MCPcopy Create free account

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

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

Functionsplitk_gemv
( A, B, BLOCK_N: int, BLOCK_K: int, reduce_threads: int, dtype: T.dtype = T.float16,
examples/gemv/example_gemv.py:80
Functionsplitk_gemv_vectorized
( A, B, BLOCK_N: int, reduce_threads: int, dtype: T.dtype = T.float16, accum_dtype: T.
examples/gemv/example_gemv.py:119
Functionsplitk_gemv_vectorized_tvm
( A, B, BLOCK_N: int, reduce_threads: int, dtype: T.dtype = T.float16, accum_dtype: T.
examples/gemv/example_gemv.py:159
Functionsqrt
(x: Union[TensorSSA, Numeric], fastmath: bool = False, **kwargs)
tilelang/contrib/cutedsl/math.py:114
Functionsqrt
Take square root of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr
tilelang/language/tir/op.py:2580
Methodsrcpath
Returns the source path of the compiled library.
tilelang/jit/adapter/cython/adapter.py:368
Functionstart_profile_intrinsic
Start profile intrinsic. Parameters ---------- id : int The intrinsic id. Returns ------- call : PrimExpr The
tilelang/language/tir/op.py:387
Functionstg128
Store 128 bits (16 bytes) to global memory using explicit PTX instructions. Usage: `T.stg128(y[i:i+4], value)` emits `tl::stg128(ptr, value)`.
tilelang/language/builtin.py:1655
Functionstg256
Store 256 bits (32 bytes) to global memory using explicit PTX instructions. Usage: `T.stg256(y[i:i+8], value)` emits `tl::stg256(ptr, value)`.
tilelang/language/builtin.py:1678
Functionstg32
Store 32 bits (4 bytes) to global memory using explicit PTX instructions. Usage: `T.stg32(y[i], value)` emits `tl::stg32(ptr, value)`. Args:
tilelang/language/builtin.py:1609
Functionstg64
Store 64 bits (8 bytes) to global memory using explicit PTX instructions. Usage: `T.stg64(y[i:i+2], value)` emits `tl::stg64(ptr, value)`. A
tilelang/language/builtin.py:1632
Methodstmatrix
(self, C_local_buf, C_buf, pid_m=None, pid_n=None)
tilelang/rocm/intrinsics/mfma_macro_generator.py:508
Methodstmatrix
(self, C_local_buf, C_buf, pid_m=None, pid_n=None)
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:565
Methodstride_A
(self)
tilelang/tileop/gemm/gemm_base.py:136
Methodstride_A
(self)
tilelang/tileop/gemm/__init__.py:86
Methodstride_A
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:108
Methodstride_B
(self)
tilelang/tileop/gemm/gemm_base.py:140
Methodstride_B
(self)
tilelang/tileop/gemm/__init__.py:90
Methodstride_B
(self)
tilelang/tileop/gemm_sp/gemm_sp_base.py:112
Methodstrides
(self)
tilelang/language/eager/builder.py:148
Functionsub2
Packed element-wise subtract (x - y).
tilelang/language/math_intrinsics.py:366
Functionsubstitute_primfunc
(prim_func, vmap)
tilelang/language/eager/builder.py:1249
Methodsupply_prog
(self, params)
examples/deepseek_v32/fp8_lighting_indexer.py:68
Methodsupply_prog
(device, _frozen_inputs=frozen_inputs, _cached_tensors_by_device=cached_tensors_by_device)
tilelang/autotuner/tuner.py:374
Functionsupply_tensors_gpu
Supply function that creates tensors on GPU for ROCm/HIP.
examples/amd/example_amd_flash_attn_fwd.py:21
Methodswizzle_atom_size
(self)
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:78
Functionsymbolic
Deprecated alias for `T.dynamic`.
tilelang/language/symbolics.py:34
Functionsync_global
Synchronize all threads in the entire grid.
tilelang/language/builtin.py:1153
Functionsync_grid
Synchronize all thread blocks in a grid. NOTE: This requires the kernel to be launched with cuLaunchCooperativeKernel to guarantee all blocks
tilelang/contrib/cutedsl/grid_sync.py:75
Functionsync_grid
Synchronize all threads in a grid.
tilelang/language/builtin.py:1162
Functionsync_thread_partial
(barrier_id=None, thread_count=None)
tilelang/contrib/cutedsl/utils.py:132
Functionsync_threads
Synchronize all threads in a block.
tilelang/language/builtin.py:954
Functionsync_warp
Synchronize all threads in a warp.
tilelang/language/builtin.py:993
Functionsyncthreads_and
Block barrier that returns non-zero only if ALL threads have a non-zero ``predicate`` (``__syncthreads_and`` on CUDA and HIP).
tilelang/language/builtin.py:1109
Functionsyncthreads_count
Block barrier that returns the number of threads whose ``predicate`` evaluates to non-zero (``__syncthreads_count`` on CUDA and HIP).
tilelang/language/builtin.py:1102
Functionsyncthreads_or
Block barrier that returns non-zero if ANY thread has a non-zero ``predicate`` (``__syncthreads_or`` on CUDA and HIP).
tilelang/language/builtin.py:1116
Functiontan
(x: Union[TensorSSA, Numeric], fastmath: bool = False, **kwargs)
tilelang/contrib/cutedsl/math.py:102
Functiontan
Take tan of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr The res
tilelang/language/tir/op.py:2385
Functiontanh
(x: Union[TensorSSA, Numeric], fastmath: bool = False)
tilelang/contrib/cutedsl/math.py:172
Functiontanh
Take hyperbolic tanh of input x. Parameters ---------- x : PrimExpr Input argument. Returns ------- y : PrimExpr
tilelang/language/tir/op.py:2289
Functiontarget
Create a target Parameters ---------- target_config : Union[Dict, str] The target configuration. host : Optional[Union[
tilelang/language/ast/ir.py:1676
Functiontarget_get_warp_size
(target: Target)
tilelang/utils/target.py:379
Functiontarget_has_async_copy
(target: Target)
tilelang/utils/target.py:363
Functiontarget_has_bulk_copy
(target: Target)
tilelang/utils/target.py:375
Functiontarget_has_ldmatrix
(target: Target)
tilelang/utils/target.py:367
Functiontarget_has_stmatrix
(target: Target)
tilelang/utils/target.py:371
Functiontarget_is_ampere
(target: Target)
tilelang/utils/target.py:339
Functiontarget_is_hopper
(target: Target)
tilelang/utils/target.py:343
Functiontarget_is_sm120
(target: Target)
tilelang/utils/target.py:347
Methodtask_generator
(self)
tilelang/autodd.py:609
Methodtask_generator
(self)
tilelang/autodd.py:635
Methodtask_generator
(self)
tilelang/autodd.py:696
Methodtask_update
(self, task: Task, is_interesting: bool)
tilelang/autodd.py:612
Methodtask_update
(self, task: Task, is_interesting: bool)
tilelang/autodd.py:648
Methodtask_update
(self, task: Task, is_interesting: bool)
tilelang/autodd.py:708
Functiontcgen05_after_thread_sync
()
tilelang/language/builtin.py:1285
Functiontcgen05_before_thread_sync
()
tilelang/language/builtin.py:1281
Functiontcgen05_cp_warpx4
Copy one or more packed scale-factor chunks from shared memory to tensor memory. The helper lowers to one or more ``tcgen05.cp.cta_group::{1,2}.3
tilelang/language/builtin.py:1312
Functiontcgen05_gemm
Explicit Blackwell TCGEN05 GEMM without an implicit wait. This is the explicit asynchronous Blackwell TCGEN5MMA counterpart to the default sy
tilelang/language/gemm_op.py:236
Functiontcgen05_gemm_blockscaled
Explicit Blackwell TCGEN05 block-scaled GEMM without an implicit wait. This is the explicit asynchronous Blackwell TCGEN5MMA block-scaled cou
tilelang/language/gemm_op.py:281
Functiontcgen05_gemm_sp
Explicit Blackwell TCGEN05 sparse GEMM without an implicit wait. This is the explicit asynchronous Blackwell TCGEN05 counterpart to the defau
tilelang/language/experimental/gemm_sp_op.py:224
Functiontcgen05_ld_32dp128bNx
Load from TMEM using 32dp128b pattern (2x 16x128b for lower/upper 16 rows). Matches tl::tmem_ld_32dp128bNx from tcgen_05_ld.h. N: x-count for
tilelang/contrib/cutedsl/gemm_tcgen05.py:558
Functiontcgen05_ld_32dp256bNx
Load from TMEM using 32dp256b pattern (2x 16x256b for lower/upper 16 rows). Matches tl::tmem_ld_32dp256bNx from tcgen_05_ld.h. N: x-count for
tilelang/contrib/cutedsl/gemm_tcgen05.py:578
Functiontcgen05_ld_32dp32bNx
Load N uint32 values from TMEM using tcgen05.ld.sync.aligned.32x32b. Matches tl::tcgen05_ld_32dp32bNx from copy_sm100.h. N: number of 32-bit
tilelang/contrib/cutedsl/gemm_tcgen05.py:523
Functiontcgen05_ld_32dp64bNx
Load from TMEM using 32dp64b pattern (2x 16x64b for lower/upper 16 rows). Matches tl::tmem_ld_32dp64bNx from tcgen_05_ld.h. N: x-count for 16
tilelang/contrib/cutedsl/gemm_tcgen05.py:540
Methodtcgen05_meta_unpacked
Return ``(atom_m, atom_n, atom_k, enable_ws, enable_2cta)`` as ints. Requires ``self.meta`` to have been set via ``get_tcgen5_mma_meta()``.
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:605
Functiontcgen05_mma_arrive
tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [mbar]; Guarded by elect_one_sync — only one thread in the warp issues the
tilelang/contrib/cutedsl/gemm_tcgen05.py:325
Functiontcgen05_mma_arrive
Signal UMMA (TCGEN05) barrier arrival for a shared-memory mbarrier pointer. Parameters ---------- mbar: tirx.Buffer | BufferLoad | PrimEx
tilelang/language/builtin.py:1264
Methodtcgen05_num_inst_m
Number of TCGEN05MMA instruction atoms along M (SS variant).
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:614
Methodtcgen05_num_inst_n
Number of TCGEN05MMA instruction atoms along N.
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:621
Methodtcgen05_num_k_atoms
Number of K-dimension micro-steps (``chunk // micro_size_k``).
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:627
Functiontcgen05_sf_warp_transpose
Warp-level transpose for one or more packed scale-factor chunks in shared memory. For 1D packed ``uint32`` scale buffers, the helper automaticall
tilelang/language/builtin.py:1345
Functiontcgen05mma_ss
tcgen05.mma.cta_group::1.kind::{kind} [tmem_c], desc_a, desc_b, desc_val, {masks}, p; Guarded by elect_one_sync — only one thread in the warp iss
tilelang/contrib/cutedsl/gemm_tcgen05.py:133
Functiontcgen05mma_ts
tcgen05.mma.cta_group::1.kind::{kind} [tmem_c], [tmem_a], desc_b, desc_val, {masks}, p;
tilelang/contrib/cutedsl/gemm_tcgen05.py:255
Functiontcgen05mma_ws_ss
tcgen05.mma.ws.cta_group::1.kind::{kind} [tmem_c], desc_a, desc_b, desc_val, p, 0;
tilelang/contrib/cutedsl/gemm_tcgen05.py:209
Functiontemporary_enable_log_propagate
()
examples/bitnet-1.58b/vllm_workspace/conftest.py:566
Functiontensor_cache
A decorator that caches the most recent result of a function with tensor inputs. This decorator will store the output of the decorated funct
examples/deepseek_v32/utils.py:32
Functiontensor_cache
A decorator that caches the most recent result of a function with tensor inputs. This decorator will store the output of the decorated funct
examples/dsa_sparse_finetune/index.py:8
Functiontensor_cache
A decorator that caches the most recent result of a function with tensor inputs. This decorator will store the output of the decorated funct
examples/kda/FLA_KDA/fla_utils.py:58
Functiontensor_cache
A decorator that caches the most recent result of a function with tensor inputs. This decorator will store the output of the decorated funct
examples/dsa_hisa/tilelang_utils.py:7
Functiontest
(A)
testing/python/issue/test_tilelang_issue_1690.py:8
Methodtest_2d_output_layout_different_order
Layouts with swapped output dimensions should not be equal.
testing/python/layout/test_tilelang_layout_equal.py:50
Methodtest_2d_output_layout_equal
Layouts with 2D output should compare correctly.
testing/python/layout/test_tilelang_layout_equal.py:44
Functiontest_abs_maps_to_fabs
Test that abs correctly maps to fabs
testing/python/fastmath/test_mathops_fastmath.py:311
Functiontest_abs_maps_to_fabs
Test that abs correctly maps to fabs
testing/python/math/test_math_fast_math.py:332
Functiontest_access_ptr_builds_tl_access_ptr_from_bufferload_1d
()
testing/python/language/test_tilelang_language_access_ptr.py:10
Functiontest_access_ptr_cp_async_codegen
Smoke-test codegen for T.access_ptr -> tl.access_ptr -> tvm_access_ptr -> cp.async.
testing/python/language/test_tilelang_language_access_ptr_codegen.py:9
Functiontest_access_ptr_defaults_to_element_extent_for_bufferload
()
testing/python/language/test_tilelang_language_access_ptr.py:27
Functiontest_access_ptr_multiplies_extents_for_2d_load
()
testing/python/language/test_tilelang_language_access_ptr.py:42
Functiontest_activemask
()
testing/python/language/test_tilelang_language_warp_vote.py:207
Functiontest_add_sub
()
testing/python/arith/test_arith_intset.py:65
Functiontest_all_sync
()
testing/python/language/test_tilelang_language_warp_vote.py:76
Functiontest_alloc_global
()
testing/python/language/test_tilelang_language_alloc.py:234
Functiontest_alloc_multi_vars_with_initializer
()
testing/python/language/test_tilelang_language_alloc.py:162
Functiontest_alloc_var
()
testing/python/language/test_tilelang_language_alloc.py:40
Functiontest_alloc_var_add
()
testing/python/language/test_tilelang_language_alloc.py:80
Functiontest_alloc_var_eq
()
testing/python/language/test_tilelang_language_alloc_var_cmp.py:44
Functiontest_alloc_var_inf_init
alloc_var(init=-T.infinity(T.float32)) — the pattern used for top1_var / top2_var in MoE topk gate kernels — must produce -inf on HIP.
testing/python/amd/test_tilelang_hip_codegen.py:428
Functiontest_alloc_var_init_correctness
All output elements must equal 7 — the initialised value.
testing/python/amd/test_tilelang_hip_codegen.py:310
Functiontest_alloc_var_init_in_hip_source
Init value must appear as `= 7;` in the generated HIP source.
testing/python/amd/test_tilelang_hip_codegen.py:292
← previousnext →5,801–5,900 of 7,254, ranked by callers