v0.1.7.post3

[LeiWang1999]

What's Changed

• [Pipeline] Refactor buffer allocation in Inject Pipeline Pass by @LeiWang1999 in [Pipeline] Refactor buffer allocation in Inject Pipeline Pass #1525
• [Dev] Fix when build local version with isolated build by @oraluben in [Dev] Fix when build local version with isolated build #1487
• [Bugfix] Skip stride check for subtype by @LeiWang1999 in [Bugfix] Skip stride check for subtype #1531
• [Lint] Enable whitespace and permission bit hooks by @XuehaiPan in [Lint] Enable whitespace and permission bit hooks #1439
• [Enhancement][Tool] Tree-style pretty ASTPrinter by @SiriusNEO in [Enhancement][Tool] Tree-style pretty ASTPrinter #1468
• [Fix] Add support for non-var complement arithmetic computation ([BUG] Layout Inference Fails for Cases Requiring Replication #1374) by @kurisu6912 in [Fix] Add support for non-var complement arithmetic computation (#1374) #1533
• [BugFix] Complete vectorized loading for common dtypes by @SiriusNEO in [BugFix] Complete vectorized loading for common dtypes #1536
• [Compat] Add CUDA version check for __nv_fp8_e8m0 type by @LeiWang1999 in [Compat] Add CUDA version check for __nv_fp8_e8m0 type #1537
• [BugFix] Fix bugs of varlen attention forward examples caused by S_q != S_kv by @hukongyi in [BugFix] Fix bugs of varlen attention forward examples caused by S_q != S_kv #1530
• [Bug] Fix hanging from reduction on sm120 by @PannenetsF in [Bug] Fix hanging from reduction on sm120 #1540
• [example] use T.dynamic instead of tvm.te.var by @botbw in [example] use T.dynamic instead of tvm.te.var #1538
• [Enhancement] Refactor KernelCache to use inheritance-based design by @sgjzfzzf in [Enhancement] Refactor KernelCache to use inheritance-based design #1483
• [Bugfix] Avoid considering local.var buffer as local by @LeiWang1999 in [Bugfix] Avoid considering local.var buffer as local #1541
• [Bugfix] Fix of T.Fill for local.var by @LeiWang1999 in [Bugfix] Fix of T.Fill for local.var #1543
• [Z3] Change z3 timeout to rlimit for determistic prove behavior by @kurisu6912 in [Z3] Change z3 timeout to rlimit for determistic prove behavior #1542
• [Feat] Adapt gemm v2 for cutedsl backend by @lucifer1004 in [Feat] Adapt gemm v2 for cutedsl backend #1544
• [Enhancement] Support larger H in deepseek sparse mla backward via split-H by @Rachmanino in [Enhancement] Support larger H in deepseek sparse mla backward via split-H #1548
• [Bugfix] Fix regression test to use installed package instead of source directory by @xwhzz in [Bugfix] Fix regression test to use installed package instead of source directory #1550
• [Refactor] Introduce layout annotations for ParallelOPNode and CopyNode by @LeiWang1999 in [Refactor] Introduce layout annotations for ParallelOPNode and CopyNode #1539
• [Script] Provide regression test script to help benchmark regression in local env by @LeiWang1999 in [Script] Provide regression test script to help benchmark regression in local env #1551
• [Typing] Update Kernel signature and add type hints for buffer operations by @clouds56 in [Typing] Update Kernel signature and add type hints for buffer operations #1545
• [CI]: Bump actions/upload-artifact from 4 to 6 by @dependabot[bot] in [CI]: Bump actions/upload-artifact from 4 to 6 #1555
• [Refactor] Use cuda capability from torch to be more generic by @oraluben in [Refactor] Use cuda capability from torch to be more generic #1557
• [CI]: Bump actions/github-script from 7 to 8 by @dependabot[bot] in [CI]: Bump actions/github-script from 7 to 8 #1556
• [Host] Provide post process to customize host code and enhance nullable check by @LeiWang1999 in [Host] Provide post process to customize host code and enhance nullable check #1562
• [Release] Build tilelang against CUDA 13.1 in CI by @oraluben in [Release] Build tilelang against CUDA 13.1 in CI #1532
• [LazyJIT] Move Type Annotations to Function Body by @kurisu6912 in [LazyJIT] Move Type Annotations to Function Body #1480
• [bugfix] fix missing clear_accum logic for gemm_sp_v2 by @botbw in [bugfix] fix missing clear_accum logic for gemm_sp_v2 #1563
• [Misc] Remove unused tl_pipeline_sync. by @c8ef in [Misc] Remove unused tl_pipeline_sync. #1566
• [Refactor] Improve scalarization handling in Pass VectorizeLoop by @LeiWang1999 in [Refactor] Improve scalarization handling in Pass VectorizeLoop #1565
• [Refactor] Simplify do_bench calls by using default warmup and rep parameters by @LeiWang1999 in [Refactor] Simplify do_bench calls by using default warmup and rep parameters #1568
• [CI] Refactor PR regression test job conditions by @xwhzz in [CI] Refactor PR regression test job conditions #1569
• [Parallel][Infer] Free-mode chooses minimal replication between buffer-based and PlanLoopPartition by @LeiWang1999 in [Parallel][Infer] Free-mode chooses minimal replication between buffer-based and PlanLoopPartition #1559
• [Refactor] Enhance deterministic ordering in shared memory allocation merge. by @LeiWang1999 in [Refactor] Enhance deterministic ordering in shared memory allocation merge. #1570
• [Enhancement] Improve equality checks in layout nodes and fragment validation by @LeiWang1999 in [Enhancement] Improve equality checks in layout nodes and fragment validation #1573
• [Feature] add kUseCooperativeLaunch tag for tvm_ffi by @silentCoder-dev in [Feature] add kUseCooperativeLaunch tag for tvm_ffi #1572
• [Refactor] Remove unnecessary logging configuration in Analyzer.py by @LeiWang1999 in [Refactor] Remove unnecessary logging configuration in Analyzer.py #1574
• [Release] Bump version to 0.1.7.post2 by @LeiWang1999 in [Release] Bump version to 0.1.7.post2 #1575
• [BugFix] Change default rounding mode for fp4 conversions by @LJC00118 in [BugFix] Change default rounding mode for fp4 conversions #1580
• [CI] Add CUDA-aware pytest scheduler + auto workers by @LeiWang1999 in [CI] Add CUDA-aware pytest scheduler + auto workers #1584
• [Enhancement] Improve performance regression output with timing and streaming by @xwhzz in [Enhancement] Improve performance regression output with timing and streaming #1585
• [Bugfix] Add kernel_global_source property to TVMFFIKernelAdapter by @haok1402 in [Bugfix] Add kernel_global_source property to TVMFFIKernelAdapter #1589
• [BugFix] Add PrimExpr substitution support for AttrStmt nodes by @LJC00118 in [BugFix] Add PrimExpr substitution support for AttrStmt nodes #1583
• [BugFix] fix tcgen5mma example by @Rachmanino in [BugFix] fix tcgen5mma example #1577
• [Refactor] Use access_ptr instead of buffer and offsets for cp async params by @LeiWang1999 in [Refactor] Use access_ptr instead of buffer and offsets for cp async params #1590
• [Layout] Support annotating loop layout in frontend by @LeiWang1999 in [Layout] Support annotating loop layout in frontend #1579
• [Typo] Rename loop layout annotation test by @LeiWang1999 in [Typo] Rename loop layout annotation test #1596
• [Fix] Add register to read A ptr in test_tilelang_language_cooperative.py by @silentCoder-dev in [Fix] Add register to read A ptr in test_tilelang_language_cooperative.py #1593
• [Feat] PDL Support by @w169q169 in [Feat] PDL Support #1494
• [Enhancement][Subtype] Enhance symbolic shape/stride handling for subtype by @LeiWang1999 in [Enhancement][Subtype] Enhance symbolic shape/stride handling for subtype #1599
• [Fix][CuteDSL] add support for tanh/tanhf (fixes [BUG] CuTe-DSL backend wrongly converts tanh to tanhf(op) as opposed to tanh(op, fastmath=True) #1595) by @lucifer1004 in [Fix][CuteDSL] add support for tanh/tanhf (fixes #1595) #1597
• [Release] Fix race condition when publishing by @oraluben in [Release] Fix race condition when publishing #1578
• Add conversion from cutlass::float_e4m3/e5m2 to tl::float_e4m3/e5m2 by @LJC00118 in Add conversion from cutlass::float_e4m3/e5m2 to tl::float_e4m3/e5m2 #1600
• [Enhancement][AMD] Add preshuffle fp8 gemm example on amd. by @Gongen-Ali in [Enhancement][AMD] Add preshuffle fp8 gemm example on amd. #1605
• [Bugfix] Mangle Single Precision Mathematical Functions of cuda math api by @silentCoder-dev in [Bugfix] Mangle Single Precision Mathematical Functions of cuda math api #1602
• [Bugfix] Open Rocm ci test and fix some bugs. by @Gongen-Ali in [Bugfix] Open Rocm ci test and fix some bugs. #1443
• [Feature] Add more curand operations & support vectorization by @silentCoder-dev in [Feature] Add more curand operations & support vectorization #1582
• [Enhancement] Allow import tilelang on CPU-only machines without CUDA libraries by @XuehaiPan in [Enhancement] Allow import tilelang on CPU-only machines without CUDA libraries #1481
• [BugFix] Add pre-commit to requirements-dev.txt by @asaadkhaja99 in [BugFix] Add pre-commit to requirements-dev.txt #1611
• [BugFix] Fix some bugs in lowering ParallelOp and VectorizeLoop by @SiriusNEO in [BugFix] Fix some bugs in lowering ParallelOp and VectorizeLoop #1607
• [Feat] Add strong checker to detect data racing in T.Parallel by @kurisu6912 in [Feat] Add strong checker to detect data racing in T.Parallel #1615
• [Feature] add T.sync_warp & T.shfl_sync; change extern pdl into intrin by @silentCoder-dev in [Feature] add T.sync_warp & T.shfl_sync; change extern pdl into intrin #1614
• [RaceChecker] RaceChecker report warning rather than error for backward compatibility by @kurisu6912 in [RaceChecker] RaceChecker report warning rather than error for backward compatibility #1620
• [BugFix] Fix ForwardRef usage in v2 frontend ([BUG] Incorrect usage of ForwardRef in v2 frontend #1619) by @kurisu6912 in [BugFix] Fix ForwardRef usage in v2 frontend (#1619) #1621
• [Refactor] Move ConstrVisitor to src/transform/common/constr_visitor.h for reuse by @silentCoder-dev in [Refactor] Move ConstrVisitor to src/transform/common/constr_visitor.h for reuse #1622
• [Feat] Improve T.reduce_absmax to use less abs call by @kurisu6912 in [Feat] Improve T.reduce_absmax to use less abs call #1626
• [Bugfix] Do not consider local.var as local buffer during LowerTileOP by @LeiWang1999 in [Bugfix] Do not consider local.var as local buffer during LowerTileOP #1628
• [Feature] Add hoist_broadcast_values pass by @silentCoder-dev in [Feature] Add hoist_broadcast_values pass #1606
• [Enhancement][CUDA] Support nvidia-cuda-nvcc as nvcc by @clouds56 in [Enhancement][CUDA] Support nvidia-cuda-nvcc as nvcc #1528
• [Bugfix] Fallback into full region when dynamic buffer read region cannot be proved by @LeiWang1999 in [Bugfix] Fallback into full region when dynamic buffer read region cannot be proved #1618
• [Feat] Allow print macro call stack in device assert by @kurisu6912 in [Feat] Allow print macro call stack in device assert #1616
• [BugFix] Correct index_map selection for transposed A matrix in MFMA Layout with k_dim==4 and open rocm-ci for gemmsr by @benenzhu in [BugFix] Correct index_map selection for transposed A matrix in MFMA Layout with k_dim==4 and open rocm-ci for gemmsr #1627
• [Example] Add Seesaw Sparse MLA Forward Kernel for DeepSeek-V3.2 by @hammersam in [Example] Add Seesaw Sparse MLA Forward Kernel for DeepSeek-V3.2 #1636
• [Bugfix] Introduce a flag to avoid unnecessary broadcast hoist and enable for let stmt by @LeiWang1999 in [Bugfix] Introduce a flag to avoid unnecessary broadcast hoist and enable for let stmt #1638
• [Refactor][CI] Reduce sparse related test time by @LeiWang1999 in [Refactor][CI] Reduce sparse related test time #1637
• [Refactor] Unify @jit and @lazy_jit into a single @jit decorator by @LeiWang1999 in [Refactor] Unify @jit and @lazy_jit into a single @jit decorator #1632
• [Bugfix] Fix pdl related intrin handling to avoid strict annotation codegen by @LeiWang1999 in [Bugfix] Fix pdl related intrin handling to avoid strict annotation codegen #1650
• [Bugfix] reverted unexpected tvm changes by @LeiWang1999 in [Bugfix] reverted unexpected tvm changes #1651
• [Bugfix] reverted unexpected tvm changes by @LeiWang1999 in [Bugfix] reverted unexpected tvm changes #1652
• [Refactor] Move dtypes.py from eager to language and add bits/bytes properties by @LeiWang1999 in [Refactor] Move dtypes.py from eager to language and add bits/bytes properties #1646
• [Feat] Allow dangling producer in wasp pipeline planning ([BUG] Another regression: "variables X are used, but are not passed in as API arguments" error in T.Pipelined #1263) by @kurisu6912 in [Feat] Allow dangling producer in wasp pipeline planning (#1263) #1647
• [bugfix] fix smem alloc for single warp reduce by @botbw in [bugfix] fix smem alloc for single warp reduce #1643
• [Example] Add attention sink varlen examples by @Rachmanino in [Example] Add attention sink varlen examples #1645
• [ASTPrinter] Fix IfThenElse printing and some format problems by @SiriusNEO in [ASTPrinter] Fix IfThenElse printing and some format problems #1640
• [CI] [pre-commit.ci] autoupdate by @pre-commit-ci[bot] in [CI] [pre-commit.ci] autoupdate #1610
• [Enhancement] Update LetStmtNode handling in loop vectorization to support variable binding overrides by @Rachmanino in [Enhancement] Update LetStmtNode handling in loop vectorization to support variable binding overrides #1649
• [Example] Remove redundant T.copy in examples/deepseek_v32/sparse_mla_fwd.py by @GoldenStain in [Example] Remove redundant T.copy in examples/deepseek_v32/sparse_mla_fwd.py #1634
• [CUDA] Introduce simulated load/store 256bits access for CUDA compatibility by @LeiWang1999 in [CUDA] Introduce simulated load/store 256bits access for CUDA compatibility  #1656
• [Enhancement] Improve unroll loop functionality for dynamic extent and corresponding test case by @LeiWang1999 in [Enhancement] Improve unroll loop functionality for dynamic extent and corresponding test case #1654
• [Bugfix] Fix missing annotations for default CallNode Visitor by @LeiWang1999 in [Bugfix] Fix missing annotations for default CallNode Visitor #1659
• [Clean] Remove unnecessary debug print by @LeiWang1999 in [Clean] Remove unnecessary debug print #1661
• [Bugfix] Fix variable scoping issue in InjectSoftwarePipeline for transitive LetStmt dependencies by @LeiWang1999 in [Bugfix] Fix variable scoping issue in InjectSoftwarePipeline for transitive LetStmt dependencies #1657
• [Refactor] Improve CallNode handling to include annotations in various operations by @LeiWang1999 in [Refactor] Improve CallNode handling to include annotations in various operations #1663
• [EagerJIT] Add Support for Parameter Only Kernel Compilation by @kurisu6912 in [EagerJIT] Add Support for Parameter Only Kernel Compilation #1664
• [AutoDD] Add Tilelang AutoDD to Reduce Buggy Program by @KEKE046 in [AutoDD] Add Tilelang AutoDD to Reduce Buggy Program #1639
• [Feature] Support cp.reduce.async.bulk.tensor by @Rachmanino in [Feature] Support cp.reduce.async.bulk.tensor #1667
• chore: update CI cutedsl version to 4.3.5 by @lucifer1004 in chore: update CI cutedsl version to 4.3.5 #1665
• [CUDA] Enhance Broadcast Codegen for Symbolic Value by @LeiWang1999 in [CUDA] Enhance Broadcast Codegen for Symbolic Value #1669
• [EagerJIT] Fix bug in handling of positional arguments by @kurisu6912 in [EagerJIT] Fix bug in handling of positional arguments #1675
• [Feature] Reimplement Threadsync with ConstrVisitor by @silentCoder-dev in [Feature] Reimplement Threadsync with ConstrVisitor #1631
• [Clean][Refactor] Phaseout Legacy Pass ParallelLoopTransformer by @LeiWang1999 in [Clean][Refactor] Phaseout Legacy Pass ParallelLoopTransformer #1672
• [Feature] Atomic Reduction Operations and Vectorization Enhancement by @LeiWang1999 in [Feature] Atomic Reduction Operations and Vectorization Enhancement #1676
• [Refactor] Move AtomicAdd Vectorization to VectorizeLoop Pass by @LeiWang1999 in [Refactor] Move AtomicAdd Vectorization to VectorizeLoop Pass #1677
• [Bugfix] Relax region analysis for complex expression by @LeiWang1999 in [Bugfix] Relax region analysis for complex expression #1679
• [Example] Add example for mHC inference kernels. by @Elevator14B in [Example] Add example for mHC inference kernels. #1684
• [Analyzer] Fix missing assume in tvm analyzer by @kurisu6912 in [Analyzer] Fix missing assume in tvm analyzer #1680
• Refactor: Use centralized do_bench from tilelang.profiler by @LeiWang1999 in Refactor: Use centralized do_bench from tilelang.profiler #1670
• [Feature] Introduce DecoupleTypeCast pass for mixed-precision vectorization by @LeiWang1999 in [Feature] Introduce DecoupleTypeCast pass for mixed-precision vectorization #1644
• [Release] Bump Version into v0.1.7.post3 by @LeiWang1999 in [Release] Bump Version into v0.1.7.post3 #1685

New Contributors

• @hukongyi made their first contribution in [BugFix] Fix bugs of varlen attention forward examples caused by S_q != S_kv #1530
• @clouds56 made their first contribution in [Typing] Update Kernel signature and add type hints for buffer operations #1545
• @c8ef made their first contribution in [Misc] Remove unused tl_pipeline_sync. #1566
• @haok1402 made their first contribution in [Bugfix] Add kernel_global_source property to TVMFFIKernelAdapter #1589
• @w169q169 made their first contribution in [Feat] PDL Support #1494
• @asaadkhaja99 made their first contribution in [BugFix] Add pre-commit to requirements-dev.txt #1611
• @hammersam made their first contribution in [Example] Add Seesaw Sparse MLA Forward Kernel for DeepSeek-V3.2 #1636
• @GoldenStain made their first contribution in [Example] Remove redundant T.copy in examples/deepseek_v32/sparse_mla_fwd.py #1634
• @KEKE046 made their first contribution in [AutoDD] Add Tilelang AutoDD to Reduce Buggy Program #1639

Full Changelog: v0.1.7.post1...v0.1.7.post3

---

This discussion was created from the release v0.1.7.post3.