MCPcopy Create free account

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

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

↓ 1 callersFunctionbenchmark_topk_sparse_attention
()
benchmark/blocksparse_attention/benchmark_library_dense_fmha.py:26
↓ 1 callersFunctionbenchmark_topk_sparse_attention
()
benchmark/blocksparse_attention/benchmark_tilelang_block_sparse_fmha.py:170
↓ 1 callersFunctionbf16_reduce
(n_t: int, n_e: int)
testing/python/amd/test_tilelang_hip_codegen.py:184
↓ 1 callersFunctionbitcast_impl
(src_val, *, loc=None, ip=None)
tilelang/contrib/cutedsl/utils.py:89
↓ 1 callersFunctionbitnet_158_int8xint2_decode
( M, N, K, in_dtype, out_dtype, accum_dtype, fast_decoding=True, n_partition=4
examples/bitnet-1.58b/kernel_benchmark/tilelang_bitnet_158_int8xint2_decode.py:67
↓ 1 callersFunctionbitnet_158_int8xint2_prefill
Create a TVM GPU prim_func implementing a block-tiled matrix multiply that multiplies dense A by compressed/interleaved low‑precision B (2-bit pa
examples/bitnet-1.58b/kernel_benchmark/tilelang_bitnet_158_int8xint2_prefill.py:73
↓ 1 callersFunctionbitwise_reduce
( M, N, block_M, block_N, name, func, clear=True, )
testing/python/math/test_math_bitwise_reduce.py:14
↓ 1 callersFunctionblock_tma_copy_kernel
()
testing/python/language/test_tilelang_language_tma_copy.py:144
↓ 1 callersFunctionblockscaled_gemm_ref
Torch reference for block-scaled MXFP8 GEMM. Args: a: [M, K] FP8 tensor b: [K, N] FP8 tensor, or [N, K] when transpose_B=True
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:582
↓ 1 callersFunctionblockscaled_mxfp4_unpacked_x_mxfp8_gemm
(M: int, N: int, K: int)
maint/gemm/correctness_evaluation_tcgen05_fp4_unpacked.py:61
↓ 1 callersFunctionblocksparse_copy_kernel
BlockSparse copy kernel using fragment for block mask indices.
testing/python/language/test_tilelang_language_let_layout.py:26
↓ 1 callersFunctionblocksparse_flashattn
(batch, heads, seq_len, dim, downsample_len, is_causal)
benchmark/blocksparse_attention/benchmark_tilelang_block_sparse_fmha.py:34
↓ 1 callersFunctionblocksparse_flashattn
(batch, heads, seq_len, dim, downsample_len, is_causal)
testing/python/transform/test_tilelang_transform_config_index_bitwidth.py:7
↓ 1 callersFunctionblocksparse_matmul
( A, B, BlockMask, block_M, block_N, block_K, num_stages, thread_num, enable_rasteration, dtype=T.float16,
examples/blocksparse_gemm/example_blocksparse_gemm.py:61
↓ 1 callersFunctionblocksparse_matmul
(M, N, K, block_M, block_N, block_K, num_stages, dtype=T.float16, accum_dtype=T.float32)
testing/python/language/test_tilelang_language_pipeline.py:123
↓ 1 callersFunctionblocksparse_matmul_global
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_any_of.py:23
↓ 1 callersFunctionblocksparse_matmul_global
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_all_of.py:23
↓ 1 callersFunctionblocksparse_matmul_local
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_any_of.py:116
↓ 1 callersFunctionblocksparse_matmul_local
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_all_of.py:116
↓ 1 callersFunctionblocksparse_matmul_shared
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_any_of.py:67
↓ 1 callersFunctionblocksparse_matmul_shared
( M, N, K, condition_dim, block_M, block_N, block_K, num_stages, thread_nu
testing/python/language/test_tilelang_language_all_of.py:67
↓ 1 callersMethodbody_sr
GEMM body_sr: A from shared memory, B from register
tilelang/contrib/cutedsl/gemm_v1.py:312
↓ 1 callersMethodbuffer
(self)
tilelang/language/eager/builder.py:157
↓ 1 callersFunctionbuggy_matmul
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/autodd/tilelang_minimized_expected.py:28
↓ 1 callersFunctionbuggy_matmul
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/autodd/tilelang_buggy.py:118
↓ 1 callersFunctionbuild_scalar_check_kernel
(target="cuda")
maint/host_checks/common.py:36
↓ 1 callersFunctionbwd
( Q, KV, dO, Indices, Lse, Delta, Offsets, TokenIndices, H, D, D_t
examples/dsa_sparse_finetune/sparse_mla_bwd.py:83
↓ 1 callersFunctioncached
Caches and reuses compiled kernels (using KernelCache class).
tilelang/cache/__init__.py:100
↓ 1 callersMethodcached
Caches and reuses compiled kernels to avoid redundant compilation. This is the ONLY place where environment variable processing, tar
tilelang/cache/kernel_cache.py:279
↓ 1 callersFunctioncal_cu_seqlen_ke_for_q
( cu_seqlens_qs: torch.LongTensor, cu_seqlens_qe: torch.LongTensor, cu_seqlens_ks: torch.LongTenso
examples/deepseek_v32/utils.py:103
↓ 1 callersFunctioncal_cu_seqlen_ke_for_q
( cu_seqlens_qs: torch.LongTensor, cu_seqlens_qe: torch.LongTensor, cu_seqlens_ks: torch.LongTenso
examples/dsa_hisa/tilelang_utils.py:235
↓ 1 callersFunctioncal_cu_seqlen_ks_for_q
( cu_seqlens_qs: torch.LongTensor, cu_seqlens_qe: torch.LongTensor, cu_seqlens_ks: torch.LongTensor, seq_l
examples/deepseek_v32/utils.py:91
↓ 1 callersFunctioncal_cu_seqlen_ks_for_q
( cu_seqlens_qs: torch.LongTensor, cu_seqlens_qe: torch.LongTensor, cu_seqlens_ks: torch.LongTensor, seq_l
examples/dsa_hisa/tilelang_utils.py:223
↓ 1 callersFunctioncal_seq_idx_from_cu_seqlens
(cu_seqlens: torch.LongTensor, seq_len: int)
examples/deepseek_v32/utils.py:75
↓ 1 callersFunctioncalc_diff
(x, y)
examples/deepseek_deepgemm/example_deepgemm_fp8_2xAcc.py:134
↓ 1 callersFunctioncalc_diff
(x, y)
examples/gemm_fp8/example_tilelang_gemm_fp8_sm100.py:68
↓ 1 callersFunctioncalc_diff
(x, y)
examples/gemm_fp8/example_tilelang_gemm_fp8_2xAcc.py:50
↓ 1 callersFunctioncalc_diff
(x, y)
examples/gemm_fp8/example_tilelang_gemm_fp8.py:7
↓ 1 callersFunctioncalc_diff
(x, y)
testing/python/kernel/test_tilelang_kernel_int8_gemm_tcgen5.py:6
↓ 1 callersFunctioncalc_sim
(x, y, name="tensor")
examples/dequantize_gemm/dequantize_utils.py:106
↓ 1 callersFunctioncalc_sim
(x, y, name="tensor")
examples/flash_decoding/example_gqa_decode.py:381
↓ 1 callersFunctioncalc_sim
(x, y, name="tensor")
examples/gdn/test_utils.py:8
↓ 1 callersFunctioncalc_sim
(x, y, name="tensor")
examples/kda/test_utils_kda.py:10
↓ 1 callersMethodcalculate
Calculate performance metrics based on the analysis. Returns: AnalysisResult: The calculated performance metrics.
tilelang/tools/Analyzer.py:164
↓ 1 callersFunctioncalculate_optimal_block_size
Calculate optimal block size - this function is completely useless
examples/autodd/tilelang_buggy.py:21
↓ 1 callersFunctioncalculate_tensor_similarity
Calculate similarity between two tensors using a normalized dot product metric. Unlike torch.testing.assert_close which uses absolute/relati
examples/deepseek_v32/utils.py:255
↓ 1 callersFunctioncalculate_tensor_similarity
Calculate similarity between two tensors using a normalized dot product metric. Unlike torch.testing.assert_close which uses absolute/relati
examples/dsa_sparse_finetune/utils.py:18
↓ 1 callersFunctioncalculate_tensor_similarity
Calculate similarity between two tensors using a normalized dot product metric. Unlike torch.testing.assert_close which uses absolute/relati
examples/dsa_hisa/tilelang_utils.py:156
↓ 1 callersFunctioncall_extern_access_ptr_mask_legalize
(access_type: str)
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:367
↓ 1 callersFunctioncall_extern_multiple_access_ptrs_legalize
()
testing/python/transform/test_tilelang_transform_legalize_safe_memory_access.py:396
↓ 1 callersFunctioncalulate_loss
(model, input, loss_fct)
examples/bitnet-1.58b/eval_ppl.py:23
↓ 1 callersFunctioncast_rs_kernel
(M_val: int)
testing/python/language/test_tilelang_cast_rounding.py:102
↓ 1 callersFunctionceil_div
Perform ceiling division of two integers. Args: x: the dividend. y: the divisor. Returns: The result of the cei
examples/cast/example_per_token_cast_to_fp8.py:44
↓ 1 callersFunctionceil_div_int
(x, y)
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:454
↓ 1 callersFunctionceil_div_int
(x, y)
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:544
↓ 1 callersFunctionceil_to_ue8m0
(x: torch.Tensor)
examples/deepseek_v32/utils.py:185
↓ 1 callersFunctionceil_to_ue8m0
(x)
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:460
↓ 1 callersFunctionceil_to_ue8m0
(x)
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:550
↓ 1 callersFunctionceil_to_ue8m0
(x: torch.Tensor)
examples/dsa_hisa/tilelang_utils.py:128
↓ 1 callersFunctionchain_equal
(N, block_size, dtype=T.float32)
testing/python/language/test_tilelang_language_chain_equal.py:10
↓ 1 callersFunctionchained_gemm
Chained GEMM: SS → tcgen05.ld → cast → tcgen05.st → MMA TS → output.
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:56
↓ 1 callersFunctioncheck_correctness
(M=256, N=8192, K=1024, seed=0)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:357
↓ 1 callersFunctioncheck_hopper
()
examples/convolution/example_convolution_autotune.py:8
↓ 1 callersFunctioncheck_hopper
()
testing/python/jit/test_tilelang_jit_tvm_ffi.py:311
↓ 1 callersFunctioncheck_hopper
()
testing/python/jit/test_tilelang_jit_nvrtc.py:313
↓ 1 callersFunctioncheck_int4_gemm_correctness
( M=1024, N=1024, K=1024, block_M=128, block_N=128, block_K=64, threads=128, )
examples/gemm_int4/example_tilelang_gemm_int4.py:70
↓ 1 callersFunctioncheck_metal_availability
()
tilelang/utils/target.py:144
↓ 1 callersFunctioncheck_nvrtc_available
Check if NVRTC backend is available. Raises ------ ImportError If cuda-python is not installed or cannot be imported
tilelang/jit/adapter/nvrtc/__init__.py:29
↓ 1 callersFunctioncheck_pytorch_version
(version_s: str = "2.4")
examples/kda/FLA_KDA/fla_utils.py:190
↓ 1 callersFunctioncheck_sm_version
(arch: str)
tilelang/carver/arch/cuda.py:8
↓ 1 callersMethodcheck_tile_shape_isvalid
Checks if the tile shapes in the TileDict are valid for the nodes in this context. Parameters: - td (TileDict): The TileDict
tilelang/carver/roller/policy/default.py:576
↓ 1 callersFunctionchunk_bwd_dv_local
( q: torch.Tensor, k: torch.Tensor, do: torch.Tensor, g: torch.Tensor = None, g_gamma: tor
examples/kda/FLA_KDA/fla_chunk_o.py:415
↓ 1 callersFunctionchunk_gla_bwd_dA
( v: torch.Tensor, do: torch.Tensor, scale: float, cu_seqlens: torch.LongTensor = None, ch
examples/kda/FLA_KDA/fla_chunk_o.py:515
↓ 1 callersFunctionchunk_gla_fwd_o_gk
( q: torch.Tensor, v: torch.Tensor, g: torch.Tensor, A: torch.Tensor, h: torch.Tensor,
examples/kda/FLA_KDA/fla_chunk_o.py:299
↓ 1 callersFunctionchunk_global_cumsum_scalar
( s: torch.Tensor, reverse: bool = False, cu_seqlens: torch.Tensor = None, scale: float = None
examples/kda/FLA_KDA/cumsum.py:329
↓ 1 callersFunctionchunk_global_cumsum_vector
( s: torch.Tensor, reverse: bool = False, cu_seqlens: torch.Tensor = None, scale: float = None
examples/kda/FLA_KDA/cumsum.py:360
↓ 1 callersFunctionchunk_kda_bwd_dqkwg
( q: torch.Tensor, k: torch.Tensor, w: torch.Tensor, v: torch.Tensor, h: torch.Tensor,
examples/kda/FLA_KDA/fla_chunk_inter.py:141
↓ 1 callersFunctionchunk_kda_bwd_intra
( q: torch.Tensor, k: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, dAqk: torch.Tenso
examples/kda/FLA_KDA/fla_chunk_intra.py:541
↓ 1 callersFunctionchunk_kda_fwd_inter_solve_fused
( q, k, gk, beta, Aqk, Akk_diag, Akk, scale, cu_seqlens: torch.LongTensor
examples/kda/FLA_KDA/fla_chunk_intra.py:611
↓ 1 callersFunctionchunk_kda_fwd_intra_token_parallel
Token-parallel implementation: each token gets its own thread block. Supports both fixed-length and variable-length sequences. Reduces wa
examples/kda/FLA_KDA/fla_chunk_intra_token_parallel.py:114
↓ 1 callersFunctionchunk_local_cumsum_vector
( g: torch.Tensor, chunk_size: int, reverse: bool = False, scale: float = None, cu_seqlens
examples/kda/FLA_KDA/cumsum.py:283
↓ 1 callersFunctionchunk_retention_fwd_kernel
( B, S, H, DK, DV, dtype: T.dtype = T.float16, scale: float = None, )
examples/linear_attention/example_retention_fwd.py:10
↓ 1 callersFunctionchunk_scan_fwd
( batch, seqlen, chunk_size, ngroups, nheads, headdim, dstate, block_M=64,
benchmark/mamba2/benchmark_mamba_chunk_scan.py:190
↓ 1 callersFunctionchunk_scan_helion
(cb, x, dt, dA_cumsum, C, states, D)
benchmark/mamba2/benchmark_mamba_chunk_scan.py:77
↓ 1 callersFunctionchunk_scan_triton
(cb, x, dt, dA_cumsum, C, states, D)
benchmark/mamba2/benchmark_mamba_chunk_scan.py:72
↓ 1 callersFunctionclamp_value_range
( N, block_N, dtype, )
testing/python/language/test_tilelang_language_clamp.py:50
↓ 1 callersFunctionclamp_within_bounds
( N, block_N, dtype, min_val=None, max_val=None, )
testing/python/language/test_tilelang_language_clamp.py:5
↓ 1 callersFunctionclean_and_maintain_logits_
( Logits, CuSeqLenKS, CuSeqLenKE, threads: int = 512, block_K: int = 4096, )
examples/dsa_hisa/clean_and_maintain_logits.py:10
↓ 1 callersFunctionclean_empty_pass
(code: str)
tilelang/autodd.py:890
↓ 1 callersFunctionclean_logits_
( Logits, CuSeqLenKS, CuSeqLenKE, threads: int = 512, block_K: int = 4096, )
examples/deepseek_v32/fp8_lighting_indexer.py:176
↓ 1 callersFunctionclear_cache
Disabled helper that previously removed the entire kernel cache. Raises: RuntimeError: Always raised to warn users to clear the cach
tilelang/cache/__init__.py:186
↓ 1 callersMethodclear_values
Clear all tracked variable bindings.
tilelang/language/frame.py:66
↓ 1 callersFunctioncli_main
(argv: "Sequence[str] | None" = None)
tilelang/autodd.py:1390
↓ 1 callersFunctioncollect_buffer_accesses
Categorizes buffers accessed in the statement by their scope. Args: statement: The TIR statement to analyze
tilelang/transform/add_bufstore_wrapper.py:44
↓ 1 callersFunctioncollect_buffer_indices
Maps each buffer to its access indices. Args: statement: The TIR statement to analyze Returns:
tilelang/transform/add_bufstore_wrapper.py:71
↓ 1 callersFunctioncollect_fragment_accesses
Collect fragment accesses in the loop body. Args: statement: The TIR statement to analyze Returns: Tuple of buffer acce
tilelang/analysis/fragment_loop_checker.py:25
↓ 1 callersFunctioncollect_vars_used_in_prim_expr
Collect the variables used in the PrimExpr.
tilelang/carver/analysis.py:220
↓ 1 callersFunctioncompare
(op_id: int, x: torch.Tensor, y: Optional[torch.Tensor] = None)
maint/precision/compare_ops.py:359
↓ 1 callersFunctioncompare
(compute_version, min_version, mode)
tilelang/testing/__init__.py:145
↓ 1 callersFunctioncompare_ab
(baseline, target, b, s_q, cache_seqlens, h_q, h_kv, d, dv, causal, dtype)
examples/deepseek_mla/benchmark_mla.py:457
← previousnext →1,901–2,000 of 7,254, ranked by callers