MCPcopy Create free account

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

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

↓ 1 callersFunctioncompare_ab
(baseline, target, b, s_q, cache_seqlens, h_q, h_kv, d, dv, causal, dtype)
examples/deepseek_mla/amd/benchmark_mla_decode_amd_aiter.py:123
↓ 1 callersFunctioncompare_ab
(baseline, target, b, s_q, cache_seqlens, h_q, h_kv, d, dv, causal, dtype)
examples/deepseek_mla/amd/benchmark_mla_decode_amd_triton.py:342
↓ 1 callersFunctioncompile_cuda
Compile cuda code with NVRTC. Parameters ---------- code : str The cuda code. target_format : Literal["ptx", "cubin"]
tilelang/contrib/nvrtc.py:14
↓ 1 callersFunctioncompile_grouped_unit_tvm_ffi
Compile one grouped unit for CUDA+tvm_ffi backend. Flow: 1. Elaborate each config into a PrimFunc. 2. Lower each PrimFunc into host/devic
tilelang/autotuner/grouped_compile.py:25
↓ 1 callersFunctioncompile_hip
Compile HIP code with hipcc. Parameters ---------- code : str The HIP code. target_format : str The target format of
tilelang/contrib/hipcc.py:37
↓ 1 callersFunctioncompile_int4_gemm
( M=1024, N=1024, K=1024, block_M=128, block_N=128, block_K=64, threads=128, p
examples/gemm_int4/example_tilelang_gemm_int4.py:41
↓ 1 callersMethodcompile_lib
Compile CUDA source to cubin using NVRTC and write output files. Output artifacts (all in temp directory): - .cu: Source code (fo
tilelang/jit/adapter/nvrtc/libgen.py:144
↓ 1 callersMethodcompile_program
Compile one candidate program using this compile configuration.
tilelang/autotuner/param.py:63
↓ 1 callersMethodcomplete_config
(self, node: PrimFuncNode)
tilelang/carver/roller/hint.py:252
↓ 1 callersFunctioncomplex_expr_2
()
testing/python/arith/test_arith_hard.py:31
↓ 1 callersFunctioncomplex_expr_3
()
testing/python/arith/test_arith_hard.py:36
↓ 1 callersFunctioncomplex_expr_4
()
testing/python/arith/test_arith_hard.py:41
↓ 1 callersFunctioncompute_correlation
(a, b, label="tensor")
examples/deepseek_v32/fp8_lighting_indexer.py:13
↓ 1 callersMethodcompute_node_stride_map
Computes the stride map for a given node based on the TileDict configuration. Parameters ---------- node : PrimFuncN
tilelang/carver/roller/policy/default.py:497
↓ 1 callersMethodcompute_tile_dict
Computes and returns a TileDict object for a given output tile configuration and reduction step map. Parameters ----------
tilelang/carver/roller/policy/default.py:537
↓ 1 callersMethodcompute_warp_partition
(self, M: int, N: int, block_size: int, target: Target, gemm_inst: str)
tilelang/ir.py:31
↓ 1 callersFunctionconstruct_inputs
(batch_sizes_list, K, M, trans_b, padding_M, device, dtype)
examples/grouped_gemm/example_grouped_gemm_fwd.py:109
↓ 1 callersFunctionconstruct_inputs
(batch_sizes_list, K, M, trans_b, padding_M, device, dtype)
examples/grouped_gemm/example_grouped_gemm_bwd.py:133
↓ 1 callersFunctionconstruct_inputs
(batch_sizes_list, K, N, block_M, device, dtype)
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:78
↓ 1 callersFunctionconstruct_strides
Construct row-major strides from shape.
tilelang/language/eager/utils.py:89
↓ 1 callersFunctionconvert_iter_expr
(expr)
testing/python/arith/test_arith_iter_affine_map.py:47
↓ 1 callersFunctionconvolution
(data, kernel, S, D, P, block_M, block_N, block_K, num_stages, threads, dtype=T.float16, accum_dtype=T.float32
examples/convolution/example_convolution.py:20
↓ 1 callersFunctionconvolution_im2col
(N, C, H, W, F, K, S, D, P, block_M, block_N, block_K, num_stages, threads, dtype=T.float16, accum_dtype=T.flo
testing/python/jit/test_tilelang_jit_tvm_ffi.py:319
↓ 1 callersFunctionconvolution_im2col
(N, C, H, W, F, K, S, D, P, block_M, block_N, block_K, num_stages, threads, dtype=T.float16, accum_dtype=T.flo
testing/python/jit/test_tilelang_jit_nvrtc.py:321
↓ 1 callersFunctioncopy_kernel
(N, QN)
testing/python/amd/test_tilelang_mxfp4_gfx950.py:70
↓ 1 callersFunctioncosine_diff
(x: torch.Tensor, y: torch.Tensor)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:333
↓ 1 callersFunctioncosine_similarity
(a, b)
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:524
↓ 1 callersFunctioncosine_similarity
(a, b)
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:629
↓ 1 callersFunctioncp_async_access_ptr_legalize
()
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:203
↓ 1 callersFunctioncp_async_access_ptr_nonzero_safe_value_legalize
()
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:249
↓ 1 callersFunctioncp_async_gs
(size, dst, src)
tilelang/contrib/cutedsl/cpasync.py:52
↓ 1 callersFunctioncp_async_shared_global
Asynchronously copy data from global memory to shared memory. :param dst: Destination pointer in shared memory :type dst: Pointer :p
tilelang/contrib/cutedsl/cpasync.py:252
↓ 1 callersMethodcreate
( cls, name: str, prim_func: PrimFunc[_P, _T], constexpr: set[Var], ir_gen: IRGenerator[_P, _T] | None
tilelang/language/eager/builder.py:1043
↓ 1 callersMethodcreate_bitblas_weights
(self, weight, weight_group=1)
examples/bitnet-1.58b/utils_quant.py:111
↓ 1 callersMethodcreate_call_func
(self, code, function_informations)
tilelang/jit/adapter/wrapper.py:789
↓ 1 callersMethodcreate_dispatch_func
(self, code, function_informations)
tilelang/jit/adapter/wrapper.py:277
↓ 1 callersMethodcreate_dispatch_func
Generate Python dispatch function that launches multiple CUDA kernels. Why two-pass design: Pass 1: Collect TMA descriptors from
tilelang/jit/adapter/nvrtc/wrapper.py:291
↓ 1 callersMethodcreate_dispatch_func
Create dispatch function - always use C++ launcher.
tilelang/jit/adapter/cutedsl/wrapper.py:1521
↓ 1 callersMethodcreate_dispatch_func_cpp_launcher
Create dispatch function using C++ launcher.
tilelang/jit/adapter/cutedsl/wrapper.py:1537
↓ 1 callersFunctioncummax_fragment
Compute cumulative maximum for fragment buffers by staging through shared memory. Args: src: Source buffer (Buffer, BufferRegion, or
tilelang/language/scan_op.py:150
↓ 1 callersFunctioncummax_fragment_test
(M, N, block_M, block_N, dim=0, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:331
↓ 1 callersFunctioncummax_fragment_test_1d
(N, block_N, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:411
↓ 1 callersFunctioncummax_smem_out_test
(M, N, block_M, block_N, dim=0, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:349
↓ 1 callersFunctioncummax_smem_test
(M, N, block_M, block_N, dim=0, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:315
↓ 1 callersFunctioncummax_smem_test_1d
(N, block_N, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:395
↓ 1 callersFunctioncumsum_fragment
Compute cumulative sum for fragment buffers by copying to shared memory first. This macro handles cumulative sum operations on fragment buff
tilelang/language/scan_op.py:42
↓ 1 callersFunctioncumsum_fragment_test
(M, N, block_M, block_N, dim=0, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:31
↓ 1 callersFunctioncumsum_fragment_test_1d
(N, block_N, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:100
↓ 1 callersFunctioncumsum_region_test_1d
Test cumsum with buffer region (slice) as input.
testing/python/language/test_tilelang_language_scan.py:182
↓ 1 callersFunctioncumsum_region_test_2d
Test cumsum with buffer region (slice) as input in 2D.
testing/python/language/test_tilelang_language_scan.py:232
↓ 1 callersFunctioncumsum_smem_test
(M, N, block_M, block_N, dim=0, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:14
↓ 1 callersFunctioncumsum_smem_test_1d
(N, block_N, reverse=False, dtype=T.float32)
testing/python/language/test_tilelang_language_scan.py:82
↓ 1 callersFunctioncustom_device_ctx
(index: int)
examples/kda/FLA_KDA/fla_utils.py:199
↓ 1 callersFunctioncustom_kernel
DeepSeek-style Mixture of Experts using Tilelang. Args: data: Tuple of (input: torch.Tensor, weights: Dict[str, torch.Tensor], confi
examples/fusedmoe/example_fusedmoe_tilelang.py:431
↓ 1 callersFunctioncustomize_op_with_parallel
(length=256, block=16, dtype=T.float32)
testing/python/analysis/test_tilelang_nested_loop_checker.py:652
↓ 1 callersFunctiondebug_print_buffer_conditional
(M=16, N=16)
testing/python/debug/test_tilelang_debug_print.py:40
↓ 1 callersFunctiondebug_print_register_files
(M=16, N=16)
testing/python/debug/test_tilelang_debug_print.py:79
↓ 1 callersFunctiondebug_print_value_conditional
(M=16, N=16)
testing/python/debug/test_tilelang_debug_print.py:60
↓ 1 callersFunctiondecorator
(func: Callable[_P, _T])
tilelang/jit/__init__.py:592
↓ 1 callersFunctiondecorator_wrapper
(func)
tilelang/language/parser/entry.py:66
↓ 1 callersFunctiondecorator_wrapper
(func)
tilelang/language/tir/entry.py:39
↓ 1 callersFunctiondeepseek_sparse_attention
( q: torch.Tensor, kv: torch.Tensor, index_q: torch.Tensor, index_k: torch.Tensor, weights
examples/dsa_sparse_finetune/dsa.py:152
↓ 1 callersFunctiondefault_cast_kernel
(M_val: int)
testing/python/language/test_tilelang_cast_rounding.py:72
↓ 1 callersFunctiondetect_iter_traits
Detect iter traits based on the pattern C[S, I, J] += A[S, I, K] * B[S, J, K] Parameters ---------- block : tirx.SBlock The block
tilelang/carver/matmul_analysis.py:237
↓ 1 callersFunctiondevice_codegen_without_compile
(device_mod: tvm.IRModule, target: Target)
tilelang/engine/lower.py:248
↓ 1 callersMethoddfs_smem_tile
(self, init_tile, rstep_map)
tilelang/carver/roller/policy/default.py:96
↓ 1 callersFunctiondisk_compile
(source, name)
tilelang/language/eager/utils.py:12
↓ 1 callersMethoddom
The iteration domain of the block.
tilelang/carver/analysis.py:68
↓ 1 callersFunctiondraw
(df: pd.DataFrame)
maint/scripts/test_perf_regression.py:69
↓ 1 callersFunctiondump_to_file
Persist parsed resource usage so it survives kernel-cache hits.
tilelang/contrib/hip_resource_info.py:123
↓ 1 callersFunctiondynamic
Create a TIR dynamic symbolic variable. Parameters: name (str): Identifier for the variable in generated TIR. dtype (str): D
tilelang/language/symbolics.py:13
↓ 1 callersFunctiondynamic_kern
(src, dst, slots)
testing/python/language/test_tilelang_language_subtype.py:349
↓ 1 callersFunctiondynamic_metadata
(field: str, settings: dict[str, object] | None = None)
version_provider.py:42
↓ 1 callersFunctiondynamic_smem_kernel
()
testing/python/runtime/test_tilelang_runtime_dynamic_shared_memory.py:10
↓ 1 callersFunctionelementwise_add
(M, N, block_M, block_N, in_dtype, out_dtype, threads)
testing/python/language/test_tilelang_language_tma_1d.py:11
↓ 1 callersFunctionelementwise_add_with_l2_cache
( M, N, block_size=256, dtype=T.float32, )
testing/python/jit/test_tilelang_jit_tvm_ffi.py:396
↓ 1 callersFunctionelementwise_add_with_l2_cache
( M, N, block_size=256, dtype=T.float32, )
testing/python/jit/test_tilelang_jit_nvrtc.py:399
↓ 1 callersFunctionencode_one
(v)
testing/python/amd/test_tilelang_mxfp4_gfx950.py:34
↓ 1 callersFunctioneval_aug_assign
(op: Operator, left: Any, sl: slice, right: Any)
tilelang/language/eager/ast.py:123
↓ 1 callersFunctionexp2
(x: Union[TensorSSA, Numeric], fastmath: bool = False, **kwargs)
tilelang/contrib/cutedsl/math.py:86
↓ 1 callersFunctionexplicit_cp_async_wait_position
A mixed TMA + explicit cp.async pipeline with cp.async consumed first.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:179
↓ 1 callersMethodexport_sources
Export generated source code to files. Parameters ---------- kernel_path : Optional[str] Destination fil
tilelang/jit/kernel.py:507
↓ 1 callersFunctionexpr_is_same
(a, b)
tilelang/carver/roller/shape_inference/tir.py:300
↓ 1 callersFunctionextrac_params
(func: tirx.PrimFunc)
tilelang/engine/lower.py:177
↓ 1 callersFunctionextract_if_condition
Extract IfThenElse condition from statement if present. Returns: A tuple of (condition, inner_body). If no IfThenElse, returns (None, stm
tilelang/transform/decouple_type_cast.py:278
↓ 1 callersFunctionextract_python_func_declaration
Extract the full Python function declaration from decorator to colon. Args: source: Source code containing the function func_name
tilelang/jit/adapter/utils.py:56
↓ 1 callersFunctionf16_reduce
(n_t: int, n_e: int)
testing/python/amd/test_tilelang_hip_codegen.py:227
↓ 1 callersFunctionfast_flashattn
( batch, heads, seq_len, dim, is_causal, groups, block_M: int, block_N: int,
examples/amd/example_amd_flash_attn_bwd.py:91
↓ 1 callersFunctionfast_flashattn
( batch, heads, seq_len, dim, is_causal, groups, block_M: int, block_N: int,
examples/amd/example_amd_flash_attn_fwd.py:109
↓ 1 callersFunctionfast_log2_ceil
(x)
examples/deepseek_v32/inference/kernel.py:17
↓ 1 callersFunctionfast_log2_ceil
Compute ceil(log2(x)) via IEEE 754 bit manipulation. Avoids slow log/ceil intrinsics.
examples/deepseek_v4/act_quant.py:6
↓ 1 callersFunctionfast_pow2
(x)
examples/deepseek_v32/inference/kernel.py:24
↓ 1 callersFunctionfast_pow2
Compute 2^x for integer x via IEEE 754 bit manipulation.
examples/deepseek_v4/act_quant.py:14
↓ 1 callersFunctionfast_round_scale
(amax, fp8_max_inv)
examples/deepseek_v32/inference/kernel.py:29
↓ 1 callersFunctionfill_static
(n: int, value: float, dtype=T.bfloat16)
testing/python/language/test_tilelang_language_int64.py:98
↓ 1 callersFunctionfill_symbolic
(value: float, dtype=T.bfloat16)
testing/python/language/test_tilelang_language_int64.py:68
↓ 1 callersFunctionfind_correction_range
Computes the range of correction dimensions for rotary positional embeddings. Args: low_rot (float): Lower bound for the
examples/deepseek_v32/inference/model.py:372
↓ 1 callersFunctionfind_decorator_annotation
Check the value of given annotation (argument name) in the prim_func decorator. Returns the value of the annotation if present, otherwise giv
tilelang/language/parser/parser.py:162
↓ 1 callersFunctionfind_first_similar_buffer
(regions: list[BufferRegion], buffer: tirx.Buffer)
tilelang/carver/matmul_analysis.py:133
↓ 1 callersFunctionfind_first_similar_region
(regions: list[BufferRegion], buffer: tirx.Buffer)
tilelang/carver/matmul_analysis.py:125
← previousnext →2,001–2,100 of 7,254, ranked by callers