MCPcopy Create free account

hub / github.com/NVIDIA/cutlass / functions

Functions19,553 in github.com/NVIDIA/cutlass

⨍ Functions 19,553◇ Types & classes 10,397↳ Endpoints 26
↓ 89 callersMethodlayout
include/cutlass/epilogue/threadblock/output_iterator_parameter.h:74
↓ 87 callersFunctionflat_divide
include/cute/layout.hpp:1633
↓ 87 callersMethodmake_fragment_C
( self, input: Any, *, loc: Optional[ir.Location] = None, ip: Optional
python/CuTeDSL/cutlass/cute/atom.py:399
↓ 87 callersMethodparse
(self, example_inputs)
python/cutlass_cppgen/epilogue/epilogue.py:167
↓ 85 callersMethodpartition_shape_C
( self, shape_mn: Shape, *, loc: Optional[ir.Location] = None, ip: Opt
python/CuTeDSL/cutlass/cute/atom.py:556
↓ 85 callersMethodstore
( self, gmem: cute.Tensor, regs: cute.Tensor, coord: cute.Tensor, tens
examples/python/CuTeDSL/cute/blackwell/kernel/attention/fmha/fmha_bwd.py:2360
↓ 84 callersMethodcompile
Compile with all the arguments to know the types during compilation while hiding the epilogue detail1s.
examples/python/CuTeDSL/cute/blackwell/efc/common_efc.py:1422
↓ 84 callersMethodgemm
Executes one GEMM
include/cutlass/gemm/kernel/gemm_universal_streamk.h:1002
↓ 84 callersMethodretile
( self, src: Any, *, loc: Optional[ir.Location] = None, ip: Optional[i
python/CuTeDSL/cutlass/cute/atom.py:879
↓ 84 callersFunctionto_grouped_schedule
(schedule, grouped)
python/cutlass_library/library.py:1003
↓ 84 callersFunctionzipped_divide
include/cute/layout.hpp:1608
↓ 83 callersMethodget_cmd_line_argument
tools/util/include/cutlass/util/command_line.h:135
↓ 83 callersMethodinitial_work_tile_info
(self, *, loc=None, ip=None)
examples/python/CuTeDSL/cute/blackwell/kernel/attention/mla/mla_helpers.py:261
↓ 83 callersMethodmma_major_mode
(self)
python/CuTeDSL/cutlass/utils/layout.py:22
↓ 83 callersMethodproducer_try_acquire
include/cutlass/pipeline/sm90_pipeline.hpp:1105
↓ 82 callersFunctioncomposition
include/cute/layout.hpp:1134
↓ 82 callersFunctionmma
include/cutlass/gemm/collective/sm100_mma_array_warpspecialized.hpp:693
↓ 82 callersMethodview
Returns a TensorRef pointing to the first element of the tensor.
include/cutlass/thread/matrix.h:152
↓ 81 callersFunctioninitialize_tensor
examples/cute/tutorial/blackwell/example_utils.hpp:97
↓ 81 callersFunctionselect_cluster_shape
include/cutlass/detail/cluster.hpp:49
↓ 80 callersMethodtensor
include/cute/tensor_impl.hpp:162
↓ 79 callersMethodconsumer_try_wait
examples/112_blackwell_ssd/utils/pipeline.h:163
↓ 79 callersFunctioncopy_if
include/cute/algorithm/copy.hpp:47
↓ 79 callersMethodget_tile_offset
Obtains the threadblock offset (in units of threadblock-scoped tiles)
include/cutlass/gemm/threadblock/threadblock_swizzle.h:131
↓ 78 callersFunctiontile_to_mma_shape
include/cute/atom/mma_traits_sm100.hpp:114
↓ 77 callersFunctionTensorEquals
< Layout function
tools/util/include/cutlass/util/reference/host/tensor_compare.h:344
↓ 77 callersFunctionb
(e)
docs/jquery.js:32
↓ 77 callersFunctionmin
Computes the minimum value from the provided arguments. This function differs from Python's built-in min() in that the return type is determi
python/CuTeDSL/cutlass/cutlass_dsl/cutlass.py:1885
↓ 77 callersMethodreshape
Reshape the tensor to a new shape. :param shape: The new shape to reshape to. :type shape: Shape :return: A new tensor with t
python/CuTeDSL/cutlass/cute/tensor.py:2261
↓ 76 callersFunctionget_grid_shape
include/cutlass/gemm/kernel/sm90_tile_scheduler_stream_k.hpp:360
↓ 76 callersMethodlayout_b
include/cute/layout_composed.hpp:83
↓ 76 callersMethodretile_D
include/cute/atom/copy_atom.hpp:395
↓ 76 callersFunctionshape_div
include/cute/int_tuple.hpp:384
↓ 75 callersFunctionget_workspace_size
test/unit/gemm/device/testbed_grouped_scheduler.h:111
↓ 75 callersFunctionpartition_fragment_C
include/cute/atom/mma_atom.hpp:575
↓ 74 callersFunctionCreateGemmUniversal3xOperator
( manifest, layouts, tile_descriptions, data_types, schedules = [[KernelScheduleType.ScheduleAuto, Epi
python/cutlass_library/generator.py:181
↓ 74 callersMethodlayout
(self)
examples/python/CuTeDSL/cute/blackwell/kernel/blockscaled_gemm/sm103_dense_blockscaled_gemm_persistent.py:1958
↓ 73 callersFunctionTensorFillRandomUniform
< Layout function
tools/util/include/cutlass/util/reference/host/tensor_fill.h:1028
↓ 73 callersMethodcreate
( params: Mamba2SSDTileSchedulerParams, block_idx: Tuple[Integer, Integer, Integer], g
examples/python/CuTeDSL/cute/blackwell/kernel/attention/mamba2_ssd/mamba2_ssd_tile_scheduler.py:125
↓ 73 callersMethodwait
include/cutlass/arch/barrier.h:370
↓ 73 callersMethodwait_for_alloc
Wait for the allocator warp to finish allocation. This method is used to synchronize the allocator warp with the other warps before retrievin
python/CuTeDSL/cutlass/utils/tmem_allocator.py:453
↓ 72 callersMethodadd_pointer_offset
Adds a pointer offset in units of Element
include/cutlass/conv/threadblock/conv2d_tile_iterator.h:114
↓ 72 callersFunctionrank
Returns the rank (dimensionality) of a tuple, layout, or tensor. The rank of a tuple is its length. For layouts and tensors, the rank is dete
python/CuTeDSL/cutlass/cute/core.py:2297
↓ 72 callersMethodretrieve_ptr
Retrieve the pointer to the allocated tensor memory. This method can be called by all warps after allocation has been performed by th
python/CuTeDSL/cutlass/utils/tmem_allocator.py:466
↓ 71 callersMethodload
Load addend source fragment from global memory
include/cutlass/epilogue/threadblock/epilogue.h:271
↓ 70 callersMethodmake_participants
( self, *, loc: Optional[ir.Location] = None, ip: Optional[ir.InsertionPoint]
python/CuTeDSL/cutlass/pipeline/sm90.py:359
↓ 69 callersFunctionrotl
include/cute/numeric/math.hpp:205
↓ 69 callersFunctionswap
include/cutlass/platform/platform.h:838
↓ 69 callersMethodvalid
Returns whether access is valid or not
include/cutlass/transform/threadblock/predicated_tile_access_iterator.h:295
↓ 68 callersMethodat
Returns a reference to the element at a given Coord
include/cutlass/thread/matrix.h:164
↓ 68 callersFunctiondomain_offset
include/cute/layout.hpp:717
↓ 68 callersFunctionfma
include/cute/arch/simd_sm100.hpp:77
↓ 68 callersMethodget
Get the context as a list of strings for error messages. :returns: Context strings like ["on argument ", "#0"] or ["on my_tuple[0][1] in argu
python/CuTeDSL/cutlass/base_dsl/tvm_ffi_builder/tvm_ffi_builder.py:46
↓ 68 callersFunctiontest_composition
test/unit/cute/core/composition.cpp:49
↓ 67 callersMethodclear
examples/41_fused_multi_head_attention/kernel_backward.h:1191
↓ 67 callersMethodload
(self)
examples/python/CuTeDSL/cute/blackwell/efc/common_efc.py:1039
↓ 67 callersFunctionmake_tmem_copy
include/cute/atom/copy_traits_sm100.hpp:286
↓ 67 callersMethodproducer_commit
include/cutlass/pipeline/sm90_pipeline.hpp:1114
↓ 66 callersFunctionBlockFillSequential
tools/util/include/cutlass/util/reference/host/tensor_fill.h:1508
↓ 66 callersMethodvisit
test/unit/gemm/device/gemm_testbed_3x_evt.hpp:406
↓ 66 callersFunctionwarpgroup_commit_batch
Marks the commit point for one or more sized batch of warpgroup MMAs.
include/cute/arch/mma_sm90_gmma.hpp:74
↓ 66 callersMethodwork_tile_to_cta_coord
include/cutlass/gemm/kernel/sm100_tile_scheduler.hpp:373
↓ 65 callersFunctionwait_on_dependent_grids
Issuing the griddepcontrol.wait instruction enforces no global memory access prior to this istruction. This ensures the correctness of global memory a
include/cutlass/arch/grid_dependency_control.h:94
↓ 65 callersFunctionwarpgroup_arrive
include/cute/arch/mma_sm90_gmma.hpp:47
↓ 64 callersMethodallocate
( self, size_or_type: int, byte_alignment: int, *, loc: Optional[ir.Lo
python/CuTeDSL/cutlass/utils/smem_allocator.py:120
↓ 64 callersFunctionclear
include/cute/container/array.hpp:352
↓ 64 callersMethodproducer_get_barrier
include/cutlass/pipeline/sm90_pipeline.hpp:1136
↓ 64 callersFunctionraw_pointer_cast
include/cute/pointer.hpp:345
↓ 64 callersMethodtransform
Transform the mma operands to the required types
include/cutlass/gemm/warp/mma_simt.h:251
↓ 63 callersFunctionceil_div
examples/41_fused_multi_head_attention/gemm_kernel_utils.h:136
↓ 63 callersMethodclear_mask
Clears the predicate set efficiently
examples/41_fused_multi_head_attention/iterators/predicated_tile_iterator_residual_last.h:845
↓ 63 callersFunctiontest_evenly_divides
test/unit/cute/core/int_tuple.cpp:87
↓ 62 callersFunctiontest_complement
test/unit/cute/core/complement.cpp:41
↓ 61 callersFunctionc
(g,e)
docs/jquery.js:32
↓ 61 callersMethodend
Returns a const iterator
tools/library/src/manifest.cpp:91
↓ 60 callersFunctioninitialize_workspace
include/cutlass/gemm/kernel/sm90_tile_scheduler.hpp:145
↓ 60 callersFunctionmax
(x, dim)
python/cutlass_cppgen/epilogue/evt_ops.py:56
↓ 59 callersFunctionfence_view_async_shared
Issue a shared memory fence for async operations
include/cutlass/arch/barrier.h:727
↓ 59 callersFunctionmul
include/cute/numeric/real.hpp:49
↓ 59 callersFunctionreverse
include/cute/algorithm/tuple_algorithms.hpp:1042
↓ 58 callersMethodpush
include/cutlass/epilogue/threadblock/epilogue.h:453
↓ 58 callersMethodvalues
include/cute/container/tuple.hpp:486
↓ 57 callersFunctionTensorFillIdentity
< Layout function
tools/util/include/cutlass/util/reference/host/tensor_fill.h:1272
↓ 57 callersMethodadvance
include/cutlass/gemm/kernel/grouped_problem_visitor.h:149
↓ 57 callersMethodclear
include/cute/container/array.hpp:179
↓ 57 callersFunctionconst_min
include/cutlass/fast_math.h:752
↓ 57 callersMethodequal
Create an equal comparison between two values.
python/CuTeDSL/cutlass/base_dsl/tvm_ffi_builder/mlir_builder.py:182
↓ 56 callersFunctionepilogue
Epilogue function that stores accumulator results directly to global memory. Used when TMA store is not enabled. :param gemm_kernel: The
python/CuTeDSL/cutlass/utils/gemm/sm100.py:298
↓ 56 callersFunctionfilter
include/cute/layout.hpp:934
↓ 56 callersFunctionmake_tiled_copy
include/cute/atom/copy_atom.hpp:493
↓ 56 callersMethodmark_compact_shape_dynamic
Implemented by subclasses.
python/CuTeDSL/cutlass/cute/typing.py:685
↓ 56 callersMethodstride
include/cute/tensor_impl.hpp:210
↓ 55 callersMethodresize
Changes the size of the view without affecting pointer or layout
include/cutlass/tensor_view.h:170
↓ 54 callersMethodcopy_from_host
tools/util/include/cutlass/util/device_memory.h:344
↓ 54 callersFunctionsync
include/cutlass/arch/barrier.h:233
↓ 54 callersMethodsynchronize
include/cutlass/transform/collective/sm90_wgmma_transpose.hpp:124
↓ 53 callersFunctionCreateGemmOperator
(manifest, layouts, tile_descriptions, data_type, \ alignment_constraints, complex_transforms = None, epilog
python/cutlass_library/generator.py:142
↓ 53 callersMethodbenchmark
test/unit/transform/device/testbed_sparse_gemm_compressor.hpp:816
↓ 53 callersFunctioncast
Cast an object to the specified numeric type. :param obj: Object to be cast :type obj: Union[bool, int, float, Value, Numeric] :param typ
python/CuTeDSL/cutlass/base_dsl/typing.py:466
↓ 53 callersFunctionget_work_k_tile_count
include/cutlass/gemm/kernel/sm90_tile_scheduler_stream_k.hpp:692
← previousnext →201–300 of 19,553, ranked by callers