MCPcopy Create free account

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

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

↓ 3 callersMethodwgmma_arrive
Emit ``warpgroup_arrive()``.
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:481
↓ 3 callersMethodwgmma_commit
Emit ``warpgroup_commit_batch()``.
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:490
↓ 3 callersMethodwgmma_rs
( self, A_region: BufferRegion, B_region: BufferRegion, C_region: BufferRegion, clear_accum: PrimExpr
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:243
↓ 3 callersMethodwgmma_wait
Emit ``warpgroup_wait(n)``.
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:499
↓ 2 callersFunctionAllReduce
AllReduce operation implementing warp/block-level reduction. Based on tl::AllReduce from reduce.h Args: reducer: Reducer operato
tilelang/contrib/cutedsl/reduce.py:515
↓ 2 callersFunctionBlockIdx
Return the current CUDA block index.
tilelang/contrib/cutedsl/threadblock_swizzle.py:34
↓ 2 callersFunctionDecoupleTypeCast
Create a TVM pass that decouples type cast vectorization constraints. This pass inserts a local buffer as an intermediate stage for vectorized
tilelang/transform/decouple_type_cast.py:585
↓ 2 callersFunctionGridDim
Return the CUDA grid dimensions.
tilelang/contrib/cutedsl/threadblock_swizzle.py:40
↓ 2 callersFunctionLetInline
LetInline Returns ------- fpass : tvm.transform.Pass The result pass
tilelang/transform/simplify.py:9
↓ 2 callersFunctionPreLowerSemanticCheck
Run backend-independent validation before lowering.
tilelang/engine/semantic_check.py:21
↓ 2 callersMethod__call__
( self, shape: ShapeType, dtype: DType = _dtypes.float32, data=None, s
tilelang/language/proxy.py:23
↓ 2 callersMethod__init__
(self, config: Dict)
examples/fusedmoe/example_fusedmoe_torch.py:44
↓ 2 callersMethod__init__
( self, config: Dict, shared_kernel: tilelang.JITKernel, routed_kernel: tilelang.JITKernel, weights: D
examples/fusedmoe/example_fusedmoe_tilelang.py:265
↓ 2 callersMethod__init__
(self, all_labels: list[int], init_proba: float = 0.93)
tilelang/autodd.py:554
↓ 2 callersMethod__init__
(self, target: Target)
tilelang/jit/adapter/wrapper.py:946
↓ 2 callersFunction_apply_lower_opaque_pipeline
(func, target, pass_configs=None)
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:50
↓ 2 callersFunction_as_expr_placeholder
(temp: ast.AST)
tilelang/autodd.py:205
↓ 2 callersFunction_assert_accepts_packed_fp4_storage
(kernel)
testing/python/issue/test_tilelang_issue_2292.py:51
↓ 2 callersFunction_assert_access_ptr_lowered
(mod)
testing/python/issue/test_tilelang_issue_2123.py:49
↓ 2 callersFunction_assert_gpu_target
(target: Target)
tilelang/carver/analysis.py:177
↓ 2 callersMethod_assign_a_shared_layout
(self, layout: Layout)
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:81
↓ 2 callersMethod_assign_b_shared_layout
(self, layout: Layout)
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:85
↓ 2 callersFunction_ast_replace
(temp: ast.expr, repl: dict[str, ast.AST])
tilelang/autodd.py:255
↓ 2 callersFunction_bench
(fn, warmup, repeats)
benchmark/matmul_metal/benchmark_matmul_metal.py:47
↓ 2 callersFunction_benchmark_kernel_ms
Measure a TileLang kernel with CUDA events and return per-run milliseconds.
testing/python/language/test_tilelang_language_pdl.py:245
↓ 2 callersFunction_buf_indices_match
Check if two (buffer, indices) pairs refer to the same access pattern.
tilelang/transform/decouple_type_cast.py:295
↓ 2 callersFunction_buffer_data
(value)
tilelang/language/tir/op.py:15
↓ 2 callersMethod_build_tir_template
Build TIR template based on the execution mode.
tilelang/language/eager/builder.py:1192
↓ 2 callersFunction_ceil_to_ue8m0
(x: torch.Tensor)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:196
↓ 2 callersFunction_check
(original, transformed)
testing/python/transform/test_tilelang_transform_let_inline.py:7
↓ 2 callersFunction_clang_cl_disabled
()
tilelang/contrib/msvc.py:26
↓ 2 callersFunction_collect_shuffle_elect
(stmt)
testing/python/transform/test_tilelang_transform_lower_shared_barrier.py:44
↓ 2 callersFunction_compile_and_check
( program, trans_A, trans_B, in_dtype, out_dtype, )
maint/gemm/correctness_evaluation_sm70.py:54
↓ 2 callersFunction_compile_cuda
(func)
testing/python/language/test_tilelang_language_sync_threads.py:6
↓ 2 callersFunction_compile_grouped_gemm_ws
(batch_sizes=(63, 77), K=128, N=128, block_M=64, block_N=64, block_K=32)
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:303
↓ 2 callersFunction_compress_bytes
(input_tensor: torch.Tensor, sparse_tensor: torch.Tensor, meta_tensor: torch.Tensor)
benchmark/matmul/benchmark_matmul_sp_compress.py:24
↓ 2 callersMethod_compute_tc_strides
( self, node: PrimFuncNode, tile: list[int], rstep: dict[str, int] | None = No
tilelang/carver/roller/policy/tensorcore.py:55
↓ 2 callersFunction_count_calls_in_stmt
(stmt: tvm.tirx.Stmt)
testing/python/transform/test_tilelang_transform_lower_ptx_async_copy.py:22
↓ 2 callersFunction_count_copy_calls_with_annotation
(func, annotation_key)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:113
↓ 2 callersFunction_count_extern_calls_with_prefix
(stmt, prefix: str)
testing/python/transform/test_tilelang_transform_inject_tcgen05_fence.py:41
↓ 2 callersFunction_count_fences
(stmt)
testing/python/transform/test_tilelang_transform_inject_fence_proxy.py:88
↓ 2 callersFunction_count_if_then_else
(stmt)
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:57
↓ 2 callersMethod_create_cast_entries
Create local cast buffers for memory accesses. Each unique (buffer, indices) pair gets its own cast buffer.
tilelang/transform/decouple_type_cast.py:452
↓ 2 callersMethod_create_copy_loops
Create vectorized copy loops between memory and cast buffers. direction: "to_memory" (cast → memory) or "from_memory" (memory → cast).
tilelang/transform/decouple_type_cast.py:488
↓ 2 callersFunction_cuda_target_from_arch
Build a CUDA target while preserving the legacy bare string fallback.
tilelang/utils/target.py:104
↓ 2 callersFunction_decode_tcgen05_formats
(desc: int)
testing/python/language/test_tilelang_language_float4_e2m1_unpacked_dtype.py:74
↓ 2 callersFunction_detect_torch_cuda_arch
Return the CUDA SM architecture detected from PyTorch, if available.
tilelang/utils/target.py:96
↓ 2 callersFunction_detect_torch_rocm_arch
()
tilelang/utils/target.py:75
↓ 2 callersMethod_determinate_swizzle_mode
(self, buffer: Buffer, layout: Layout)
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:190
↓ 2 callersFunction_dtype_total_bits
(dtype: dt.dtype)
tilelang/language/eager/builder.py:51
↓ 2 callersFunction_empty_kernel_with_binding_variants
(use_tuple_binding: bool = False)
testing/python/issue/test_tilelang_issue_830.py:51
↓ 2 callersFunction_ensure_torch_dtypes
(*dtype_names)
maint/gemm/correctness_evaluation_sm70.py:233
↓ 2 callersFunction_find_tool
Find a CUDA binary in PATH or under CUDA_HOME/bin.
tilelang/contrib/nvcc.py:227
↓ 2 callersFunction_flatten_to_2d
Keep last dim, merge all preceding dims into one row dim.
tilelang/tools/plot_layout.py:324
↓ 2 callersFunction_format_options
(options: Iterable[str])
tilelang/jit/execution_backend.py:62
↓ 2 callersFunction_fp4_lut
(device: str = "cuda")
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:179
↓ 2 callersFunction_fp4_lut
(device: torch.device)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:241
↓ 2 callersFunction_fp4_tma_descriptor_init_block
(host_source, desc_name)
testing/python/language/test_tilelang_language_tma_copy.py:328
↓ 2 callersFunction_generate_dtype_cases
Generate dtype test cases for given K values and num_threads.
maint/gemm/correctness_evaluation.py:392
↓ 2 callersMethod_get_base_key
()
tilelang/cache/kernel_cache.py:129
↓ 2 callersFunction_get_device_kernel_script
(detect_inplace: bool)
testing/python/components/test_storage_rewrite_detect_inplace.py:44
↓ 2 callersMethod_get_frontend_cache_path
(self, frontend_key: str)
tilelang/cache/kernel_cache.py:423
↓ 2 callersMethod_get_frontend_cache_root
()
tilelang/cache/kernel_cache.py:205
↓ 2 callersMethod_get_implementation_class
Get the appropriate implementation class for the given GEMM instruction key. Args: gemm_inst: The selected backend-specific GEMM
tilelang/tileop/gemm/__init__.py:160
↓ 2 callersMethod_get_implementation_class
(self, gemm_inst: str, target: Target)
tilelang/tileop/gemm_sp/__init__.py:70
↓ 2 callersMethod_get_missing_complete_result_files
Return complete-result files missing from a candidate cache path.
tilelang/autotuner/param.py:576
↓ 2 callersFunction_get_node_span
Return the span (lineno, col, end_lineno, end_col) for a doc node.
tilelang/language/overrides/parser.py:12
↓ 2 callersMethod_get_output_tile_map
Handles multiple output cases by mapping output nodes to their respective tile configurations. Parameters ----------
tilelang/carver/roller/policy/default.py:170
↓ 2 callersMethod_get_params
(self, with_output=False)
tilelang/profiler/__init__.py:97
↓ 2 callersMethod_get_ptx
Compile and return PTX for the current kernel (CUDA only). Parameters ---------- verbose : Optional[bool]
tilelang/jit/kernel.py:702
↓ 2 callersFunction_get_rdna_tuning_config
(mcpu: str | None)
tilelang/carver/arch/rdna.py:45
↓ 2 callersMethod_get_required_kernel_files
Return backend-specific files required to reload a kernel.
tilelang/autotuner/param.py:551
↓ 2 callersMethod_get_sass
Compile and return SASS for the current kernel (CUDA only). Parameters ---------- verbose : Optional[bool]
tilelang/jit/kernel.py:766
↓ 2 callersFunction_get_sig_line
(code: str)
testing/python/components/test_cuda_restrict_codegen.py:6
↓ 2 callersFunction_get_sm90_cuda_device
Return an SM90+ CUDA device for runtime PDL tests.
testing/python/language/test_tilelang_language_pdl.py:160
↓ 2 callersFunction_has_op_call
(func, op_name)
testing/python/issue/test_tilelang_issue_2123.py:36
↓ 2 callersMethod_host_entry_func
Return the lowered host entry PrimFunc, not the generated Python wrapper.
tilelang/jit/adapter/cutedsl/wrapper.py:846
↓ 2 callersFunction_host_entry_gvar_and_func
Return the GlobalVar and PrimFunc selected as the wrapper host entry.
testing/python/jit/test_tilelang_jit_cutedsl_host_codegen.py:36
↓ 2 callersFunction_imax
(a, b, *, loc=None, ip=None)
tilelang/contrib/cutedsl/reduce.py:100
↓ 2 callersFunction_imin
(a, b, *, loc=None, ip=None)
tilelang/contrib/cutedsl/reduce.py:74
↓ 2 callersFunction_index_dtype
(buf: tirx.Buffer)
tilelang/language/builtin.py:172
↓ 2 callersMethod_infer_jit_mode
Infer the JIT execution mode based on function behavior. Returns "lazy" if the function explicitly returns a PrimFunc, or "e
tilelang/jit/__init__.py:317
↓ 2 callersMethod_init_with_output_nodes
(self, output_nodes: list[OutputNode])
tilelang/carver/roller/policy/default.py:57
↓ 2 callersMethod_is_complete_cache_dir
(self, cache_path: str)
tilelang/cache/kernel_cache.py:769
↓ 2 callersMethod_is_complete_result_dir
Return whether a cache directory contains all required files.
tilelang/autotuner/param.py:581
↓ 2 callersFunction_is_equal
(a, b)
examples/deepseek_v32/utils.py:22
↓ 2 callersFunction_is_freeze_call
Detect ``__freeze__(expr)``.
tilelang/autodd.py:64
↓ 2 callersFunction_is_freeze_with
Detect ``with __freeze__: body`` (no ``as`` clause).
tilelang/autodd.py:53
↓ 2 callersFunction_is_op_call
(node, op_name: str)
testing/python/issue/test_tilelang_issue_sm120_tma_smem_alignment.py:126
↓ 2 callersFunction_kernel_multi_alloc_var_init
Two local variables with different init values, summed into output.
testing/python/amd/test_tilelang_hip_codegen.py:323
↓ 2 callersFunction_kernel_sync_warp_codegen
Minimal kernel that exercises T.sync_warp().
testing/python/amd/test_tilelang_hip_codegen.py:549
↓ 2 callersMethod_legalize_to_buffer_region
(obj)
tilelang/rocm/intrinsics/wmma_macro_generator.py:495
↓ 2 callersMethod_legalize_to_buffer_region
Convert Buffer/BufferRegion/BufferLoad to a BufferRegion. - Buffer -> full-region BufferRegion covering entire shape - Buffe
tilelang/rocm/intrinsics/mfma_macro_generator.py:769
↓ 2 callersMethod_legalize_to_buffer_region
Convert Buffer/BufferRegion/BufferLoad to a BufferRegion. - Buffer -> full-region BufferRegion covering entire shape - Buffe
tilelang/cuda/intrinsics/macro/mma_sm70_macro_generator.py:506
↓ 2 callersFunction_linux_compile
(output, objects, options, compile_cmd, cwd=None, ccache_env=None, compile_shared=False)
tilelang/contrib/cc.py:370
↓ 2 callersFunction_load_from_src
Load elements from src_values, handling both TensorSSA and _Pointer types.
tilelang/contrib/cutedsl/atomic.py:167
↓ 2 callersFunction_make_async_tcgen05_kernel
(gemm_op)
testing/python/language/test_tilelang_language_tcgen05_gemm.py:34
↓ 2 callersFunction_make_auto_vec_fma_kernel
Build a kernel that lets CUDA codegen fuse mul + add into tl::fma2.
testing/python/cuda/test_cuda_f32x2_intrinsics.py:126
↓ 2 callersFunction_make_binary_kernel
Build a kernel: C[idx] = op(A[idx], B[idx]).
testing/python/cuda/test_cuda_f32x2_intrinsics.py:39
↓ 2 callersFunction_make_buggy_arena_kernel
2-stage pipelined bf16 GEMM with a tiny T.float32 shared scalar whose lifetime *overlaps* the TMA-loaded tiles. The scalar is written before t
testing/python/issue/test_tilelang_issue_sm120_tma_smem_alignment.py:44
← previousnext →801–900 of 7,254, ranked by callers