MCPcopy Create free account

hub / github.com/NVIDIA/cutlass / functions

Functions19,553 in github.com/NVIDIA/cutlass

↓ 36 callersFunctionis_dynamic_expression
Given the `value`, check if itself is an IR value or recursively go through it to check if it contains IR value
python/CuTeDSL/cutlass/base_dsl/dsl.py:93
↓ 36 callersMethodphase
include/cutlass/pipeline/sm90_pipeline.hpp:193
↓ 36 callersFunctionwarp_uniform
examples/41_fused_multi_head_attention/gemm_kernel_utils.h:233
↓ 35 callersMethod_to_ir
(self)
python/CuTeDSL/cutlass/cute/nvgpu/tcgen05/mma.py:134
↓ 35 callersMethodexec
Requires an active DSL preprocessor session.
python/CuTeDSL/cutlass/base_dsl/ast_preprocessor.py:653
↓ 35 callersMethodinit
include/cutlass/arch/barrier.h:355
↓ 35 callersFunctionis_tuple
(x)
python/pycute/int_tuple.py:47
↓ 35 callersFunctionpartition_shape_C
include/cute/atom/mma_atom.hpp:561
↓ 35 callersMethodraw_data
include/cutlass/array.h:415
↓ 35 callersFunctionstore_with_pointer_offset
Store a fragment to memory
include/cutlass/transform/threadblock/regular_tile_iterator_tensor_op_sm70.h:237
↓ 35 callersFunctiontma_descriptor_replace_dims_strides_in_shared_mem
Replace tensor dims and strides for GEMMs by bringing the tensormap from GMEM into the shared memory
include/cute/arch/copy_sm90_desc.hpp:358
↓ 34 callersFunctiona
(e)
docs/jquery.js:23
↓ 34 callersMethodallocate
Allocate and initialize barrier memory on the GPU. This function allocates device memory for the barrier. It must be called
examples/python/CuTeDSL/dsl_tutorials/cooperative_launch.py:142
↓ 34 callersFunctionapply
include/cute/algorithm/tuple_algorithms.hpp:71
↓ 34 callersFunctionas_numeric
Convert a Python primitive value to a Numeric type. :param obj: Python primitive value to convert :type obj: Union[bool, int, float] :ret
python/CuTeDSL/cutlass/base_dsl/typing.py:1597
↓ 34 callersMethodbitcast
Reinterpret the bits of this tensor as a different element type. Total bit width is preserved; the element count adjusts proportionally.
python/CuTeDSL/cutlass/cute/tensor.py:2448
↓ 34 callersFunctionenum_auto
()
python/cutlass_cppgen/backend/library.py:62
↓ 34 callersMethodload
examples/41_fused_multi_head_attention/gemm/mma_from_smem.h:292
↓ 34 callersFunctionmax
Computes the maximum value from the provided arguments. This function differs from Python's built-in max() in that the return type is determi
python/CuTeDSL/cutlass/cutlass_dsl/cutlass.py:1948
↓ 34 callersFunctionreal
include/cutlass/complex.h:393
↓ 34 callersFunctionsynclog_alloc
include/cutlass/arch/synclog.hpp:65
↓ 34 callersFunctionsynclog_condition_emit
include/cutlass/arch/synclog.hpp:226
↓ 34 callersFunctionsynclog_emit_prefix
include/cutlass/arch/synclog.hpp:77
↓ 34 callersFunctionsynclog_print_prefix
include/cutlass/arch/synclog.hpp:246
↓ 33 callersMethodadvance_to_next_work
include/cutlass/gemm/kernel/sm100_tile_scheduler.hpp:440
↓ 33 callersMethodappend
Appends an operation and takes ownership
tools/library/include/cutlass/library/manifest.h:93
↓ 33 callersFunctionfind_tmem_tensor_col_offset
include/cutlass/detail/sm100_tmem_helper.hpp:48
↓ 33 callersFunctiongenerate_data_types_from_math_instruction
(math_instruction, element_source = None, element_dest = None, element_epilogue = None)
python/cutlass_library/sm90_utils.py:718
↓ 33 callersMethodget_grid_shape
( params: MLAStaticTileSchedulerParams, max_active_clusters: int, *, loc=None,
examples/python/CuTeDSL/cute/blackwell/kernel/attention/mla/mla_helpers.py:209
↓ 33 callersMethodretile_S
include/cute/atom/copy_atom.hpp:386
↓ 33 callersFunctionrun
()
examples/python/CuTeDSL/dsl_tutorials/inline_ptx.py:185
↓ 32 callersFunctionceil_div
include/cute/layout.hpp:1592
↓ 32 callersFunctionflatten
include/cute/layout.hpp:529
↓ 32 callersMethodget_arch_enum
Get the arch enum from the environment variable
python/CuTeDSL/cutlass/base_dsl/dsl.py:2574
↓ 32 callersMethodinit_masks
include/cutlass/pipeline/sm100_pipeline.hpp:152
↓ 32 callersFunctionsum
include/cute/int_tuple.hpp:286
↓ 32 callersFunctiontest_right_inverse
test/unit/cute/core/inverse_right.cpp:47
↓ 31 callersFunctionTensorForEach
< rank of index space
tools/util/include/cutlass/util/reference/host/tensor_foreach.h:93
↓ 31 callersMethodconvert
( self, dQ_acc: cute.Tensor, dQ: cute.Tensor, count: Int32, d_dim: Int
examples/python/CuTeDSL/cute/blackwell/kernel/attention/fmha/fmha_bwd.py:1197
↓ 31 callersMethodis_producer_load_needed
include/cutlass/epilogue/collective/detail.hpp:313
↓ 31 callersMethodstore
Stores a fragment to memory at the location pointed to by the iterator
include/cutlass/gemm/warp/mma_tensor_op_tile_iterator_wmma.h:777
↓ 31 callersMethodwait
Wait for all thread blocks to arrive at the barrier. This method blocks (spins) until all thread blocks have called arrive()
examples/python/CuTeDSL/dsl_tutorials/cooperative_launch.py:411
↓ 30 callersFunctioncomposition
include/cute/layout_composed.hpp:361
↓ 30 callersMethodcreate
( tile_sched_params, bid, grid_dim, cluster_tile_shape_mnk, search_sta
examples/python/CuTeDSL/cute/hopper/kernel/grouped_gemm/grouped_gemm.py:232
↓ 30 callersMethoddivide
Alias for `div` to match the interface of FastDivmodU64
include/cutlass/fast_math.h:396
↓ 30 callersFunctionget
Unpacks an element from memory
include/cutlass/subbyte_reference.h:180
↓ 30 callersMethodinit_masks
examples/77_blackwell_fmha/common/pipeline_mla.hpp:110
↓ 30 callersFunctionmake_coord_tensor
include/cute/tensor_impl.hpp:482
↓ 30 callersFunctionpartition_shape_A
include/cute/atom/mma_atom.hpp:588
↓ 30 callersMethodvalid
Returns true if element is in a defined state
examples/03_visualize_layout/visualize_layout.h:148
↓ 30 callersMethodvalid
Returns whether access is valid or not
examples/41_fused_multi_head_attention/iterators/predicated_tile_access_iterator_residual_last.h:876
↓ 30 callersFunctionweakly_congruent
include/cute/int_tuple.hpp:454
↓ 29 callersMethodconditional_consumer_try_wait
(self, b_consumer_state, b_pipeline, C)
examples/python/CuTeDSL/cute/blackwell/kernel/attention/mamba2_ssd/mamba2_ssd.py:2918
↓ 29 callersMethodget_instantiation_level
(self, pruned_level=0, default_level=111, exhaustive_level=9992)
python/cutlass_library/manifest.py:573
↓ 29 callersMethodkn
Obtains a Coord<2> from GemmCoord
include/cutlass/gemm_coord.h:186
↓ 29 callersFunctionsynclog_setup
include/cutlass/arch/synclog.hpp:279
↓ 29 callersMethodtranspose
examples/13_two_tensor_op_fusion/kernel/b2b_gemm.h:432
↓ 28 callersFunctionCreateConv2dOperator
(manifest, layout, tile_descriptions, data_type, alignment_constraints, \ conv_kinds = [ConvKind.Fprop, Conv
python/cutlass_library/generator.py:573
↓ 28 callersFunctionTensorHash
test/unit/conv/cache_testbed_output.h:540
↓ 28 callersMethod_apply_op
( self, op: Callable, other: object, flip: bool = False, *, lo
python/CuTeDSL/cutlass/cute/tensor.py:1521
↓ 28 callersFunctionadd
examples/44_multi_gemm_ir_and_codegen/leaky_bias.h:36
↓ 28 callersMethodadd_pointer_offset
Adds a pointer offset in units of Element
examples/41_fused_multi_head_attention/iterators/predicated_tile_iterator_residual_last.h:814
↓ 28 callersFunctionbasis_get
include/cute/numeric/arithmetic_tuple.hpp:262
↓ 28 callersMethodclone
Create a new Producer instance with the same state.
python/CuTeDSL/cutlass/pipeline/sm90.py:973
↓ 28 callersFunctioncompute_epilogue
Returns whether the block assigned this work should compute the epilogue for the corresponding output tile. For the case of stream-K, this should only
include/cutlass/gemm/kernel/sm90_tile_scheduler_stream_k.hpp:583
↓ 28 callersFunctionconst_expr
This function is used to check if the expression is a python value. If the expression is a python value, return the boolean value of the expr
python/CuTeDSL/cutlass/base_dsl/ast_helpers.py:401
↓ 28 callersFunctionconst_max
include/cutlass/fast_math.h:757
↓ 28 callersMethodcount
include/cutlass/epilogue/threadblock/predicated_tile_iterator_params.h:84
↓ 28 callersFunctioncp_async_wait<0>
include/cutlass/arch/memory_sm80.h:461
↓ 28 callersMethodget
(self)
python/cutlass_cppgen/backend/compiler.py:102
↓ 28 callersMethodget_hier_coord
include/cute/layout.hpp:250
↓ 28 callersMethodmake_fragment_A
include/cute/atom/mma_atom.hpp:146
↓ 28 callersFunctionpartition_shape_B
include/cute/atom/mma_atom.hpp:601
↓ 28 callersFunctionprefetch_tma_descriptors
Issue Tma Descriptor Prefetch -- ideally from a single thread for best performance
include/cutlass/gemm/collective/sm103_blockscaled_mma_warpspecialized.hpp:546
↓ 28 callersMethodrank
Return the rank of the tensor.
python/CuTeDSL/cutlass/base_dsl/runtime/tensor_descriptor.py:142
↓ 28 callersMethodverify
Verify the results
test/python/cutlass/evt/utils/evt_testbed.py:140
↓ 27 callersFunction_enhance_enum_with_str_mapping
Enhance an IntEnum class with automatic string-to-enum conversion. Builds a reverse mapping from __str__() output to enum members and adds
python/CuTeDSL/cutlass/cute/arch/nvvm_wrappers.py:57
↓ 27 callersMethod_to_ir
(self)
python/CuTeDSL/cutlass/cute/nvgpu/common.py:255
↓ 27 callersMethoddevice_view
Accesses the tensor reference pointing to data
tools/util/include/cutlass/util/host_tensor.h:324
↓ 27 callersFunctionget_cutlass_status
//////////////////////////////////////////////////////////////////////////////////////////// Converts a cuDNN status to cutlass::Status
tools/profiler/src/cudnn_helpers.cpp:45
↓ 27 callersMethodget_problem_sizes
(self, alignment, k=None, batch_count=[3,])
test/python/cutlass/evt/utils/evt_testbed.py:221
↓ 27 callersMethodinitial_work_tile_info
include/cutlass/gemm/kernel/sm90_tile_scheduler_group.hpp:614
↓ 27 callersFunctionmake_tiled_copy_S
include/cute/atom/copy_atom.hpp:573
↓ 27 callersMethodset_kgroup_index
Notify the iterator which k-group it is currently pointing to. This does not advance the iterator. Rather, it overrides its internal tracking with co
include/cutlass/conv/warp/mma_depthwise_simt_tile_iterator.h:244
↓ 26 callersFunctionf
(e)
docs/jquery.js:23
↓ 26 callersMethodis_m_major_c
(self)
python/CuTeDSL/cutlass/utils/layout.py:47
↓ 26 callersFunctionmake_identity_layout
include/cute/layout.hpp:483
↓ 26 callersFunctionmake_tiled_copy_D
include/cute/atom/copy_atom.hpp:584
↓ 26 callersMethodset_iteration_index
Overrides the internal iteration index
include/cutlass/transform/threadblock/predicated_tile_access_iterator.h:238
↓ 26 callersMethodset_k_partition
Functionally required for serial reduction in the epilogue
include/cutlass/epilogue/thread/conversion_op.h:97
↓ 26 callersFunctionstore_with_pointer_offset
Stores a fragment to memory at the location pointed to by the iterator
include/cutlass/gemm/warp/mma_simt_tile_iterator.h:507
↓ 26 callersFunctiontest_left_inverse
test/unit/cute/core/inverse_left.cpp:47
↓ 26 callersFunctiontiled_product
include/cute/layout.hpp:1696
↓ 26 callersFunctiontransform_apply
include/cute/algorithm/tuple_algorithms.hpp:125
↓ 26 callersMethoduniform
Constructs a matrix from a uniform element
include/cutlass/matrix.h:114
↓ 25 callersFunctionadd
include/cute/numeric/real.hpp:40
↓ 25 callersMethodarrive
Arrive at the barrier without waiting. This signals that the calling thread block has reached the barrier point, but does no
examples/python/CuTeDSL/dsl_tutorials/cooperative_launch.py:252
↓ 25 callersFunctionblock_id_in_cluster
Returns the relative dim3 block rank local to the cluster.
include/cute/arch/cluster_sm90.hpp:126
↓ 25 callersFunctioncheck_contiguous_16B_alignment
( dtype: type[cutlass.Numeric], is_mode0_major: bool, tensor_shape: tuple[int, int],
python/CuTeDSL/cutlass/utils/mixed_input_helpers.py:714
↓ 25 callersFunctionlazy_import
(mod_name: str)
python/cutlass_cppgen/utils/lazy_import.py:35
← previousnext →401–500 of 19,553, ranked by callers