Module burn_wgpu::kernel::reduce

Reduction algorithms

Structs

• MeanDimAutotuneOperationSet
  Set of mean_dim implementations available for autotune Autotune key is given by concatenating the closest upper power of 2 of dim to reduce, and product of others
• RecursiveSumRaw
  Generated kernel from wgsl file.
• ReduceAutotuneKey
  Autotune key representative of reduce versions
• ReductionArgsRaw
  Generated kernel from wgsl file.
• ReductionDimRaw
  Generated kernel from wgsl file.
• ReductionDimSharedMemoryRaw
  Generated kernel from wgsl file.
• SumDimAutotuneOperationSet
  Set of sum_dim implementations available for autotune Autotune key is given by concatenating the closest upper power of 2 of dim to reduce, and product of others

Functions

• argmax
  Execute the argmax kernel.
• argmin
  Execute the argmin kernel.
• init_reduce_output
  Creates an empty output tensor with reduce output shape
• mean_dim
  Execute the mean dim kernel.
• mean_dim_autotune
  Executes autotune on mean_dim operation
• mean_dim_shared_memory
  Execute the mean dim kernel leveraging shared memory Probably more efficient on tensors where the dimension to reduced is much larger than the others
• sum
  Sum all elements in the input buffer.
• sum_dim
  Execute the sum dim kernel.
• sum_dim_autotune
  Executes autotune on sum_dim operation
• sum_dim_shared_memory
  Execute the sum dim kernel leveraging shared memory Probably more efficient on tensors where the dimension to reduced is much larger than the others