MCPcopy Create free account

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

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

Functioncall_packed
Build expression by call an external packed function. The argument to packed function can be Expr or Buffer. The argument is the correspondin
tilelang/language/tir/op.py:25
Functioncall_packed_lowered
Lowered version of call packed. The argument to packed function can be Expr or Buffer. The argument is the corresponding POD type when Expr is
tilelang/language/tir/op.py:81
Functioncall_pure_extern
Build expression by calling a pure extern function. Parameters ---------- dtype : str The data type of the result. func_name
tilelang/language/tir/op.py:163
Functioncall_tir
Performs a call into another PrimFunc in the same IRModule Returns ------- call : PrimExpr The call expression.
tilelang/language/tir/op.py:376
Functioncallback
(node)
tilelang/carver/matmul_analysis.py:31
Methodcallback
(code, _)
testing/python/cache/test_tilelang_kernel_cache.py:69
Functioncallback_rocm_bitcode_path
Utility function to find ROCm device library bitcodes Parameters ---------- rocdl_dir : str The path to rocm library directory
tilelang/contrib/rocm.py:129
Functioncallback_rocm_link
Links object file generated from LLVM to HSA Code Object Parameters ---------- obj_bin : bytearray The object file Return
tilelang/contrib/rocm.py:105
Methodcan_free
(node, out_id)
tilelang/carver/roller/policy/default.py:474
Functioncaplog_vllm
(temporary_enable_log_propagate, caplog)
examples/bitnet-1.58b/vllm_workspace/conftest.py:576
Functioncdiv
(a, b)
examples/gemm_streamk/example_tilelang_gemm_streamk.py:8
Functioncdiv
(a: int, b: int)
tilelang/math/__init__.py:5
Functionceil
Take ceil of float input x. Parameters ---------- x : PrimExpr Input argument. span : Optional[Span] The location of
tilelang/language/tir/op.py:2648
Functionceildiv
Generic ceildiv operator. Parameters ---------- lhs : object The left operand. rhs : object The right operand. sp
tilelang/language/tir/op.py:3438
Functionceildiv_kernel
(A: T.Tensor((1,), T.int32))
testing/python/language/test_tilelang_language_ceildiv.py:9
Functioncheck_func_with_dynamic
(func)
tilelang/carver/analysis.py:169
Functioncheck_hopper
()
testing/python/jit/test_tilelang_jit_cutedsl.py:376
Functioncheck_logprobs_close
Compare the logprobs of two sequences generated by different models, which should be similar but not necessarily equal.
examples/bitnet-1.58b/vllm_workspace/utils.py:24
Functioncheck_outputs_equal
Compare the two sequences generated by different models, which should be equal.
examples/bitnet-1.58b/vllm_workspace/utils.py:6
Methodcheck_tile_shape_isvalid
(self, td: TileDict)
tilelang/carver/roller/policy/tensorcore.py:213
Methodcheck_tunable_argument_value
(key, parameters, key_args_tuple)
tilelang/autotuner/tuner.py:981
Functionchunk_bwd_kernel_dv_local
( q, k, g, g_gamma, A, do, dv, cu_seqlens, chunk_indices, scale, T
examples/kda/FLA_KDA/fla_chunk_o.py:360
Functionchunk_gated_delta_rule_bwd_kernel_dhu_blockdim64
( q, k, w, g, gk, dht, dh0, do, dh, dv, dv2, cu_seqlens, c
examples/kda/FLA_KDA/fla_chunk_delta.py:246
Functionchunk_gated_delta_rule_fwd_kernel_h_blockdim64
( k, v, w, v_new, g, gk, h, h0, ht, cu_seqlens, chunk_offsets,
examples/kda/FLA_KDA/fla_chunk_delta.py:33
Functionchunk_gla_bwd_kernel_dA
( v, do, dA, cu_seqlens, chunk_indices, scale, T, H: tl.constexpr, V: tl.c
examples/kda/FLA_KDA/fla_chunk_o.py:477
Functionchunk_gla_bwd_kernel_dv
( k, g, A, do, dh, dv, cu_seqlens, chunk_indices, T, H: tl.constexpr,
examples/kda/FLA_KDA/fla_chunk_o.py:118
Functionchunk_gla_bwd_kernel_inter
( q, k, v, g, h, do, dh, dq, dk, dq2, dk2, dg, cu_seqlens,
examples/kda/FLA_KDA/fla_chunk_o.py:192
Functionchunk_gla_fwd_kernel_o
( q, v, g, h, o, A, cu_seqlens, chunk_indices, scale, T, H: tl.con
examples/kda/FLA_KDA/fla_chunk_o.py:32
Functionchunk_global_cumsum
( s: torch.Tensor, reverse: bool = False, cu_seqlens: torch.Tensor = None, scale: float = None
examples/kda/FLA_KDA/cumsum.py:394
Functionchunk_global_cumsum_scalar_kernel
( s, o, scale, cu_seqlens, T, B: tl.constexpr, H: tl.constexpr, BT: tl.constex
examples/kda/FLA_KDA/cumsum.py:138
Functionchunk_global_cumsum_vector_kernel
( s, o, scale, cu_seqlens, T, B: tl.constexpr, H: tl.constexpr, S: tl.constexp
examples/kda/FLA_KDA/cumsum.py:200
Functionchunk_kda_bwd_kernel_inter
( q, k, v, g, h, do, dh, dq, dk, dv, dw, dg, cu_seqlens,
examples/kda/FLA_KDA/fla_chunk_inter.py:31
Functionchunk_kda_bwd_kernel_intra
( q, k, g, beta, dAqk, dAkk, dq, dq2, dk, dk2, dg, dg2, db
examples/kda/FLA_KDA/fla_chunk_intra.py:340
Functionchunk_kda_fwd_kernel_inter_solve_fused
Fused kernel: compute inter-subchunk Akk + solve_tril in one pass. Prerequisite: token_parallel has already computed diagonal Akk blocks in A
examples/kda/FLA_KDA/fla_chunk_intra.py:32
Functionchunk_kda_fwd_kernel_intra_token_parallel
( q, k, g, beta, Aqk, Akk, scale, cu_seqlens, N, T, H: tl.constexp
examples/kda/FLA_KDA/fla_chunk_intra_token_parallel.py:22
Functionchunk_local_cumsum_scalar_kernel
( s, o, scale, cu_seqlens, chunk_indices, T, B: tl.constexpr, H: tl.constexpr,
examples/kda/FLA_KDA/cumsum.py:25
Functionchunk_local_cumsum_vector_kernel
( s, o, scale, cu_seqlens, chunk_indices, T, B: tl.constexpr, H: tl.constexpr,
examples/kda/FLA_KDA/cumsum.py:78
Functionchunk_retention_fwd
( Q: T.Tensor([B, S, H, DK], dtype), # type: ignore K: T.Tensor([B, S, H, DK], dtype), # typ
examples/linear_attention/example_retention_fwd.py:31
Functionchunk_scan_fwd
( cb, x, dt, dA_cumsum, C, prev_states, D, block_M=64, block_N=64, blo
examples/linear_attention/example_mamba_chunk_scan.py:76
Functionchunk_scan_triton
(cb, x, dt, dA_cumsum, C, states, D)
examples/linear_attention/example_mamba_chunk_scan.py:10
Functionchunk_state_fwd
(B, x, dt, dA_cumsum, block_M=64, block_N=64, block_K=64, num_stages=2, threads=128)
examples/linear_attention/example_mamba_chunk_state.py:55
Functionchunk_state_triton
(B, x, dt, dA_cumsum)
examples/linear_attention/example_mamba_chunk_state.py:11
Functionclamp
Clamps the input value dst between [min_val, max_val] Args: dst: Input value to be clamped min_val: Minimum value max_val
tilelang/language/customize.py:31
Methodclass_attributes
Returns the class attributes in dictionary form. Returns: dict: Dictionary of class attributes.
tilelang/carver/template/gemv.py:135
Methodclass_attributes
Returns the class attributes in dictionary form. Returns: dict: Dictionary of class attributes.
tilelang/carver/template/flashattention.py:154
Methodclass_attributes
Returns the class attributes in dictionary form. Returns: dict: Dictionary of class attributes.
tilelang/carver/template/matmul.py:152
Methodclass_attributes
(self)
tilelang/carver/template/general_reduce.py:115
Methodclass_attributes
Returns the class attributes in dictionary form. Returns: dict: Dictionary of class attributes.
tilelang/carver/template/conv.py:189
Methodclass_attributes
Returns class attributes as a dictionary. Returns: dict: A dictionary representation of the class attributes.
tilelang/carver/template/elementwise.py:77
Functionclc_get_first_ctaid_x
Return the x coordinate of the first CTA in a successful CLC response.
tilelang/language/cluster.py:81
Functionclc_get_first_ctaid_y
Return the y coordinate of the first CTA in a successful CLC response.
tilelang/language/cluster.py:90
Functionclc_get_first_ctaid_z
Return the z coordinate of the first CTA in a successful CLC response.
tilelang/language/cluster.py:99
Functionclc_is_canceled
Return 1 when the CLC query successfully canceled a future launch.
tilelang/language/cluster.py:72
Functionclc_try_cancel
Issue a single-CTA cluster launch control query.
tilelang/language/cluster.py:52
Functionclc_try_cancel_multicast
Issue a cluster-wide multicast cluster launch control query.
tilelang/language/cluster.py:62
Functionclean_cache_env
Provide isolated cache environment for each test. Creates isolated cache/tmp directories to ensure: - No interference from previous test runs
testing/python/cache/test_tilelang_kernel_cache.py:100
Methodcleanup
Explicitly cleanup this adapter's CUDA resources. This method can be called explicitly to immediately release CUDA resources without
tilelang/jit/adapter/cutedsl/adapter.py:454
Functioncleanup_fixture
(should_do_global_cleanup_after_test: bool)
examples/bitnet-1.58b/vllm_workspace/conftest.py:103
Functionclear
Clear a buffer by filling it with zeros. Args: buffer: Either a TVM buffer or a variable that contains a buffer region Returns:
tilelang/language/fill_op.py:40
Methodclear_accum
(self)
tilelang/tileop/gemm/gemm_base.py:152
Methodclear_accum
(self)
tilelang/tileop/gemm/__init__.py:102
Methodclear_accum
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:124
Methodclear_cache
Clears the entire kernel cache, including both in-memory and disk cache.
tilelang/cache/kernel_cache.py:403
Functioncluster_arrive
Issue barrier.cluster.arrive.aligned.
tilelang/language/cluster.py:32
Functioncluster_arrive_relaxed
Issue barrier.cluster.arrive.relaxed.aligned.
tilelang/language/cluster.py:27
Functioncluster_sync
Issue cluster barrier arrive + wait (full synchronization).
tilelang/language/cluster.py:42
Functioncluster_wait
Issue barrier.cluster.wait.aligned.
tilelang/language/cluster.py:37
Functionclz
Count leading zero bits of an integer x. Parameters ---------- x : PrimExpr Input 32 or 64 bit integer. The result is und
tilelang/language/tir/op.py:2612
Functioncollect_assumes
(stmt)
testing/python/transform/test_tilelang_transform_split_host_device.py:44
Functioncollect_vars_from_expr
(expr)
testing/python/transform/test_tilelang_transform_split_host_device.py:51
Functioncomm_reducer
Create a CommReducer from lambda inputs/outputs and the identities Parameters ---------- combiner : Callable A binary functi
tilelang/language/ast/ir.py:1636
Methodcompile_lib
(self, timeout: float = None)
tilelang/jit/adapter/libgen.py:60
Methodcompile_unit
(unit_items: list[tuple[int, dict[str, Any]]])
tilelang/autotuner/tuner.py:572
Functioncompress_32bit_ordered_metadata
( dense: T.Tensor([S, D], dtype), nonzero: T.Tensor([S, D * elem // group], dtype),
tilelang/utils/sparse.py:126
Functioncompress_8bit_16bit_ordered_metadata
( dense: T.Tensor([S, D], dtype), nonzero: T.Tensor([S, D * elem // group], dtype),
tilelang/utils/sparse.py:54
Methodcompute_func
(*spatial_indices)
tilelang/carver/template/general_reduce.py:75
Methodcompute_node_stride_map
(self, node: PrimFuncNode, td: TileDict)
tilelang/carver/roller/policy/tensorcore.py:243
Methodcompute_strides_from_shape
(self, shape: list[int])
tilelang/carver/roller/hint.py:48
Methodcompute_warp_partition
(self, M: int, N: int, block_size: int, target: Target, gemm_inst: str)
tilelang/ir.py:42
Methodcondense_rep_var
Condense or fold the replicate variable into the existing iteration space. This operation may be used to reduce dimensionality if the
tilelang/layout/fragment.py:153
Functionconv
( data: T.Tensor((N, H, W, C), T.float16), weight: T.Tensor((KH, KW, C, F), T.float16),
testing/python/transform/test_tilelang_transform_im2col_fallback.py:18
Functionconv
( data: T.Tensor((N, H, W, C), T.float16), weight: T.Tensor((KH, KW, C, F), T.float16),
testing/python/transform/test_tilelang_transform_pipeline_barrier_ownership.py:107
Functionconvert_func
Convert a tvm function into one that accepts a tensor from another framework, provided the other framework supports DLPACK Parameters
tilelang/contrib/dlpack.py:22
Methodconvert_tokens_to_string
Converts a sequence of tokens (string) in a single string.
examples/bitnet-1.58b/tokenization_bitnet.py:292
Functionconvolution
( data, kernel, S, D, P, block_M, block_N, block_K, num_stages, thread_num, enable_rasteration, dtype=T.fl
examples/convolution/example_convolution_autotune.py:77
Functioncopy1
(A, B)
testing/python/language/test_tilelang_language_eager_jit.py:106
Functioncopy2
( A: T.Tensor[[128, 128], T.float32], B: T.Tensor[[128, 128], T.float32], )
testing/python/language/test_tilelang_language_eager_jit.py:113
Functioncopy3
(A, B)
testing/python/language/test_tilelang_language_eager_jit.py:120
Functioncopy4
(A, B)
testing/python/language/test_tilelang_language_eager_jit.py:127
Functioncopy5
(A, B)
testing/python/language/test_tilelang_language_eager_jit.py:135
Functioncopy6
(A, B)
testing/python/language/test_tilelang_language_eager_jit.py:142
Functioncopy_cluster
Cluster-aware copy for TMA multicast or SM-to-SM shared-memory copy. Args: src: Source memory region. dst: Destination memory reg
tilelang/language/copy_op.py:137
Functioncopy_loop_num_stages_one
( x: T.Tensor((M, K), T.float16), y: T.Tensor((M, K), T.float16), )
testing/python/issue/test_tilelang_issue_tma_no_ws.py:79
Functioncopy_loop_num_stages_zero
( x: T.Tensor((M, K), T.float16), y: T.Tensor((M, K), T.float16), )
testing/python/issue/test_tilelang_issue_tma_no_ws.py:34
Functioncopy_with_layout_kernel
(A, B, loop_layout)
testing/python/layout/test_tilelang_annotate_loop_layout.py:50
Functioncopy_with_padded_layout
( x: T.Tensor((M, K), "float16"), y: T.Tensor((M, K), "float16"), )
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:505
Functioncopysign
Change the sign of x1 to that of x2, element-wise. Parameters ---------- x1 : PrimExpr Input argument. x2 : PrimExpr
tilelang/language/tir/op.py:2894
Functioncopysignf
(x: Union[TensorSSA, Numeric], y: Union[TensorSSA, Numeric], fastmath: bool = False)
tilelang/contrib/cutedsl/math.py:132
Functioncos
(x: Union[TensorSSA, Numeric], fastmath: bool = False, **kwargs)
tilelang/contrib/cutedsl/math.py:106
Functioncos
Take cos of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The res
tilelang/language/tir/op.py:2401
← previousnext →4,201–4,300 of 7,254, ranked by callers