MCPcopy Create free account

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

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

↓ 1 callersFunctiontest_sparse_mla_fwd_pipelined
( B=1, S=4096, SKV=8192, H=128, HKV=1, DQK=576, DV=512, topk=2048, dtype=torch.bfloat16, q_start_s_index=1
examples/deepseek_v32/sparse_mla_fwd_pipelined.py:421
↓ 1 callersFunctiontest_ss_gemm_bf16_baseline
()
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:177
↓ 1 callersFunctiontest_sta_attention
()
testing/python/transform/test_tilelang_transform_config_index_bitwidth.py:149
↓ 1 callersFunctiontest_storage_rewrite_preserves_scope
lexical_alloc_scope should survive StorageRewrite without crashing.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:312
↓ 1 callersFunctiontest_tiled_ws_correctness
End-to-end correctness test: pipelined GEMM via tiled WS.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:386
↓ 1 callersFunctiontest_tiled_ws_does_not_clone_local_var_into_producer_branch
Producer branch should reuse shared local.var state instead of cloning it.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:620
↓ 1 callersFunctiontest_tiled_ws_explicit_cp_async_wait_precedes_first_consumer_read
Explicit cp.async destinations must pull the consumer wait earlier.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:544
↓ 1 callersFunctiontest_tiled_ws_incompatible_layout_blocks_ws
A non-swizzle, non-linear layout on ALL TMA copy targets should block WS. If every copy that could be a TMA producer has an incompatible layout,
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:488
↓ 1 callersFunctiontest_tiled_ws_keeps_shared_prelude_local_vars_for_grouped_gemm
Shared-prelude grouped-gemm indices must stay outside WS branches.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:604
↓ 1 callersFunctiontest_tiled_ws_places_producer_in_first_warp_group
Auto-WS should put the producer in the low threadIdx.x partition.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:310
↓ 1 callersFunctiontest_tiled_ws_sinks_preloop_tma_waits_into_consumer
Pre-loop TMA loads should not emit immediate waits in the common prelude.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:528
↓ 1 callersFunctiontest_tiled_ws_stage1_dynamic_loop_start
Stage-1 tiled WS should handle dynamic pipeline loop bounds.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:342
↓ 1 callersFunctiontest_tiled_ws_stage3
Pipelined GEMM with 3 stages.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:405
↓ 1 callersFunctiontest_tiled_ws_swizzled_layout_allows_ws
Swizzled layout on a TMA copy target should NOT block warp specialization. Swizzled layouts are valid TMA layouts (TMA supports 32B/64B/128B swiz
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:438
↓ 1 callersFunctiontest_tilelang_copy_mask_copy_range
()
testing/python/language/test_tilelang_language_mask_op.py:140
↓ 1 callersFunctiontest_tma_descriptor_init_after_alloc_global
()
testing/python/transform/test_tilelang_transform_lower_hopper_intrin.py:57
↓ 1 callersFunctiontest_tma_store_2_stages
()
testing/python/language/test_tilelang_language_tma_store.py:190
↓ 1 callersFunctiontest_topk_selector
(batch=64, seq_len=32 * 1024, topk=2048)
examples/deepseek_v32/topk_selector.py:183
↓ 1 callersFunctiontest_topk_sparse_attention
()
examples/blocksparse_attention/example_tilelang_block_sparse_attn.py:129
↓ 1 callersFunctiontest_topk_sparse_attention
()
examples/blocksparse_attention/block_sparse_attn_triton.py:254
↓ 1 callersFunctiontest_topk_sparse_attention
()
examples/seer_attention/block_sparse_attn_tilelang.py:132
↓ 1 callersFunctiontest_topk_sparse_attention
()
examples/seer_attention/block_sparse_attn_triton.py:249
↓ 1 callersFunctiontest_topk_sparse_attention_qlen_lt_klen
()
examples/seer_attention/block_sparse_attn_tilelang.py:177
↓ 1 callersFunctiontest_topk_sparse_attention_qlt_kl
()
examples/blocksparse_attention/block_sparse_attn_triton.py:302
↓ 1 callersFunctiontest_topk_sparse_attention_qlt_kl
()
examples/seer_attention/block_sparse_attn_triton.py:297
↓ 1 callersFunctiontest_unroll_decl_buffer_defs_are_fresh
()
testing/python/transform/test_tilelang_transform_unroll_loop.py:7
↓ 1 callersFunctiontest_varlen_decode_main
Test decode kernel with variable sequence lengths.
examples/flash_decoding/example_gqa_decode_varlen_logits.py:169
↓ 1 callersFunctionthread_id_shared_access_64x16_to_32x32_layout_C_m_n
Return (m, n) = (row, col) for the 32x32 MFMA output register layout. For v_mfma_i32_32x32x32_i8 (gfx950), each wave-64 lane holds 16 output
tilelang/rocm/intrinsics/mfma_layout.py:148
↓ 1 callersFunctionthread_id_shared_access_64x4_to_16x16_layout_C_n_m
(thread_id, local_id)
tilelang/rocm/intrinsics/mfma_layout.py:77
↓ 1 callersFunctiontile_atomic_add_expr_program
(M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:346
↓ 1 callersFunctiontile_atomic_add_program
(K, M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:321
↓ 1 callersFunctiontile_atomic_add_scalar_program
(dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:372
↓ 1 callersFunctiontile_atomic_max_expr_program
(M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:587
↓ 1 callersFunctiontile_atomic_max_program
(K, M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:525
↓ 1 callersFunctiontile_atomic_min_program
(K, M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:556
↓ 1 callersFunctiontilelang_chunk_bwd_intra
( # task config B, S, H, DK, input_dtype, output_dtype, accum_dtype, gate_
examples/kda/chunk_bwd_intra.py:81
↓ 1 callersFunctiontilelang_chunk_bwd_kernel_dv_local
( B, S, H, DV, scale, input_dtype, da_dtype, do_dtype, chunk_size, blo
examples/kda/chunk_bwd_gla_dA.py:52
↓ 1 callersFunctiontilelang_chunk_bwd_kernel_dv_local
( B, S, H, DV, input_dtype, output_dtype, do_dtype, chunk_size, block_DV=1
examples/kda/chunk_bwd_dv.py:56
↓ 1 callersFunctiontilelang_chunk_fwd_o
( # task config B, S, H, DK, DV, input_dtype, output_dtype, accum_dtype,
examples/kda/chunk_o.py:62
↓ 1 callersFunctiontilelang_chunk_gated_delta_rule_bwd_dhu
( # task config B, S, H, DK, DV, input_dtype, output_dtype, accum_dtype,
examples/kda/chunk_delta_bwd.py:78
↓ 1 callersFunctiontilelang_chunk_gated_delta_rule_fwd_h
( # task config B, S, H, DK, DV, input_dtype, output_dtype, accum_dtype,
examples/kda/chunk_delta_h_fwd.py:80
↓ 1 callersFunctiontilelang_chunk_kda_fwd_inter_fused
( B, S, H, DK, input_dtype, output_dtype, accum_dtype, gate_dtype, chunk_s
examples/kda/chunk_inter_solve_fused.py:53
↓ 1 callersFunctiontilelang_chunk_kda_fwd_intra_token_parallel
( B, S, H, DK, input_dtype, output_dtype, accum_dtype, gate_dtype, chunk_s
examples/kda/chunk_intra_token_parallel.py:60
↓ 1 callersFunctiontilelang_composable_copy
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_composable_index.py:9
↓ 1 callersFunctiontilelang_copy
(M, N, block_M, block_N, dtype=T.float16, pad_value=0)
testing/python/language/test_tilelang_language_annotate_safe_value.py:9
↓ 1 callersFunctiontilelang_copy_buffer_load_with_parallel
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:110
↓ 1 callersFunctiontilelang_copy_bufferload
(num_tokens, dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:82
↓ 1 callersFunctiontilelang_copy_mask_copy
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_mask_op.py:43
↓ 1 callersFunctiontilelang_copy_mask_copy_range
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_mask_op.py:112
↓ 1 callersFunctiontilelang_copy_mask_parallel
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_mask_op.py:8
↓ 1 callersFunctiontilelang_copy_mask_parallel_range
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_mask_op.py:77
↓ 1 callersFunctiontilelang_copy_shape_mismatched
(M, N, src_dtype=T.float16, dst_dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:140
↓ 1 callersFunctiontilelang_copy_with_stride
(M, N, NN, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:47
↓ 1 callersFunctiontilelang_if_range
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_if_range.py:10
↓ 1 callersFunctiontilelang_kernel_block_mask
( BlockIndices, BlockCounts, block_size, dtype=T.int32, )
examples/deepseek_nsa/example_tilelang_nsa_bwd.py:517
↓ 1 callersFunctiontilelang_kernel_bwd_dqkv
( Q, K, V, LSE_slc, Delta_slc, DO_slc, DQ, DK, DV, BlockMask, dim,
examples/deepseek_nsa/example_tilelang_nsa_bwd.py:319
↓ 1 callersFunctiontilelang_kernel_fwd
( Q, K, V, BlockIndices, dim, is_causal, scale=None, block_size=64, groups
examples/deepseek_nsa/example_tilelang_nsa_bwd.py:23
↓ 1 callersFunctiontilelang_kernel_preprocess
( O, dO, dtype=T.float16, accum_dtype=T.float32, blk=32, )
examples/deepseek_nsa/example_tilelang_nsa_bwd.py:481
↓ 1 callersFunctiontilelang_rand_1d
(M=1024, seed=42)
examples/rand/rand_uint.py:9
↓ 1 callersFunctiontilelang_rand_1d
(M=1024, seed=42, generator="curandStatePhilox4_32_10_t")
testing/python/language/test_tilelang_language_rand.py:9
↓ 1 callersFunctiontilelang_recompute_w_u_fwd
( # task config B, S, H, DK, DV, input_dtype, output_dtype, gate_dtype,
examples/kda/wy_fast.py:55
↓ 1 callersFunctiontilelang_ternary
(M, N, block_M, block_N, dtype=T.float16)
testing/python/language/test_tilelang_language_ternary.py:10
↓ 1 callersFunctiontilelang_transpose
Kernel: read tile from A into shared, transpose in shared, write to B. A is (M, N), B is (M, N). B = A.T.T = A when block_M == M and block_N
testing/python/language/test_tilelang_language_transpose.py:9
↓ 1 callersFunctiontilelang_transpose_square
Simpler test: square transpose with single tile.
testing/python/language/test_tilelang_language_transpose.py:59
↓ 1 callersFunctiontilelang_wy_fast_bwd
( # task config B, S, H, DK, DV, input_dtype, output_dtype, accum_dtype,
examples/kda/wy_fast_bwd.py:81
↓ 1 callersFunctiontir_op_with_parallel
(length=256, block=16, dtype=T.float32)
testing/python/analysis/test_tilelang_nested_loop_checker.py:637
↓ 1 callersFunctiontl_indexer_bwd_impl
( IndexQ, Weights, IndexK, AttnScore, IndexScore, TopkIndices, Offsets, TokenI
examples/dsa_sparse_finetune/indexer_bwd.py:20
↓ 1 callersFunctiontl_indexer_topk_reducesum_impl
( heads: int, dim: int, topk: int, sm_scale: Optional[float] = None, block_K: int = 32,
examples/dsa_sparse_finetune/indexer_topk_reducesum.py:21
↓ 1 callersFunctiontl_matmul
( M, N, K, in_dtype, out_dtype, accum_dtype, )
examples/bitnet-1.58b/kernel_benchmark/tl_int8xint8.py:32
↓ 1 callersFunctiontl_matmul
( M, N, K, block_M, block_N, block_K, num_stages, k_pack=2, num_threads=25
examples/gemm_fp8/example_tilelang_gemm_amd_fp8_preshuffle.py:39
↓ 1 callersFunctiontl_matmul
Build a TileLang MFMA kernel for ``A @ B^T`` (with optional preshuffle). The (block_*_warps, warp_*_tiles, chunk, num_stages, panel_size) paramet
testing/python/amd/test_tilelang_gemm_mfma_preshuffle.py:16
↓ 1 callersFunctiontl_matmul
( M, N, K, in_dtype, out_dtype, accum_dtype, a_transposed=False, b_transposed=
testing/python/amd/test_tilelang_gemm_mfma_intrinsic.py:17
↓ 1 callersFunctiontl_matmul_simt
( M, N, K, in_dtype, out_dtype, accum_dtype, )
testing/python/kernel/test_tilelang_kernel_gemm_simt.py:29
↓ 1 callersFunctiontl_matmul_with_ladder_weight_only_transform_block_reduce_int4
( M, N, K, in_dtype, out_dtype, accum_dtype, transform_b, )
examples/dequantize_gemm/example_dequant_gemm_fine_grained.py:137
↓ 1 callersFunctiontl_sparse_mla_topk_reducesum_impl
( Q, KV, Indices, Lse, Offsets, TokenIndices, heads, dim, tail_dim, to
examples/dsa_sparse_finetune/sparse_mla_topk_reducesum.py:19
↓ 1 callersFunctiontl_tanh
Triton tanh implementation using sigmoid.
maint/precision/compare_ops.py:112
↓ 1 callersFunctiontl_topk_impl
(input, index, starts, ends, in_dtype=T.float32, out_dtype=T.int32)
examples/deepseek_v32/topk_selector.py:28
↓ 1 callersFunctiontma_atomic_add_program
(out, explicit_swizzle=False)
testing/python/language/test_tilelang_language_atomic.py:189
↓ 1 callersMethodto_dict
(self)
tilelang/carver/roller/hint.py:193
↓ 1 callersMethodtop
Return the item at the top of the stack without removing it. Returns: The item at the top of the stack. Raises:
tilelang/autotuner/capture.py:41
↓ 1 callersFunctiontopo_sort_get_layer
(node, topo_layer)
tilelang/carver/roller/node.py:548
↓ 1 callersFunctiontorch_convert
(tensor)
examples/dequantize_gemm/example_dequant_gemm_w4a8.py:68
↓ 1 callersFunctiontorch_gmm
Perform grouped matrix multiplication using PyTorch. Args: a (torch.Tensor): Input tensor of shape (N, K). b (torch.Tensor):
examples/grouped_gemm/example_grouped_gemm_fwd.py:8
↓ 1 callersFunctiontorch_grouped_gemm_ptr
(a_list, b_list)
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:17
↓ 1 callersFunctiontorch_sparse_attention
Reference: gather KV by indices, compute attention with sink.
examples/deepseek_v4/sparse_attn_fwd_sm90.py:162
↓ 1 callersFunctiontransform
(x)
testing/python/language/test_tilelang_language_frontend_v2.py:410
↓ 1 callersMethodtraverse_dependencies
(self, compute)
tilelang/carver/roller/shape_inference/tir.py:109
↓ 1 callersFunctiontrigger_cpp_side_error
()
testing/python/language/test_tilelang_memory_leak.py:53
↓ 1 callersFunctiontriton_libdevice_op
LibDevice Triton operation interface.
maint/precision/compare_ops.py:298
↓ 1 callersFunctiontriton_op
Standard Triton operation interface.
maint/precision/compare_ops.py:283
↓ 1 callersFunctiontriton_program
(Q, K, V, Sinks, window_size: Optional[int] = None)
examples/attention_sink/benchmark_mha_sink_fwd.py:97
↓ 1 callersFunctiontvm_tuple
Create a tuple structure in value field of AttrStmt Parameters ---------- value : Expr The value in tuple. Returns -----
tilelang/language/tir/op.py:415
↓ 1 callersFunctiontwo_kernel_program
Create a program with two lowered host kernel call sites.
testing/python/jit/test_tilelang_jit_cutedsl_host_codegen.py:64
↓ 1 callersFunctiontwo_warp_reduce
()
testing/python/amd/test_tilelang_hip_codegen.py:137
↓ 1 callersFunctiontype_annotation
Create a type annotation expression Parameters ---------- dtype : Expr The data type. Returns ------- call : PrimExp
tilelang/language/tir/op.py:652
↓ 1 callersFunctionunroll
The unrolled For statement. Parameters ---------- start : PrimExpr The minimum value of iteration. stop : PrimExpr T
tilelang/language/loop.py:235
↓ 1 callersMethodupdate_host_func
Store generated Python launch code for later file write. Called by adapter after wrapper generates the launch code. This is the bridg
tilelang/jit/adapter/nvrtc/libgen.py:97
↓ 1 callersMethodupdate_host_func
(self, host_func: str)
tilelang/jit/adapter/cutedsl/libgen.py:37
↓ 1 callersMethodupdate_launcher_cpp_code
(self, launcher_cpp_code: str)
tilelang/jit/adapter/cutedsl/libgen.py:40
↓ 1 callersMethodupdate_launcher_lib_name
(self, launcher_lib_name: str)
tilelang/jit/adapter/cutedsl/libgen.py:43
← previousnext →3,201–3,300 of 7,254, ranked by callers