MCPcopy Create free account

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

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

↓ 1 callersFunctionprepare_output
( B, S, H, DV, chunk_size, d_type, )
examples/kda/chunk_bwd_gla_dA.py:27
↓ 1 callersFunctionprepare_output
( B, S, H, DV, chunk_size, output_dtype, )
examples/kda/chunk_bwd_dv.py:31
↓ 1 callersFunctionprepare_sequence_ids
(cu_seqlens: torch.LongTensor)
examples/dsa_sparse_finetune/index.py:75
↓ 1 callersFunctionprepare_sequence_ids
(cu_seqlens: torch.LongTensor)
examples/dsa_hisa/tilelang_utils.py:74
↓ 1 callersFunctionprepare_wy_repr_bwd
( k: torch.Tensor, v: torch.Tensor, beta: torch.Tensor, gk: torch.Tensor, A: torch.Tensor,
examples/kda/FLA_KDA/fla_wy_fast.py:257
↓ 1 callersFunctionpreprocess
( O, dO, block_ND=32, num_stages=5, dtype=T.bfloat16, accum_dtype=T.float32, )
examples/dsa_sparse_finetune/sparse_mla_bwd.py:11
↓ 1 callersFunctionprim_func
(func: Callable[_P, _T] = None, *, eager_jit: bool = False)
tilelang/language/eager/builder.py:1271
↓ 1 callersMethodprim_func_arg
(self, name, value)
tilelang/language/eager/builder.py:731
↓ 1 callersFunctionprint_fragment_buffer_with_condition
Conditionally prints the values of a flattened TIR buffer if the condition is True. Parameters: condition (tirx.PrimExpr): A TIR exp
tilelang/language/print_op.py:78
↓ 1 callersFunctionprint_fragment_format
Format fragment layout information into a human-readable string. Parameters ---------- layout : T.Fragment The fragment layo
tilelang/analysis/layout_visual.py:14
↓ 1 callersFunctionprint_global_buffer_with_condition
Conditionally prints the values of a flattened TIR buffer if the condition is True.
tilelang/language/print_op.py:46
↓ 1 callersFunctionprint_local_buffer_with_condition
Conditionally prints the values of a flattened TIR buffer if the condition is True. Parameters: condition (tirx.PrimExpr): A TIR exp
tilelang/language/print_op.py:110
↓ 1 callersFunctionprint_msg
Prints a message string.
tilelang/language/print_op.py:100
↓ 1 callersFunctionprint_shared_buffer_with_condition
Conditionally prints the values of a flattened TIR buffer if the condition is True. Parameters: condition (tirx.PrimExpr): A TIR exp
tilelang/language/print_op.py:58
↓ 1 callersFunctionprint_var
Prints the value of a TIR primitive expression (PrimExpr) for debugging purposes. Parameters: var (tirx.PrimExpr): The variable or e
tilelang/language/print_op.py:16
↓ 1 callersFunctionprofile
(model, input_data)
examples/bitnet-1.58b/benchmark_inference_latency.py:12
↓ 1 callersMethodpropagate_outputs
(self, tile, rstep: dict | None = None)
tilelang/carver/roller/node.py:379
↓ 1 callersFunctionptr
The pointer declaration function. Parameters ---------- dtype : str The data type of the pointer. storage_scope : str
tilelang/language/ast/ir.py:1530
↓ 1 callersFunctionptx_arrive_barrier
TVM intrinsic for ptx barrier arrival using mbarrier.arrive https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchroniza
tilelang/language/tir/op.py:1927
↓ 1 callersFunctionptx_arrive_barrier_expect_tx
TVM intrinsic for ptx barrier arrival with expect tx using mbarrier.arrive.expect_tx https://docs.nvidia.com/cuda/parallel-thread-execution/index.
tilelang/language/tir/op.py:1944
↓ 1 callersFunctionptx_arrive_cluster_barrier
Arrive at a shared barrier in cluster. Args: mbarrier: BarrierType The memory barrier to arrive at cta_id: int | Var
tilelang/language/builtin.py:436
↓ 1 callersMethodpush
Pushes an item onto the top of the stack.
tilelang/language/kernel.py:38
↓ 1 callersMethodpush
Push an item onto the stack and update variable mapping if applicable. Args: item: The frame object to push onto the stack
tilelang/language/frame.py:25
↓ 1 callersMethodpush
Push an item onto the top of the stack. Args: item: The item to be pushed onto the stack.
tilelang/autotuner/capture.py:20
↓ 1 callersFunctionqwq
()
testing/python/issue/test_tilelang_issue_1678.py:9
↓ 1 callersFunctionqwq
()
testing/python/issue/test_tilelang_issue_1601.py:9
↓ 1 callersFunctionqwq
()
testing/python/issue/test_tilelang_issue_1604.py:8
↓ 1 callersFunctionqwq
(dtype=torch.float8_e4m3fn)
testing/python/transform/test_tilelang_transform_hoist_broadcast_values.py:13
↓ 1 callersMethodrecommend_block_size
Recommends optimal block sizes based on the TileDict configuration. Parameters ---------- td : TileDict
tilelang/carver/roller/policy/default.py:599
↓ 1 callersFunctionreduce_after_reshape_test
(N, M, dtype)
testing/python/language/test_tilelang_language_reshape.py:205
↓ 1 callersFunctionreduce_kernel
()
testing/python/amd/test_tilelang_hip_codegen.py:100
↓ 1 callersFunctionreduce_macro
(buffer: tirx.Buffer, out: tirx.Buffer, reduce_type: str, dim: int, clear: bool)
tilelang/language/reduce_op.py:65
↓ 1 callersFunctionreduce_ref
(Q, K, V, mask, glse, Output_partial)
examples/flash_decoding/example_gqa_decode.py:356
↓ 1 callersMethodreduction_step_for_dtype_bits
(self, bits: int)
tilelang/carver/arch/rdna.py:29
↓ 1 callersFunctionref_attention
Compute reference attention output and weights. Args: q: [b, q_heads, head_size] k, v: [b, kv_heads, max_seqlen, head_size]
examples/flash_decoding/example_gqa_decode_varlen_logits.py:133
↓ 1 callersFunctionref_blocksparse_copy
Reference implementation for blocksparse copy.
testing/python/language/test_tilelang_language_let_layout.py:52
↓ 1 callersFunctionref_deepgemm_fp8
(A_fp8, B_fp8, A_scale, B_scale, out_dtype)
examples/deepseek_deepgemm/example_deepgemm_fp8_2xAcc.py:106
↓ 1 callersFunctionref_deepseek_sparse_attention
( q: torch.Tensor, kv: torch.Tensor, index_q: torch.Tensor, index_k: torch.Tensor, weights
examples/dsa_sparse_finetune/dsa.py:82
↓ 1 callersFunctionref_deepseek_sparse_attention_innner
( q: torch.Tensor, kv: torch.Tensor, index_q: torch.Tensor, index_k: torch.Tensor, weights
examples/dsa_sparse_finetune/dsa.py:28
↓ 1 callersFunctionref_fp8_block_mean_pooling
Spec: per-token dequant + per-block mean (dividing by actual valid count). Returns the f32 mean (caller can compare against fp8*scale re-quant of
examples/dsa_hisa/fp8_block_mean_pooling.py:85
↓ 1 callersFunctionref_fp8_block_sparse_mqa
( q_fp8: torch.Tensor, k_fp8: torch.Tensor, k_scale: torch.Tensor, topk_block_index: torch.Ten
examples/dsa_hisa/block_sparse_mqa_fp8.py:134
↓ 1 callersFunctionref_fp8_mqa_logits
(q: torch.Tensor, kv: torch.Tensor, weights: torch.Tensor, cu_seqlen_ks: torch.Tensor, cu_seqlen_ke: torch.Ten
examples/deepseek_v32/fp8_lighting_indexer.py:222
↓ 1 callersFunctionref_index_score
(Q: torch.Tensor, Weights: torch.Tensor, K: torch.Tensor, topk: int, offsets: torch.Tensor)
examples/dsa_sparse_finetune/indexer_topk_reducesum.py:208
↓ 1 callersFunctionref_indexer_bwd
( Q: torch.Tensor, Weights: torch.Tensor, K: torch.Tensor, TopkIndices: torch.Tensor, AttnScore: torch.Ten
examples/dsa_sparse_finetune/indexer_bwd.py:185
↓ 1 callersFunctionref_kernel
Reference implementation of DeepSeek-style Mixture of Experts using PyTorch. Args: data: Tuple of (input: torch.Tensor, weights: Dic
examples/fusedmoe/example_fusedmoe_torch.py:94
↓ 1 callersFunctionref_matmul
(A, B, trans_A, trans_B)
testing/python/cpu/test_tilelang_cpu_tgemm.py:47
↓ 1 callersFunctionref_mla
()
examples/deepseek_mla/example_mla_decode_paged.py:236
↓ 1 callersFunctionref_mla
()
examples/deepseek_mla/benchmark_mla.py:39
↓ 1 callersFunctionref_mla
()
examples/deepseek_mla/amd/benchmark_mla_decode_amd_aiter.py:44
↓ 1 callersFunctionref_mla
()
examples/deepseek_mla/amd/benchmark_mla_decode_amd_triton.py:35
↓ 1 callersFunctionref_mla_topk_softmax
(Q: torch.Tensor, K: torch.Tensor, TopkIndices: torch.Tensor, offsets: torch.Tensor)
examples/dsa_sparse_finetune/sparse_mla_topk_reducesum.py:174
↓ 1 callersFunctionref_per_token_cast_to_fp8
(x: torch.Tensor)
examples/cast/example_group_per_split_token_cast_to_fp8.py:127
↓ 1 callersFunctionref_pool_mqa_fp8
Spec: for each (m, n), logits[m, n] = sum_h ReLU(q[m,h] . k[n] * k_scale[n]) * w[m,h]. Computes the full dense [M, Nb] grid — caller is responsibl
examples/dsa_hisa/pool_mqa_fp8.py:139
↓ 1 callersFunctionref_program
(A, B, BlockMask, block_M, block_N, block_K)
examples/blocksparse_gemm/example_blocksparse_gemm.py:41
↓ 1 callersFunctionref_program
(A, B)
examples/dynamic_shape/example_dynamic.py:72
↓ 1 callersFunctionref_program
(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, scale: Optional[float] = None)
examples/linear_attention/example_linear_attn_bwd.py:133
↓ 1 callersFunctionref_program
(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, scale: Optional[float] = None)
examples/linear_attention/example_linear_attn_fwd.py:94
↓ 1 callersFunctionref_program
(stride, padding, dilation)
examples/convolution/example_convolution.py:8
↓ 1 callersFunctionref_program
(stride, padding, dilation)
examples/convolution/example_convolution_autotune.py:16
↓ 1 callersFunctionref_program
(a, b, batch_sizes)
examples/grouped_gemm/example_grouped_gemm_bwd.py:111
↓ 1 callersFunctionref_program
(query, key, value, mask)
examples/flash_decoding/example_gqa_decode.py:260
↓ 1 callersFunctionref_program
(Q, K, V, is_causal)
examples/flash_attention/example_mha_bwd_bhsd.py:287
↓ 1 callersFunctionref_program
(Q, K, V, is_causal, groups=1)
examples/flash_attention/example_gqa_bwd_tma_reduce.py:416
↓ 1 callersFunctionref_program
(Q, K, V, is_causal, groups=1)
examples/flash_attention/example_gqa_bwd.py:409
↓ 1 callersFunctionref_program
(Q, K, V, is_causal)
examples/flash_attention/example_mha_bwd_bshd.py:284
↓ 1 callersFunctionref_program
(Q, K, V, padding_mask, is_causal, groups=1)
examples/flash_attention/example_gqa_bwd_tma_reduce_varlen.py:593
↓ 1 callersFunctionref_program
Reference implementation for varlen attention with sinks.
examples/attention_sink/example_gqa_sink_bwd_varlen.py:592
↓ 1 callersFunctionref_program
Reference implementation for varlen attention with sinks.
examples/attention_sink/example_gqa_sink_fwd_varlen.py:212
↓ 1 callersFunctionref_program
( query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, sinks: torch.Tensor, slidin
examples/attention_sink/example_gqa_sink_bwd_bhsd.py:380
↓ 1 callersFunctionref_program
(x, y)
examples/elementwise/example_elementwise_add.py:7
↓ 1 callersFunctionref_program
(x: torch.Tensor)
examples/hadamard_transform/example_hadamard.py:124
↓ 1 callersFunctionref_program
(A, B)
examples/gemm_sm100/gemm_tcgen5mma_ws_clc.py:360
↓ 1 callersFunctionref_program
CPU reference computation to avoid cuBLAS issues on Blackwell.
examples/flash_attention_sm100/mha_fwd_bshd.py:396
↓ 1 callersFunctionref_program
CPU reference: K/V [b,s,head_kv,d], expand to heads for einsum.
examples/flash_attention_sm100/gqa_fwd_bshd.py:401
↓ 1 callersFunctionref_program
(q, q_pe, kv, k_pe, glse, Output_partial)
examples/deepseek_mla/amd/benchmark_mla_decode_amd_tilelang.py:205
↓ 1 callersFunctionref_program
(A, B)
examples/gemm_fp8/example_tilelang_gemm_amd.py:9
↓ 1 callersFunctionref_program
(x, y)
testing/python/language/test_tilelang_language_tma_1d.py:6
↓ 1 callersFunctionref_program
(A, B)
testing/python/language/test_tilelang_language_pipeline.py:93
↓ 1 callersFunctionref_program
(a, b)
testing/python/language/test_tilelang_language_ptr.py:117
↓ 1 callersFunctionref_program
(A)
testing/python/language/test_tilelang_language_clamp.py:41
↓ 1 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_cutedsl.py:206
↓ 1 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_gemm_cython.py:147
↓ 1 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_gemm.py:95
↓ 1 callersFunctionref_program
(A, B)
testing/python/jit/test_tilelang_jit_callback.py:212
↓ 1 callersFunctionref_program
A reference matrix multiplication program, used to compare performance. Parameters ---------- A : torch.Tensor The matrix wi
testing/python/autotune/test_tilelang_autotune_with_inputs.py:13
↓ 1 callersFunctionref_program
A reference matrix multiplication program, used to compare performance. Parameters ---------- A : torch.Tensor The matrix wi
testing/python/autotune/test_tilelang_autotune_eager_mode.py:13
↓ 1 callersFunctionref_program_fa
(query, key, value, cache_seqlens)
examples/blocksparse_attention/example_triton_sparse_gqa_decode_varlen_mask.py:306
↓ 1 callersFunctionref_program_fa
(query, key, value, cache_seqlens)
examples/blocksparse_attention/example_triton_sparse_gqa_decode_varlen_indice.py:310
↓ 1 callersFunctionref_program_torch
(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:267
↓ 1 callersFunctionref_program_torch
(query, key, value, block_mask, cache_seqlens, max_cache_seqlen, num_blocks, block_size)
examples/blocksparse_attention/example_triton_sparse_gqa_decode_varlen_mask.py:271
↓ 1 callersFunctionref_program_torch
(query, key, value, block_mask, cache_seqlens, max_cache_seqlen, num_blocks, block_size)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_mask.py:257
↓ 1 callersFunctionref_program_torch
(query, key, value, block_indices, cache_seqlens, max_cache_seqlen, num_blocks, block_size)
examples/blocksparse_attention/example_triton_sparse_gqa_decode_varlen_indice.py:275
↓ 1 callersFunctionref_program_torch_paged
Paged version of sparse attention reference implementation. Args: query: [batch, heads, dim] key_cache: [num_pages, page_blo
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_paged.py:225
↓ 1 callersFunctionref_sparse_mla_bwd_interface
(q, kv, o, do, indices, lse, sm_scale=None, is_casual=True)
examples/deepseek_v32/sparse_mla_bwd.py:270
↓ 1 callersFunctionref_sparse_mla_bwd_interface
(q, kv, o, do, indices, lse, offsets, sm_scale=None, is_casual=True)
examples/dsa_sparse_finetune/sparse_mla_bwd.py:281
↓ 1 callersFunctionref_sparse_mla_fwd_interface
(q, kv, indices, q_start_index_s, kv_stride=4, sm_scale=None, is_casual=True)
examples/deepseek_v32/sparse_mla_fwd_pipelined.py:382
↓ 1 callersFunctionref_sparse_mla_fwd_interface
(q, kv, indices, q_start_index_s, kv_stride=1, sm_scale=None, is_casual=True)
examples/deepseek_v32/sparse_mla_fwd_seesaw.py:551
↓ 1 callersFunctionref_split_program
(Q, K, V, mask, glse=None, Output_partial=None)
examples/flash_decoding/example_gqa_decode.py:372
↓ 1 callersFunctionreference_bitwise_reduce
(name, a)
testing/python/math/test_math_bitwise_reduce.py:68
↓ 1 callersFunctionregion
Create a tl.region call for a BufferLoad and extents.
tilelang/language/utils.py:12
← previousnext →2,801–2,900 of 7,254, ranked by callers