MCPcopy Create free account

hub / github.com/tile-ai/tilelang / functions

Functions7,254 in github.com/tile-ai/tilelang

↓ 1 callersMethodupdate_lib_code
(self, code: str)
tilelang/jit/adapter/wrapper.py:572
↓ 1 callersMethodupdate_lib_code
(self, code: str)
tilelang/jit/adapter/wrapper.py:874
↓ 1 callersMethodupdate_lib_code
(self, code: str)
tilelang/jit/adapter/wrapper.py:927
↓ 1 callersMethodupdate_lib_code
Update library code and generate host dispatch function. Entry point for code generation. Walks the host IR to extract kernel call si
tilelang/jit/adapter/nvrtc/wrapper.py:544
↓ 1 callersFunctionvalid_loop_not_frag
(dtype: T.dtype = T.bfloat16, accum_dtype: T.dtype = T.float32, num_threads: int = 128)
testing/python/analysis/test_tilelang_fragment_loop_checker.py:94
↓ 1 callersFunctionvalid_loop_not_use_loop_var
(dtype: T.dtype = T.bfloat16, accum_dtype: T.dtype = T.float32, num_threads: int = 128)
testing/python/analysis/test_tilelang_fragment_loop_checker.py:72
↓ 1 callersFunctionvalid_loop_serial
(dtype: T.dtype = T.bfloat16, accum_dtype: T.dtype = T.float32, num_threads: int = 128)
testing/python/analysis/test_tilelang_fragment_loop_checker.py:115
↓ 1 callersMethodvalidate
(self)
examples/autodd/tilelang_buggy.py:78
↓ 1 callersFunctionvalidate_gemm_ab_dtypes
Validate mixed A/B dtypes for GEMM tile ops. TS variants (A in TMEM) skip validation because A/B dtypes may legitimately differ. Only TCGEN05
tilelang/language/dtypes.py:313
↓ 1 callersFunctionvalidate_parameters
Validate parameters - redundant check
examples/autodd/tilelang_buggy.py:40
↓ 1 callersFunctionvalidate_tensor_match
(a, b, tolerance=1e-8, tensor_name="tensor", should_raise=True)
examples/deepseek_v32/fp8_lighting_indexer.py:23
↓ 1 callersFunctionvar_init_kernel
(iv, tld)
testing/python/amd/test_tilelang_hip_codegen.py:406
↓ 1 callersFunctionvectorize_access_legalize
(M: int = 64, N: int = 64, M_offset: int = 2, N_offset: int = 2)
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:81
↓ 1 callersFunctionvectorize_access_legalize
(M: int = 64, N: int = 64)
testing/python/transform/test_tilelang_transform_legalize_vectorized_loop.py:7
↓ 1 callersFunctionvectorize_access_with_atmoic_add_legalize
(M: int = 64, N: int = 64, M_offset: int = 2, N_offset: int = 2)
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:121
↓ 1 callersFunctionvectorize_test
(N, M, stride_A, stride_B)
testing/python/language/test_tilelang_language_vectorize.py:10
↓ 1 callersFunctionvectorize_test_all_dtypes
(dtype, vec_num)
testing/python/language/test_tilelang_language_vectorize.py:117
↓ 1 callersFunctionvectorize_test_invariant_index
(N, M, K)
testing/python/language/test_tilelang_language_vectorize.py:63
↓ 1 callersFunctionvectorized
The vectorized For statement. Parameters ---------- start : PrimExpr The minimum value of iteration. stop : PrimExpr
tilelang/language/loop.py:329
↓ 1 callersFunctionvectorized_cast_kernel
(M: int, dtype_A: str, dtype_B: str)
testing/python/language/test_tilelang_language_vectorized_cast.py:8
↓ 1 callersFunctionverify_outputs
(outputs, refs, batch_sizes_list, atol=1e-2, rtol=1e-2)
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:96
↓ 1 callersFunctionvertical_slash_sparse_attention
( query: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD] key: torch.Tensor, # [BATCH, N_HEADS, N_CTX
examples/minference/example_vertical_slash_sparse_attn.py:484
↓ 1 callersFunctionview_shape_mismatch_test
(N, M, dtype, new_dtype=None)
testing/python/language/test_tilelang_language_view.py:57
↓ 1 callersFunctionview_test
(N, M, dtype, new_dtype=None)
testing/python/language/test_tilelang_language_view.py:9
↓ 1 callersMethodvisit
(self, node: ast.AST, parent: "ast.AST | None", field: "str | None", inside_list: bool)
tilelang/autodd.py:433
↓ 1 callersMethodvisit
(self, node: ast.AST, parent: "ast.AST | None", field: "str | None", inside_list: bool)
tilelang/autodd.py:495
↓ 1 callersFunctionwalk_indice
(expr)
tilelang/carver/roller/shape_inference/tir.py:311
↓ 1 callersFunctionweight_dequant
(weight, scale)
examples/deepseek_v32/inference/model.py:514
↓ 1 callersMethodwgmma_rs_atom
Emit a single WGMMA RS instruction for atom ``(inst_m_idx, inst_n_idx, ki)``. Must be called between a ``wgmma_fence_a``/``wgmma_fence_c``/``
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:510
↓ 1 callersFunctionwhile_loop
()
testing/python/language/test_tilelang_language_frontend_v2.py:325
↓ 1 callersMethodwith_default_adapter
(self, adapter: BaseKernelAdapter)
tilelang/profiler/__init__.py:58
↓ 1 callersMethodworker
(self, wid: int)
tilelang/autodd.py:1198
↓ 1 callersFunctionwrap_label
(s: str, width: int = 42)
maint/scripts/test_perf_regression.py:111
MethodA
(self)
tilelang/tileop/gemm/gemm_base.py:112
MethodA
(self)
tilelang/tileop/gemm/__init__.py:42
MethodA
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:76
MethodAPtr
(self)
tilelang/tileop/gemm/__init__.py:54
MethodARegion
(self)
tilelang/tileop/gemm/gemm_base.py:124
MethodARegion
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:92
FunctionASTPrinter
A visitor pass that renders the TileLang AST hierarchy in a visual tree format. Comparing with TL script, this printer is more suitable for
tilelang/analysis/ast_printer.py:109
MethodA_base_offsets
Get base offsets for each dimension of A region
tilelang/tileop/gemm/gemm_base.py:221
FunctionAddWrapperForSingleBufStore
Creates a TVM pass that wraps single buffer stores with parallel loops. This transformation adds T.Parallel wrappers around buffer stores th
tilelang/transform/add_bufstore_wrapper.py:6
FunctionAnnotateDeviceRegions
AnnotateDeviceRegions Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:217
FunctionAnnotateReadOnlyParams
Annotate read-only handle parameters for PrimFuncs. Adds attribute `tl.readonly_param_indices` listing param indices that are never written,
tilelang/transform/__init__.py:239
FunctionAnnotateWarpGroupRegAlloc
Inject set_max_nreg calls into warp-specialized functions. This pass analyzes the function to collect register hints from set_max_nreg and no
tilelang/cuda/transform/__init__.py:128
FunctionAssert
Create an assertion statement. Parameters ---------- condition : PrimExpr The PrimExpr to test. message : str The ou
tilelang/language/ast/ir.py:852
FunctionAtomicAddRet
Perform atomic addition and return the previous value. This is the same as AtomicAdd since nvvm.atomicrmw always returns old value.
tilelang/contrib/cutedsl/atomic.py:154
FunctionAtomicAddx2
Vectorized atomic add for 2 consecutive elements. Uses PTX atom.add.v2.f32 for float32 or atom.add.noftz.v2.f16 for float16. Args: d
tilelang/contrib/cutedsl/atomic.py:176
FunctionAtomicAddx4
Vectorized atomic add for 4 consecutive float32 elements. Uses PTX atom.global.add.v4.f32 for true vectorized atomic operation on SM90+. Arg
tilelang/contrib/cutedsl/atomic.py:241
FunctionAtomicLoad
Perform atomic load with specified memory ordering. Args: ptr: Pointer to load from memory_order: TileLang memory order ID (0=rel
tilelang/contrib/cutedsl/atomic.py:463
FunctionAtomicMaxRet
Perform atomic maximum and return the previous value.
tilelang/contrib/cutedsl/atomic.py:355
FunctionAtomicMinRet
Perform atomic minimum and return the previous value.
tilelang/contrib/cutedsl/atomic.py:433
FunctionAtomicStore
Perform atomic store with specified memory ordering. Args: ptr: Pointer to store to value: Value to store memory_order: T
tilelang/contrib/cutedsl/atomic.py:562
MethodB
(self)
tilelang/tileop/gemm/gemm_base.py:116
MethodB
(self)
tilelang/tileop/gemm/__init__.py:46
MethodB
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:84
MethodBPtr
(self)
tilelang/tileop/gemm/__init__.py:58
MethodBRegion
(self)
tilelang/tileop/gemm/gemm_base.py:128
MethodBRegion
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:100
MethodB_base_offsets
Get base offsets for each dimension of B region
tilelang/tileop/gemm/gemm_base.py:226
MethodC
(self)
tilelang/tileop/gemm/gemm_base.py:120
MethodC
(self)
tilelang/tileop/gemm/__init__.py:50
MethodC
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:88
FunctionCPUPassPipelineBody
(mod: IRModule, target: Target)
tilelang/cpu/pipeline.py:18
MethodCPtr
(self)
tilelang/tileop/gemm/__init__.py:62
MethodCRegion
(self)
tilelang/tileop/gemm/gemm_base.py:132
MethodCRegion
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:104
FunctionCUDAPassPipelineBody
(mod: IRModule, target: Target)
tilelang/cuda/pipeline.py:141
FunctionCUDASourceCodeKernel
Launch a kernel from CUDA source code or a CUDA source file. The code must follows the following rules: 1. The kernel source must be a valid
tilelang/language/kernel.py:372
MethodC_base_offsets
Get base offsets for each dimension of C region
tilelang/tileop/gemm/gemm_base.py:231
MethodC_coords
(self)
tilelang/tileop/gemm/gemm_base.py:180
FunctionClusterPlanning
ClusterPlanning Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:19
FunctionColumnMajorInterleaved
(i: int, j: int)
tilelang/layout/gemm_sp.py:138
FunctionConfigIndexBitwidth
Config index bitwidth. Returns ------- fpass : tvm.transform.Pass The result pass ----
tilelang/transform/__init__.py:289
MethodE
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:80
MethodERegion
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:96
FunctionElse
Create an else. Returns ------- res : frame.ElseFrame The result ElseFrame.
tilelang/language/ast/ir.py:1122
FunctionFlattenBuffer
FlattenBuffer Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:301
FunctionFragmentLoopChecker
When using T.Parallel over a local/fragment buffer, there are several restrictions: to ensure that the parallelization is valid. 1. The
tilelang/analysis/fragment_loop_checker.py:100
FunctionFuseMBarrierArriveExpectTx
Fuse simple expect_tx -> TMA issue -> arrive back into arrive_and_expect_tx.
tilelang/cuda/transform/__init__.py:67
FunctionHoistBroadcastValues
TVM Pass: HoistBroadcastValues. This pass scans the TIR for Broadcast operations involving immediate constants (IntImm, FloatImm). It ex
tilelang/transform/hoist_broadcast_values.py:71
FunctionHoistGlobalBufferAllocations
Hoist global buffer allocations to the top of the block (host side). Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:334
FunctionHoistNonRestrictParams
()
tilelang/transform/__init__.py:345
FunctionIf
Create an if node. Parameters ---------- condition : PrimExpr The condition of if statement, executes the true branch if the cond
tilelang/language/ast/ir.py:1092
FunctionIfStmtBinding
IfStmtBinding Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:140
FunctionInjectAssumes
Inject Assumes for natural shape boundary conditions. And convert Assumes in Evaluate(Call(...)) form (tvm builtin assume call) to AttrNode form.
tilelang/transform/__init__.py:101
FunctionInjectFenceProxy
InjectFenceProxy Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/cuda/transform/__init__.py:99
FunctionInjectPTXAsyncCopy
Deprecated alias of `LowerPTXAsyncCopy`.
tilelang/transform/__init__.py:284
FunctionInjectSoftwarePipeline
InjectSoftwarePipeline Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:79
FunctionInjectTcgen05Fence
Inject tcgen05.fence::before_thread_sync / after_thread_sync at conservative TCGEN05/TMEM synchronization boundaries on Blackwell (SM100+) tar
tilelang/cuda/transform/__init__.py:110
FunctionInstructionAnnotation
Annotate tile operations with coarse-grained instruction kind. This pass runs before LayoutInference and LowerTileOp. It adds a ``tl_instruc
tilelang/transform/__init__.py:41
MethodK
(self)
tilelang/tileop/gemm/gemm_base.py:82
MethodK
(self)
tilelang/tileop/gemm/__init__.py:74
MethodK
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:42
FunctionKernel
Tools to quickly construct a GPU kernel launch frame. Parameters ---------- blocks : int A list of extent, can be 1-3 dimension,
tilelang/language/kernel.py:267
FunctionLayoutInference
LayoutInference Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:57
FunctionLayoutReducer
Return a TVM transform pass that performs layout reduction/normalization. This wrapper delegates to the underlying FFI implementation and re
tilelang/transform/__init__.py:388
FunctionLayoutVisual
(formats: str | Sequence[str] = "")
tilelang/analysis/layout_visual.py:103
FunctionLegalizeNegativeIndex
Legalize negative indices in buffer loads. Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:90
FunctionLegalizeSafeMemoryAccess
LegalizeLoopVectorize Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/__init__.py:184
← previousnext →3,301–3,400 of 7,254, ranked by callers