MCPcopy Create free account

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

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

↓ 1 callersFunctionstatic_kern
(src, dst, slots)
testing/python/language/test_tilelang_language_subtype.py:335
↓ 1 callersMethodstep
(self)
tilelang/carver/roller/hint.py:244
↓ 1 callersFunctionstepped_serial
()
testing/python/language/test_tilelang_language_frontend_v2.py:259
↓ 1 callersFunctionstepped_serial_neg
()
testing/python/language/test_tilelang_language_frontend_v2.py:275
↓ 1 callersFunctionstg128_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:71
↓ 1 callersFunctionstg128_pred_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:183
↓ 1 callersFunctionstg256_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:101
↓ 1 callersFunctionstg256_pred_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:211
↓ 1 callersFunctionstg32_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:14
↓ 1 callersFunctionstg32_pred_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:130
↓ 1 callersFunctionstg64_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:42
↓ 1 callersFunctionstg64_pred_kernel
(X, Y)
testing/python/language/test_tilelang_language_stg.py:156
↓ 1 callersMethodstop_proc
(self)
tilelang/autodd.py:787
↓ 1 callersMethodstop_workers
(self)
tilelang/autodd.py:1215
↓ 1 callersFunctionstore_frontend_cache
( frontend_key_data: dict, kernel_key: str, *, out_idx: list[int] | int | None = None, tar
tilelang/cache/__init__.py:161
↓ 1 callersMethodsubmit_result
(self, task: Task, is_interested: bool)
tilelang/autodd.py:1182
↓ 1 callersFunctionsubstitute_buffer
(buf)
tilelang/language/eager/builder.py:1255
↓ 1 callersFunctionsummarize_error
Summarize and print error statistics for an implementation.
maint/precision/compare_ops.py:340
↓ 1 callersFunctionsupply_prog
(args)
examples/gemm_fp8/example_tilelang_gemm_amd.py:17
↓ 1 callersFunctionswap_idx
(A)
testing/python/language/test_tilelang_language_frontend_v2.py:304
↓ 1 callersFunctionswap_var
(A)
testing/python/language/test_tilelang_language_frontend_v2.py:295
↓ 1 callersFunctionsync_warp_cond_kernel
()
testing/python/amd/test_tilelang_hip_codegen.py:589
↓ 1 callersFunctiontarget_is_metal
(target: Target)
tilelang/utils/target.py:327
↓ 1 callersMethodtask_update
(self, task: Task, is_interesting: bool)
tilelang/autodd.py:736
↓ 1 callersMethodtcgen05_blockscaled_atom
Emit a single TCGEN05MMA block-scaled SS instruction. Parameters ---------- desc_a, desc_b : Buffer Initialized A
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:1040
↓ 1 callersMethodtcgen05_ts_atom
Emit a single TCGEN05MMA TS instruction for atom ``(inst_m_idx, inst_n_idx, ki)``. A resides in tensor memory; B in shared memory. P
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:945
↓ 1 callersMethodtcgen05mma_ss
Emit the SS (Shared-Shared) variant of TCGEN5MMA. Reads operand A and B from shared memory via a descriptor. Parameters ----
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:225
↓ 1 callersMethodtcgen05mma_ts
Emit the TS (TensorMemory-Shared) variant of TCGEN5MMA. Reads operand A directly from tensor memory (TMEM) and operand B from shared
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:269
↓ 1 callersFunctiontensor_null_test
(M, N, K, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32, with_bias=False)
testing/python/jit/test_tilelang_jit_nullptr.py:9
↓ 1 callersMethodtensorcore_legalization
(self)
tilelang/carver/roller/hint.py:231
↓ 1 callersFunctiontest
(n: int, hidden_size: int, hc_mult: int)
examples/deepseek_mhc/example_mhc_pre.py:386
↓ 1 callersFunctiontest
()
testing/python/issue/test_tilelang_issue_1744.py:25
↓ 1 callersFunctiontest_all_attrs_together_lazy
annotate_pass_configs, annotate_compile_flags, and out_idx should all work together.
testing/python/language/test_tilelang_language_func_attrs.py:32
↓ 1 callersFunctiontest_annotations_after_tensor_type
Annotations placed after tensor type annotations should work.
testing/python/language/test_tilelang_language_func_attrs.py:181
↓ 1 callersFunctiontest_annotations_before_tensor_type
Annotations placed before tensor type annotations should work.
testing/python/language/test_tilelang_language_func_attrs.py:158
↓ 1 callersFunctiontest_assert_matmul
()
testing/python/kernel/test_tilelang_kernel_int8_gemm_tcgen5.py:74
↓ 1 callersFunctiontest_assert_tl_matmul
()
examples/gemm_fp8/example_tilelang_gemm_amd_fp8_preshuffle.py:220
↓ 1 callersFunctiontest_c2d_im2col_alias_warns_and_uses_new_tileop
()
testing/python/transform/test_tilelang_transform_im2col_fallback.py:69
↓ 1 callersFunctiontest_chained_gemm_128
()
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:183
↓ 1 callersFunctiontest_chained_gemm_256
()
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:189
↓ 1 callersFunctiontest_chained_gemm_non_power_of_2_k
()
testing/python/kernel/test_tilelang_kernel_bf16_gemm_tcgen5_ts.py:195
↓ 1 callersFunctiontest_clean_and_maintain_logits
Correctness + speed test where `M` query rows are packed from `num_seqs` equal-length causal sequences. Per-row ``cu_ks / cu_ke`` is derived f
examples/dsa_hisa/clean_and_maintain_logits.py:67
↓ 1 callersFunctiontest_codegen_emits_braces
The generated CUDA source should contain scoped { } blocks for explicitly marked allocs.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:348
↓ 1 callersFunctiontest_codegen_emits_const_for_readonly_params
()
testing/python/transform/test_readonly_param_const_codegen.py:19
↓ 1 callersFunctiontest_codegen_skips_redundant_top_level_braces
The outermost top-level lexical scope should not emit a redundant brace block.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:375
↓ 1 callersFunctiontest_compile_flags_only_lazy
annotate_compile_flags should work standalone.
testing/python/language/test_tilelang_language_func_attrs.py:137
↓ 1 callersFunctiontest_convert
(N, K, block_N, block_K, in_dtype, num_bits=4, threads=128)
examples/dequantize_gemm/example_dequant_gemm_fp4_hopper.py:45
↓ 1 callersFunctiontest_correctness
( BATCH: int = 1, H: int = 8, N_CTX: int = 512, N_KV: int = 1024, D_HEAD: int = 128, T
examples/deepseek_v4/sparse_attn_fwd_sm90.py:190
↓ 1 callersFunctiontest_dealloc_before_thread_return_keeps_auto_dealloc
Dealloc on non-fallthrough path (before thread_return) does NOT suppress auto-dealloc.
testing/python/transform/test_tilelang_transform_lower_shared_tmem.py:69
↓ 1 callersFunctiontest_disable_reuse_data_integrity
Allocate multiple shared buffers, copy data in, compute sum, copy out. Verifies data integrity when shared memory reuse is disabled.
testing/python/transform/test_tilelang_transform_disable_memory_reuse.py:77
↓ 1 callersFunctiontest_disable_reuse_no_overlap
Two sequential buffers must NOT share the same offset when reuse is disabled. a_shared is used (copy in, copy out), then b_shared is used (copy i
testing/python/transform/test_tilelang_transform_disable_memory_reuse.py:95
↓ 1 callersFunctiontest_eager_mode_attrs
Eager mode should support annotate_pass_configs and out_idx via T.empty.
testing/python/language/test_tilelang_language_func_attrs.py:58
↓ 1 callersFunctiontest_example_gemv
()
examples/gemv/test_example_gemv.py:4
↓ 1 callersFunctiontest_example_wy_fast_compilation
()
examples/gdn/test_example_gdn_compilation.py:26
↓ 1 callersFunctiontest_explicit_deallocate_only_suppresses_matching_buffer
Only the explicitly-deallocated buffer skips auto-dealloc; others keep it.
testing/python/transform/test_tilelang_transform_lower_shared_tmem.py:48
↓ 1 callersFunctiontest_explicit_deallocate_tmem_suppresses_auto_dealloc
Explicit T.deallocate_tmem on fallthrough suppresses auto-dealloc.
testing/python/transform/test_tilelang_transform_lower_shared_tmem.py:29
↓ 1 callersFunctiontest_fill_static
()
testing/python/language/test_tilelang_language_int64.py:121
↓ 1 callersFunctiontest_fill_symbolic
()
testing/python/language/test_tilelang_language_int64.py:92
↓ 1 callersFunctiontest_float4_e2m1_unpacked_distinct_from_packed
()
testing/python/language/test_tilelang_language_float4_e2m1_unpacked_dtype.py:32
↓ 1 callersFunctiontest_float4_e2m1_unpacked_dtype_helpers
()
testing/python/language/test_tilelang_language_float4_e2m1_unpacked_dtype.py:40
↓ 1 callersFunctiontest_float4_e2m1_unpacked_dtype_properties
()
testing/python/language/test_tilelang_language_float4_e2m1_unpacked_dtype.py:13
↓ 1 callersFunctiontest_float4_e2m1_unpacked_tir_dtype_tag
()
testing/python/language/test_tilelang_language_float4_e2m1_unpacked_dtype.py:21
↓ 1 callersFunctiontest_fp4_act_quant
Test FP4 act quant: tilelang vs PyTorch reference via dequantized comparison.
examples/deepseek_v4/act_quant.py:324
↓ 1 callersFunctiontest_fp8_act_quant
Test FP8 act quant: tilelang vs PyTorch reference.
examples/deepseek_v4/act_quant.py:307
↓ 1 callersFunctiontest_fp8_block_mean_pooling
Correctness + speed test with `num_seqs` sequences of equal length packed into the flat K buffer. NOTE: the flat mean-pool kernel is sequence
examples/dsa_hisa/fp8_block_mean_pooling.py:99
↓ 1 callersFunctiontest_fp8_block_sparse_mqa
Correctness + speed test packing `num_seqs` equal-length causal sequences into the [M, H, D] Q and [M, D] K tensors. Each query sees only the
examples/dsa_hisa/block_sparse_mqa_fp8.py:165
↓ 1 callersFunctiontest_fp8_lighting_indexer
(S=4096, SKV=8192, H=32, HKV=1, D=64, kv_stride=1)
examples/deepseek_v32/fp8_lighting_indexer.py:240
↓ 1 callersFunctiontest_gemm_fp8
(M, N, K)
examples/gemm_fp8/example_tilelang_gemm_amd.py:107
↓ 1 callersFunctiontest_hisa
End-to-end smoke + speed test packing `num_seqs` equal-length causal sequences into the flat [M, H, D] Q and [N=M, D] K tensors. Per-token ``
examples/dsa_hisa/hisa.py:135
↓ 1 callersFunctiontest_im2col_uses_simt_fallback_before_hopper
()
testing/python/transform/test_tilelang_transform_im2col_fallback.py:56
↓ 1 callersFunctiontest_im2col_uses_tma_on_hopper
()
testing/python/transform/test_tilelang_transform_im2col_fallback.py:63
↓ 1 callersFunctiontest_kernel
( B=1, S=2048, H=16, D=512, tail_D=64, index_D=128, topk=64, )
examples/dsa_sparse_finetune/dsa.py:166
↓ 1 callersFunctiontest_kernel
( B=1, S=2048, H=64, D=128, topk=64, )
examples/dsa_sparse_finetune/indexer_topk_reducesum.py:232
↓ 1 callersFunctiontest_kernel
( B=1, S=2048, H=16, D=128, topk=64, )
examples/dsa_sparse_finetune/indexer_bwd.py:218
↓ 1 callersFunctiontest_kernel
( B=1, S=2048, H=16, D=512, tail_D=64, topk=128, )
examples/dsa_sparse_finetune/sparse_mla_topk_reducesum.py:201
↓ 1 callersFunctiontest_lower_opaque_block_inserts_lexical_alloc_scope_for_explicit_block
An explicitly annotated block should produce a lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:68
↓ 1 callersFunctiontest_lower_opaque_block_inserts_scope_for_gemm_descriptor_alloc
Lowered WGMMA descriptor buffers inside a loop should trigger lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:158
↓ 1 callersFunctiontest_lower_opaque_block_marks_explicit_top_level_local_alloc
A top-level explicitly annotated local alloc should get lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:218
↓ 1 callersFunctiontest_lower_opaque_block_skips_empty_alloc
A block without alloc_buffers should not produce a lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:131
↓ 1 callersFunctiontest_lower_opaque_block_skips_fragment_alloc
A fragment alloc should not force lexical_alloc_scope by itself.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:247
↓ 1 callersFunctiontest_lower_opaque_block_skips_fragment_root_in_disable_ws_pipeline
A fragment root block should not force lexical_alloc_scope in disable-ws pipeline.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:275
↓ 1 callersFunctiontest_lower_opaque_block_skips_local_var_only_alloc
A block that allocates only local.var should not get lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:189
↓ 1 callersFunctiontest_lower_opaque_block_skips_unmarked_local_alloc
An unmarked local-alloc block should not produce a lexical_alloc_scope.
testing/python/transform/test_tilelang_transform_lexical_alloc_scope.py:102
↓ 1 callersFunctiontest_matmul
()
testing/python/language/test_tilelang_language_clear.py:53
↓ 1 callersFunctiontest_matmul
()
testing/python/language/test_tilelang_language_alias.py:52
↓ 1 callersFunctiontest_named_barrier_arrive_api
T.named_barrier_arrive emits the CUDA named-barrier arrive helper.
testing/python/issue/test_tilelang_issue_2188.py:13
↓ 1 callersFunctiontest_no_out_idx_when_not_using_empty
When T.empty is not used, tilelang_out_idx attr should not be present.
testing/python/language/test_tilelang_language_func_attrs.py:95
↓ 1 callersFunctiontest_nonws_im2col_tma_num_stages_3_uses_pipeline_barrier
Non-WS pipelined im2col TMA with num_stages=3 must use pipeline_mbar[3].
testing/python/transform/test_tilelang_transform_pipeline_barrier_ownership.py:96
↓ 1 callersFunctiontest_nonws_plain_copy_gemm_num_stages_1_stays_sync
num_stages=1 should also keep non-WS plain T.copy on the sync path.
testing/python/transform/test_tilelang_transform_pipeline_barrier_ownership.py:62
↓ 1 callersFunctiontest_nonws_plain_copy_gemm_num_stages_3_stays_sync
Non-WS pipelined GEMM should not auto-upgrade plain T.copy to TMA.
testing/python/transform/test_tilelang_transform_pipeline_barrier_ownership.py:28
↓ 1 callersFunctiontest_out_idx_conflict_detection
Specifying both T.empty return and external out_idx should raise ValueError.
testing/python/language/test_tilelang_language_func_attrs.py:78
↓ 1 callersFunctiontest_out_idx_via_attr_lazy
out_idx should be stored as PrimFunc attr when using T.empty + return.
testing/python/language/test_tilelang_language_func_attrs.py:10
↓ 1 callersFunctiontest_pass_configs_only_lazy
annotate_pass_configs should work without T.empty or annotate_compile_flags.
testing/python/language/test_tilelang_language_func_attrs.py:115
↓ 1 callersFunctiontest_pool_mqa_fp8
Correctness + speed test packing `num_seqs` equal-length causal sequences into the [M, H, D] Q tensor. Per-query ``cu_seqlen_blocked_ks/ke``
examples/dsa_hisa/pool_mqa_fp8.py:155
↓ 1 callersFunctiontest_round_trip_error
Round-trip sanity: quantize then dequantize, check MSE.
examples/deepseek_v4/act_quant.py:357
↓ 1 callersFunctiontest_run_dequantize_gemm
()
examples/dequantize_gemm/example_dequant_gemm_fine_grained.py:426
↓ 1 callersFunctiontest_sparse_mla_bwd
(B=1, S=4096, SKV=8192, H=64, HKV=1, DQKV=576, DV=512, topk=2048, dtype=torch.bfloat16, check_correctness=True
examples/deepseek_v32/sparse_mla_bwd.py:282
↓ 1 callersFunctiontest_sparse_mla_bwd
(B=1, S=2048, H=64, HKV=1, DQKV=576, DV=512, topk=512, dtype=torch.bfloat16, check_correctness=True)
examples/dsa_sparse_finetune/sparse_mla_bwd.py:293
↓ 1 callersFunctiontest_sparse_mla_fwd
( B=1, S=4096, SKV=8192, H=128, HKV=1, DQK=576, DV=512, topk=2048, dtype=t
examples/deepseek_v32/sparse_mla_fwd.py:230
↓ 1 callersFunctiontest_sparse_mla_fwd
( B=1, S=4096, H=128, HKV=1, DQK=576, DV=512, topk=2048, dtype=torch.bfloat16,
examples/dsa_sparse_finetune/sparse_mla_fwd.py:260
← previousnext →3,101–3,200 of 7,254, ranked by callers