MCPcopy Create free account

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

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

↓ 1 callersMethodget_thread_binding
Return the thread index expression for the current kernel.
tilelang/metal/intrinsics/metal_macro_generator.py:49
↓ 1 callersMethodget_thread_bindings
Returns the thread binding for the given dimension. dim=0 corresponds to threadIdx.x, dim=1 to threadIdx.y, and dim=2 to threadIdx.z.
tilelang/language/kernel.py:216
↓ 1 callersFunctionget_thread_extents
Returns all three thread extents.
tilelang/language/kernel.py:469
↓ 1 callersMethodget_thread_extents
Returns the thread extents for all three dimensions.
tilelang/language/kernel.py:203
↓ 1 callersMethodget_tir
(self, *args, **kwargs)
tilelang/language/eager/builder.py:1222
↓ 1 callersFunctionget_tma_aligned_size
Global memory address of TMA must be 16-byte aligned. Since we use column-major layout for the LHS scaling tensor, the M-axis of the
examples/cast/example_group_per_split_token_cast_to_fp8.py:73
↓ 1 callersMethodget_tunner
(self)
tilelang/autotuner/tuner.py:1233
↓ 1 callersFunctionget_used_variables
Collects all variables used in the given operation. Args: operation: The TIR operation to analyze
tilelang/transform/add_bufstore_wrapper.py:25
↓ 1 callersMethodget_value
Get the value bound to a variable in any active frame. Args: var (Var): The variable to look up Returns: The
tilelang/language/frame.py:185
↓ 1 callersFunctionget_windows_runtime_dll_dirs
Return Windows-only DLL directories shipped with sibling Python packages. Currently locates ``tvm_ffi`` and ``z3`` install dirs so their DLLs res
tilelang/env.py:434
↓ 1 callersFunctionget_wmma_a_fragment_forward_func
Return the fragment forward function for A layout.
tilelang/rocm/intrinsics/wmma_layout.py:397
↓ 1 callersFunctionget_wmma_a_layout_funcs
Return (forward_map, reverse_map) for A layout.
tilelang/rocm/intrinsics/wmma_layout.py:323
↓ 1 callersFunctionget_wmma_b_fragment_forward_func
Return the fragment forward function for B layout.
tilelang/rocm/intrinsics/wmma_layout.py:406
↓ 1 callersFunctionget_wmma_b_layout_funcs
Return (forward_map, reverse_map) for B layout.
tilelang/rocm/intrinsics/wmma_layout.py:348
↓ 1 callersFunctionget_wmma_c_layout_funcs
Return (forward_map, reverse_map) for C/D layout.
tilelang/rocm/intrinsics/wmma_layout.py:373
↓ 1 callersFunctionget_wmma_fragment_replicate_count
Return the fragment replicate count used for logical one-to-many owners.
tilelang/rocm/intrinsics/wmma_layout.py:415
↓ 1 callersFunctionget_wmma_store_index_map_func
Return the (thread_id, local_id) -> (row, col) store map.
tilelang/rocm/intrinsics/wmma_layout.py:388
↓ 1 callersFunctionget_wrong_kernel
(M: int = 4096)
testing/python/issue/test_tilelang_issue_1374.py:8
↓ 1 callersFunctiongrid_sync
(N=1024)
testing/python/language/test_tilelang_language_cooperative.py:8
↓ 1 callersFunctiongrouped_blockscaled_gemm_ref
(a, b, sfa_packed, sfb_packed, offsets, sf_granularity_k=128, transpose_B=True)
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:487
↓ 1 callersFunctiongrouped_gemm_bwd
args: a (torch.Tensor): Input tensor of shape (M, K). b (torch.Tensor): Input tensor of shape (G, K, N).
examples/grouped_gemm/example_grouped_gemm_bwd.py:152
↓ 1 callersFunctiongrouped_gemm_fwd
args: a (torch.Tensor): Input tensor of shape (M, K). b (torch.Tensor): Input tensor of shape (G, K, N).
examples/grouped_gemm/example_grouped_gemm_bwd.py:9
↓ 1 callersFunctiongrouped_gemm_inputs
(batch_sizes, K, N, block_M, dtype="float16")
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:278
↓ 1 callersFunctiongrouped_gemm_padded_pipelined
Grouped GEMM with padded M tiles to exercise WS shared-prelude local vars.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:207
↓ 1 callersFunctiongrouped_gemm_ptr
(batch_sizes_list, K, N, block_M, block_N, block_K, num_stages=2, threads=128, dtype=T.float16)
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:26
↓ 1 callersFunctiongrouped_gemm_reference
(A, B, batch_sizes)
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:266
↓ 1 callersFunctionguarded_prelude_tma_postloop_scalar
Nested guarded pipeline whose post-loop consumer uses a prelude scalar.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:148
↓ 1 callersFunctionhandle
Create a TIR var that represents a pointer. Parameters ---------- dtype: str The data type of the pointer. storage_scope: st
tilelang/language/ast/ir.py:1464
↓ 1 callersFunctionhas_internal_prim_func
(func: Callable[_P, _T])
tilelang/language/eager/ast.py:635
↓ 1 callersMethodhas_value
Check if a variable has a binding in any active frame. Args: var (Var): The variable to check Returns: bool:
tilelang/language/frame.py:197
↓ 1 callersFunctionhave_mbarrier
Whether hardware mbarrier support (the cutlass Barrier type, tl_shuffle_elect, fence_barrier_init) is available on the target. Available on H
tilelang/contrib/nvcc.py:598
↓ 1 callersFunctionhave_pdl
(target)
tilelang/contrib/nvcc.py:590
↓ 1 callersFunctionhave_tma
Whether TMA support is provided in the specified compute capability or not Parameters ---------- target : tvm.target.Target The c
tilelang/contrib/nvcc.py:564
↓ 1 callersFunctionhipcc_remark_flag
The clang flag callers should pass to hipcc to enable the remark output we parse here.
tilelang/contrib/hip_resource_info.py:58
↓ 1 callersFunctionim2col
Perform im2col transformation for 2D convolution. Args: img (tirx.Buffer): Input image buffer col (tirx.Buffer): Output column bu
tilelang/language/copy_op.py:534
↓ 1 callersMethodimport_from_file
Dynamically import Python module from file path. Standard importlib pattern for loading modules outside sys.path. Used to import gene
tilelang/jit/adapter/nvrtc/libgen.py:77
↓ 1 callersMethodimport_from_file
(module_name, file_path)
tilelang/jit/adapter/cutedsl/libgen.py:31
↓ 1 callersFunctioninf_init_kernel
()
testing/python/amd/test_tilelang_hip_codegen.py:436
↓ 1 callersMethodinfer
(self, shape: dict[str, list[arith.ConstIntBound]], rstep: dict[str, int] = None, targets=None)
tilelang/carver/roller/shape_inference/tir.py:228
↓ 1 callersMethodinfer_layout
Infer the layout for the GEMM operation based on target architecture.
tilelang/tileop/gemm/__init__.py:121
↓ 1 callersMethodinfer_layout
(self, target: Target, thread_nums: int)
tilelang/tileop/gemm_sp/__init__.py:56
↓ 1 callersMethodinfer_node_smem_usage
Infers the shared memory usage of a node given a TileDict configuration. Parameters ---------- td : TileDict
tilelang/carver/roller/policy/default.py:434
↓ 1 callersMethodinfer_tensorcore_axis
(self)
tilelang/carver/roller/node.py:407
↓ 1 callersFunctioninitialize_cuda
Initialize CUDA and load the custom operators module.
maint/precision/compare_ops.py:60
↓ 1 callersMethodinitialize_function
Placeholder method that should be implemented by subclasses. This method is responsible for initializing the function. Raise
tilelang/carver/template/base.py:117
↓ 1 callersFunctioninterleave_weight
(qweight, nbits=4, target_dtype=T.float16)
examples/bitnet-1.58b/kernel_benchmark/tilelang_bitnet_158_int8xint2_decode.py:197
↓ 1 callersFunctioninterleave_weight
(qweight, nbits=4, target_dtype=T.float16)
examples/bitnet-1.58b/kernel_benchmark/tilelang_bitnet_158_int8xint2_prefill.py:320
↓ 1 callersFunctioninvalid_loop_with_complex_dataflow
(dtype: T.dtype = T.bfloat16, accum_dtype: T.dtype = T.float32, num_threads: int = 128)
testing/python/analysis/test_tilelang_fragment_loop_checker.py:51
↓ 1 callersMethodir_pass
Traverse and transform the IR module to extract performance-related information. Returns: self: The Analyzer instance.
tilelang/tools/Analyzer.py:108
↓ 1 callersFunctionis_ada_arch
(arch: TileDevice)
tilelang/carver/arch/cuda.py:32
↓ 1 callersMethodis_cache_globally_disabled
(self)
tilelang/env.py:364
↓ 1 callersFunctionis_cdna_arch
(arch: TileDevice)
tilelang/carver/arch/cdna.py:11
↓ 1 callersFunctionis_cuda_event
(event)
tilelang/profiler/bench.py:203
↓ 1 callersFunctionis_dequantize
(block: SBlockRV)
tilelang/carver/matmul_analysis.py:422
↓ 1 callersMethodis_enabled
Return current cache state.
tilelang/env.py:197
↓ 1 callersFunctionis_float4_e2m1_unpacked
Whether *value* is the 8-bit FP4 E2M1 unpacked shared-memory variant.
tilelang/language/dtypes.py:294
↓ 1 callersFunctionis_float4_e2m1fn
Whether *value* is the packed 4-bit FP4 E2M1 variant.
tilelang/language/dtypes.py:289
↓ 1 callersMethodis_full_col
Check if the policy is a full column partitioning. Returns: bool: True if the policy is full column, False otherwise.
tilelang/tileop/base.py:32
↓ 1 callersMethodis_full_row
Check if the policy is a full row partitioning. Returns: bool: True if the policy is full row, False otherwise.
tilelang/tileop/base.py:23
↓ 1 callersMethodis_gemm_ss
(self)
tilelang/metal/op/gemm/gemm_metal.py:17
↓ 1 callersFunctionis_global
Check if the buffer is in the global memory scope. Args: buffer: The TVM buffer, BufferLoad, or BufferRegion to check. Returns:
tilelang/utils/language.py:34
↓ 1 callersFunctionis_hip
()
benchmark/blocksparse_attention/benchmark_triton_block_sparse_fmha.py:10
↓ 1 callersFunctionis_hip
()
examples/blocksparse_attention/block_sparse_attn_triton.py:9
↓ 1 callersFunctionis_hip
()
examples/seer_attention/block_sparse_attn_triton.py:10
↓ 1 callersFunctionis_hopper_arch
(arch: TileDevice)
tilelang/carver/arch/cuda.py:39
↓ 1 callersFunctionis_local
Check if the buffer is in the local memory scope. Args: buffer: The TVM buffer, BufferLoad, or BufferRegion to check. Returns:
tilelang/utils/language.py:94
↓ 1 callersFunctionis_local_var
Check if the buffer is in the local.var memory scope. Args: buffer: The TVM buffer, BufferLoad, or BufferRegion to check. Retur
tilelang/utils/language.py:136
↓ 1 callersFunctionis_pipelined_for
Check if a for loop is pipelined.
tilelang/analysis/nested_loop_checker.py:11
↓ 1 callersMethodis_reduction
Whether the block is a reduction workload.
tilelang/carver/analysis.py:100
↓ 1 callersMethodis_running_autodd
Return True if we are running under `python -m tilelang.autodd`.
tilelang/env.py:388
↓ 1 callersFunctionis_shared_dynamic
Check if the buffer is in the dynamic shared memory scope. Args: buffer: The TVM buffer, BufferLoad, or BufferRegion to check.
tilelang/utils/language.py:66
↓ 1 callersFunctionis_spatial
(block: SBlockRV)
tilelang/carver/analysis.py:270
↓ 1 callersMethodis_square
Check if the policy is a square partitioning. Returns: bool: True if the policy is square, False otherwise.
tilelang/tileop/base.py:14
↓ 1 callersFunctionis_symbolic_expr
Check if the expression is a symbolic expression. A symbolic expression can be a simple tvm.Var, or an tvm.PrimExpr containing tvm.Var.
tilelang/jit/adapter/cython/adapter.py:31
↓ 1 callersFunctionis_tensorcore_supported_precision
(in_dtype: str, accum_dtype: str, arch: TileDevice)
tilelang/carver/arch/cuda.py:100
↓ 1 callersFunctionis_tile_op
Check if a call is a tile-op
tilelang/analysis/nested_loop_checker.py:18
↓ 1 callersFunctionis_transpose_block
(block_stmt: tirx.SBlock)
tilelang/carver/matmul_analysis.py:468
↓ 1 callersMethodjit_compile
(**config_arg)
tilelang/autotuner/tuner.py:1291
↓ 1 callersFunctionkernel
( K: T.Tensor(K_shape, dtype=input_dtype), V: T.Tensor(V_shape, dtype=input_dtype), Be
examples/gdn/example_wy_fast.py:76
↓ 1 callersFunctionkernel
( # input K: T.Tensor(K_shape, dtype=input_dtype), V: T.Tensor(V_shape, dtype=input_dt
examples/gdn/example_wy_fast_bwd_split.py:136
↓ 1 callersFunctionkernel
( K: T.Tensor(K_shape, dtype=input_dtype), W: T.Tensor(W_shape, dtype=input_dtype), U:
examples/gdn/example_chunk_delta_h.py:138
↓ 1 callersFunctionkernel
( # Input Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dt
examples/gdn/example_chunk_delta_bwd.py:227
↓ 1 callersFunctionkernel
( Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dtype), V:
examples/gdn/example_chunk_o.py:90
↓ 1 callersFunctionkernel
( # input Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dt
examples/gdn/example_chunk_o_bwd.py:156
↓ 1 callersFunctionkernel
( K: T.Tensor(K_shape, dtype=input_dtype), Beta: T.Tensor(Beta_shape, dtype=input_dtype),
examples/gdn/example_chunk_scaled_dot_kkt.py:78
↓ 1 callersFunctionkernel
( Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dtype), GK
examples/kda/chunk_inter_solve_fused.py:91
↓ 1 callersFunctionkernel
( Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dtype), GK
examples/kda/chunk_intra_token_parallel.py:85
↓ 1 callersFunctionkernel
( # input K: T.Tensor(K_shape, dtype=input_dtype), V: T.Tensor(V_shape, dtype=input_dt
examples/kda/wy_fast_bwd.py:118
↓ 1 callersFunctionkernel
( # Input Q: T.Tensor(Q_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dt
examples/kda/chunk_delta_bwd.py:118
↓ 1 callersFunctionkernel
( K: T.Tensor(K_shape, dtype=input_dtype), W: T.Tensor(W_shape, dtype=input_dtype), U:
examples/kda/chunk_delta_h_fwd.py:116
↓ 1 callersFunctionkernel
( Q: T.Tensor(K_shape, dtype=input_dtype), K: T.Tensor(K_shape, dtype=input_dtype), V:
examples/kda/chunk_bwd_dqkwg.py:90
↓ 1 callersFunctionkernel
( Q: T.Tensor(Q_shape, dtype=input_dtype), # type: ignore V: T.Tensor(V_shape, dtype=input_dt
examples/kda/chunk_o.py:92
↓ 1 callersFunctionkernel
( # input q: T.Tensor(K_shape, dtype=input_dtype), k: T.Tensor(K_shape, dtype=input_dt
examples/kda/chunk_bwd_intra.py:118
↓ 1 callersFunctionkernel
( DO: T.Tensor(DO_shape, dtype=do_dtype), V: T.Tensor(V_shape, dtype=input_dtype), dA:
examples/kda/chunk_bwd_gla_dA.py:72
↓ 1 callersFunctionkernel
( DO: T.Tensor(DO_shape, dtype=do_dtype), A: T.Tensor(A_shape, dtype=input_dtype), dv:
examples/kda/chunk_bwd_dv.py:74
↓ 1 callersFunctionkernel
(A)
testing/python/language/test_tilelang_language_func_attrs.py:14
↓ 1 callersFunctionkernel
(A: T.Tensor((N), T.float32))
testing/python/language/test_tilelang_language_cooperative.py:12
↓ 1 callersFunctionkernel
(A: T.Tensor((N,), T.float32), s: T.float32)
testing/python/autotune/test_tilelang_autotune_scalar_inputs.py:13
↓ 1 callersFunctionkernel
(A, B, C, D)
testing/python/transform/test_tilelang_transform_disable_memory_reuse.py:27
↓ 1 callersFunctionkernel_activemask
All 32 threads are active → lower 32 bits of activemask must all be set.
testing/python/language/test_tilelang_language_warp_vote.py:191
← previousnext →2,301–2,400 of 7,254, ranked by callers