MCPcopy Create free account

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

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

↓ 3 callersFunctionget_runtime
(num_repeats=1)
examples/bitnet-1.58b/benchmark_inference_latency.py:20
↓ 3 callersFunctionget_runtime
(num_repeats=1)
examples/bitnet-1.58b/benchmark_generate.py:57
↓ 3 callersFunctionget_runtime
(num_repeats=1)
examples/bitnet-1.58b/eval_gpu_memory.py:20
↓ 3 callersFunctionget_runtime
(num_repeats=1)
examples/bitnet-1.58b/benchmark_model_10k_loops.py:27
↓ 3 callersFunctionget_runtime
(num_repeats=1)
examples/bitnet-1.58b/eval_correctness.py:55
↓ 3 callersMethodget_shape
(self, id: int = 0)
tilelang/carver/roller/node.py:157
↓ 3 callersMethodget_store_index_map
Return the store index map. The forward map is (thread_id, local_id) -> (i, j), which is affine. The inverse map is (i, j) -> (thread
tilelang/rocm/intrinsics/wmma_macro_generator.py:157
↓ 3 callersMethodget_store_index_map
(self, inverse: bool = False)
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:229
↓ 3 callersFunctionget_tensorized_func_and_tags
transform function to matmul if necessary (e.g. transform conv2d with im2col)
tilelang/carver/matmul_analysis.py:513
↓ 3 callersMethodget_thread_binding
(self)
tilelang/rocm/intrinsics/wmma_macro_generator.py:126
↓ 3 callersFunctionget_thread_bindings
Returns all three thread bindings.
tilelang/language/kernel.py:445
↓ 3 callersFunctionget_windows_compiler
Resolve the preferred Windows host compiler. Returns ``(path, kind)`` where ``kind`` is ``"clang-cl"`` or ``"msvc"``. Set the ``TILELANG_DISA
tilelang/contrib/msvc.py:108
↓ 3 callersFunctioninfer_layout
Infer the layout based on the region and the kind of buffer kind: "A", "B", "C"
tilelang/carver/matmul_analysis.py:356
↓ 3 callersMethodinfer_shared_layout
(self, continuity: int)
tilelang/tileop/gemm_sp/gemm_sp_wgmma.py:24
↓ 3 callersMethodinfer_shared_layout
Infer a standard shared-memory swizzle layout for TCGEN05 operands.
tilelang/cuda/op/gemm/gemm_tcgen05.py:57
↓ 3 callersMethodinfer_shared_layout
Infer the swizzle layout for shared memory based on continuity. WGMMA can directly use shared memory as input, so the swizzle layout must
tilelang/cuda/op/gemm/gemm_wgmma.py:28
↓ 3 callersMethodinfer_shared_layout
(self, continuity: int)
tilelang/cuda/op/gemm_sp/gemm_sp_wgmma.py:24
↓ 3 callersMethodinit_tcgen05_a_desc
Emit TIR to initialize a pre-allocated TCGEN05 A descriptor (SS variant). Parameters ---------- desc_a : Buffer A
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:791
↓ 3 callersMethodinit_wgmma_b_desc
Emit TIR to initialize a pre-allocated WGMMA B descriptor. Parameters ---------- desc_b : Buffer A descriptor buf
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:403
↓ 3 callersFunctionis_cpu_target
(target: Target)
tilelang/jit/adapter/utils.py:105
↓ 3 callersFunctionis_cuda_tensorcore_target
(target: Target)
tilelang/carver/matmul_analysis.py:546
↓ 3 callersFunctionis_float4
Whether *value* is any FP4 E2M1 logical variant (packed or unpacked).
tilelang/language/dtypes.py:299
↓ 3 callersMethodis_float4_e2m1fn
(self)
tilelang/language/dtypes.py:20
↓ 3 callersFunctionis_float8_dtype
(dtype: torch.dtype)
tilelang/utils/tensor.py:21
↓ 3 callersFunctionis_metal_target
(target: Target)
tilelang/jit/adapter/utils.py:109
↓ 3 callersMethodis_placeholder
(self)
tilelang/carver/roller/node.py:122
↓ 3 callersFunctionkernel
The actual kernel to compute C = A @ B^T. Parameters ---------- block_M : int Block size in M dimension.
benchmark/matmul/benchmark_matmul_sp.py:127
↓ 3 callersFunctionkernel
( A_ptrs: T.Tensor([batch_count], T.ptr), B_ptrs: T.Tensor([batch_count], T.ptr), C_pt
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:42
↓ 3 callersFunctionkernel
(out: T.Tensor[(num_tokens,), T.float32])
testing/python/issue/test_tilelang_issue_1993.py:14
↓ 3 callersFunctionlayer_norm
(x, gamma, beta, eps=1e-5)
examples/norm/layernorm.py:165
↓ 3 callersMethodload_frontend_cached
( self, frontend_key: str, *, target: str | Target = "auto", target_ho
tilelang/cache/kernel_cache.py:635
↓ 3 callersFunctionmake_dq_layout
(dQ)
examples/flash_attention/example_gqa_bwd.py:119
↓ 3 callersFunctionmake_linear_layout
Create a row-major linear layout for any dimension. Args: buffer_or_load_or_region: BufferLikeType Returns: Layout: A r
tilelang/layout/swizzle.py:129
↓ 3 callersFunctionmake_mma_swizzle_layout
(shared_buf, is_smooth: bool = False)
tilelang/cuda/intrinsics/layout/mma_layout.py:241
↓ 3 callersFunctionmake_ptr_table
(tensors)
examples/grouped_gemm/example_grouped_gemm_fwd_ptr.py:11
↓ 3 callersMethodmake_smem_layout_AB
(dtype, major_mode: LayoutEnum, tiler_mk, stages=1)
tilelang/contrib/cutedsl/gemm_v1.py:451
↓ 3 callersFunctionmake_volta_swizzled_layout
(buffer: BufferLikeType, is_a: bool = True, k_inner: bool = True)
tilelang/layout/swizzle.py:72
↓ 3 callersFunctionmatch_buffer
The buffer match function. Note ---- This function will perform different behavior, depending on the type of param. If the param is a
tilelang/language/ast/ir.py:246
↓ 3 callersFunctionmatch_declare_kernel
(source: str, annotation: str = "__global__")
tilelang/jit/adapter/utils.py:30
↓ 3 callersFunctionmatmul
Construct and return a tiled matrix-multiply TIR kernel that multiplies A (shape MxK) by a quantized B (shape Nx(QK)) and writes an MxN output in
examples/dequantize_gemm/example_dequant_gemm_bf16_mxfp4_hopper.py:86
↓ 3 callersFunctionmatmul
( M, N, K, in_dtype, out_dtype, accum_dtype, num_bits=4, block_M=128, bloc
examples/dequantize_gemm/example_dequant_gemm_fp4_hopper.py:110
↓ 3 callersFunctionmatmul
Builds a parameterized TileLang/TIR matrix-multiplication kernel that dequantizes 4-bit FP inputs to BF16 on-the-fly and computes C = A @ B^T.
examples/dequantize_gemm/example_dequant_gemm_bf16_fp4_hopper.py:46
↓ 3 callersFunctionmatmul
( M, N, K, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_
testing/python/jit/test_tilelang_jit_cutedsl.py:22
↓ 3 callersFunctionmatmul
( M, N, K, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_
testing/python/jit/test_tilelang_jit_tvm_ffi.py:9
↓ 3 callersFunctionmatmul
( M, N, K, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_
testing/python/jit/test_tilelang_jit_nvrtc.py:9
↓ 3 callersFunctionmatmul_dynamic_mnk
( A, B, block_M, block_N, block_K, trans_A, trans_B, in_dtype, out_dtype,
examples/dynamic_shape/example_dynamic.py:6
↓ 3 callersFunctionmatmul_int8xint4
( M, N, K, in_dtype, out_dtype, accum_dtype, num_bits=4, block_M=32, block
examples/dequantize_gemm/example_dequant_gemm_w4a8.py:87
↓ 3 callersFunctionmatmul_pipelined
A simple pipelined GEMM using T.copy + T.gemm tile ops.
testing/python/transform/test_tilelang_transform_producer_consumer_ws.py:11
↓ 3 callersFunctionmqa_attn_return_logits_interface
(q, kv, kv_scales, weights, cu_seqlen_ks, cu_seqlen_ke, clean_logits=True)
examples/deepseek_v32/fp8_lighting_indexer.py:205
↓ 3 callersFunctionpack_sf_u8_to_u32_1d
(sf_u8)
examples/blockscaled_gemm_sm100/gemm_mxfp8_blockscaled_1d1d.py:525
↓ 3 callersMethodpar_compile
Parallel compile multiple TileLang PrimFunc with TVM and build JITKernels. Parameters ---------- configs : Iterable[U
tilelang/jit/__init__.py:340
↓ 3 callersFunctionparse_function_call_args
Parse function call arguments from a kernel declaration. Args: declaration: The kernel function declaration string. function
tilelang/jit/adapter/utils.py:314
↓ 3 callersFunctionparse_tma_descriptor_args
Parse TMA descriptor arguments into structured parameters. Args: tma_descriptor_args: Dictionary mapping TMA descriptor variables to
tilelang/jit/adapter/utils.py:388
↓ 3 callersFunctionper_token_cast_to_fp8
(X, blk_m)
examples/cast/example_per_token_cast_to_fp8.py:9
↓ 3 callersFunctionpool_mqa_attn_return_logits_fp8_interface
Raw kernel invocation; zero-inits logits so positions the kernel doesn't touch are 0 (matches the ref).
examples/dsa_hisa/pool_mqa_fp8.py:109
↓ 3 callersFunctionprepare_cumsum_output
( B, S, H, dtype, )
examples/gdn/example_cumsum.py:84
↓ 3 callersFunctionprepare_output
( B, S, H, DK, DV, chunk_size, output_dtype, gate_dtype, state_dtype, )
examples/gdn/example_wy_fast_bwd_split.py:77
↓ 3 callersFunctionrandint_semi_sparse
Generate a random semi-sparse integer tensor. The generated tensor will have 2:4 sparsity along the K dimension. Args: M (int): Numbe
tilelang/utils/sparse.py:294
↓ 3 callersFunctionref_program
( query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, sinks: torch.Tensor, slidin
examples/attention_sink/example_mha_sink_fwd_bhsd.py:136
↓ 3 callersFunctionref_program
(x: torch.Tensor, batch_sizes: torch.Tensor)
examples/cast/example_group_per_split_token_cast_to_fp8.py:140
↓ 3 callersFunctionref_program
(A, B, BlockMask, block_M, block_N, block_K)
testing/python/language/test_tilelang_language_any_of.py:7
↓ 3 callersFunctionref_program
(A, B, BlockMask, block_M, block_N, block_K)
testing/python/language/test_tilelang_language_all_of.py:7
↓ 3 callersFunctionref_program_fa
(query, kcache, vcache, cache_seqlens, block_table)
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_paged.py:308
↓ 3 callersFunctionregion_exist_in_list
(a, list)
tilelang/carver/roller/shape_inference/tir.py:299
↓ 3 callersMethodregister_callback
Register postproc callback for the given backend.
testing/python/cache/test_tilelang_kernel_cache.py:64
↓ 3 callersFunctionresolve_execution_backend
Resolve an execution backend string to a concrete backend. - Supports the alias "dlpack" -> "tvm_ffi". - Supports the sentinel "auto" which s
tilelang/jit/execution_backend.py:66
↓ 3 callersMethodrun
Perform 1D cumulative sum. Args: src: Source pointer dst: Destination pointer N: Number of eleme
tilelang/contrib/cutedsl/reduce.py:292
↓ 3 callersFunctionrun_atomic_different_memory_orders
(M, N, block_M, block_N, dtype=T.float32)
testing/python/language/test_tilelang_language_atomic.py:111
↓ 3 callersFunctionrun_autotune
(M, N, K, M_value=None, N_value=None, K_value=None, return_kernel=False)
testing/python/autotune/test_tilelang_autotune_eager_mode.py:93
↓ 3 callersFunctionrun_blocksparse_matmul
(num_stages)
testing/python/language/test_tilelang_language_pipeline.py:157
↓ 3 callersFunctionrun_cutedsl_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_cutedsl.py:320
↓ 3 callersFunctionrun_elementwise_add
(M, N)
testing/python/language/test_tilelang_language_tma_1d.py:30
↓ 3 callersFunctionrun_elementwise_recommend_hints
(shape: List[int] = None, dtype: T.dtype = T.float16, topk: int = 20)
testing/python/carver/test_tilelang_carver_recommend_hints.py:29
↓ 3 callersFunctionrun_gemm
( M, N, K, trans_A, trans_B, in_dtype, out_dtype, dtypeAccum, block_M,
testing/python/amd/test_tilelang_test_amd.py:52
↓ 3 callersFunctionrun_gemv_recommend_hints
( N: int = 1024, K: int = 1024, in_dtype: T.dtype = T.float16, out_dtype: T.dtype = T.float16, accum_dtype
testing/python/carver/test_tilelang_carver_recommend_hints.py:80
↓ 3 callersFunctionrun_general_reduction_recommend_hints
(structure: str = "SSR", shape: List[int] = None, dtype: T.dtype = T.float16, topk: int = 20)
testing/python/carver/test_tilelang_carver_recommend_hints.py:8
↓ 3 callersFunctionrun_matmul_recommend_hints
( M: int = 1024, N: int = 1024, K: int = 1024, in_dtype: T.dtype = T.float16, out_dtype: T
testing/python/carver/test_tilelang_carver_recommend_hints.py:49
↓ 3 callersFunctionrun_nvrtc_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_nvrtc.py:259
↓ 3 callersFunctionrun_passes
(func: tvm.tirx.PrimFunc)
testing/python/transform/test_tilelang_transform_thread_sync.py:8
↓ 3 callersFunctionrun_tilelang_composable_copy
(M=1024, N=1024, block_M=128, block_N=128, dtype=T.float16)
testing/python/language/test_tilelang_language_composable_index.py:28
↓ 3 callersFunctionrun_tilelang_copy
(M=1024, N=1024, block_M=128, block_N=128, dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:27
↓ 3 callersFunctionrun_tilelang_copy_fp4
(M=1024, N=1024, block_M=128, block_N=128, src_dtype=T.float4_e2m1fn, dst_dtype=T.float4_e2m1fn)
testing/python/language/test_tilelang_language_copy.py:188
↓ 3 callersFunctionrun_tilelang_transpose_square
(M=256, block_M=128, dtype=T.float16)
testing/python/language/test_tilelang_language_transpose.py:84
↓ 3 callersFunctionrun_tvm_ffi_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_tvm_ffi.py:256
↓ 3 callersFunctionrun_vectorized_cast
Run the vectorized cast kernel and check the correctness. Args: src_dtype: The source data type. dst_dtype: The destination data t
testing/python/language/test_tilelang_language_vectorized_cast.py:43
↓ 3 callersMethodset_shape
(self, shape: list[int], id=0, overwrite=False)
tilelang/carver/roller/node.py:160
↓ 3 callersMethodset_tag
(self, k: str, v: Any = True)
tilelang/carver/roller/node.py:111
↓ 3 callersFunctionsimple_fuse0
(x)
testing/python/arith/test_arith_iter_affine_map.py:1075
↓ 3 callersFunctionsparse_mla_bwd
(q, kv, o, do, indices, lse, offsets, sm_scale=None, is_casual=True, return_kernel=False, delta=None)
examples/dsa_sparse_finetune/sparse_mla_bwd.py:254
↓ 3 callersMethodstore_frontend_cache
(self, frontend_key: str, kernel_key: str, *, verbose: bool = False)
tilelang/cache/kernel_cache.py:701
↓ 3 callersFunctionstrided_kernel
(x)
testing/python/language/test_tilelang_language_subtype.py:27
↓ 3 callersFunctiontarget_is_turing
(target: Target)
tilelang/utils/target.py:335
↓ 3 callersFunctiontarget_is_volta
(target: Target)
tilelang/utils/target.py:331
↓ 3 callersFunctiontest_rand_1d
(M, seed)
examples/rand/rand_uint.py:42
↓ 3 callersFunctiontest_sparse_mla_fwd_pipelined
( B=1, S=4096, SKV=8192, H=128, HKV=1, DQK=576, DV=512, topk=2048, dtype=t
examples/deepseek_v32/sparse_mla_fwd_seesaw.py:591
↓ 3 callersFunctiontilelang_copy
(M, N, block_M, block_N, src_dtype=T.float16, dst_dtype=T.float16)
testing/python/language/test_tilelang_language_copy.py:11
↓ 3 callersFunctiontorch_compress
Reference 2:4 sparse compressor in pure PyTorch with natural row-major metadata. Modified from https://github.com/pytorch/pytorch/blob/bfa6895a34
tilelang/utils/sparse.py:191
↓ 3 callersMethodunwrap_value
Unwrap some tilelang objects to get their inner value
tilelang/language/eager/builder.py:511
↓ 3 callersMethodvisit
(self, node: ast.AST, parent: "ast.AST | None", field: "str | None", inside_list: bool)
tilelang/autodd.py:407
↓ 3 callersMethodvisit
(self, node: ast.AST)
tilelang/language/eager/ast.py:618
← previousnext →701–800 of 7,254, ranked by callers