MCPcopy Create free account

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

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

↓ 2 callersFunctionpreprocess
( O, dO, block_ND=32, num_stages=5, dtype=T.bfloat16, accum_dtype=T.float32, )
examples/deepseek_v32/sparse_mla_bwd.py:9
↓ 2 callersFunctionprint_red_warning
(msg)
examples/flash_decoding/example_gqa_decode.py:377
↓ 2 callersMethodprint_with_clip
(self, s: str)
tilelang/analysis/ast_printer.py:27
↓ 2 callersFunctionprobe
(A, tmp: bool)
testing/python/language/test_tilelang_language_frontend_v2.py:457
↓ 2 callersMethodpropagate_inputs_on_reduction
(self, tile, rstep: dict | None = None)
tilelang/carver/roller/node.py:358
↓ 2 callersFunctionquantize_fp8_with_packed_ue8m0
DeepGEMM-style per-token FP8 quantization with UE8M0 scale factors. Returns: x_fp8: [MN, K] in float8_e4m3fn sf_packed_u32: flatt
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:535
↓ 2 callersFunctionquantize_fp8_with_packed_ue8m0_rows
(x, gran_k=128)
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:453
↓ 2 callersFunctionquantize_mxfp4_with_packed_ue8m0
(x: torch.Tensor, gran_k: int = 128)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:268
↓ 2 callersFunctionquantize_mxfp8_with_packed_ue8m0
(x: torch.Tensor, gran_k: int = 128)
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:289
↓ 2 callersFunctionrecompute_w_u_fwd
( k: torch.Tensor, v: torch.Tensor, beta: torch.Tensor, A: torch.Tensor, q: torch.Tensor =
examples/kda/FLA_KDA/fla_wy_fast.py:210
↓ 2 callersFunctionref_clean_and_maintain_logits
Pure torch equivalent. Returns a new tensor (doesn't mutate the input).
examples/dsa_hisa/clean_and_maintain_logits.py:50
↓ 2 callersFunctionref_program
(Q, K, V, is_causal, groups=1)
examples/amd/example_amd_flash_attn_bwd.py:23
↓ 2 callersFunctionref_program
( query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, sinks: torch.Tensor, slidin
examples/attention_sink/example_mha_sink_bwd_bhsd.py:384
↓ 2 callersFunctionref_program
(x: torch.Tensor)
examples/cast/example_per_token_cast_to_fp8.py:58
↓ 2 callersFunctionref_program
(logits, top_k)
examples/topk/example_topk.py:53
↓ 2 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_tvm_ffi.py:145
↓ 2 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_nvrtc.py:145
↓ 2 callersFunctionref_program_fa
(query, key, value, block_indices, cache_seqlens, max_cache_seqlen, num_blocks, block_size)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_indice.py:301
↓ 2 callersFunctionref_program_fa
(query, key, value, block_indices, cache_seqlens, max_cache_seqlen, num_blocks, block_size)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_mask.py:292
↓ 2 callersFunctionref_sparse_mla_fwd_interface
(q, kv, indices, sm_scale=None, is_casual=True)
examples/deepseek_v32/sparse_mla_fwd.py:193
↓ 2 callersFunctionref_sparse_mla_fwd_interface
(Q, KV, Indices, offsets, sm_scale=None, is_casual=True)
examples/dsa_sparse_finetune/sparse_mla_fwd.py:214
↓ 2 callersFunctionregister_c_postproc
Register a post-processing function for C host code generation. This callback intercepts C host code emitted by TileLang just before it is wr
tilelang/engine/callback.py:29
↓ 2 callersFunctionregister_cuda_postproc
Register a post-processing function for CUDA code generation. Args: func: A callable that takes generated code (str) and target (Target)
tilelang/engine/callback.py:7
↓ 2 callersFunctionregister_gemm_sp_impl
Register a backend-specific GEMM_SP Python implementation class.
tilelang/tileop/gemm_sp/registry.py:23
↓ 2 callersFunctionregister_hip_postproc
Register a post-processing function for HIP code generation. Args: func: A callable that takes generated code (str) and target (Target) a
tilelang/engine/callback.py:18
↓ 2 callersFunctionregister_metal_postproc
Register a post-processing function for Metal code generation. Args: func: A callable that takes generated code (str) and target (Target)
tilelang/engine/callback.py:44
↓ 2 callersFunctionrepeat_kv
This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he
examples/flash_decoding/example_gqa_decode_varlen_logits.py:11
↓ 2 callersFunctionreshape
(w)
tilelang/quantize/utils.py:34
↓ 2 callersFunctionresolve_pipeline
Resolve the lowering pipeline from a TVM target.
tilelang/backend/pass_pipeline/pipeline.py:46
↓ 2 callersMethodret
(self, value: Any)
tilelang/language/eager/ast.py:229
↓ 2 callersFunctionretrieve_buffer_and_offset
Retrieve the underlying buffer together with its logical element offset. - Buffer -> (buffer, 0) - BufferRegion -> (buffer, offset from
tilelang/utils/language.py:364
↓ 2 callersFunctionretrieve_func_from_module
(ir_module: IRModule)
tilelang/carver/utils.py:86
↓ 2 callersFunctionrocm_warp_size_for_arch
(arch: str | None)
tilelang/utils/target.py:48
↓ 2 callersFunctionrotate_activation
(x: torch.Tensor)
examples/deepseek_v32/inference/model.py:439
↓ 2 callersFunctionrotate_half
Rotates half the hidden dims of the input.
examples/bitnet-1.58b/modeling_bitnet.py:160
↓ 2 callersFunctionround
Round elements of the array to the nearest integer. Parameters ---------- x : PrimExpr Input argument. rounding_mode : str
tilelang/language/tir/op.py:2793
↓ 2 callersFunctionrun_atomic_addx2
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_atomic.py:75
↓ 2 callersFunctionrun_autotune
(M, N, K, M_value=None, N_value=None, K_value=None)
testing/python/autotune/test_tilelang_autotune_with_inputs.py:103
↓ 2 callersFunctionrun_blocksparse_copy
Run blocksparse copy test with given parameters.
testing/python/language/test_tilelang_language_let_layout.py:67
↓ 2 callersFunctionrun_clamp
( N, block_N, dtype, min=None, max=None, )
testing/python/language/test_tilelang_language_clamp.py:29
↓ 2 callersFunctionrun_clamp_value_range
( N, block_N, dtype, )
testing/python/language/test_tilelang_language_clamp.py:79
↓ 2 callersFunctionrun_cmd
(cmd, env=None)
maint/scripts/test_perf_regression.py:43
↓ 2 callersFunctionrun_cython_dynamic_shape
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M, block_N, block_K, num_stages=3, num
testing/python/jit/test_tilelang_jit_gemm_cython.py:259
↓ 2 callersFunctionrun_fmha_recommend_hints
( batch_size: int = 4, num_heads: int = 32, seq_length: int = 512, seq_kv_length: int = 512,
testing/python/carver/test_tilelang_carver_recommend_hints.py:105
↓ 2 callersFunctionrun_fp8_fp4_gemm_1d1d
( a_fp4, b_fp8, sfa, sfb, out_dtype=T.bfloat16, )
examples/deepseek_v4/fp8_fp4_gemm_1d1d_sm100.py:339
↓ 2 callersFunctionrun_gemm
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
maint/gemm/correctness_evaluation_sm70.py:93
↓ 2 callersFunctionrun_gemm
( M, N, K, in_dtype, out_dtype, dtypeAccum, block_M, block_N, block_K,
examples/dequantize_gemm/example_dequant_gemm_fine_grained.py:93
↓ 2 callersFunctionrun_gemm
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
testing/python/components/test_tilelang_pass_config_disable_warp_specialized.py:52
↓ 2 callersFunctionrun_gemm_pipeline_test
(N, block_M=128, block_N=128, block_K=32)
testing/python/issue/test_tilelang_issue_96.py:38
↓ 2 callersFunctionrun_gemm_rs
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
maint/gemm/correctness_evaluation_sm70.py:176
↓ 2 callersFunctionrun_gemm_ss
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py:64
↓ 2 callersFunctionrun_gemm_tma_copy
(num_stages)
testing/python/language/test_tilelang_language_tma_copy.py:65
↓ 2 callersFunctionrun_gemm_tma_copy_store
(num_stages)
testing/python/language/test_tilelang_language_tma_copy.py:227
↓ 2 callersFunctionrun_gemm_tma_store
(num_stages)
testing/python/language/test_tilelang_language_tma_store.py:59
↓ 2 callersFunctionrun_get_lane_id
(num_threads: int = 128, warp_size: Optional[int] = None)
testing/python/language/test_tilelang_language_get_warp_info.py:84
↓ 2 callersFunctionrun_get_warp_group_idx
( num_threads: int = 128, warp_size: Optional[int] = None, warps_per_group: Optional[int] = None,
testing/python/language/test_tilelang_language_get_warp_info.py:117
↓ 2 callersFunctionrun_get_warp_idx
(num_threads: int = 128, warp_size: Optional[int] = None)
testing/python/language/test_tilelang_language_get_warp_info.py:106
↓ 2 callersFunctionrun_get_warp_idx_sync
(num_threads: int = 128, warp_size: Optional[int] = None)
testing/python/language/test_tilelang_language_get_warp_info.py:95
↓ 2 callersFunctionrun_grouped_mxfp8_blockscaled_gemm
( a, b, sfa_flat, sfb_flat, offsets, max_M_per_E, transpose_B=True, persistent
examples/blockscaled_gemm_sm100/grouped_gemm_mxfp8_blockscaled_1d1d.py:536
↓ 2 callersFunctionrun_reduce_after_reshape
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:225
↓ 2 callersFunctionrun_reshape
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:21
↓ 2 callersFunctionrun_reshape_fragment
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:142
↓ 2 callersFunctionrun_reshape_layout_transform_shared
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:185
↓ 2 callersFunctionrun_reshape_smem_1d_2_2d
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:61
↓ 2 callersFunctionrun_reshape_smem_2d_2_1d
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:100
↓ 2 callersFunctionrun_shuffle_elect
(num_threads: int = 128, thread_extent: int = 64)
testing/python/language/test_tilelang_language_get_warp_info.py:136
↓ 2 callersFunctionrun_single_arg_mathop_test
Test single-argument mathops. T.exp should generate expf (non-fastmath), T.__exp should generate __expf (fastmath)
testing/python/fastmath/test_mathops_fastmath.py:53
↓ 2 callersFunctionrun_single_arg_mathop_test
Test single-argument mathops. T.exp should generate expf (non-fastmath), T.__exp should generate __expf (fastmath)
testing/python/math/test_math_fast_math.py:53
↓ 2 callersFunctionrun_tilelang_copy_with_stride
(M=1024, N=1024, NN=2048, block_M=128, block_N=128, dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:61
↓ 2 callersFunctionrun_tilelang_grouped_gemm_ptr
( batch_sizes_list, K, N, block_M, block_N, block_K, num_stages=2, threads=128
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:119
↓ 2 callersFunctionset_cache_kernel
( S, D, pos_ty="int64", dtype=T.float32, )
testing/python/issue/test_tilelang_issue_1115.py:8
↓ 2 callersMethodset_dtype
(self, dtype: tvm.DataType, id=0)
tilelang/carver/roller/node.py:147
↓ 2 callersMethodset_dtype
(self, dtype: tvm.DataType, id=0)
tilelang/carver/roller/node.py:319
↓ 2 callersMethodset_inputs
(self, i: int, edge: Edge)
tilelang/carver/roller/node.py:136
↓ 2 callersMethodset_kernel_parameters
(self, k_parameters: tuple[tuple[Any, ...], tuple[tuple[str, Any], ...]], f_parameters: dict[str, Any])
tilelang/autotuner/tuner.py:412
↓ 2 callersFunctionset_max_nreg
Set the maximum number of registers to use. Detailed Documentation: https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-inst
tilelang/language/builtin.py:380
↓ 2 callersMethodset_mode
Set the JIT execution mode (internal use only).
tilelang/language/eager/builder.py:1232
↓ 2 callersFunctionshared_symbolic_strided_kernel
Kernel with shared symbolic variable in strides.
testing/python/language/test_tilelang_language_subtype.py:143
↓ 2 callersMethodshould_cleanup_temp_files
(self)
tilelang/env.py:373
↓ 2 callersFunctionshuffle_weight
( x: torch.Tensor, layout=(16, 32), k_pack=1, is_transpose=False, )
testing/python/amd/test_tilelang_gemm_mfma_preshuffle.py:251
↓ 2 callersFunctionside_effect
(expr: PrimExpr)
tilelang/utils/language.py:585
↓ 2 callersMethodsimdgroup_copy
Copy between simdgroup local buffers and shared/global memory.
tilelang/metal/intrinsics/metal_macro_generator.py:186
↓ 2 callersFunctionsoftmax_kernel
(X, BLOCK_M=1, BLOCK_N=8192, dtype: T.dtype = T.float16)
examples/online_softmax/online_softmax.py:8
↓ 2 callersFunctionsparse_attn_fwd
( batch, heads, seq_len, seq_len_kv, dim, topk, sm_scale=None, block_N=64,
examples/deepseek_v4/sparse_attn_fwd_sm90.py:27
↓ 2 callersFunctionsparse_mla_bwd
(q, kv, o, do, indices, lse, sm_scale=None, is_casual=True, return_kernel=False, delta=None)
examples/deepseek_v32/sparse_mla_bwd.py:244
↓ 2 callersFunctionsparse_mla_fwd
( Q, KV, Indices, heads, dim, tail_dim, topk, kv_group=1, sm_scale=None,
examples/deepseek_v32/sparse_mla_fwd.py:15
↓ 2 callersFunctionsparse_mla_fwd
( Q, KV, Indices, q_start_index_s, heads, dim, tail_dim, topk, kv_stride,
examples/deepseek_v32/sparse_mla_fwd_pipelined.py:13
↓ 2 callersFunctionsparse_mla_topk_reducesum_interface
( q: torch.Tensor, kv: torch.Tensor, topk_indices: torch.Tensor, lse: torch.Tensor, offset
examples/dsa_sparse_finetune/sparse_mla_topk_reducesum.py:151
↓ 2 callersMethodstart_proc
(self)
tilelang/autodd.py:778
↓ 2 callersMethodstart_workers
(self)
tilelang/autodd.py:1209
↓ 2 callersMethodstmatrix
(self, C_local_buf, C_buf, pid_m=None, pid_n=None)
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:613
↓ 2 callersMethodsubroutine
(A_data: T.handle("float32"))
testing/python/transform/test_tilelang_transform_make_packed_api.py:113
↓ 2 callersFunctionsum_all_diagonal_matrix
(mat: torch.tensor)
examples/minference/example_vertical_slash_sparse_attn.py:552
↓ 2 callersMethodswizzle_atom_size
(self)
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:81
↓ 2 callersFunctionsymbolic_last_dim_kernel
Kernel with symbolic variable in the last dimension.
testing/python/language/test_tilelang_language_subtype.py:111
↓ 2 callersFunctionsymbolic_last_dim_strided_kernel
Kernel with symbolic variable in both shape and stride of last dimension.
testing/python/language/test_tilelang_language_subtype.py:121
↓ 2 callersFunctiontarget_get_rdna_generation
(target: Target)
tilelang/utils/target.py:383
↓ 2 callersFunctiontarget_is_cdna
(target: Target)
tilelang/utils/target.py:351
↓ 2 callersFunctiontarget_is_hip
(target: Target)
tilelang/utils/target.py:323
↓ 2 callersFunctiontarget_is_rdna
(target: Target)
tilelang/utils/target.py:355
← previousnext →1,301–1,400 of 7,254, ranked by callers