MCPcopy Create free account

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

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

Method_warp_mma_blockscaled
(A_buf, B_buf, C_local_buf, sfa_data, sfb_data, mbar)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:424
Method_warp_mma_sp
(A_local_buf, E_local_buf, B_local_buf, C_local_buf)
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:521
Method_warp_mma_ss
(A_buf, B_buf, C_local_buf, mbar)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:255
Method_warp_mma_ts
(a_data, B_buf, C_local_buf, mbar)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:309
Method_warp_stmatrix_global
(C_local_buf, C_buf, thread_binding)
tilelang/rocm/intrinsics/wmma_macro_generator.py:365
Method_warp_stmatrix_global
(C_local_buf, C_buf, thread_binding)
tilelang/rocm/intrinsics/mfma_macro_generator.py:549
Method_warp_stmatrix_global
(C_local_buf, C_buf, thread_binding)
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:652
Method_warp_stmatrix_global
(C_local_buf, C_buf, thread_binding)
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:603
Method_warp_stmatrix_shared
(C_local_buf, C_buf, thread_binding)
tilelang/rocm/intrinsics/wmma_macro_generator.py:349
Method_warp_stmatrix_shared
(C_local_buf, C_buf, thread_binding)
tilelang/rocm/intrinsics/mfma_macro_generator.py:534
Method_warp_stmatrix_shared
(C_local_buf, C_buf, thread_binding)
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:635
Method_warp_stmatrix_shared
(C_local_buf, C_buf, thread_binding)
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:586
Method_warp_wmma
(A_local_buf, B_local_buf, C_local_buf)
tilelang/rocm/intrinsics/wmma_macro_generator.py:306
Function_watchdog
()
tilelang/autotuner/tuner.py:121
Function_worker_loop
(input_queue, output_queue)
tilelang/autodd.py:740
Method_wrap_forward_from_prebuild_lib
High-level wrapper for kernel execution. Handles: 1. Input validation 2. Output tensor allocation 3. Dynamic shape re
tilelang/jit/adapter/nvrtc/adapter.py:218
Method_wrap_forward_from_prebuild_lib
High-level wrapper for kernel execution. Handles: 1. Input validation 2. Output tensor allocation 3. Dynamic shape re
tilelang/jit/adapter/cutedsl/adapter.py:326
Function_wrapper
(*args)
tilelang/contrib/dlpack.py:54
Function_wrapper
(*args, **kwargs)
tilelang/utils/deprecated.py:33
Methoda_dtype
A operand dtype.
tilelang/tileop/gemm/gemm_base.py:94
Methoda_dtype
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:62
Functionabs2
Packed element-wise absolute value.
tilelang/language/math_intrinsics.py:407
Functionaccess_ptr
Create a TileLang `tl.access_ptr` from a buffer-like base location. This is a frontend convenience wrapper that keeps a `BufferLoad` argument
tilelang/language/builtin.py:93
Methodaccum_dtype
(self)
tilelang/tileop/gemm/gemm_base.py:104
Methodaccum_dtype
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:72
Functionacos
Take acos of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The re
tilelang/language/tir/op.py:2433
Functionacosh
Take acos of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The re
tilelang/language/tir/op.py:2449
Functionact_quant_kernel_
( X: T.Tensor[(M, N), in_dtype], Y: T.Tensor[(M, N), out_dtype], S: T.Tensor[(M, T.cei
examples/deepseek_v32/inference/kernel.py:44
Functionactivemask
Return a ``uint64`` bitmask of currently active (non-exited) lanes. Lowers to ``__activemask()`` (zero-extended to ``uint64``) on CUDA and ``
tilelang/language/builtin.py:1088
Functionadd2
Packed element-wise add (x + y).
tilelang/language/math_intrinsics.py:358
Methodadd_node
(self, node)
maint/scripts/pytest_cuda_scheduler.py:98
Methodadd_to_queue
(tile)
tilelang/carver/roller/policy/default.py:114
Functionaddress_of
Returns the address of an element in the buffer Parameters ---------- buffer_load: BufferLoad The buffer load. span : Option
tilelang/language/tir/op.py:481
Functionafter
(A: T.Tensor((1024, 32), T.float16), B: T.Tensor((32, 1024), T.float16), C: T.Tensor((1024, 1024), T.float16))
testing/python/transform/test_tilelang_transform_cluster_planning.py:42
Functionafter
(A: T.Tensor((1024,), T.float32))
testing/python/transform/test_tilelang_transform_legalize_negative_index.py:28
Functionafter
()
testing/python/transform/test_tilelang_transform_hoist_broadcast_values.py:75
Functionafter
()
testing/python/transform/test_tilelang_transform_inject_fence_proxy.py:48
Functionafter
( A: T.Tensor((1024, 32), T.float32), B: T.Tensor((32, 1024), T.float32), C: T.Tensor(
testing/python/transform/test_tilelang_transform_pipeline_planning.py:66
Functionafter
()
testing/python/transform/test_tilelang_transform_inject_tcgen05_fence.py:84
Functionafter
(b: T.Tensor[(16,), T.float4_e2m1fn])
testing/python/transform/test_tilelang_transform_decouple_type_cast.py:28
Functionall_of
Check if all elements in the buffer are true. Args: buffer: Either a TVM buffer or buffer region to be checked Returns: A TV
tilelang/language/logical.py:51
Functionall_sync
Non-zero only if ALL active lanes in ``mask`` have a non-zero predicate. Lowers to ``__all_sync(mask, predicate)`` on CUDA and ``__all(predicate)
tilelang/language/builtin.py:1042
Functionalloc_barrier
Allocate a barrier buffer. Args: arrive_count (int | list[int]): The number of threads that need to arrive at each barrier Returns:
tilelang/language/allocate.py:177
Functionalloc_buffer
The buffer allocation function. Parameters ---------- shape : ShapeType | PrimExpr | Integral The type of the buffer prior to fla
tilelang/language/ast/ir.py:443
Functionalloc_cluster_barrier
Allocate a cluster barrier buffer. Args: arrive_count (int | list[int]): The number of threads that need to arrive at each barrier R
tilelang/language/allocate.py:205
Functionalloc_global
Allocate a global memory buffer as a global workspace. NOTE(chaofan): Memory allocated in this way doesn't go through torch allocator. Instead,
tilelang/language/allocate.py:154
Functionalloc_local
Allocate a local memory buffer for thread-private storage. Args: shape (tuple): The shape of the buffer to allocate dtype (str):
tilelang/language/allocate.py:52
Functionalloc_reducer
Allocate a reducer buffer. Modifications needs to conform with `op`, such as `op="sum"` requires `reducer[...] += ...` and `op="max"
tilelang/language/allocate.py:263
Functionalloc_tcgen05_instr_desc
(dtype: DType = _dtypes.uint32)
tilelang/language/allocate.py:334
Functionalloc_tcgen05_smem_desc
(dtype: DType = _dtypes.uint64)
tilelang/language/allocate.py:325
Functionalloc_tmem
Allocate a Tensor Memory (TMEM) buffer for use with 5th generation Tensor Core operations (e.g., TCGEN5.MMA). TMEM is a dedicated on-chip me
tilelang/language/allocate.py:228
Functionalloc_var
(dtype: DType, init: PrimExpr | int | float, scope: str = "local.var")
tilelang/language/allocate.py:81
Functionalloc_wgmma_desc
(dtype: DType = _dtypes.uint64)
tilelang/language/allocate.py:321
Functionallocate
Allocate node. Parameters ---------- extents : List[PrimExpr] The extents of the allocate. dtype : str The data type
tilelang/language/ast/ir.py:980
Functionallocate_const
Allocate constant node. Parameters ---------- data : List[PrimExpr] The data associated with the constant. dtype : str
tilelang/language/ast/ir.py:1013
Methodallow_f8f6f4_mixed_dtypes
(self)
tilelang/tileop/gemm/gemm_base.py:36
Methodallow_f8f6f4_mixed_dtypes
(self)
tilelang/cuda/op/gemm/gemm_tcgen05.py:54
Functionallow_global_thread_synchronization
(pass_ctx: PassContext | None = None)
tilelang/backend/pass_pipeline/pipeline_utils.py:17
Functionannotate_compile_flags
Annotate additional device compile flags inside a function body. The flags will be merged with any externally provided compile_flags at
tilelang/language/eager/builder.py:965
Functionannotate_consumer_reg_alloc
Annotate the consumer reg alloc.
tilelang/language/builtin.py:417
Functionannotate_layout
Annotate the layout of the buffer.
tilelang/language/annotations.py:29
Functionannotate_min_blocks_per_sm
Annotate the minimum number of thread blocks per SM (multiprocessor). When set, this value is passed as the second argument of ``__launch_bou
tilelang/language/annotations.py:62
Functionannotate_pass_configs
Annotate pass configuration inside a function body. The configs will be merged with any externally provided pass_configs at compilation
tilelang/language/eager/builder.py:987
Functionannotate_producer_reg_dealloc
Annotate the producer reg dealloc.
tilelang/language/builtin.py:408
Functionannotate_restrict_buffers
Mark the given buffer parameters as non-restrict. This annotation tells codegen to omit the `__restrict__` qualifier for the specified kernel
tilelang/language/annotations.py:84
Functionannotate_safe_value
Annotate the safe value of the buffer.
tilelang/language/annotations.py:45
Functionany_of
Check if any element in the buffer is true. Args: buffer: Either a TVM buffer or buffer region to be checked Returns: A TVM
tilelang/language/logical.py:12
Functionany_sync
Non-zero if ANY active lane in ``mask`` has a non-zero ``predicate``. Lowers to ``__any_sync(mask, predicate)`` on CUDA and ``__any(predicate)``
tilelang/language/builtin.py:1022
Functionanylist_getitem
Returns an item from any list. list_handle: Var The handle to anylist index : int The index Returns ------- call :
tilelang/language/tir/op.py:3544
Functionanylist_resetitem
Reset an item from any list. list_handle: Var The handle to anylist index : int The index Returns ------- call : P
tilelang/language/tir/op.py:3558
Functionanylist_setitem_call_cpacked
Set anylist item by result of packed call. list_handle: Var The handle to anylist index : int The index func_name: str
tilelang/language/tir/op.py:3590
Functionanylist_setitem_call_packed
Set anylist item by result of packed call. list_handle: Var The handle to anylist index : int The index func_name: str
tilelang/language/tir/op.py:3572
Methodapply
(self, target_labels: set[int])
tilelang/autodd.py:630
Functionarange_semi_sparse
Generate a semi-sparse tensor with values from 0 to M*K-1. The generated tensor will have 2:4 sparsity along the K dimension. Args: M
tilelang/utils/sparse.py:324
Methodarch
Returns the current architecture. Returns: TileDevice: The architecture of this template.
tilelang/carver/template/base.py:166
Functionarg
The PrimFunc arguments adding function. Parameters ---------- name : str The name of the argument. var : Union[Var, Buffer]
tilelang/language/ast/ir.py:189
Functionarray_reduce
Reduce an array of integers to a single integer. Args: array (List[int]): The array of integers to reduce. Returns: int
tilelang/utils/language.py:157
Functionasin
Take asin of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The re
tilelang/language/tir/op.py:2497
Functionasinh
Take asinh of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The r
tilelang/language/tir/op.py:2513
Functionassert_close
(prefix, ref, tri, ratio, warning=False, err_atol=1e-6)
examples/kda/FLA_KDA/fla_utils.py:44
Methodassert_consistent
Checks for kernel consistency across multiple runs. Args: repeat: Number of times to repeat the consistency check
tilelang/profiler/__init__.py:195
Methodassert_expr
(self, cond: Any, msg: Any)
tilelang/language/eager/ast.py:235
Methodassert_expr
(self, cond, msg=None)
tilelang/language/eager/builder.py:689
Functionassert_similar
(x, y, eps=1e-8, name="tensor", data="", raise_assert=True)
examples/dequantize_gemm/dequantize_utils.py:116
Functionassert_similar
(x, y, eps=1e-8, name="tensor", data="", raise_assert=True)
examples/kda/test_utils_kda.py:20
Functionassert_tensors_similar
Assert that two tensors are similar using a global similarity metric. Key differences from torch.testing.assert_close: - torch.testing.a
examples/dsa_hisa/tilelang_utils.py:186
Methodassign_compile_flags
Accept compile flags without invoking the real generator.
testing/python/jit/test_tilelang_jit_cutedsl_host_codegen.py:227
Methodassign_pass_configs
(self, pass_configs: dict[str, Any] | None = None)
tilelang/jit/adapter/libgen.py:41
Methodassign_slice
(self, lval: Any, sl: slice, value: Any, annot: Any = empty)
tilelang/language/eager/ast.py:208
Methodassign_slice
(self, lval: Any, sl: slice, value: Any, annot=BaseBuilder.empty)
tilelang/language/eager/builder.py:557
Methodassigned
(self)
maint/scripts/pytest_cuda_scheduler.py:66
Functionassume
Provide a true statement that can be used for simplifications Parameters ---------- cond : Expr The constraint condition. Ret
tilelang/language/tir/op.py:349
Functionasync_copy
Asynchronous copy primitive lowered through cp.async. This operator is intended for explicitly asynchronous global->shared copy. The backend
tilelang/language/copy_op.py:190
Functionatan
Take atan of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The re
tilelang/language/tir/op.py:2529
Functionatan2
Take arctan2(x1, x2). Parameters ---------- x1 : PrimExpr Input argument. x2 : PrimExpr Input argument. Returns
tilelang/language/tir/op.py:2561
Functionatanh
Take atanh of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The r
tilelang/language/tir/op.py:2545
Functionatomic_add
(A: T.Tensor((K, M, N), dtype), B: T.Tensor((M, N), dtype))
testing/python/language/test_tilelang_language_atomic.py:13
Functionatomic_add
Atomically add `value` into `dst`, returning a handle to the operation. Supports scalar/addressed extern atomic add when neither argument ex
tilelang/language/atomic.py:185
Functionatomic_addx2
(A: T.Tensor((M, N), dtype), B: T.Tensor((M, N), dtype))
testing/python/language/test_tilelang_language_atomic.py:65
Functionatomic_addx2
Perform an atomic addition operation with double-width operands. Args: dst (Buffer): Destination buffer where the atomic addition will be
tilelang/language/atomic.py:281
← previousnext →4,001–4,100 of 7,254, ranked by callers