MCPcopy Create free account

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

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

↓ 1 callersFunction_issue1846_fill_scalar
(fill, M)
testing/python/issue/test_tilelang_issue_1846.py:9
↓ 1 callersFunction_issue1993_dynamic_grid
()
testing/python/issue/test_tilelang_issue_1993.py:10
↓ 1 callersFunction_issue1993_static_grid
()
testing/python/issue/test_tilelang_issue_1993.py:22
↓ 1 callersFunction_iter_kind
(i: tirx.IterVar)
tilelang/carver/analysis.py:134
↓ 1 callersFunction_kernel_alloc_var_count
Counter initialised to 0, incremented 5 times in a loop.
testing/python/amd/test_tilelang_hip_codegen.py:359
↓ 1 callersFunction_kernel_sync_grid_codegen
Kernel that calls T.sync_grid() to trigger cooperative groups codegen.
testing/python/amd/test_tilelang_hip_codegen.py:633
↓ 1 callersFunction_kernel_tr16
(X, Out)
testing/python/amd/test_tilelang_ds_read_tr.py:37
↓ 1 callersFunction_kernel_tr8
(X, Out)
testing/python/amd/test_tilelang_ds_read_tr.py:56
↓ 1 callersFunction_kernel_vector_exp2_codegen
T.vectorized(VEC) makes T.exp2 lower as tirx.exp2 on float32xVEC.
testing/python/amd/test_tilelang_hip_codegen.py:770
↓ 1 callersFunction_layernorm_bwd
(N, D, blk_m=32, threads=256, in_dtype="bfloat16")
examples/norm/layernorm.py:62
↓ 1 callersFunction_layernorm_fwd
(N, D, eps=1e-5, blk_m=1, threads=256, in_dtype="bfloat16", out_dtype="bfloat16")
examples/norm/layernorm.py:7
↓ 1 callersFunction_lazy_load_lib
()
tilelang/__init__.py:128
↓ 1 callersMethod_legalize_info
(self)
tilelang/carver/roller/policy/tensorcore.py:30
↓ 1 callersMethod_legalize_result_idx
(self, result_idx: list[int] | None = None)
tilelang/profiler/__init__.py:42
↓ 1 callersFunction_library_dir
(cu_dir: pathlib.Path)
cmake/find_pip_cuda.py:179
↓ 1 callersFunction_linux_ar
(output, inputs, ar)
tilelang/contrib/cc.py:137
↓ 1 callersFunction_load_cuda_source
(source_code_or_path: str | os.PathLike[str])
tilelang/language/kernel.py:351
↓ 1 callersFunction_load_cudart
Load the CUDA runtime library, searching for the version-suffixed DLL on Windows.
tilelang/carver/arch/driver/cuda_driver.py:55
↓ 1 callersMethod_load_kernel_from_disk
Loads a previously compiled kernel from disk cache. Args: key (str): The hash key identifying the kernel. ta
tilelang/autotuner/param.py:293
↓ 1 callersMethod_load_result_from_disk
(self, key)
tilelang/autotuner/tuner.py:468
↓ 1 callersFunction_load_tile_lang_lib
Load Tile Lang lib
tilelang/__init__.py:177
↓ 1 callersFunction_logical_shape_value
(value, buffer: Buffer, dim: int)
tilelang/language/eager/builder.py:67
↓ 1 callersFunction_logical_stride_value
(value, buffer: Buffer, dim: int)
tilelang/language/eager/builder.py:74
↓ 1 callersMethod_lookup_dynamic_symbolic
Look up a tirx.Var in the dynamic symbolic map. Falls back to name-based lookup when object identity doesn't match.
tilelang/jit/adapter/nvrtc/adapter.py:186
↓ 1 callersMethod_lookup_dynamic_symbolic
Look up a tirx.Var in the dynamic symbolic map. Falls back to name-based lookup when object identity doesn't match (can happen when p
tilelang/jit/adapter/cutedsl/adapter.py:227
↓ 1 callersFunction_lop3_sub_f16x2_signed
LOP3 + sub.f16x2 for signed i4 to f16x2 decode.
tilelang/contrib/cutedsl/quantize.py:74
↓ 1 callersFunction_lop3_sub_f16x2_unsigned
LOP3 + sub.f16x2 for unsigned i4 to f16x2 decode. PTX equivalent: shr.b32 shifted, i4s, shift; lop3.b32 h, shifted, BOTTOM_M
tilelang/contrib/cutedsl/quantize.py:36
↓ 1 callersMethod_lower_blockscaled
Lower block-scaled MXFP8 GEMM to TIR. Block-scaled GEMM follows explicit-async TCGEN5MMA semantics: the MMA issue posts completion to
tilelang/cuda/op/gemm/gemm_tcgen05.py:229
↓ 1 callersFunction_make_data_integrity_kernel
()
testing/python/transform/test_tilelang_transform_disable_memory_reuse.py:20
↓ 1 callersFunction_make_fake_nvrtc_kernel
(tmp_path)
testing/python/cache/test_tilelang_kernel_cache_atomic_save.py:47
↓ 1 callersFunction_make_input
(M, N, dtype)
testing/python/language/test_tilelang_language_reduce.py:15
↓ 1 callersFunction_make_kernel
(M, N)
testing/python/issue/test_tilelang_issue_1210.py:6
↓ 1 callersFunction_make_kernel_if_cond
(M, N)
testing/python/issue/test_tilelang_issue_1210.py:25
↓ 1 callersFunction_make_simdgroup_gemm_func
(M, N, K, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
testing/python/metal/test_metal_simdgroup_store.py:16
↓ 1 callersMethod_make_tiled_mma
(self, warp_m, warp_n)
tilelang/contrib/cutedsl/gemm_v1.py:280
↓ 1 callersMethod_make_unique_name
Generate a unique name with incrementing counter.
tilelang/transform/decouple_type_cast.py:339
↓ 1 callersFunction_make_vec_add_autotuned
Create a tiny CuTeDSL autotuned vector-add kernel.
testing/python/autotune/test_tilelang_autotune_cutedsl_cache.py:74
↓ 1 callersMethod_make_vectorized_loop
Create a vectorized For loop based on the original.
tilelang/transform/decouple_type_cast.py:475
↓ 1 callersFunction_match_any
Return True if *nodeid* contains any of the *patterns*.
examples/conftest.py:41
↓ 1 callersFunction_materialize_pointer_from_addr
(addr: PrimExpr, dtype: DType, storage_scope: str = "global")
tilelang/language/proxy.py:298
↓ 1 callersFunction_matmul_kernel_async
Return a prim_func using T.async_copy (explicit async, no pipeline) for codegen tests.
testing/python/amd/test_tilelang_gfx950_copy_async.py:75
↓ 1 callersFunction_merge_two_bounds
(x: arith.ConstIntBound, y: arith.ConstIntBound)
tilelang/carver/roller/shape_inference/common.py:14
↓ 1 callersFunction_min_version_str
()
tilelang/jit/adapter/cutedsl/checks.py:26
↓ 1 callersFunction_mla_attn
( q_nope, q_pe, kv_c_cache, k_pe_cache, attn_logits, req_to_tokens, b_seq_len,
examples/deepseek_mla/benchmark_mla.py:223
↓ 1 callersFunction_mla_attn
( q_nope, q_pe, kv_c_cache, k_pe_cache, attn_logits, req_to_tokens, b_seq_len,
examples/deepseek_mla/amd/benchmark_mla_decode_amd_triton.py:145
↓ 1 callersFunction_mla_softmax_reducev
( logits, o, b_seq_len, num_kv_splits, )
examples/deepseek_mla/benchmark_mla.py:325
↓ 1 callersFunction_mla_softmax_reducev
( logits, o, b_seq_len, num_kv_splits, )
examples/deepseek_mla/amd/benchmark_mla_decode_amd_triton.py:248
↓ 1 callersFunction_module_has_shared_barrier
Whether any function allocates a shared.barrier / shared.cluster_barrier buffer (i.e. uses ``T.alloc_barrier``).
tilelang/cuda/pipeline.py:49
↓ 1 callersFunction_nearest_fp4_nibble
Round each float in [-6, 6] to the nearest FP4 E2M1 nibble value (uint8 0-15). Nibble encoding for float4_e2m1fn: 0-7: +{0, 0.5, 1, 1.5, 2,
examples/deepseek_v4/act_quant.py:217
↓ 1 callersMethod_normalize_cached_text_source
(source: CachedTextSourceLike)
tilelang/jit/adapter/base.py:112
↓ 1 callersFunction_normalize_explicit_path
(path: pathlib.Path)
cmake/find_pip_cuda.py:54
↓ 1 callersFunction_normalize_flat_binds
Inline flat Bind statements according to sequential dominance. ``Bind`` no longer has a body. A bind in ``SeqStmt`` dominates only later sibl
tilelang/transform/decouple_type_cast.py:220
↓ 1 callersMethod_normalize_function_informations
Normalize legacy function metadata maps into ordered call-site metadata.
tilelang/jit/adapter/cutedsl/wrapper.py:1526
↓ 1 callersMethod_normalize_gfx950_f16_bf16_kpack
(self)
tilelang/rocm/intrinsics/mfma_macro_generator.py:217
↓ 1 callersFunction_normalize_option
(option: str)
tilelang/contrib/msvc.py:248
↓ 1 callersFunction_nvcc_name
()
cmake/find_pip_cuda.py:24
↓ 1 callersFunction_pack_fp4_nibbles
Pack pairs of FP4 nibbles into uint8 bytes. nibbles: (M, N) uint8 tensor, each element in 0-15. Returns: (M, N//2) uint8 tensor. Low nibb
examples/deepseek_v4/act_quant.py:233
↓ 1 callersFunction_packed_fp4_to_float
(packed: torch.Tensor, logical_k: int)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:245
↓ 1 callersMethod_parse_arg_annot
(self, stmt: ast.stmt, arg_names: set[str])
tilelang/language/eager/ast.py:510
↓ 1 callersFunction_parse_formats
Parse the formats parameter into a list of format strings.
tilelang/tools/plot_layout.py:64
↓ 1 callersFunction_parse_table
(output: str)
maint/scripts/regression_all.py:27
↓ 1 callersFunction_parse_version_triple
Parse a best-effort (major, minor, patch) triple from a version string. We intentionally avoid importing heavy/optional version parsers. For our
tilelang/jit/adapter/cutedsl/checks.py:14
↓ 1 callersFunction_patch_source_for_msvc
(path: str, tmp_dir: str)
tilelang/contrib/msvc.py:260
↓ 1 callersFunction_plot_fragment_layout
( layout: T.Fragment, save_directory="./tmp", name: str = "layout", colormap: str = "RdPu",
tilelang/tools/plot_layout.py:112
↓ 1 callersFunction_plot_layout_map
Visualize a Layout object as a 2D grid showing position mappings. By default (`view="input"`), the grid represents the input space (viewed a
tilelang/tools/plot_layout.py:271
↓ 1 callersMethod_prepare_compile_execution
( self, config_args: list[dict[str, Any]], grouped_compile_active: bool, group
tilelang/autotuner/tuner.py:533
↓ 1 callersFunction_preprocess_freeze_comments
Convert ``# autodd: freeze`` comment annotations to ``with __freeze__:`` blocks. Supports two forms: **Block form** – wrap a group of statem
tilelang/autodd.py:989
↓ 1 callersFunction_print_pdl_benchmark_environment
Print exact local environment details for opt-in PDL benchmark runs.
testing/python/language/test_tilelang_language_pdl.py:221
↓ 1 callersFunction_quantize_f32_to_packed_fp4
Quantize float32 (m, k) to packed fp4 bytes (m, k//2).
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:199
↓ 1 callersFunction_quantize_float_to_fp4_packed
(x: torch.Tensor)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:258
↓ 1 callersFunction_quantize_mxfp4_with_packed_ue8m0
(x, gran_k: int = 128)
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:219
↓ 1 callersFunction_quantize_mxfp8_with_packed_ue8m0
(x, gran_k: int = 128)
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:122
↓ 1 callersFunction_random_fp4_tensor
Create random packed fp4 tensor for logical shape (m, k).
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:212
↓ 1 callersFunction_reduce_op
(T, op, src, dst, dim, batch=1)
testing/python/language/test_tilelang_language_reduce.py:36
↓ 1 callersFunction_remap_match_buffer
Remap buffer inside a MatchBufferRegion using buf_map.
tilelang/metal/transform/metal_fragment_to_simdgroup.py:83
↓ 1 callersMethod_remove_incomplete_cache_dir
(self, cache_path: str)
tilelang/cache/kernel_cache.py:772
↓ 1 callersMethod_remove_incomplete_result_dir
Remove a stale incomplete result directory when present.
tilelang/autotuner/param.py:586
↓ 1 callersFunction_remove_leading_ident
(source: str)
tilelang/language/eager/utils.py:27
↓ 1 callersMethod_replace_access
Replace memory accesses with cast buffer accesses.
tilelang/transform/decouple_type_cast.py:549
↓ 1 callersFunction_require_backend_available
(backend: str)
testing/python/cache/test_tilelang_kernel_cache.py:46
↓ 1 callersFunction_require_cuda_tensor
(shape, dtype)
testing/python/layout/test_tilelang_layout_fused_replicate.py:37
↓ 1 callersFunction_require_cuda_tensor
(shape, dtype)
testing/python/runtime/test_tilelang_runtime_dynamic_shared_memory.py:25
↓ 1 callersFunction_require_sm100
()
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:14
↓ 1 callersFunction_reset_results
()
tilelang/testing/perf_regression.py:48
↓ 1 callersFunction_resolve_artifact_paths
(temp, file_name, target_format, kernels_output_dir=None)
tilelang/contrib/hipcc.py:25
↓ 1 callersFunction_resolve_artifact_paths
(temp, file_name, target_format, kernels_output_dir=None)
tilelang/contrib/nvcc.py:36
↓ 1 callersMethod_resolve_benchmark_devices
( self, benchmark_multi_gpu: bool, benchmark_devices: list[int] | None, target
tilelang/autotuner/tuner.py:796
↓ 1 callersMethod_resolve_grouped_compile_mode
( self, enable_grouped_compile: bool, group_compile_size: int, )
tilelang/autotuner/tuner.py:494
↓ 1 callersMethod_resolve_num_compile_workers
(self)
tilelang/autotuner/tuner.py:512
↓ 1 callersFunction_resolve_warps_per_group
(warps_per_group: Optional[int])
testing/python/language/test_tilelang_language_get_warp_info.py:18
↓ 1 callersFunction_resolve_windows_compiler
Return (compiler_path, kind) where kind is 'clang-cl' or 'msvc'. Cached so repeated JIT compiles don't re-spawn vswhere or rescan PATH.
tilelang/contrib/msvc.py:92
↓ 1 callersFunction_results_to_jsonable
()
tilelang/testing/perf_regression.py:28
↓ 1 callersMethod_rewrite
(self, node: ast.AST)
tilelang/autodd.py:372
↓ 1 callersFunction_rewrite_scope
(body, var_map)
tilelang/metal/transform/metal_fragment_to_simdgroup.py:97
↓ 1 callersMethod_rope_scaling_validation
Validate the `rope_scaling` configuration.
examples/bitnet-1.58b/configuration_bitnet.py:175
↓ 1 callersFunction_run_command_summary
Return all non-empty command output lines joined for compact benchmark logs.
testing/python/language/test_tilelang_language_pdl.py:202
↓ 1 callersFunction_run_wgmma_atom
(M, N, K, in_dtype, out_dtype, accum_dtype)
testing/python/language/test_tilelang_language_atom_mma.py:215
↓ 1 callersFunction_run_with_sigalrm
POSIX path: ``SIGALRM`` interrupts ``func`` synchronously inside the same thread. The runaway call actually stops, releasing CPU/GPU resources
tilelang/autotuner/tuner.py:51
↓ 1 callersFunction_run_with_thread
Cross-platform fallback when ``SIGALRM`` is unavailable (Windows) or when called outside the main thread (``signal.signal`` is restricted to t
tilelang/autotuner/tuner.py:102
↓ 1 callersFunction_rw_mask
(access_type: str | int)
tilelang/language/builtin.py:162
← previousnext →1,701–1,800 of 7,254, ranked by callers