MCPcopy Create free account

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

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

↓ 4 callersMethodmake_mma_store_layout
Create a layout function for storing MMA results into a fragment buffer. This layout is used in conjunction with `inverse_mma_store_l
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:559
↓ 4 callersFunctionmake_quarter_bank_swizzled_layout
Args: buffer: BufferLikeType Examples: make_quarter_bank_swizzled_layout(buffer)
tilelang/layout/swizzle.py:118
↓ 4 callersFunctionmake_source_kernel
(source_code_or_path: str | os.PathLike[str], entry_name: str)
testing/python/language/test_tilelang_language_source_kernel.py:24
↓ 4 callersFunctionmake_swizzle_layout
(shared_buf)
examples/bitnet-1.58b/kernel_benchmark/tl_int8xint8.py:16
↓ 4 callersMethodmake_wmma_load_layout
(self, local_buf: Buffer, matrix: Literal["A", "B"] = "A")
tilelang/rocm/intrinsics/wmma_macro_generator.py:385
↓ 4 callersMethodmake_wmma_store_layout
(self, local_buf: Buffer)
tilelang/rocm/intrinsics/wmma_macro_generator.py:462
↓ 4 callersMethodmap_forward_thread
Get the thread mapping expression for a given set of argument indices. Parameters ---------- indices : list of PrimE
tilelang/layout/fragment.py:166
↓ 4 callersFunctionmatmul
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/warp_specialize/example_warp_specialize_gemm_softpipe_stage2.py:6
↓ 4 callersFunctionmatmul
( M, N, K, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_
testing/python/jit/test_tilelang_jit_gemm_cython.py:9
↓ 4 callersFunctionmatmul_warp_specialize_copy_0_gemm_1
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/warp_specialize/example_warp_specialize_gemm_copy_0_gemm_1.py:6
↓ 4 callersFunctionmatmul_warp_specialize_copy_1_gemm_0
(A, B, block_M, block_N, block_K, dtype=T.float16, accum_dtype=T.float32)
examples/warp_specialize/example_warp_specialize_gemm_copy_1_gemm_0.py:6
↓ 4 callersMethodmma_sp
(self, A_local_buf: Buffer, E_local_buf: Buffer, B_local_buf: Buffer, C_local_buf: Buffer, k_inner: PrimExpr =
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:498
↓ 4 callersFunctionparallel_nsa_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, o_slc: torch.Tensor, o_swa: Optional[
examples/deepseek_nsa/benchmark/benchmark_nsa_fwd.py:131
↓ 4 callersFunctionprepare_ks_ke_from_cu_seqlens
( cu_seqlens: torch.LongTensor, )
examples/dsa_hisa/tilelang_utils.py:98
↓ 4 callersFunctionprepare_lens
(cu_seqlens: torch.LongTensor)
examples/dsa_hisa/tilelang_utils.py:49
↓ 4 callersFunctionprepare_position_ids
(cu_seqlens: torch.LongTensor)
examples/dsa_hisa/tilelang_utils.py:69
↓ 4 callersMethodprim_func
(self, name)
tilelang/language/eager/builder.py:226
↓ 4 callersFunctionprint_red_warning
(message)
examples/dequantize_gemm/dequantize_utils.py:102
↓ 4 callersFunctionprint_red_warning
(message)
examples/gdn/test_utils.py:4
↓ 4 callersFunctionprint_red_warning
(message)
examples/kda/test_utils_kda.py:6
↓ 4 callersMethodpropagate_reduction_inputs
(self, shape, rstep: dict | None = None)
tilelang/carver/roller/node.py:392
↓ 4 callersFunctionpythonic_expr
Converts a TVM PrimExpr into a Python-style string, correctly handling operator precedence. Args: expr: The TVM PrimExpr to convert.
tilelang/jit/adapter/utils.py:155
↓ 4 callersMethodquantize
(self, fuse_qkv=True, fuse_gateup=True)
examples/bitnet-1.58b/modeling_bitnet.py:1345
↓ 4 callersFunctionrandn_semi_sparse
Generate a random semi-sparse tensor. The generated tensor will have 2:4 sparsity along the K dimension. Args: M (int): Number of row
tilelang/utils/sparse.py:274
↓ 4 callersMethodreduce
The reduced block axis defining function. Parameters ---------- dom : Union[Range, List[PrimExpr], Tuple[PrimExpr]]
tilelang/language/ast/ir.py:564
↓ 4 callersFunctionref_program
(x, gamma, beta, eps)
examples/norm/layernorm.py:169
↓ 4 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/bitnet-1.58b/modeling_bitnet.py:272
↓ 4 callersFunctionrequires_gfx950
Skip the test when the current ROCm target is not gfx950.
testing/python/amd/test_tilelang_ds_read_tr.py:21
↓ 4 callersFunctionresolve_gemm_impl
Resolve the registered implementation class for a GEMM instruction key.
tilelang/tileop/gemm/registry.py:38
↓ 4 callersFunctionretrieve_func_from_module
Retrieve the single PrimFunc from an IRModule. Args: ir_module (IRModule): The TVM IRModule to extract the function from.
tilelang/utils/language.py:170
↓ 4 callersMethodrun
Run the auto-tuning process. Args: warmup: Number of warmup iterations. rep: Number of repetitions for timing.
tilelang/autotuner/tuner.py:861
↓ 4 callersFunctionrun_atomic_add_auto_vectorized_unit_test
(vec_size: int, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:236
↓ 4 callersFunctionrun_ceildiv
(a=128, b=32)
testing/python/language/test_tilelang_language_ceildiv.py:16
↓ 4 callersFunctionrun_ceildiv_dyn
(a=128, b=32)
testing/python/language/test_tilelang_language_ceildiv.py:40
↓ 4 callersFunctionrun_cummax_1d
(N, block_N, reverse=False, dtype=T.float32, scope="smem")
testing/python/language/test_tilelang_language_scan.py:429
↓ 4 callersFunctionrun_cumsum_1d
(N, block_N, reverse=False, dtype=T.float32, scope="smem")
testing/python/language/test_tilelang_language_scan.py:120
↓ 4 callersFunctionrun_cumsum_region_1d
Run test for cumsum with region input.
testing/python/language/test_tilelang_language_scan.py:206
↓ 4 callersFunctionrun_cumsum_region_2d
Run test for cumsum with 2D region input.
testing/python/language/test_tilelang_language_scan.py:261
↓ 4 callersFunctionrun_gemm
( order, stage, )
testing/python/language/test_tilelang_language_pipeline.py:53
↓ 4 callersFunctionrun_gemm_codegen
(M, N, K, block_M, block_N, block_K, trans_A=False, trans_B=False)
testing/python/cpu/test_tilelang_cpu_tgemm.py:55
↓ 4 callersFunctionrun_gemm_compile
(M, N, K, block_M, block_N, block_K, trans_A=False, trans_B=False, dtype=T.float32)
testing/python/cpu/test_tilelang_cpu_tgemm.py:65
↓ 4 callersFunctionrun_gemm_rr
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
maint/gemm/correctness_evaluation.py:349
↓ 4 callersFunctionrun_gemm_rs
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
maint/gemm/correctness_evaluation.py:177
↓ 4 callersFunctionrun_gemm_sr
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
maint/gemm/correctness_evaluation.py:261
↓ 4 callersFunctionrun_split_host_device_passes
Run the necessary passes before and including SplitHostDevice.
testing/python/transform/test_tilelang_transform_split_host_device.py:12
↓ 4 callersFunctionrun_vectorize
(N, M, stride_A, stride_B)
testing/python/language/test_tilelang_language_vectorize.py:26
↓ 4 callersFunctionrun_vectorize_invariant_index
(N, M, K)
testing/python/language/test_tilelang_language_vectorize.py:80
↓ 4 callersFunctionrun_view
(N, M, dtype, new_dtype=None)
testing/python/language/test_tilelang_language_view.py:33
↓ 4 callersMethodset_compile_args
Set compilation arguments for the auto-tuner. Args: out_idx: List of output tensor indices. target: Target platform.
tilelang/autotuner/tuner.py:275
↓ 4 callersMethodset_profile_args
Set profiling arguments for the auto-tuner. Args: supply_type: Type of tensor supply mechanism. Ignored if `supply_prog` is provi
tilelang/autotuner/tuner.py:329
↓ 4 callersFunctionshould_disable_shared_memory_reuse
(pass_ctx: PassContext | None = None)
tilelang/backend/pass_pipeline/pipeline_utils.py:49
↓ 4 callersFunctionshould_enable_aggressive_merge
(pass_ctx: PassContext | None = None, target: Target | None = None)
tilelang/backend/pass_pipeline/pipeline_utils.py:24
↓ 4 callersFunctionshould_enable_race_check
(pass_ctx: PassContext | None = None)
tilelang/backend/pass_pipeline/pipeline_utils.py:43
↓ 4 callersFunctionshould_force_let_inline
(pass_ctx: PassContext | None = None)
tilelang/backend/pass_pipeline/pipeline_utils.py:31
↓ 4 callersFunctionsingle_kernel_program
Create a single-kernel CuTeDSL program.
testing/python/jit/test_tilelang_jit_cutedsl_host_codegen.py:45
↓ 4 callersFunctionsparse_mla_fwd_interface
(q, kv, indices, sm_scale=None, return_p_sum: bool = False, d_v=512, block_I=64, num_stages=2, threads=256)
examples/deepseek_v32/sparse_mla_fwd.py:171
↓ 4 callersFunctionsparse_mla_fwd_interface
( q, kv, indices, offsets, sm_scale=None, return_p_sum: bool = False, d_v=512, block_I=32, num_stages=2, t
examples/dsa_sparse_finetune/sparse_mla_fwd.py:174
↓ 4 callersFunctionsub
(v)
tilelang/language/eager/builder.py:1252
↓ 4 callersMethodtcgen05_atom_arrive
Emit ``tcgen05_mma_arrive(mbar)``.
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:829
↓ 4 callersFunctiontl_topk
(input, starts, ends, topk)
examples/deepseek_v32/topk_selector.py:176
↓ 4 callersFunctiontl_topk
(logits, topk, blk_m, threads=128)
examples/topk/example_topk.py:18
↓ 4 callersFunctionunwrap_expr
unwrap expr and convert it into PrimExpr like
tilelang/language/eager/builder.py:36
↓ 4 callersMethodweight_quant
(weight)
examples/bitnet-1.58b/utils_quant.py:150
↓ 4 callersFunctionwith_rocm_target_attrs
(target: Target)
tilelang/utils/target.py:58
↓ 4 callersMethodwmma
(self, A_local_buf: Buffer, B_local_buf: Buffer, C_local_buf: Buffer, k_inner: PrimExpr | None = 0)
tilelang/rocm/intrinsics/wmma_macro_generator.py:283
↓ 4 callersMethodwrap
(self, c_source: str)
tilelang/jit/adapter/wrapper.py:966
↓ 3 callersFunctionIsRDNA
()
examples/amd/example_amd_flash_attn_fwd.py:12
↓ 3 callersFunctionPrefetch
( K: T.Tensor(shape, dtype), V: T.Tensor(shape, dtype), K_shared: T.Shared
examples/minference/example_vertical_slash_sparse_attn.py:40
↓ 3 callersMethod__init__
(self, tags: dict | None = None, name: str = "Node")
tilelang/carver/roller/node.py:96
↓ 3 callersMethod__init__
(self)
tilelang/carver/roller/rasterization.py:7
↓ 3 callersMethod__init__
(self)
tilelang/transform/decouple_type_cast.py:102
↓ 3 callersFunction_apply
(func)
testing/python/transform/test_tilelang_transform_fuse_mbarrier_arrive_expect_tx.py:12
↓ 3 callersFunction_apply
(func)
testing/python/transform/test_tilelang_transform_lower_shared_tmem.py:11
↓ 3 callersFunction_apply
(func)
testing/python/transform/test_tilelang_transform_lower_shared_barrier.py:13
↓ 3 callersFunction_as_stmt_placeholder
(temp: ast.AST)
tilelang/autodd.py:212
↓ 3 callersFunction_assert_mma_sync_shape
(source, a_type, c_type, m, n, k)
testing/python/kernel/test_tilelang_kernel_gemm_sm75.py:30
↓ 3 callersFunction_assert_runs_packed_fp4_storage
(kernel, dtype)
testing/python/issue/test_tilelang_issue_2292.py:45
↓ 3 callersFunction_assert_tma_destinations_128_aligned
(arch: str, mod)
testing/python/issue/test_tilelang_issue_sm120_tma_smem_alignment.py:148
↓ 3 callersFunction_bench_with_cupti
Benchmark using CUPTI profiler for detailed kernel timing.
tilelang/profiler/bench.py:177
↓ 3 callersFunction_bytes_to_elements
(byte_count: int, elem_bits: int)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:89
↓ 3 callersFunction_check
(original, transformed)
testing/python/transform/test_tilelang_transform_hoist_broadcast_values.py:56
↓ 3 callersFunction_check
(original, expected, target=sm100_target)
testing/python/transform/test_tilelang_transform_inject_tcgen05_fence.py:21
↓ 3 callersFunction_check_with_config
(original, transformed, config)
testing/python/transform/test_tilelang_transform_loop_unswitching.py:15
↓ 3 callersFunction_collect_wait_args
(func)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:82
↓ 3 callersFunction_count_attrs_and_calls
(func)
testing/python/transform/test_tilelang_transform_Inject_software_pipeline.py:41
↓ 3 callersFunction_count_calls
(func: tvm.tirx.PrimFunc)
testing/python/transform/test_tilelang_transform_lower_tile_op.py:10
↓ 3 callersMethod_determinate_swizzle_mode
(self, buffer, layout: Layout)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:188
↓ 3 callersMethod_determinate_swizzle_mode
(self, buffer: Buffer, layout: Layout)
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:104
↓ 3 callersFunction_find_compute_scope
(func)
testing/python/transform/test_tilelang_transform_make_packed_api.py:41
↓ 3 callersFunction_find_nvcc
(cu_dir: pathlib.Path)
cmake/find_pip_cuda.py:28
↓ 3 callersFunction_gemm_impl
Shared GEMM implementation. Returns a call_intrin handle for the given op key.
tilelang/language/gemm_op.py:22
↓ 3 callersFunction_gemm_sp_impl
Shared sparse GEMM implementation. Returns a call_intrin handle for the given op key.
tilelang/language/experimental/gemm_sp_op.py:20
↓ 3 callersMethod_get_cache_root
()
tilelang/cache/kernel_cache.py:201
↓ 3 callersFunction_get_current_stack
()
tilelang/language/kernel.py:81
↓ 3 callersFunction_get_current_stack
()
tilelang/autotuner/capture.py:81
↓ 3 callersMethod_get_dtype_abbrv
(self, dtype: str)
tilelang/cuda/intrinsics/macro/mma_sm70_macro_generator.py:108
↓ 3 callersMethod_get_kernel_lib_file
Return the cache filename for one backend's executable artifact.
tilelang/autotuner/param.py:540
↓ 3 callersMethod_get_missing_complete_cache_files
(self, cache_path: str)
tilelang/cache/kernel_cache.py:766
↓ 3 callersMethod_get_warp_indices
Compute (warp_m, warp_n) from the current thread binding.
tilelang/metal/intrinsics/metal_macro_generator.py:58
↓ 3 callersFunction_kernel_alloc_var_init
Kernel that initialises a local int32 variable to 7 and writes it out.
testing/python/amd/test_tilelang_hip_codegen.py:278
← previousnext →501–600 of 7,254, ranked by callers