MCPcopy Create free account

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

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

↓ 3 callersFunction_kind_for
(dtype_str)
tilelang/contrib/cutedsl/gemm_tcgen05.py:115
↓ 3 callersMethod_legalize_to_buffer_region
(obj: Buffer | BufferLoad | BufferRegion)
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:481
↓ 3 callersMethod_lookup_type
(self, dtype: str | Any)
tilelang/jit/adapter/wrapper.py:268
↓ 3 callersFunction_lower_cutedsl_for_pdl
Lower a PDL program through CuTeDSL and build its host wrapper.
testing/python/language/test_tilelang_language_pdl.py:288
↓ 3 callersFunction_lower_for
(arch: str)
testing/python/issue/test_tilelang_issue_sm120_tma_smem_alignment.py:110
↓ 3 callersFunction_lower_to_cuda_source
(func, arch)
testing/python/transform/test_tilelang_transform_im2col_fallback.py:47
↓ 3 callersFunction_make_gemm_kernel
(M, N, K, block_M, block_N, block_K, in_dtype, out_dtype, accum_dtype)
testing/python/kernel/test_tilelang_kernel_gemm_sm75.py:8
↓ 3 callersFunction_make_im2col_kernel
(use_deprecated_alias=False)
testing/python/transform/test_tilelang_transform_im2col_fallback.py:9
↓ 3 callersFunction_nd_to_flat
Convert an N-D index to a row-major flat index.
tilelang/tools/plot_layout.py:339
↓ 3 callersFunction_normalize_copy_regions
( src: BufferLikeType, dst: BufferLikeType )
tilelang/language/copy_op.py:17
↓ 3 callersFunction_normalize_primexpr_args
(args)
tilelang/language/tir/op.py:21
↓ 3 callersMethod_parse_buffer_nd
Extract (buffer, extra_indices, row_offset, col_offset, stride) from Buffer or BufferRegion. For 2D buffers, extra_indices is an empty tuple.
tilelang/metal/intrinsics/metal_macro_generator.py:70
↓ 3 callersMethod_parse_phase1_key
(self, *args, **kwargs)
tilelang/language/eager/builder.py:1141
↓ 3 callersMethod_post_init
(self)
tilelang/jit/adapter/base.py:108
↓ 3 callersMethod_post_process_weights
(self)
examples/bitnet-1.58b/modeling_bitnet.py:1373
↓ 3 callersMethod_primary_resource_usage
(self)
tilelang/jit/kernel.py:648
↓ 3 callersFunction_remap_buffer_region
Remap buffer inside a BufferRegion using buf_map.
tilelang/metal/transform/metal_fragment_to_simdgroup.py:73
↓ 3 callersFunction_resolve
(dim, provided, name)
testing/python/autotune/test_tilelang_autotune_with_inputs.py:106
↓ 3 callersFunction_resolve
(dim, provided, name)
testing/python/autotune/test_tilelang_autotune_eager_mode.py:96
↓ 3 callersFunction_resolve_cache_dispatch
( target: str | Target | None, execution_backend: Literal["auto", "tvm_ffi", "cython", "nvrtc", "torch
tilelang/cache/__init__.py:42
↓ 3 callersMethod_sanitize_path_component
(component: str)
tilelang/cache/kernel_cache.py:141
↓ 3 callersFunction_save_plot
Save the current matplotlib figure in the specified format(s).
tilelang/tools/plot_layout.py:80
↓ 3 callersFunction_scalar_type_name
(scalar_type)
tilelang/contrib/cutedsl/math.py:37
↓ 3 callersFunction_shared_layout_continuity
(buffer, is_k_major: bool, k_extent: int, mn_extent: int)
tilelang/cuda/op/gemm/gemm_tcgen05.py:34
↓ 3 callersFunction_target_kind
(target: Target)
tilelang/jit/execution_backend.py:21
↓ 3 callersFunction_tir_u8_to_f4_to_f16
(nbit: int, val: tirx.PrimExpr, pos: tirx.PrimExpr, dtype: str)
examples/dequantize_gemm/example_dequant_gemm_fp4_hopper.py:10
↓ 3 callersMethod_try_eval
(self, node: ast.expr)
tilelang/language/eager/ast.py:507
↓ 3 callersMethod_use_fp64_store_index_map
(self)
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:225
↓ 3 callersFunction_warp_prefix_max_forward
Warp-level inclusive prefix max (forward).
tilelang/contrib/cutedsl/reduce.py:245
↓ 3 callersFunction_warp_prefix_max_reverse
Warp-level inclusive prefix max (reverse).
tilelang/contrib/cutedsl/reduce.py:260
↓ 3 callersFunction_warp_prefix_sum_forward
Warp-level inclusive prefix sum (forward). Uses shfl.up to propagate values from lower lanes.
tilelang/contrib/cutedsl/reduce.py:206
↓ 3 callersFunction_warp_prefix_sum_reverse
Warp-level inclusive prefix sum (reverse). Uses shfl.down to propagate values from higher lanes.
tilelang/contrib/cutedsl/reduce.py:225
↓ 3 callersFunction_write_complete_kernel_cache_entry
( cache: KernelCache, key: str, device_source: str = "// device kernel", host_source: str = "/
testing/python/cache/test_tilelang_kernel_cache_atomic_save.py:54
↓ 3 callersFunction_xdist_enabled
(config: Config)
maint/scripts/pytest_cuda_scheduler.py:146
↓ 3 callersFunctionact_quant
Quantizes the input tensor `x` using block-wise quantization. Args: x (torch.Tensor): The input tensor to be quantized. Must be cont
examples/deepseek_v32/inference/kernel.py:81
↓ 3 callersFunctionalloc_descriptor
Allocate a descriptor buffer for WGMMA and TCGEN5.MMA. Args: kind: The descriptor kind, one of "wgmma", "tcgen05" ("utcmma" as alias).
tilelang/language/allocate.py:302
↓ 3 callersFunctionallowed_backends_for_target
Return allowed execution backends for a given TVM target kind. include_unavailable: if False, this will filter out backends that are known to
tilelang/jit/execution_backend.py:26
↓ 3 callersFunctionapply_rewrites
(tree: ast.AST, target_labels: set[int])
tilelang/autodd.py:529
↓ 3 callersFunctionapply_rotary_pos_emb
Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T
examples/bitnet-1.58b/modeling_bitnet.py:167
↓ 3 callersFunctionassert_call_extern_access_ptr_mask_legalize
(access_type: str)
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:467
↓ 3 callersFunctionassert_chained_gemm
Compile and run a chained GEMM (SS + TS), verifying tcgen05.st and mma_ts presence.
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:155
↓ 3 callersFunctionassert_gemm
( M, N, K, block_M, block_N, block_K, dtype=T.float32, accum_dtype=T.float32,
testing/python/metal/test_metal_codegen.py:36
↓ 3 callersFunctionassert_metal_codegen
( M, N, K, block_M, block_N, block_K, dtype=T.float32, accum_dtype=T.float32,
testing/python/metal/test_metal_codegen_linux.py:40
↓ 3 callersFunctionassert_tensors_similar
Assert that two tensors are similar using a global similarity metric. Key differences from torch.testing.assert_close: - torch.testing.a
examples/deepseek_v32/utils.py:285
↓ 3 callersFunctionassert_tensors_similar
Assert that two tensors are similar using a global similarity metric. Key differences from torch.testing.assert_close: - torch.testing.a
examples/dsa_sparse_finetune/utils.py:48
↓ 3 callersFunctionassert_tl_matmul_correctness
(M, N, K, in_dtype, out_dtype, accum_dtype)
examples/bitnet-1.58b/kernel_benchmark/tl_int8xint8.py:179
↓ 3 callersFunctionassert_tl_matmul_correctness
(M, N, K, in_dtype, out_dtype, accum_dtype)
testing/python/kernel/test_tilelang_kernel_gemm_simt.py:132
↓ 3 callersMethodassign
(self, spec: object)
maint/scripts/pytest_cuda_scheduler.py:69
↓ 3 callersMethodassign_device_module
(self, device_mod: IRModule)
tilelang/jit/adapter/wrapper.py:962
↓ 3 callersMethodassign_host_module
(self, host_mod: IRModule)
tilelang/jit/adapter/wrapper.py:959
↓ 3 callersMethodassign_optimized_module
(self, scheduled_ir_module: IRModule)
tilelang/jit/adapter/wrapper.py:953
↓ 3 callersFunctioncal_seq_idx_for_q
(cu_seqlens_qs: torch.LongTensor, cu_seqlens_qe: torch.LongTensor, seq_len: int)
examples/deepseek_v32/utils.py:83
↓ 3 callersFunctioncheck_fastmath_usage
Check source for fastmath/non-fastmath versions
testing/python/fastmath/test_mathops_fastmath.py:23
↓ 3 callersFunctioncheck_fastmath_usage
Check source for fastmath/non-fastmath versions
testing/python/math/test_math_fast_math.py:23
↓ 3 callersFunctioncheck_non_fastmath_usage
Check that source uses non-fastmath versions (no __ prefix)
testing/python/fastmath/test_mathops_fastmath.py:48
↓ 3 callersFunctioncheck_non_fastmath_usage
Check that source uses non-fastmath versions (no __ prefix)
testing/python/math/test_math_fast_math.py:48
↓ 3 callersFunctionchunk_bwd_dqkwg
( B, S, H, DK, DV, scale, chunk_size, input_dtype, gate_dtype, block_D
examples/kda/chunk_bwd_dqkwg.py:68
↓ 3 callersFunctionchunk_gated_delta_rule_bwd_dhu
( q: torch.Tensor, k: torch.Tensor, w: torch.Tensor, do: torch.Tensor, dv: torch.Tensor,
examples/kda/FLA_KDA/fla_chunk_delta.py:524
↓ 3 callersFunctioncleanup
()
examples/bitnet-1.58b/vllm_workspace/conftest.py:81
↓ 3 callersFunctioncoalesced_tensor_shape
(subtensor: list[int], tensor: list[int], transaction_size: int)
tilelang/carver/roller/policy/common.py:41
↓ 3 callersFunctioncollect_assume_vars
Collect all variables used in assume statements.
testing/python/transform/test_tilelang_transform_split_host_device.py:38
↓ 3 callersFunctioncollect_block_iter_vars_used_in_access_region
Collect the block iter variables used in the access region of a buffer region.
tilelang/carver/analysis.py:209
↓ 3 callersFunctioncomm_reducer
Create a commutative reducer for reduction. Parameters ---------- fcombine : function(Expr -> Expr -> Expr) A binary function whi
tilelang/language/tir/op.py:3458
↓ 3 callersMethodcompute_tcgen05_instr_desc
Compute the 64-bit instruction descriptor using current meta. Requires ``self.meta`` to have been set via ``get_tcgen5_mma_meta()``.
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:817
↓ 3 callersMethodcompute_wgmma_b_desc_params
Compute B descriptor parameters from the B shared buffer region. This is a pure-Python helper -- no TIR code is emitted. The returned
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:315
↓ 3 callersFunctioncond
()
testing/python/language/test_tilelang_language_frontend_v2.py:449
↓ 3 callersFunctiondebug_tensor_comparison
(tensor1, tensor2, name, rtol=1e-3, atol=1e-3)
examples/amd/example_amd_flash_attn_bwd.py:447
↓ 3 callersFunctiondiv
Compute a / b as in C/C++ semantics. Parameters ---------- a : PrimExpr The left hand operand, known to be non-negative. b :
tilelang/language/tir/op.py:3261
↓ 3 callersFunctiondot
(x1, x2, y1, y2, buf, out, tilesize, n_stream)
examples/deepseek_mhc/example_mhc_bwd.py:67
↓ 3 callersFunctionexample_tensor_annot
()
testing/python/language/test_tilelang_language_annot.py:11
↓ 3 callersFunctionf
(x: torch.Tensor)
examples/deepseek_v32/utils.py:167
↓ 3 callersFunctionfill
Fill a buffer or buffer region with a specified value. Args: buffer: Either a TVM buffer or buffer region to be filled value: The
tilelang/language/fill_op.py:10
↓ 3 callersFunctionfind_cuda_path
Utility function to find cuda path Returns ------- path : str Path to cuda root.
tilelang/contrib/nvcc.py:297
↓ 3 callersFunctionflashattn
(batch, heads, groups, seqlen_kv, dim, block_N, block_H, num_split, num_stages, threads)
examples/flash_decoding/example_gqa_decode.py:50
↓ 3 callersFunctionflashattn
(batch, heads, seq_len, dim, is_causal, block_M=64, block_N=64, num_stages=1, threads=128)
examples/flash_attention/example_mha_fwd_bshd.py:23
↓ 3 callersFunctionflashattn
(batch, heads, seq_len, dim, is_causal, groups=1, block_M=64, block_N=64, num_stages=0, threads=128)
examples/flash_attention/example_gqa_fwd_bshd.py:68
↓ 3 callersFunctionflashattn
(batch_size, UQ, UKV, heads, dim, is_causal, block_M=64, block_N=64, num_stages=1, threads=128)
examples/flash_attention/example_mha_fwd_varlen.py:27
↓ 3 callersFunctionflashattn
(batch, heads, seq_q, seq_kv, dim, is_causal, block_M=64, block_N=64, num_stages=1, threads=128)
examples/flash_attention/example_mha_fwd_bhsd.py:23
↓ 3 callersFunctionflashattn
(batch, heads, heads_kv, dim, dim_v, block_N, block_H, num_stages, threads)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_mask.py:19
↓ 3 callersFunctionfn
()
examples/deepseek_v32/sparse_mla_fwd_seesaw.py:628
↓ 3 callersMethodforward
(self, query, key, value, block_indices, cache_seqlens, block_table)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_paged.py:185
↓ 3 callersFunctionfp4_dequant_to_float
Dequantize FP4 packed tensor back to float32 for verification. quant: shape (M, N//2), dtype float4_e2m1fn_x2. scale: shape (M, num_blocks),
examples/deepseek_v4/act_quant.py:271
↓ 3 callersFunctionfp8_native_block_mean_pooling_interface
(k: torch.Tensor, k_scale: torch.Tensor, k_block_size: int)
examples/dsa_hisa/fp8_block_mean_pooling.py:81
↓ 3 callersFunctionfp8_native_block_sparse_mqa_attn_return_logits_interface
( q: torch.Tensor, k: torch.Tensor, k_scale: torch.Tensor, topk_block_index: torch.Tensor,
examples/dsa_hisa/block_sparse_mqa_fp8.py:106
↓ 3 callersMethodfrom_kernel
Create an AutoTuner instance from a kernel function. Args: kernel: The kernel function to auto-tune. configs: List of
tilelang/autotuner/tuner.py:263
↓ 3 callersMethodfrom_prim_func
(cls, func: tvm.tirx.PrimFunc, arch: TileDevice, tags: dict | None = None, name: str = "PrimFuncNode")
tilelang/carver/roller/policy/default.py:40
↓ 3 callersFunctionget_access_axes
(region: list[Range])
tilelang/carver/matmul_analysis.py:256
↓ 3 callersFunctionget_cc
Return the path to the default C/C++ compiler. Returns ------- out: Optional[str] The path to the default C/C++ compiler, or None
tilelang/contrib/cc.py:41
↓ 3 callersFunctionget_device_call
(is_device_c: bool = False)
tilelang/engine/lower.py:47
↓ 3 callersMethodget_dynamic_symbolic_set
(self, prim_func)
tilelang/jit/adapter/wrapper.py:859
↓ 3 callersFunctionget_kernel
()
testing/python/transform/test_nullable_buffer_params.py:11
↓ 3 callersMethodget_launcher_cpp_code
Get the generated C++ launcher code.
tilelang/jit/adapter/cutedsl/wrapper.py:1616
↓ 3 callersMethodget_ldmatrix_index_map
Return (forward, reverse) index maps for shared→local loading. The actual layout functions are chosen during __init__ based on rdna_gen:
tilelang/rocm/intrinsics/wmma_macro_generator.py:146
↓ 3 callersFunctionget_ldmatrix_offset_b
( matrix: Literal["B"], row_idx, col_idx, stride, dtype: Literal["float16", "int8"] = "flo
tilelang/cuda/intrinsics/layout/mma_sp_layout.py:150
↓ 3 callersFunctionget_logical_id_32bit
(thread_id: int)
tilelang/cuda/intrinsics/layout/mma_sp_layout.py:71
↓ 3 callersFunctionget_logical_id_8bit
(thread_id: int)
tilelang/cuda/intrinsics/layout/mma_sp_layout.py:100
↓ 3 callersFunctionget_msvc_subprocess_env
Return the resolved MSVC subprocess environment. Memoized via :func:`functools.cache` (matching ``contrib/cc.py`` and ``cache/kernel_cache.py
tilelang/contrib/msvc.py:206
↓ 3 callersFunctionget_nvcc_compiler
Get the path to the nvcc compiler
tilelang/contrib/nvcc.py:611
↓ 3 callersFunctionget_persisting_l2_cache_max_size
(device_id: int = 0)
tilelang/carver/arch/driver/cuda_driver.py:106
↓ 3 callersMethodget_region_base_offsets
Get the base offset (start index) for each dimension from a BufferRegion. For example, if region is A_shared[ko % 2, 0:128, 0:64],
tilelang/tileop/gemm/gemm_base.py:203
← previousnext →601–700 of 7,254, ranked by callers