P1

flagos-ai/FlagTree#1218 · 1 assignee · Opened Sep 18, 2026
[BUG][Ascend] Launcher workspace size overflows in 32-bit multiplication before uint64_t assignment Open

Difficulty 3/5 1-2 days Newbie friendliness 78/100

flagos-ai/FlagTree#1194 · 2 comments · Opened Sep 16, 2026
[MThreads][TLE] Reusing a TME descriptor across K iterations crashes llc: cannot allocate physical SH register for TME operand Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1183 · 1 comment · Opened Sep 16, 2026
Improve integer divisibility specialization to enable matmul `cp.async` pipelining Open

Difficulty 4/5 3-5 days Newbie friendliness 72/100

flagos-ai/FlagTree#1182 · Opened Sep 16, 2026
[BUG] Incorrect matmul results with num_warps=8: missing cross-warp-group barrier after WGMMA wait Open

Difficulty 3/5 1-2 days Newbie friendliness 75/100

flagos-ai/FlagTree#1180 · Opened Sep 16, 2026
[Bug][metax][TLE] smem support is incomplete: scalar-indexed `local_ptr` asserts at LoadStoreOpToLLVM.cpp:427, smem atomics are unsupported, and smem+histogram silently corrupts data Open

metax

Difficulty 5/5 Over a week Newbie friendliness 35/100

flagos-ai/FlagTree#1152 · Opened Sep 11, 2026
[iluvatar] runtime path imports pytest: triton/spec/iluvatar/triton/testing.py::nvsmi reaches triton._internal_testing Open

Difficulty 2/5 1-3 hours Newbie friendliness 84/100

flagos-ai/FlagTree#1142 · Opened Sep 10, 2026
[BUG][ascend] uint16 and uint32 `>>` are lowered as arithmetic shifts, silently corrupting every value with the top bit set Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1121 · 3 comments · Opened Sep 7, 2026
【FR】Bug: FlagTree Ascend backend fails to lower vLLM apply_top_k_top_p_triton (TritonIR→LinalgIR, () → tensor<1xf32> cannot be legalized) Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1117 · Opened Sep 7, 2026
[DOC][TLE.DSA][ascend] tle.dsa.local_ptr is documented in tle.md but absent from every release Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

flagos-ai/FlagTree#1111 · 1 comment · Opened Sep 7, 2026
[Bugs][Iluvatar] vllm 在安装了tree的环境中启动时 triton报错了 Open

iluvatar

Difficulty 4/5 3-5 days Newbie friendliness 52/100

flagos-ai/FlagTree#1093 · 1 comment · Opened Sep 3, 2026
Help, please! Source code compile. Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

flagos-ai/FlagTree#1088 · 2 comments · Opened Sep 2, 2026
[BUILD] Release tags never trigger build-deb/build-rpm: tags are pushed with GITHUB_TOKEN by the delivery workflows Open

Difficulty 3/5 1-2 days Newbie friendliness 75/100

flagos-ai/FlagTree#1062 · Opened Aug 28, 2026
[KunlunXin] repeat_interleave_self_int 在总元素数超过 4M 时产生错误结果 Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1057 · Opened Aug 26, 2026
[XPU] tl.store 在向量计算之后写入标量时，实际写入的是过期的寄存器值 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

flagos-ai/FlagTree#1056 · Opened Aug 26, 2026
[BUG][hygon] Scalar atomic_add(..., 0) poll in a while-loop causing cross-block spin-barrier deadlock on hcu Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

flagos-ai/FlagTree#1051 · Opened Aug 26, 2026
[BUG][TLE][PPU] PPU0010 async AIU load rejects i32 operands and lacks B32 lowering Open

Difficulty 4/5 3-5 days Newbie friendliness 42/100

flagos-ai/FlagTree#1050 · Opened Aug 25, 2026
Ascend backend: decouple torch_npu so PrivateUse1 plugins can use it Open

ascend

Difficulty 5/5 Over a week Newbie friendliness 35/100

flagos-ai/FlagTree#1046 · Opened Aug 25, 2026
[BUG][hygon] tl.load(..., volatile=True) causing cross-block spin-barrier deadlock on hygon Open

hcu

Difficulty 4/5 3-5 days Newbie friendliness 45/100

flagos-ai/FlagTree#1030 · 1 comment · Opened Aug 21, 2026
[BUG][ppu] tl.atomic_add(..., sem="acquire") causing cross-block spin-barrier deadlock on ppu Open

ppu

Difficulty 4/5 3-5 days Newbie friendliness 48/100

flagos-ai/FlagTree#1029 · 2 comments · Opened Aug 21, 2026
fix(nvidia): nvidia driver is_active() fires on Ascend NPU under torch_npu TORCH_TRANSFER_TO_NPU=1 Open

ascend

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

flagos-ai/FlagTree#1022 · 1 comment · Opened Aug 21, 2026
[BUG][TLE.DSA][ascend] wrong result when replace tle.dsa.extract_slice with tle.dsa.extract_element Open

ascend

Difficulty 4/5 3-5 days Newbie friendliness 64/100

flagos-ai/FlagTree#1012 · 3 comments · Opened Aug 19, 2026
[BUG][mthreads]kernel crashes on device illegal memory access + core dump Open

mthreads

Difficulty 5/5 Over a week Newbie friendliness 35/100

flagos-ai/FlagTree#1010 · Opened Aug 18, 2026
[BUG][ppu] fp8e4nv is gated by an NVIDIA compute-capability threshold, at both the frontend and in ConvertTritonGPUToLLVMPPU Open

ppu

Difficulty 4/5 3-5 days Newbie friendliness 52/100

flagos-ai/FlagTree#1005 · Opened Aug 18, 2026
[Bug][npu] not supported float8_e4m3fn on ascend npu Open

ascend

Difficulty 4/5 3-5 days Newbie friendliness 55/100

flagos-ai/FlagTree#987 · 1 comment · Opened Aug 14, 2026
[BUG][mthreads] llc non-termination on giant IR: FFT N=250 naive DFT baseline compile hangs >15min Open

mthreads

Difficulty 4/5 3-5 days Newbie friendliness 42/100

flagos-ai/FlagTree#984 · Opened Aug 13, 2026
[Bug][ppu] not supported float8_e4m3fn on ppu backend Open

ppu

Difficulty 3/5 1-2 days Newbie friendliness 68/100

flagos-ai/FlagTree#975 · Opened Aug 13, 2026
[Kunlunxin] 昆仑芯使用FlageTree Triton3.0 在测试vllm Triton attention时也会报错 Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

flagos-ai/FlagTree#935 · Opened Aug 7, 2026

flagos-ai/FlagTree

Get new issues in your inbox