MCPcopy Create free account

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

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

↓ 1 callersFunction_check_is_cuda_or_cdna
()
tilelang/testing/__init__.py:46
↓ 1 callersFunction_check_is_gfx950
()
tilelang/testing/__init__.py:30
↓ 1 callersMethod_cleanup_module
Cleanup a single adapter instance's CUDA module and contexts. This is called automatically when the adapter instance is garbage collected.
tilelang/jit/adapter/cutedsl/adapter.py:441
↓ 1 callersMethod_cleanup_stale_staging_dirs
Remove stale entries from the dedicated staging root. These are left behind when a process crashes mid-save.
tilelang/cache/kernel_cache.py:213
↓ 1 callersMethod_clear_disk_cache
Removes all cached kernels from disk. Note: This operation will delete the current kernel-cache namespace and recreate i
tilelang/cache/kernel_cache.py:714
↓ 1 callersFunction_collect_barrier_blocks
(stmt)
testing/python/transform/test_tilelang_transform_lower_shared_barrier.py:48
↓ 1 callersFunction_collect_consumers
(sch: Schedule, block: SBlockRV)
tilelang/carver/matmul_analysis.py:53
↓ 1 callersFunction_collect_direct_bind_names
(mod)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:449
↓ 1 callersFunction_collect_external_cuda_kernel_names
(source: str)
tilelang/engine/lower.py:58
↓ 1 callersFunction_collect_fragment_gemm_accum_vars
(body: tir.Stmt)
tilelang/metal/transform/metal_fragment_to_simdgroup.py:39
↓ 1 callersMethod_collect_function_args
Collect all function arguments from primary function. Returns: Tuple of (function_args, buffer_args)
tilelang/jit/adapter/cutedsl/wrapper.py:895
↓ 1 callersFunction_collect_producers
(sch: Schedule, block: SBlockRV)
tilelang/carver/matmul_analysis.py:45
↓ 1 callersFunction_collect_storage_syncs
(stmt)
testing/python/transform/test_tilelang_transform_lower_shared_barrier.py:32
↓ 1 callersFunction_collect_tma_loads
(mod)
testing/python/issue/test_tilelang_issue_sm120_tma_smem_alignment.py:130
↓ 1 callersFunction_collect_tvm_ffi_exports
(path: str)
tilelang/contrib/msvc.py:278
↓ 1 callersFunction_compare_attributes
Checks if the attributes of two tensors match. Always checks - the :attr:`~torch.Tensor.shape`, - whether both inputs are quantized or not
tilelang/utils/tensor.py:122
↓ 1 callersFunction_compile_and_check
( program, trans_A, trans_B, in_dtype, out_dtype, )
maint/gemm/correctness_evaluation_tcgen05.py:56
↓ 1 callersFunction_compile_and_check
(program, out_dtype)
maint/gemm/correctness_evaluation_tcgen05_2cta.py:78
↓ 1 callersMethod_compile_and_create_adapter
Compiles the given TileLang PrimFunc using TVM and creates a kernel adapter. Parameters ---------- tilelang_func : t
tilelang/jit/kernel.py:205
↓ 1 callersFunction_compile_kernel_with_inplace
()
testing/python/components/test_storage_rewrite_detect_inplace.py:28
↓ 1 callersFunction_compile_kernel_without_inplace
()
testing/python/components/test_storage_rewrite_detect_inplace.py:7
↓ 1 callersFunction_compile_tvm_ffi
(func, pass_configs=None)
testing/python/issue/test_tilelang_issue_ws_simt_copy_full_producer_extent.py:8
↓ 1 callersFunction_compile_tvm_ffi
(func, *, target=CUDA_SM90_TARGET, target_host="c", pass_configs=None)
testing/python/runtime/test_tilelang_runtime_tma_validation.py:15
↓ 1 callersFunction_compress_fn
(D, dtype, meta_dtype, block_M=_BLOCK_M, block_K=_BLOCK_K, elem_per_thread=_ELEM_PER_THREAD)
tilelang/utils/sparse.py:44
↓ 1 callersMethod_compute_memory_traffic
Computes the memory traffic for a given output tile configuration. Parameters ---------- output_tile : List[int]
tilelang/carver/roller/policy/default.py:398
↓ 1 callersMethod_compute_stride_map
Computes the stride map for all nodes in a TileDict. Parameters ---------- td : TileDict The TileDict ob
tilelang/carver/roller/policy/default.py:517
↓ 1 callersFunction_compute_version
Return the package version without being polluted by unrelated installs. Preference order: 1) If running from a source checkout (VERSION file
tilelang/__init__.py:10
↓ 1 callersMethod_construct_strides
(shape: tuple[Any])
tilelang/language/proxy.py:158
↓ 1 callersMethod_construct_unique_name2dep
This is a workaround for the issue that we have two same ops' fuse case. See https://github.com/apache/tvm/issues/16433
tilelang/carver/roller/shape_inference/tir.py:83
↓ 1 callersFunction_contains_seq_stmt
Check if statement contains SeqStmt (multiple statements). When the For body has SeqStmt, the transformation is more complex and we skip the
tilelang/transform/decouple_type_cast.py:124
↓ 1 callersMethod_convert_torch_func
(self)
tilelang/jit/adapter/base.py:56
↓ 1 callersFunction_count_allocate_inside_attr
Count flat AllocBuffer nodes nested inside the given AttrStmt.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:32
↓ 1 callersFunction_cpu_ref_chained
Compute chained GEMM reference on CPU to avoid cuBLAS issues on Blackwell.
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:134
↓ 1 callersMethod_create_adapter_from_database
( self, params: list[KernelParam], result_idx: list[int] | int, target: str |
tilelang/jit/kernel.py:341
↓ 1 callersFunction_cumsum_view_infer_layout
(hidden)
testing/python/issue/test_tilelang_issue_1001.py:10
↓ 1 callersFunction_decorator
(func: Callable)
tilelang/language/parser/entry.py:144
↓ 1 callersFunction_decorator
(func: Callable)
tilelang/language/tir/entry.py:103
↓ 1 callersFunction_default_meta_dtype
(dtype)
testing/python/utils/test_compress_utils.py:84
↓ 1 callersFunction_disable_rocm_tvm_ffi_torch_c_dlpack
(torch_module)
tilelang/__init__.py:109
↓ 1 callersFunction_discover_bench_files
(examples_root: Path)
maint/scripts/regression_all.py:64
↓ 1 callersFunction_discover_cuda_devices
Discover available CUDA devices, then select only ceil(n/2) devices. - Priority of discovery: 1) ENV: PYTEST_XDIST_CUDA_DEVICES
maint/scripts/pytest_cuda_scheduler.py:109
↓ 1 callersFunction_do_alloc
(dst_val, ncols_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:367
↓ 1 callersFunction_do_arrive
(bar_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:332
↓ 1 callersFunction_do_dealloc
(tptr_val, ncols_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:395
↓ 1 callersFunction_do_mma
(c_val, da_val, db_val, dv_val, sc_val, m0_val, m1_val, m2_val, m3_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:167
↓ 1 callersFunction_do_mma_ts
(c_val, a_val, db_val, dv_val, sc_val, m0_val, m1_val, m2_val, m3_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:283
↓ 1 callersFunction_do_mma_ws
(c_val, da_val, db_val, dv_val, sc_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/gemm_tcgen05.py:227
↓ 1 callersFunction_emit_results
Emit results for parent collectors. Default output remains the historical text format. Set `TL_PERF_REGRESSION_FORMAT=json` to emit a single
tilelang/testing/perf_regression.py:32
↓ 1 callersFunction_emit_tmem_ld_segment
Emit one tcgen05.ld inline asm for a power-of-2 segment. Called during @cute.jit compilation — emits MLIR ops directly. ptx_type: PTX inst
tilelang/contrib/cutedsl/gemm_tcgen05.py:427
↓ 1 callersFunction_empty_kernel
()
testing/python/issue/test_tilelang_issue_830.py:10
↓ 1 callersFunction_empty_with_dead_code_kernel
()
testing/python/issue/test_tilelang_issue_830.py:33
↓ 1 callersFunction_ensure_cuda_stub
Create a minimal Linux libcuda.so stub for build-time -lcuda linking.
cmake/find_pip_cuda.py:156
↓ 1 callersFunction_ensure_global_counter
Declare the global counter variable at gpu.module level if not already done. Creates: @__grid_sync_ctr = internal global i32 0, addrspace(1)
tilelang/contrib/cutedsl/grid_sync.py:43
↓ 1 callersMethod_ensure_jit_functions
( self, )
tilelang/autotuner/tuner.py:483
↓ 1 callersFunction_ensure_lib_symlinks
Create Linux symlinks that CMake / nvcc expect but pip packages omit.
cmake/find_pip_cuda.py:133
↓ 1 callersFunction_equalize_attributes
Equalizes some attributes of two tensors for value comparison. If ``actual`` and ``expected`` are ... - ... not on the same :attr:`~torch.Tens
tilelang/utils/tensor.py:163
↓ 1 callersFunction_examples_root
()
maint/scripts/regression_all.py:59
↓ 1 callersMethod_expand_reduce_axis
Expands the reduction axis in the TileDict based on shared memory limits. Parameters ---------- td : TileDict
tilelang/carver/roller/policy/default.py:331
↓ 1 callersFunction_expr_depends_on_var
Check if an expression references the given Var.
tilelang/transform/decouple_type_cast.py:141
↓ 1 callersFunction_extract_buffer_var_from_region
(region_call)
tilelang/metal/transform/metal_fragment_to_simdgroup.py:28
↓ 1 callersFunction_extract_dependent_region
(block_analyzer, block: SBlockRV)
tilelang/carver/roller/shape_inference/tir.py:334
↓ 1 callersMethod_extract_func_call_args
Extract function call arguments from Python function declaration.
tilelang/jit/adapter/cutedsl/wrapper.py:925
↓ 1 callersFunction_fadd_f32
(a: Float32, b: Float32, *, rounding: str = "rn", loc=None, ip=None)
tilelang/contrib/cutedsl/ieee_math.py:31
↓ 1 callersFunction_fdiv_f32
(a: Float32, b: Float32, *, rounding: str = "rn", loc=None, ip=None)
tilelang/contrib/cutedsl/ieee_math.py:133
↓ 1 callersFunction_fill_with_dynamic_region_kernel
()
testing/python/issue/test_tilelang_issue_1008.py:24
↓ 1 callersFunction_fill_with_static_region_kernel
()
testing/python/issue/test_tilelang_issue_1008.py:10
↓ 1 callersFunction_find_alloc_buffer
(func, data)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:144
↓ 1 callersFunction_find_block
(stmt: tvm.tirx.Stmt, name_hint: str)
testing/python/transform/test_tilelang_transform_plan_update_buffer_allocation_location.py:19
↓ 1 callersFunction_find_block_with_layout_map
(func)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:132
↓ 1 callersFunction_find_clang_cl
(compiler_env: dict[str, str] | None)
tilelang/contrib/msvc.py:83
↓ 1 callersFunction_find_cu_dir
Find the nvidia/cu<ver> directory from pip-installed CUDA packages.
cmake/find_pip_cuda.py:123
↓ 1 callersFunction_find_cuda_home
Find the CUDA install path. Adapted from https://github.com/pytorch/pytorch/blob/main/torch/utils/cpp_extension.py
tilelang/env.py:114
↓ 1 callersFunction_find_first_for
(stmt: tvm.tirx.Stmt)
testing/python/transform/test_tilelang_transform_plan_update_buffer_allocation_location.py:31
↓ 1 callersFunction_find_frozen_line_set
Return the set of indices into *nonempty_lines* that belong to frozen regions. A line is considered frozen if it falls within the source span of
tilelang/autodd.py:1092
↓ 1 callersFunction_find_gpu_module_from_ip
Find gpu.module by walking up from current insertion point. Hierarchy: block -> KernelOp -> gpu.module -> builtin.module Uses bounded for-loo
tilelang/contrib/cutedsl/grid_sync.py:25
↓ 1 callersFunction_find_if_with_set_max_nreg
(func, then_call, else_call)
testing/python/transform/test_tilelang_transform_inject_set_max_nreg.py:22
↓ 1 callersFunction_find_import_libs
()
tilelang/contrib/msvc.py:289
↓ 1 callersFunction_find_output_file
(args: list[str])
cmake/ccache_strip_pep517.py:38
↓ 1 callersMethod_find_path_recursive
Recursive helper function for find_path_from_source.
tilelang/carver/roller/shape_inference/tir.py:142
↓ 1 callersFunction_find_pipelined_loop
(func)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:101
↓ 1 callersFunction_find_rocm_home
Find the ROCM install path.
tilelang/env.py:166
↓ 1 callersFunction_find_vsdevcmd
()
tilelang/contrib/msvc.py:118
↓ 1 callersMethod_flash_attention_forward
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token first unpad the input, th
examples/bitnet-1.58b/modeling_bitnet.py:644
↓ 1 callersFunction_fmaf_f32
(a: Float32, b: Float32, c: Float32, *, rounding: str = "rn", loc=None, ip=None)
tilelang/contrib/cutedsl/ieee_math.py:82
↓ 1 callersFunction_fmax
(a, b, c=None, *, loc=None, ip=None)
tilelang/contrib/cutedsl/reduce.py:86
↓ 1 callersFunction_fmin
(a, b, c=None, *, loc=None, ip=None)
tilelang/contrib/cutedsl/reduce.py:60
↓ 1 callersFunction_fmul_f32
(a: Float32, b: Float32, *, rounding: str = "rn", loc=None, ip=None)
tilelang/contrib/cutedsl/ieee_math.py:65
↓ 1 callersMethod_format_version_namespace
(version: str)
tilelang/cache/kernel_cache.py:147
↓ 1 callersFunction_forward
(ctx, q, k, v, block_sparse_mask, sm_scale, BLOCK_M=64, BLOCK_N=64, num_warps=None, num_stages=1, out=None)
benchmark/blocksparse_attention/benchmark_triton_block_sparse_fmha.py:194
↓ 1 callersFunction_forward
(ctx, q, k, v, block_sparse_mask, sm_scale, BLOCK_M=64, BLOCK_N=64, num_warps=None, num_stages=1, out=None)
examples/blocksparse_attention/block_sparse_attn_triton.py:194
↓ 1 callersFunction_forward
(ctx, q, k, v, block_sparse_mask, sm_scale, BLOCK_M=64, BLOCK_N=64, num_warps=None, num_stages=1, out=None)
examples/seer_attention/block_sparse_attn_triton.py:189
↓ 1 callersMethod_forward_from_prebuild_lib
Low-level function to call the compiled CUDA kernel.
tilelang/jit/adapter/nvrtc/adapter.py:214
↓ 1 callersMethod_forward_from_prebuild_lib
Low-level function to call the compiled CUDA kernel. Args: *args: Kernel arguments (tensors and scalars) stream: CUDA
tilelang/jit/adapter/cutedsl/adapter.py:271
↓ 1 callersFunction_fp4_encode
Encode a list of floats to packed uint8 FP4 E2M1 (2 per byte).
testing/python/amd/test_tilelang_mxfp4_gfx950.py:30
↓ 1 callersFunction_fp4_lazy_kernel
()
testing/python/issue/test_tilelang_issue_2292.py:36
↓ 1 callersFunction_fp4_to_bf16_twiddling_r0
Convert 8 FP4 to 8 BF16, return output 0.
tilelang/contrib/cutedsl/quantize.py:144
↓ 1 callersFunction_fp4_to_bf16_twiddling_r1
Convert 8 FP4 to 8 BF16, return output 1.
tilelang/contrib/cutedsl/quantize.py:166
↓ 1 callersFunction_fp4_to_bf16_twiddling_r2
Convert 8 FP4 to 8 BF16, return output 2.
tilelang/contrib/cutedsl/quantize.py:189
↓ 1 callersFunction_fp4_to_bf16_twiddling_r3
Convert 8 FP4 to 8 BF16, return output 3.
tilelang/contrib/cutedsl/quantize.py:212
↓ 1 callersFunction_frcp_f32
(a: Float32, *, rounding: str = "rn", loc=None, ip=None)
tilelang/contrib/cutedsl/ieee_math.py:99
← previousnext →1,501–1,600 of 7,254, ranked by callers