MCPcopy Create free account

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

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

↓ 2 callersFunctionis_global_or_shared_buffer
Check if a buffer is a global or shared buffer.
tilelang/transform/decouple_type_cast.py:81
↓ 2 callersFunctionis_identity_or_transpose_block
(block_stmt: tirx.SBlock)
tilelang/carver/matmul_analysis.py:435
↓ 2 callersMethodis_injective
Whether the block is injective, i.e. all its iteration domains are injective.
tilelang/carver/analysis.py:76
↓ 2 callersMethodis_light_import
Return True if we are running in light import mode.
tilelang/env.py:393
↓ 2 callersFunctionis_metal_simdgroup
Check if the buffer is in the Metal simdgroup scope. Args: buffer: The TVM buffer, BufferLoad, or BufferRegion to check. Return
tilelang/utils/language.py:122
↓ 2 callersFunctionis_pow_of_2
(n)
examples/hadamard_transform/example_hadamard.py:12
↓ 2 callersMethodis_print_on_compilation_enabled
(self)
tilelang/env.py:370
↓ 2 callersFunctionis_reduction
(block: SBlockRV)
tilelang/carver/analysis.py:265
↓ 2 callersMethodis_swizzle_128b
(self)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:65
↓ 2 callersMethodis_swizzle_128b
(self)
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:68
↓ 2 callersMethodis_swizzle_32b
(self)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:59
↓ 2 callersMethodis_swizzle_32b
(self)
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:62
↓ 2 callersMethodis_swizzle_64b
(self)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:62
↓ 2 callersMethodis_swizzle_64b
(self)
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:65
↓ 2 callersFunctionis_tile_operation_loop
Determines if a For loop is a tile operation. Args: loop: The For loop to check Returns:
tilelang/transform/add_bufstore_wrapper.py:90
↓ 2 callersMethodis_unsigned
Checks if the parameter represents an unsigned integer type. Returns: bool: True if parameter is an unsigned integer typ
tilelang/engine/param.py:79
↓ 2 callersMethodis_valid
(self)
tilelang/carver/roller/hint.py:68
↓ 2 callersFunctionis_volta_arch
(arch: TileDevice)
tilelang/carver/arch/cuda.py:17
↓ 2 callersFunctionissue_2123_atomic_load_repro
(num_tiles, threads=32)
testing/python/issue/test_tilelang_issue_2123.py:11
↓ 2 callersFunctionkernel
(M, N, dtype, op, src_scope, dst_scope, threads, batch)
testing/python/language/test_tilelang_language_reduce.py:101
↓ 2 callersFunctionkernel
( A: T.Tensor((M, N), dtype), B: T.Tensor((M, N), dtype), C: T.Tensor((M,
testing/python/jit/test_tilelang_jit_tvm_ffi.py:403
↓ 2 callersFunctionkernel
The actual kernel to compute C = A @ B^T. Parameters ---------- block_M : int Block size in M dimension.
testing/python/autotune/test_tilelang_autotune.py:95
↓ 2 callersFunctionkernels_with_pdl_pipeline
Create a two-kernel TileLang pipeline using PDL trigger and sync.
testing/python/language/test_tilelang_language_pdl.py:54
↓ 2 callersFunctionkernels_with_pdl_sync
Create a TileLang kernel that waits for PDL dependencies.
testing/python/language/test_tilelang_language_pdl.py:34
↓ 2 callersFunctionkernels_with_pdl_trigger
Create a TileLang kernel that triggers dependent PDL launches.
testing/python/language/test_tilelang_language_pdl.py:14
↓ 2 callersMethodldmatrix_a
Load matrix A tiles from shared memory into simdgroup local buffers.
tilelang/metal/intrinsics/metal_macro_generator.py:90
↓ 2 callersMethodldmatrix_b
Load matrix B tiles from shared memory into simdgroup local buffers.
tilelang/metal/intrinsics/metal_macro_generator.py:127
↓ 2 callersMethodldmatrix_e
(self, E_local_buf: Buffer, E_shared_buf: Buffer, inst_i: PrimExpr, warp_m: PrimExpr, ki: PrimExpr, ki_slot: P
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:431
↓ 2 callersMethodload_lib
Load compiled cubin and Python launcher into memory. Why two loads: 1. Import Python module for launch logic 2. Load
tilelang/jit/adapter/nvrtc/libgen.py:108
↓ 2 callersFunctionlower_to_host_device_ir
Lower input TIR to split host/device IRModules without backend codegen.
tilelang/engine/lower.py:272
↓ 2 callersFunctionmain
()
examples/gemm_streamk/example_tilelang_gemm_streamk.py:146
↓ 2 callersFunctionmain
Kernel entry for the tiled, pipelined matmul used by the generated prim_func. This function implements a block-wise GEMM over a 2D g
examples/dequantize_gemm/example_dequant_gemm_bf16_fp4_hopper.py:295
↓ 2 callersFunctionmake_aligned_tensor
(ptr: cute.Pointer, layout: cute.Layout, align_bytes: int, swizzle=False)
tilelang/contrib/cutedsl/gemm_v1.py:20
↓ 2 callersFunctionmake_buf
()
testing/python/language/test_tilelang_language_subtype.py:319
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/flash_attention/example_mha_bwd_bhsd.py:121
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/flash_attention/example_mha_bwd_bshd.py:118
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/attention_sink/example_gqa_sink_bwd_varlen.py:265
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/attention_sink/example_mha_sink_bwd_bhsd.py:167
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/attention_sink/example_gqa_sink_bwd_bhsd.py:168
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/flash_attention_sm100/gqa_bwd_bshd.py:123
↓ 2 callersFunctionmake_dq_layout
(dQ)
examples/flash_attention_sm100/mha_bwd_bshd.py:118
↓ 2 callersMethodmake_mma_load_layout
(self, local_buf: Buffer, matrix: str = "A")
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:503
↓ 2 callersFunctionmake_swizzle_layout
(shared_buf)
testing/python/language/test_tilelang_language_atom_mma.py:19
↓ 2 callersMethodmalloc
(self, size)
tilelang/carver/roller/bestfit.py:28
↓ 2 callersMethodmatch_placeholders
(self, node: "ast.AST | list[ast.AST]")
tilelang/autodd.py:305
↓ 2 callersMethodmatch_placeholders
(self, node: ast.AST)
tilelang/autodd.py:352
↓ 2 callersFunctionmatmul
(M, N, K, block_M, block_N, block_K, num_stages, thread_num, enable_rasteration, dtype=T.float16, accum_dtype=
examples/gemm/example_gemm_autotune.py:188
↓ 2 callersFunctionmatmul
(M, N, K, block_M, block_N, block_K, num_stages, thread_num, enable_rasteration, dtype=T.float16, accum_dtype=
examples/gemm/example_gemm_advanced_autotune.py:246
↓ 2 callersFunctionmatmul
(A, B, C, block_M, block_N, block_K, split_k, dtype=T.float16, accum_dtype=T.float32, out_dtype=T.float32)
examples/gemm_splitk/example_tilelang_gemm_splitk.py:6
↓ 2 callersFunctionmatmul
(A, B, C, block_M, block_N, block_K, split_k, dtype=T.float16, accum_dtype=T.float32, out_dtype=T.float32)
examples/gemm_splitk/example_tilelang_gemm_splitk_vectorize_atomicadd.py:6
↓ 2 callersFunctionmatmul
( A, B, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_dtype,
examples/gemm_sm100/gemm_tcgen5mma.py:13
↓ 2 callersFunctionmatmul
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/gemm_sm100/gemm_mma.py:13
↓ 2 callersFunctionmatmul
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/warp_specialize/example_warp_specialize_gemm_barrierpipe_stage2.py:6
↓ 2 callersFunctionmatmul
(M, N, K, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
testing/python/language/test_tilelang_language_cluster.py:7
↓ 2 callersFunctionmatmul
(M, N, K, block_M, block_N, block_K, trans_A=False, trans_B=False, dtype=T.float32, accum_dtype=T.float32)
testing/python/cpu/test_tilelang_cpu_tgemm.py:15
↓ 2 callersFunctionmatmul_do_not_spec
(M, N, K, threads=128)
testing/python/autotune/test_tilelang_autotune_do_not_specialize.py:36
↓ 2 callersFunctionmatmul_nested_pipa
( M, N, K, block_M, block_N, block_K, in_dtype, out_dtype, accum_dtype,
testing/python/analysis/test_tilelang_nested_loop_checker.py:364
↓ 2 callersFunctionmatmul_simdgroup
(M, N, K, block_M=64, block_N=64, block_K=32, dtype=T.float16, accum_dtype=T.float32)
benchmark/matmul_metal/benchmark_matmul_metal.py:21
↓ 2 callersFunctionmatmul_warp_specialize_copy_gemm_0_1
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/warp_specialize/example_warp_specialize_gemm_copy_gemm_0_1.py:6
↓ 2 callersFunctionmatvec_A
(R, x1, x2, buf, y1, y2, tilesize, n_stream)
examples/deepseek_mhc/example_mhc_bwd.py:52
↓ 2 callersFunctionmaybe_contiguous
(x)
examples/attention_sink/example_gqa_sink_bwd_bhsd.py:514
↓ 2 callersMethodmerge
(self, other)
tilelang/carver/roller/bestfit.py:13
↓ 2 callersFunctionmhc_post_tilelang
(a, b, c, d, x, hc: int, hidden: int, n_thr: int = 128, h_blk: int = 1024)
examples/deepseek_mhc/example_mhc_post.py:12
↓ 2 callersFunctionmhc_pre_big_fuse_tilelang
Deeply fused kernels, everything other than gemm & sqrsum in mHC pre block.
examples/deepseek_mhc/example_mhc_pre.py:11
↓ 2 callersFunctionmhc_pre_gemm_sqrsum_tilelang
Not highly optimized TileLang implementation of fused gemm and sqrsum in mHC pre block.
examples/deepseek_mhc/example_mhc_pre.py:135
↓ 2 callersFunctionmla_decode_tilelang
(batch, h_q, h_kv, max_seqlen_pad, dv, dpe, block_N, block_H, num_split, block_size, softmax_scale=None)
examples/deepseek_mla/example_mla_decode_paged.py:16
↓ 2 callersMethodmma_atom
Emit a single MMA atom for tile (inst_m_idx, inst_n_idx). This is the atomic building block of ``mma()``. Calling this method for ev
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:524
↓ 2 callersFunctionmma_store_index_map
(thread_id, local_id)
tilelang/cuda/intrinsics/layout/utils.py:82
↓ 2 callersFunctionnaive_nsa
r""" Args: q (torch.Tensor): Queries of shape `[B, T, HQ, K]` if `head_first=False` else `[B, HQ, T, K]`. k (torch.Ten
examples/deepseek_nsa/benchmark/benchmark_nsa_fwd.py:306
↓ 2 callersFunctionnative_sparse_attention
(Q, K, V, BlockIndices, dim, is_causal, scale=None, block_size=64, groups=1, selected_blocks=16)
examples/deepseek_nsa/example_tilelang_nsa_fwd.py:18
↓ 2 callersFunctionnative_sparse_attention
( Q, K, V, BlockIndices, dim, # Embedding dimension per head scale=None, block_si
examples/deepseek_nsa/example_tilelang_nsa_decode.py:20
↓ 2 callersFunctionnormalize_cutedsl_target
(target: TargetLike)
tilelang/utils/target.py:191
↓ 2 callersFunctionnormalize_rocm_arch
(arch: str | None)
tilelang/utils/target.py:33
↓ 2 callersFunctionpack_int4
(tensor: torch.Tensor)
examples/gemm_int4/example_tilelang_gemm_int4.py:59
↓ 2 callersFunctionpack_sf_u8_to_u32_1d
(sf_u8: torch.Tensor)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:202
↓ 2 callersFunctionparallel_nsa_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, o_slc: torch.Tensor, lse_slc: torch.T
examples/deepseek_nsa/example_triton_nsa_bwd.py:701
↓ 2 callersFunctionparallel_nsa_bwd_preprocess
(o: torch.Tensor, do: torch.Tensor)
examples/deepseek_nsa/example_triton_nsa_bwd.py:688
↓ 2 callersFunctionparallel_nsa_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, block_indices: torch.LongTensor, bloc
examples/deepseek_nsa/example_triton_nsa_bwd.py:162
↓ 2 callersFunctionparallel_nsa_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, block_indices: torch.LongTensor, bloc
examples/deepseek_nsa/example_triton_nsa_fwd.py:137
↓ 2 callersFunctionparse_compute_version
Parse compute capability string to divide major and minor version Parameters ---------- compute_version : str compute capability
tilelang/contrib/rocm.py:179
↓ 2 callersFunctionparse_output
(output)
maint/scripts/test_perf_regression.py:27
↓ 2 callersFunctionper_custom_dims_cast_to_fp8
(x: torch.Tensor, dims: Tuple[int], use_ue8m0: bool)
examples/deepseek_v32/utils.py:190
↓ 2 callersFunctionpointer_table_multi_copy_test
(G, N, dtype=T.float16)
testing/python/language/test_tilelang_language_ptr.py:56
↓ 2 callersMethodpost_proc
(self, text)
tilelang/autodd.py:1195
↓ 2 callersFunctionpow_of_int
Fast power operation than pow(float, float). Args: x (PrimExpr): Base value y (int): Exponent value
tilelang/language/tir/op.py:3028
↓ 2 callersFunctionprepare_cumsum_input
( B, S, H, dtype, )
examples/gdn/example_cumsum.py:74
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, output_dtype, )
examples/gdn/example_wy_fast.py:34
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, state_dtype, )
examples/gdn/example_chunk_delta_h.py:70
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, gate_dtype, state_dtype, )
examples/gdn/example_chunk_delta_bwd.py:90
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, )
examples/gdn/example_chunk_o.py:45
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, gate_dtype, state_dtype,
examples/gdn/example_chunk_o_bwd.py:82
↓ 2 callersFunctionprepare_output
( B, S, H, chunk_size, dtype, )
examples/gdn/example_chunk_scaled_dot_kkt.py:40
↓ 2 callersFunctionprepare_output
( B, S, H, chunk_size, sub_chunk_size, output_dtype, )
examples/kda/chunk_inter_solve_fused.py:40
↓ 2 callersFunctionprepare_output
( B, S, H, chunk_size, sub_chunk_size, output_dtype, )
examples/kda/chunk_intra_token_parallel.py:33
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, state_dtype, )
examples/kda/chunk_delta_h_fwd.py:48
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, use_qg, use_kg, output_dtype, )
examples/kda/wy_fast.py:24
↓ 2 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, )
examples/kda/chunk_o.py:34
↓ 2 callersFunctionprepare_position_ids
(cu_seqlens: torch.LongTensor)
examples/dsa_sparse_finetune/index.py:70
↓ 2 callersFunctionprepend_dll_search_path
Prepend ``paths`` to ``%PATH%`` on Windows, skipping entries already present. Used by Windows DLL discovery: PATH is consulted by ``LoadLibrary``
tilelang/env.py:51
↓ 2 callersFunctionprepend_pythonpath
(path)
tilelang/env.py:464
← previousnext →1,201–1,300 of 7,254, ranked by callers