NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #2350 most downloaded on PyPI
A tile level programming language to generate high performance code.
Latest stable
0.1.15
Last release 1 months ago
30 Sep 2026
Ships fairly regularly
a new release about every 4 weeks
Nearly every release is documented
notes for 22 of 22 stable releases
1 version withdrawn
withdrawn after publishing
2 years old
24 releases · first in 2025
Drag across the ruler to compare, or tap a release to open it.
T.symbolic remains available as a deprecated alias; use T.dynamic for new code (#3216).
T.gemm_blockscaled semantics with dedicated backend dispatch, improved SM100 instruction selection, and expanded SM120 fragment support.enumerate, zip, comprehensions, and generator expressions.tilelang.ascend.language and target="ascend" for Huawei Ascend 950 (dav-3510).T.SimdVF and T.SimtVF regions.tvm_ffi and Cython execution, PyTorch NPU tensors and streams, and NPU profiling.See the Ascend 950 guide for installation and usage. Ascend A2/A3 support remains in the community-maintained TileLang-Ascend projects.
TL_ENABLE_AUTO_WARP_SPECIALIZATION: "role_based"; add GEMM and FlashAttention examples (#3059, #3185).T.fma and T.fmul, and complete additional FP16/BF16 math bridges (#3134, #3132, #3163).for ... in range(...) (#3230).else clauses (#3141, #3232, #3262, #3142).T.copy / T.async_copy coalesced_width as a hint and clamp it to the achievable vector width (#3246).uint64 argument mappings (#3207, #3229).sync and group parameters from T.Pipelined, and k_pack from T.gemm_sp (#3202).T.gemm(k_pack=...) should import tilelang.rocm.language (#3203).threads= explicitly in T.Kernel (#3186).T.symbolic remains available as a deprecated alias; use T.dynamic for new code (#3216).TILELANG_CACHE_VERIFY_HASH; binary artifact hash verification is now mandatory. Legacy cache formats are rebuilt automatically (#3143, #3177).Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.14...v0.1.15
AutoSchedule to AutoWarpSpecialization by @Yongqi-Zhuo in https://github.com/tile-ai/tilelang/pull/3185Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.14...v0.1.15
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Reducer v2 (#2940, #3093, #3043, #3044, #3079, #3100): T.alloc_reducer reworked into first-class deferred reduction epochs whose physical lowering is
T.alloc_reducer reworked into first-class deferred reduction epochs whose physical lowering is planned by layout inference (via a first-class PartialFragment layout). Adds loop-scoped epochs, conditional reducer finalization, and automatic vectorization of contiguous reducer updates.T.unroll(explicit=True) for early explicit unrolling (#2859)cluster_mask on T.tma_copy (#2932)T.gemm tile dimensions with a clear message (#3113), validate T.gemm k_pack arguments (#3094), reject non-positive arrive_count in alloc_barrier/alloc_cluster_barrier (#3112), reject T.Parallel indexing of local buffers (#3041), reject break in fully expanded loops (#3078)tcgen05.alloc arenas (#2831); support half-subpartition (M=64) TMEM tiles in tcgen05.ld/st (#2880); fix ld/st segment pointer advancement in b32 columns (#2952)int4x2/uint4x2 codegen (#3036); 16-bit CUTLASS type overloads for fast-math, __ldg, and htan intrinsics (#3097, #3077, #3028, #2894)T.{reads,writes} for T.tma_{gather4,scatter4} (#3053); separate TMA atomic-add dtype support from layout encoding (#2846)tl.sync_warp on HIP (#2872), reject sub-wavefront block sizes instead of crashing (#2918), resolve versioned device properties in the HIP stub (#2919); emit #line directives for the HIP target (#3058)T.Parallel loops (#3121); scalarize Select in automatic vectorization (#3060)VerifyBufferInit, a general buffer-initialization check (#2956)#line directives from TIR spans (#3048)get_parent_locals frame self-reference leak (#2934)project() (#3102)tcgen05.alloc arenas by @Rachmanino in https://github.com/tile-ai/tilelang/pull/2831T.unroll(explicit=True) for early explicit unrolling by @Yongqi-Zhuo in https://github.com/tile-ai/tilelang/pull/2859T.Parallel indexing of local buffers by @SiriusNEO in https://github.com/tile-ai/tilelang/pull/3041T.{reads,writes} for T.tma_{gather4,scatter4} by @Yongqi-Zhuo in https://github.com/tile-ai/tilelang/pull/3053T.Parallel loops by @Yongqi-Zhuo in https://github.com/tile-ai/tilelang/pull/3121Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.13...v0.1.14
> Breaking changes: this release removes several legacy APIs and packages. See Backend, API & Refactors before upgrading.
This release contains 138 commits (79 bug fixes plus features, refactors, and examples) accumulated since v0.1.12 (2026-07-08 → 2026-08-02).
The headline work is a multi-backend language-dialect refactor that replaces the runtime-activated language facade with static per-backend re-exports, alongside two major new hardware paths: SM120 NVF4 block-scale MMA for Blackwell and Metal 4 (M5) cooperative-tensor GEMM. On top of that, a large batch of correctness fixes landed across reductions/scans, atomics, TMA/copy lowering, loop-step preservation, and FP encoding edge cases.
Breaking changes: this release removes several legacy APIs and packages. See Backend, API & Refactors before upgrading.
T.mma_gemm_blockscaled now routes the packed-scale SM120 path internally through package-pingpong lowering, with an optimized non-persistent example (8192³ measured at ~1527 TFLOPS on SM120). The public micro_pipeline strategy knob was removed from the API.T.gemm (#2252) — TileLang-owned cooperative-tensor intrinsics, Metal 4 MPP matmul2d shader emission, and a shape-aware instruction selector that keeps the simdgroup fallback for fragment accumulators and unsupported tiles.from tilelang.cuda.language import * re-export; CUDA/Metal/ROCm dialects now build on tilelang.language.common with per-backend TIR overlays (details below).T.gemm now works on older architectures instead of erroring out.__hfma (#2769).sm_100a (#2691).fp32x2 ops usable as reducers (#2637).pass_profile pass-config option (with a configurable threshold) (#2622).lower-trace support for debugging (new doc: docs/tools/lower_trace.md) (#2725).T.assume conditions are now enforced at runtime (#2655).CanProve (#2772).The runtime-activated language facade has been replaced by a static re-export architecture:
.pyi stubs + generator, py.typed, the globals()-based __all__ scraping, and _activate_cuda_facade().cuda / metal / rocm) now build on tilelang.language.common with per-backend TIR overlays.mma/wgmma/mfma macro generators are pinned to the dtypes leaf and no longer touch the half-initialized facade during bootstrap.tilelang/cuda/intrinsics/sparse_layout.py (a dtypes-only leaf).Follow-up fixes: ROCm intrinsic resolution (#2779), rng_init (#2776), and shared-intrinsic resolution across backends.
tilelang.common package removed (#2810).tilelang package (#2761).apache-tvm-ffi 0.1.12, while keeping 0.1.11 compatibility (#2795); lower bound raised to >=0.1.11 (#2736).ptxas register-usage level is cast to int before building the nvcc command (#2641).Int64Promoter extracted into a common header (#2558).actions/setup-python 6 → 7 (#2773); transformers bumped in examples/bitnet-1.58b (#2658).For nodes (#2752).LoopUnswitching (#2585).warp_reduce no longer truncates int64/uint64 to 32 bits on sm_80+ (#2782).reduce_sum over-counts on straddle layouts fixed (#2424).nan_propagate honored in reduce max/min/absmax clear=False write-back (#2788).T.atomic_max/T.atomic_min no longer silently corrupt fp32 values (#2780).return_prev supported for scalar atomic_min/atomic_max (#2672), atomic_addx2 with BufferRegion destinations (#2753), and HIP vector atomic add (#2712).T.atomic_addx4 return type guarded for sliced destinations (#2590).T.infinity supported for float8_e5m2 (#2671).T.pow/T.power fixed for constant integer exponent y <= 0 (#2677).T.__exp computes e**x, not 2**x (docstring + CuTeDSL codegen) (#2696).x2 operand dtypes rejected (#2802); floating-point predicates rejected in vote intrinsics (#2800); alloc_var initializer dtype preserved (#2801); invalid dtypes rejected in T.dp4a (#2652).T.copy path casts to the destination dtype (#2771).== +1 (#2649); vectorized Select constraint handling fixed (#052e6741).st.bulk destination emitted as a shared write to fix a missing barrier and compilation-introduced races (#2700).increase_descriptor_offset guard (#2675).T.transpose swaps only the final two axes (#2757); contracting shared-buffer layouts rejected in T.annotate_layout (#2719); unused fragment buffers allowed without layouts (#2717); shared-TMEM buffer pointer types checked before dereference (#2794).vec_type in common.h for CPU codegen (#2768).Bind modeling fixed in parallel race checks (#2665).T.serial (#2674).T.gemm rejected instead of silently producing wrong results (#2724).DataType args no longer break compilation on ROCm (#2726).T.Kernel (#2653).fragment spelling corrected (#2695); BufferStore cast warning context improved (#2733).pass_configs supported in autotuning (#2496).None (#2657).topk_selector memory-access optimization with thread coarsening — ~1.9× faster with identical results (#2659).T.mma_gemm_blockscaled (#2364).<details> <summary>138 commits (click to expand)</summary>
e0f0ac90 [CUDA] Refactor TMA atomic add layout validation
6b81bb87 [BugFix] Correctly preserve loop step when unrolling loops (#2835)
09526a27 Revert "[BugFix] Preserve loop step when unrolling loops" (#2834)
56a0f729 [BugFix] Reject unsupported TMA atomic add dtypes (#2830)
bdb769ae [Enhancement] Aggregate VerifyParallelLoop race diagnostics with span (#2806)
e01c498b [JIT] Remove legacy DLPack execution backend (#2816)
2bb0def9 [BugFix] Fix two-instance modeling in ThreadSync cross-thread race checks (#2805)
8f34abf4 [CUDA] Add SM120 NVF4 block-scale MMA support (#2364)
3e4a0544 [Carver] Remove unused shape inference module (#2813)
e18d9699 [CUDA][Reduce] Simplify scalar AllReduce thread range analysis (#2814)
50481cce [Refactor] Remove intrinsic compatibility facade (#2812)
21e8c064 [BugFix] Reject unsupported fast-math input dtypes (#2804)
0bc1913d [BugFix] Resolve partial scalar reduce barrier participation (#2777)
5b1f3218 [BugFix] Reject uncovered warp partitions in T.gemm instead of producing silently wrong results (#2724)
7fd95363 [CUDA] Extend the GEMM FMA fallback to SM75 (#2811)
dd92b781 [Refactor] Remove unused tilelang.common package (#2810)
6c3dd971 [BugFix] Skip descriptor TMA for device-bound copy bases (#2803)
e41fadbe [BugFix] Fix warp_reduce truncating int64/uint64 to 32 bits on sm_80+ (#2782)
2a06036f [BugFix] Reject mixed packed x2 operand dtypes (#2802)
32e02e6c [Fix] Refine architecture guards (#2790)
bc9515fe [BugFix] Prevent autotuner cache reuse across different outputs and validation settings (#2793)
2c84f4f9 [BugFix] Check shared-TMEM buffer pointer types before dereference (#2794)
28f70338 [Metal] Add line-level threadgroup qualifier scanning (pass 5) (#2796)
51f88a88 [BugFix] Reject floating-point predicates in vote intrinsics (#2800)
b1b605da [BugFix] Preserve alloc_var initializer dtype (#2801)
1cb4d4f3 [CUDA] Support arbitrary TMEM layouts (#2785)
4086ba8e [FFI] Support apache-tvm-ffi 0.1.12 (#2795)
7ec5adbe [BugFix] Pass buffer row stride to 2D scan kernel to fix silent miscomputation (#2620)
6171343c [TIR][Language] Add typed vector lane extraction API (#2789)
1545f006 [Metal] M5 Cooperative Tensor T.gemm (#2252)
aaf68d2e [Metal] Add 16-byte alignment padding to shared/threadgroup memory (#2786)
940b1061 [BugFix] Honor nan_propagate in reduce max/min/absmax clear=False write-back (#2788)
a42bbc3c [TIR] Inject source spans into tirx IR and surface source locations in compiler errors (#2751)
500c3686 [CUDA][Transform] Fix PCWS index dtype handling (#2783)
eb31994a [BugFix] Preserve loop step when unrolling loops (#2784)
51fbfc7e [BugFix] Fallback non-16B cluster bulk copies (#2683)
b5e3eb93 [Enhancement] More compile-time guards for architecture-specific CUDA intrinsics (#2781)
2a17fffd [TileOP] Add SM70 GEMM FMA fallback (#2339)
aa7df867 [BugFix] Fix fp16/bf16 T.atomic_max/atomic_min silently corrupting fp32 values (#2780)
9fb75728 [BugFix] Fix ROCm intrinsic resolution after language dialect refactor (#2779)
92072ab2 [BugFix] Cast to the destination dtype in the scalar T.copy path (#2771)
a69708c1 [BugFix] Reject non-power-of-two AllReduce widths (#2611)
30aac1ac [CUDA][Reduce] Fix packed AllReduce workspace pointer (#2778)
fc517bdc [BugFix] Fix rng_init after language dialect refactor (#2776)
eceb0e66 [BugFix] Marshal NVRTC scalar parameters and dynamic strides (#2756)
5ef1500e [BugFix] Handle strided global buffers correctly in 1D TMA copies (#2746)
1dc86d71 [BugFix] Add arithmetic operators to vec_type in common.h for CPU codegen (#2768)
22a2452a [BugFix] Add threadgroup address space qualifier in Metal codegen for shared memory pointer arithmetic (#2770)
9609d3a5 [BugFix] Preserve explicit loop steps when transforms rebuild For nodes (#2752)
b049f87d [CI]: Bump actions/setup-python from 6 to 7 (#2773)
7cb4b1d9 [Enhancement] Fix nondeterministic CanProve (#2772)
c6294f07 [BugFix] Add pre-SM80 fallback for bf16 __hfma (#2769)
8ad82fa0 [Quality] Fixes typings in ast frontend (#2520)
8bb3300d [BugFix] Preserve if condition evaluation during fan-out (#2764)
e9240d68 [Refactor] Move example-only helpers out of tilelang package (#2761)
1591d368 [BugFix] Preserve re-evaluation of mutable if conditions (#2744)
f862dc38 [Language][Backend] Language dialect for multi-backends (#2734)
ab1d2df4 [BugFix] Support BufferRegion destinations in atomic_addx2 return_prev (#2753)
2b4dd803 [BugFix] Make T.transpose swap only the final two axes (#2757)
235077cb [BugFix] Preserve loop steps during unswitching (#2741)
390d208d [BugFix] Use callee global symbols for cross-target calls (#2740)
192ddea6 [BugFix] Gate TMEM and TMA builtins by CUDA architecture (#2743)
aae97c0e [TIR][Python] Add typing wrappers for DSL ops (#2739)
0c88682f [BugFix] Emit Metal barriers for dynamic shared memory (#2738)
bff1b9a3 [Feature] Support stochastic FP32 to FP16/BF16 casts (#2735)
22baf2e2 [FEATURE] Add block-causal attention for dLLM example (#2499)
a443dde9 [BugFix] Reject contracting shared-buffer layouts in T.annotate_layout (#2719)
dff136d4 [CUDA][Pipeline] Fix 1D TMA selection for versioned layouts (#2737)
f84825db [Build] Raise apache-tvm-ffi lower bound to 0.1.11 (#2736)
512d51f5 [BugFix] Honor explicit row strides in Metal GEMM (#2730)
322a9cbb [BugFix] Isolate cross-compiler options per invocation (#2728)
e4e110e5 [Example][DeepSeek-V3.2] Adaptive threads for sparse MLA backward (#2592)
24a023c6 [BugFix][Transform] Never classify side-effecting binds as replayable (atomics re-executed at every use site since v0.1.11) (#2651)
1ea7530f [Cherry][TIRx] Improve BufferStore cast warning context (#2733)
052e6741 [TIR][Analyzer] Fix vectorized Select constraint handling (#2731)
9d819c3f [BugFix] Fix MFMA DataType args causing compilation failure on ROCm (#2726)
cc106fa2 [Feature] Add lower-trace support for debugging & rebased (#2725)
96900c7d [Autotune] Support early_stop in decorator mode and add decorator example (#2729)
bd5ca2f0 [BugFix] Preserve buffer element offsets in access pointers (#2727)
923c8a7d [Autotune] Add early stop to skip slow configs during benchmark (#2723)
f8d8cd4b [BugFix] Use blockDim as workspace stride in batch AllReduce (#2621)
ac576c63 [BugFix] Fix FP4 dequant symbolic exponent clamp (#2656)
25c0a155 [Transform] Replace CPU fallback thread placeholder with a constant-zero logical thread index (#2718)
c4c5ec59 [Example][Opt] deepseek_v32 topk_selector kernel memory access optimization (~1.9× faster) (#2659)
8cfc90e0 [BugFix] Fix HIP predicated dword copy zero fill (#2721)
88e007d4 [BugFix] Guard T.atomic_addx4 return type for sliced destinations (#2590)
656c287a [BugFix] Fix wrong offset when scanning a non-zero-offset buffer sub-region (#2680)
ce9ff0c1 [BugFix] Gate stochastic FP4/FP8 casts on sm_100a (#2691)
39c5b4ef [BugFix] Correct IEEE math intrinsic names for fp64/fp16/bf16 (#2619)
cb26539a [BugFix] Fix T.pow/T.power for constant integer exponent y <= 0 (#2677)
ffeda9f3 [BugFix] Pack 32-lane 8-bit CUDA vectors correctly (#2701)
30221e20 [Bugfix] Emit st.bulk destination as a shared write to fix missing barrier and potential compilation-introduced races (#2700)
172f6fbf [TIR][Transform] Allow unused fragment buffers without layouts (#2717)
8cdd4d62 [BugFix] Prevent partial 1-D TMA stores from bypassing bounds checks (#2716)
46f3b31a [BugFix][WS] Fix pipeline replacement under persistent T.serial (#2674)
4433981c [Enhancement] Add local buffer reduction lowering (#2693)
335afcf8 [BugFix] Decode FP8 E4M3 special encodings correctly (#2710)
e3c3048f [BugFix] Correct the spelling of fragment (#2695)
134f9c2e [BugFix] Implement atomic load and store for HIP (#2711)
bfe126d1 [BugFix] Support return_prev for HIP vector atomic add (#2712)
1354b610 [BugFix] Map CUDA atomic add consume ordering to acquire PTX (#2713)
c5c10d26 [Fix] T.__exp must compute e**x, not 2**x (docstring + CuTeDSL codegen) (#2696)
a1e3aee7 [BugFix] Fix operator precedence in increase_descriptor_offset guard (#2675)
6f70e5c9 [BugFix] reject float dtype in bitwise reduce with an actionable error (#2676)
0c7f14c3 [BugFix] Fix WGMMA C-store layout for multiple warpgroups along M (#2663)
4aea435b [BugFix] Reject unsafe non-warp-multiple partial thread sync (#2679)
9754ac44 [BugFix] Preserve bf16 NaN and Inf during RNE packing (#2690)
61f968b6 [BugFix] Track GEMM accumulator writes in warp-specialization liveness collector (#2685)
4a5cf099 [Feature] Add compiler pass timing profiling (#2622)
9ff4ef8d [BugFix] Respect safe value in tile copy OOB (#2636)
31755e71 [BugFix] Support runtime-dependent vector negative indices (#2654)
66837430 [JIT][TVM-FFI] Reuse executable across kernel launches (#2686)
4f442a39 [BugFix] Reject invalid atomic load and store memory orders (#2666)
c5e53076 [BugFix] Return previous value for scalar atomic_min/atomic_max (#2672)
51e6c69f [BugFix] Zero-extend signed int32 lanes in 256-bit vector pack (#2673)
d36ec37c [BugFix] Support T.infinity for float8_e5m2 (#2671)
3af87b7f [TIR][Transform] Fix flat Bind modeling in parallel race checks (#2665)
8164c9a0 [CUDA][Scan] Enable pipelining for multi-segment scans (#2664)
70548a17 [BugFix][CUDA][HIP] Correct packed shared memory allocation sizes (#2660)
effebb6b [BugFix] Fix segfault when tunable params default to None (#2657)
917cd2b9 Bump transformers from 5.3.0 to 5.5.0 in /examples/bitnet-1.58b (#2658)
1ac5a01a [TIR][Runtime] Enforce host-evaluable assumptions at runtime (#2655)
6ed02aee [Autotune] Support per-config pass_configs in autotuning (#2496)
d1ccb925 [BugFix] Fix 1D bulk TMA transfer alignment check (#2646)
207f3a75 [BugFix] Reject non-positive thread extents in T.Kernel (#2653)
c8c49d50 [BugFix] Reject invalid dtypes in T.dp4a (#2652)
28101de3 [Arith] Gate canonical-simplify LT Case 2 on extra scale == +1 (#2649)
6c09e889 fix: cast ptxas register usage level to int before building the nvcc command (#2641)
2047357e [Reduce][Codegen] Fix thread-segment projection for packed layouts (#2647)
5dfa0f5d [BugFix] Fix grouped reduce_sum over-counts on straddle layout (#2424)
1e075149 [CUDA] Support fp32x2 ops as reducers (#2637)
2f4d0fe7 [BugFix] Fix metal stream bridge (#2639)
228c7c04 [Feature] Support iket profiler for CUDA backend (#2515)
bd764738 [BugFix] Preserve guard identity in LoopUnswitching (#2585)
3b37333c [Refactor] Extract shared Int64Promoter into common header (#2558)
6611eaec [Doc] Update SKILL.md to support editable installs and clarify development workflow (#2533)
250c1fc9 [TIR][Codegen] Preserve decoupled cast buffer scope (#2545)
8533d2a9 [Language][Scheduler] Expose scalar tile scheduler state (#2553)
</details>
T.__exp must compute ex, not 2x (docstring + CuTeDSL codegen) by @Hughshine in https://github.com/tile-ai/tilelang/pull/2696T.atomic_max/atomic_min silently corrupting fp32 values by @jjppp in https://github.com/tile-ai/tilelang/pull/2780Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.12...v0.1.13
Summary of the main changes between v0.1.11 and v0.1.12 (91 commits).
Summary of the main changes between v0.1.11 and v0.1.12 (91 commits).
pass_visualizer structure-tree pass browser (#2449) and pass-diff display for debugging (#2375)st.bulk shared-memory zero fill on SM100+ (#2403), stmatrix m16n8 on Blackwell (#2417), SM75 MMA dispatchers for FP16 accumulation and UINT8 (#2392)-ccbin support for choosing the C++ compiler (#2348), TILELANG_VERBOSE env var to control compile output (#2453)T.Persistent dropping tiles when the last dim isn't a multiple of group_size (#2455), sign-extension bugs in packed uint32 decode (#2500) and make_int negative int8 lanes (#2438), PTX v4 atomics for fp16/bf16 atomic_addx4 (#2492), vectorized atomic_add dtype mismatch (#2414), bf16 exp self-recursion (#2402) and rsqrt overload (#2386)do_bench gained a cache_size option (#2531)T.view / T.reshape enhancements (#2450), better T.assume/loop-bound handling to eliminate redundant boundary checks (#2502), improved diagnostics for T.serial fragment access (#2462)This release centers on the new LLVM backend and backend-registry refactor, major TMA/GMMA layout flexibility on CUDA, fp8/fp16/bf16 cast performance, and a large batch of correctness fixes across layout inference, atomics, and pipelining.
-ccbin arguments to specify C++ compiler by @Triang-jyed-driung in https://github.com/tile-ai/tilelang/pull/2348st.bulk for shared zero fill on SM100+ by @Rachmanino in https://github.com/tile-ai/tilelang/pull/2403__shfl_sync from tl_shuffle_elect by @Yongqi-Zhuo in https://github.com/tile-ai/tilelang/pull/2445T.view and T.reshape by @bucket-xv in https://github.com/tile-ai/tilelang/pull/2450TILELANG_VERBOSE environment var to control the compile output info by @bucket-xv in https://github.com/tile-ai/tilelang/pull/2453Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.11...v0.1.12
Fix atomic_load access_ptr lowering for dynamic indices by @VitalyAnkh in https://github.com/tile-ai/tilelang/pull/2157
T.tma_copy for flexible TMA copy by @Rachmanino in https://github.com/tile-ai/tilelang/pull/2205Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.10...v0.1.11
This release focuses on broader backend support, new GPU instructions, compiler pipeline improvements, and release/build stability.
This release focuses on broader backend support, new GPU instructions, compiler pipeline improvements, and release/build stability.
.shared::cta in TMA copy paths only on CUDA 12.8+ by @ColmaLiu in https://github.com/tile-ai/tilelang/pull/2087T.sync_threads() by @bucket-xv in https://github.com/tile-ai/tilelang/pull/2197do_not_specialize for autotune by @Rachmanino in https://github.com/tile-ai/tilelang/pull/2084T.tma_copy SIMT fallback by @Rachmanino in https://github.com/tile-ai/tilelang/pull/2242Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.9...v0.1.10
Re-enable deprecated TL_DISABLE_TMA_LOWER pass config for TMA store by @LJC00118 in https://github.com/tile-ai/tilelang/pull/2024
InjectFenceProxy by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1850ProveFragmentContains to avoid false layout conflicts by @LJC00118 in https://github.com/tile-ai/tilelang/pull/1950wgmma_gemm and tcgen05_gemm functions by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1949gemm_streamk example on SM90 by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1969minBlocksPerMultiprocessor in __launch_bounds__ by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1979annotations parameter to alloc_buffer in tilelang/language/ast/ir.py by @Copilot in https://github.com/tile-ai/tilelang/pull/1996TL_DISABLE_TMA_LOWER pass config for TMA store by @LJC00118 in https://github.com/tile-ai/tilelang/pull/2024DecoupleTypeCast Pass by @LJC00118 in https://github.com/tile-ai/tilelang/pull/2026T.gemm by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/2063map_torch_type with T.dtype.as_torch by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/2075Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.8...v0.1.9
[Refactor] Enhance T.alloc_barrier with new features and deprecate legacy mbarrier related intrinsics by @Rachmanino in https://github.com/tile-ai/til…
T.make_tensor not on the top of prim_func by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1412T.__ldg by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1414compile_flags to ffi compilation path with pass_configs by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1434pytest.mark.parameterize to speedup parallel testing by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1447T.annotate_restrict_buffers by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1428test_tilelang_language_rand.py by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1464kDisableDynamicTailSplit and kDynamicAlignment as they are legacy by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1486alloc_local statement in examples and introduce processing for floating fragment buffers by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1495ctypes by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1510TargetIsCuda for all cuda target by @oraluben in https://github.com/tile-ai/tilelang/pull/1522S_q != S_kv by @hukongyi in https://github.com/tile-ai/tilelang/pull/1530local.var buffer as local by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1541T.Fill for local.var by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1543H in deepseek sparse mla backward via split-H by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1548ParallelOPNode and CopyNode by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1539tl_pipeline_sync. by @c8ef in https://github.com/tile-ai/tilelang/pull/1566test_tilelang_language_cooperative.py by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1593import tilelang on CPU-only machines without CUDA libraries by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1481T.sync_warp & T.shfl_sync; change extern pdl into intrin by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1614ForwardRef usage in v2 frontend (#1619) by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1621ConstrVisitor to src/transform/common/constr_visitor.h for reuse by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1622T.reduce_absmax to use less abs call by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1626nvidia-cuda-nvcc as nvcc by @clouds56 in https://github.com/tile-ai/tilelang/pull/1528k_dim==4 and open rocm-ci for gemmsr by @benenzhu in https://github.com/tile-ai/tilelang/pull/1627examples/deepseek_v32/sparse_mla_fwd.py by @GoldenStain in https://github.com/tile-ai/tilelang/pull/1634cp.reduce.async.bulk.tensor by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1667Threadsync with ConstrVisitor by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1631ParallelLoopTransformer by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1672thread_sync by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1682cute::elect_one_sync() for slightly better performance by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1703RewriteUnsafeSelect Pass by @LJC00118 in https://github.com/tile-ai/tilelang/pull/1705ptx_stmatrix by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1689T.alloc_barrier with new features and deprecate legacy mbarrier related intrinsics by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1733.kind::i8 by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1764_typing.py to dist check by @oraluben in https://github.com/tile-ai/tilelang/pull/1803value -= 1 by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1825local.var as local buffers when deciding vectorization for stable actions by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1835T.access_of to combine T.address_of and access_ptr by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1827mbar by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1774Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.7...v0.1.8
[Pipeline] Refactor buffer allocation in Inject Pipeline Pass by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1525
S_q != S_kv by @hukongyi in https://github.com/tile-ai/tilelang/pull/1530local.var buffer as local by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1541T.Fill for local.var by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1543H in deepseek sparse mla backward via split-H by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1548ParallelOPNode and CopyNode by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1539tl_pipeline_sync. by @c8ef in https://github.com/tile-ai/tilelang/pull/1566test_tilelang_language_cooperative.py by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1593import tilelang on CPU-only machines without CUDA libraries by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1481T.sync_warp & T.shfl_sync; change extern pdl into intrin by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1614ForwardRef usage in v2 frontend (#1619) by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1621ConstrVisitor to src/transform/common/constr_visitor.h for reuse by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1622T.reduce_absmax to use less abs call by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1626nvidia-cuda-nvcc as nvcc by @clouds56 in https://github.com/tile-ai/tilelang/pull/1528k_dim==4 and open rocm-ci for gemmsr by @benenzhu in https://github.com/tile-ai/tilelang/pull/1627examples/deepseek_v32/sparse_mla_fwd.py by @GoldenStain in https://github.com/tile-ai/tilelang/pull/1634cp.reduce.async.bulk.tensor by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1667Threadsync with ConstrVisitor by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1631ParallelLoopTransformer by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1672Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.7.post1...v0.1.7.post3
[Pipeline] Refactor buffer allocation in Inject Pipeline Pass by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1525
S_q != S_kv by @hukongyi in https://github.com/tile-ai/tilelang/pull/1530local.var buffer as local by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1541T.Fill for local.var by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1543H in deepseek sparse mla backward via split-H by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1548ParallelOPNode and CopyNode by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1539tl_pipeline_sync. by @c8ef in https://github.com/tile-ai/tilelang/pull/1566Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.7.post1...0.1.7.post2
[Bugfix][Build] Update CMake configuration to remove project root injection for sys.path by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1
T.make_tensor not on the top of prim_func by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1412T.__ldg by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1414compile_flags to ffi compilation path with pass_configs by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1434pytest.mark.parameterize to speedup parallel testing by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1447T.annotate_restrict_buffers by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1428test_tilelang_language_rand.py by @silentCoder-dev in https://github.com/tile-ai/tilelang/pull/1464kDisableDynamicTailSplit and kDynamicAlignment as they are legacy by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1486alloc_local statement in examples and introduce processing for floating fragment buffers by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1495ctypes by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1510TargetIsCuda for all cuda target by @oraluben in https://github.com/tile-ai/tilelang/pull/1522Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.7...v0.1.7.post1
[Enhancement] Deprecate split&sum in attn bwd examples on Hopper by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1065
seq_q<seq_kv in flash attention examples by @Rachmanino in https://github.com/tile-ai/tilelang/pull/864B[i,j] = c[i] + A[i,j] by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/798ExprDeepEqual instead of StructuralEqual when merge consecutive If stmt by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/876T.ieee_rsqrt and related high precision op by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/882atomic_add performance for bwd examples by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/940bfloat16 and user-defined sm_scale in attention sink examples by @Rachmanino in https://github.com/tile-ai/tilelang/pull/924pre-commit integration by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/955CumSum1D by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/978T.alloc_var for AugAssign and AnnAsign by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/979InjectFenceProxy and expose some warp group primitives in frontend by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/977access_ptr("r") instead of access_ptr("w") for correct pipeline analysis by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/983torch.accelerator.synchronize() to torch.cuda.synchronize() by @yyttt6 in https://github.com/tile-ai/tilelang/pull/987LowerIntrin from tvm into tilelang by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/999LowerIntrin by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1035TL_LIBS by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1038T.get_warp_idx_sync and T.shuffle_elect for efficient thread election by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/989has_simt_copy to decide whether to insert set_max_nreg by @chengyupku in https://github.com/tile-ai/tilelang/pull/982LegalizeSafeMemoryAccess to support recursive load/store rewrite by @SiriusNEO in https://github.com/tile-ai/tilelang/pull/1050T.Parallel with dynamic extents by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/990tileang.clear_cache() by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1077T.dynamic instead of T.symbolic by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1076T.reduce_ with shared memory input/output by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1080tilelang_cython and relocate its path by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1086tilelang.disable_cache() calls from examples and tests by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1088TL_STORAGE_REWRITE_DETECT_INPLACE by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1089alloc_var(dtype, init=x) by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1092cuTensorMapEncodeIm2col call by @chengyupku in https://github.com/tile-ai/tilelang/pull/1094format.sh and add clang-tidy to GHA workflow by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1044ldmatrix and update mamba scan kernel by @chengyupku in https://github.com/tile-ai/tilelang/pull/1104format.sh by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1102T.ptr and T.Tensor by @xwhzz in https://github.com/tile-ai/tilelang/pull/1114fence_barrier_init primitive after mbarrier init by @chengyupku in https://github.com/tile-ai/tilelang/pull/1121format.sh and introduce loop carry thread sync unit test by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1153T.warpgroup_fence_operand for nvcc code motion by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/986cibuildwheel and reduce size of sdist by @oraluben in https://github.com/tile-ai/tilelang/pull/1171tl.infinity operator for infinity handling of bfloat16 by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1175ccache for CIBW on Linux by @oraluben in https://github.com/tile-ai/tilelang/pull/1184T.serial with step and negative step by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1188reduce.h by @LJC00118 in https://github.com/tile-ai/tilelang/pull/1204libtvm as a dep of libtilelang by @oraluben in https://github.com/tile-ai/tilelang/pull/1215CompleteBufferFragment by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1226builder.py by @LJC00118 in https://github.com/tile-ai/tilelang/pull/1235-inf instead of clearing accumulators. by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1222from __future__ import annotations for python 3.8 by @oraluben in https://github.com/tile-ai/tilelang/pull/1273int64_t static and dynamic shape. by @Elevator14B in https://github.com/tile-ai/tilelang/pull/1218T.view/reshape by @SiriusNEO in https://github.com/tile-ai/tilelang/pull/1277T.print for bool type by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1279T.Tensor(n * 2 + 1) in function annotation by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1285T.Var annotation by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1291T.assume handling by @LJC00118 in https://github.com/tile-ai/tilelang/pull/1292T.print by @xwhzz in https://github.com/tile-ai/tilelang/pull/1329NormalizeToBufferRegion and MakeAccessPtrFromRegion to utils by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1333not operator in frontend (#1347) by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/1348T.gemm_sp_v2 on sm80 and sm89 by @botbw in https://github.com/tile-ai/tilelang/pull/1056Full Changelog: https://github.com/tile-ai/tilelang/compare/0.1.6...v0.1.7
[Enhancement] Deprecate split&sum in attn bwd examples on Hopper by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1065
The Last Release for Python 3.8 (without tvm-ffi) 🚀
seq_q<seq_kv in flash attention examples by @Rachmanino in https://github.com/tile-ai/tilelang/pull/864B[i,j] = c[i] + A[i,j] by @kurisu6912 in https://github.com/tile-ai/tilelang/pull/798ExprDeepEqual instead of StructuralEqual when merge consecutive If stmt by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/876T.ieee_rsqrt and related high precision op by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/882atomic_add performance for bwd examples by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/940bfloat16 and user-defined sm_scale in attention sink examples by @Rachmanino in https://github.com/tile-ai/tilelang/pull/924pre-commit integration by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/955CumSum1D by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/978T.alloc_var for AugAssign and AnnAsign by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/979InjectFenceProxy and expose some warp group primitives in frontend by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/977access_ptr("r") instead of access_ptr("w") for correct pipeline analysis by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/983torch.accelerator.synchronize() to torch.cuda.synchronize() by @yyttt6 in https://github.com/tile-ai/tilelang/pull/987LowerIntrin from tvm into tilelang by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/999LowerIntrin by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1035TL_LIBS by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1038T.get_warp_idx_sync and T.shuffle_elect for efficient thread election by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/989has_simt_copy to decide whether to insert set_max_nreg by @chengyupku in https://github.com/tile-ai/tilelang/pull/982LegalizeSafeMemoryAccess to support recursive load/store rewrite by @SiriusNEO in https://github.com/tile-ai/tilelang/pull/1050T.Parallel with dynamic extents by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/990tileang.clear_cache() by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1077T.dynamic instead of T.symbolic by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1076T.reduce_ with shared memory input/output by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1080tilelang_cython and relocate its path by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1086tilelang.disable_cache() calls from examples and tests by @Rachmanino in https://github.com/tile-ai/tilelang/pull/1088TL_STORAGE_REWRITE_DETECT_INPLACE by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1089alloc_var(dtype, init=x) by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1092cuTensorMapEncodeIm2col call by @chengyupku in https://github.com/tile-ai/tilelang/pull/1094format.sh and add clang-tidy to GHA workflow by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1044ldmatrix and update mamba scan kernel by @chengyupku in https://github.com/tile-ai/tilelang/pull/1104format.sh by @XuehaiPan in https://github.com/tile-ai/tilelang/pull/1102T.ptr and T.Tensor by @xwhzz in https://github.com/tile-ai/tilelang/pull/1114fence_barrier_init primitive after mbarrier init by @chengyupku in https://github.com/tile-ai/tilelang/pull/1121format.sh and introduce loop carry thread sync unit test by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/1153Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.6.post1...v0.1.6.post2
In version 0.1.6, libgcc and libg++ were statically linked to improve version compatibility. However, this could introduce certain unpredictable risks
In version 0.1.6, libgcc and libg++ were statically linked to improve version compatibility. However, this could introduce certain unpredictable risks in some programs. In post1, this process was reworked based on the PyTorch build workflow, eliminating the risks while ensuring better compatibility. This is the reason for releasing version 0.1.6.post1.
[Bugfix] Added missing thread offsets and other information to reduce by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/646
--ptxas-options=--register-usage-level=10 option by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/684StridedTensor to support non contigious torch inputs by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/722fix by @coderabbitai[bot] in https://github.com/tile-ai/tilelang/pull/726mxfp4 by @coderabbitai[bot] in https://github.com/tile-ai/tilelang/pull/732main by @coderabbitai[bot] in https://github.com/tile-ai/tilelang/pull/745disable_cache in some tests by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/755Operator into TileOperator and with tvm reflection by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/763alloc_reducer to separate inter and intra warp reduction by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/757pytile_0826 by @coderabbitai[bot] in https://github.com/tile-ai/tilelang/pull/770reducer_0825 by @coderabbitai[bot] in https://github.com/tile-ai/tilelang/pull/772T.rsqrt(x) into cuda intrin instead of 1 / T.sqrt(x) by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/781T.gemm_v2 by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/793alloc_reducer definition to the python side by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/802ENABLE_FAST_MATH by default by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/846local.var by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/851Full Changelog: https://github.com/tile-ai/tilelang/commits/0.1.6
[Release] Bump version from 0.1.3 into 0.1.4 by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/375
attrs and decorator by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/411T.cumsum by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/423T.reduce_sum by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/436power_of_int with pow_of_int for consistency by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/468SM75_U16x4_LDSM_N when warp_n is 8 by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/498jit to _JitImplementation to support @tilelang.jit by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/502T.view by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/509T.annotate_l2_hit_ratio via cudaStreamSetAttribute by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/539Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.4...v0.1.5
[Refactor] Deprecated T.Buffer as arguments and rename related calls into T.Tensor by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/281
T.clear for let binding by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/268T.ptr and T.Tensor by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/276T.Buffer as arguments and rename related calls into T.Tensor by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/281T.gemm with transpose_b=False for amd backend by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/327Transpose_A=True and GEMM_RS for hip backend by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/331kernel.func with kernel in mla benchmark scripts by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/354disable_warp_specialize to disable auto specialization on hopper by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/357max_workers limit to available CPUs by @tth37 in https://github.com/tile-ai/tilelang/pull/368T.any_of and T.all_of to reduce a bool arrary by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/371T.Parallel with local register assignment by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/395Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.3...v0.1.4
[Docker] Add libstdcxx-ng-12 to Dockerfiles for CUDA versions by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/160
T.thread_binding with T.get_thread_binding in examples and test cases by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/163T.copy for scalar datatypes by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/190T.reshape and T.view by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/212minBlocksPerMultiprocesor by @cherichy in https://github.com/tile-ai/tilelang/pull/248T.alloc_var to define a variable like int var; by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/255Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.2...v0.1.3
The v0.1.2 prebuild package used a legacy cython file, which may lead to some bugs.
The v0.1.2 prebuild package used a legacy cython file, which may lead to some bugs.
T.thread_binding with T.get_thread_binding in examples and test cases by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/163Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.2...v0.1.2.post1
[Dev] Add MLA and GQA decode examples by @chengyupku in https://github.com/tile-ai/tilelang/pull/109
T.print for fragment scope by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/130T.gemm support for fp8 by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/144Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.1...v0.1.2
[Doc] Update release news by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/80
InjectPtxAsyncCopy Pass behind ThreadSync Pass by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/97Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.1.0...v0.1.1
[LICENSE] Add LICENSE for flashinfer by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/19
AnnotateDeviceRegions and MakePackedAPI by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/26KernelFrame.Current by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/34LayoutInference and LowerTileOp on loop tail split functionality by @tzj-fxz in https://github.com/tile-ai/tilelang/pull/29T.print for buffer and variables logging on frontend by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/45pull_request_target for documen… by @xwhzz in https://github.com/tile-ai/tilelang/pull/48LoopVectorize Pass from upstream tvm by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/62plot_layout to visualize the fragment layout by @LeiWang1999 in https://github.com/tile-ai/tilelang/pull/68Full Changelog: https://github.com/tile-ai/tilelang/compare/v0.0.1...v0.1.0
Pre-release for the v0.0.1. Under testing, Only cuda prebuilt are provided.
Pre-release for the v0.0.1. Under testing, Only cuda prebuilt are provided.
Full Changelog: https://github.com/tile-ai/tilelang/commits/v0.0.1
Nothing published for this version
One column per month.