triton.language.tensor_descriptor¶

class triton.language.tensor_descriptor(self, handle, shape: List[tensor], strides: List[tensor], block_type: block_type)¶

A descriptor representing a tensor in global memory.

__init__(self, handle, shape: List[tensor], strides: List[tensor], block_type: block_type)¶: Not called by user code.

Methods

`__init__`(self, handle, shape, strides, ...)	Not called by user code.
`atomic_add`(self, offsets, value[, _semantic])
`atomic_and`(self, offsets, value[, _semantic])
`atomic_max`(self, offsets, value[, _semantic])
`atomic_min`(self, offsets, value[, _semantic])
`atomic_or`(self, offsets, value[, _semantic])
`atomic_xor`(self, offsets, value[, _semantic])
`gather`(self, *args[, _semantic])	Gather multiple descriptors worth of data
`load`(self, offsets[, _semantic])	Load a block from the descriptor starting at the given element offsets.
`scatter`(self, value, *args[, _semantic])	Scatter multiple descriptors worth of data
`store`(self, offsets, value[, _semantic])	Store a block from the descriptor starting at the given element offsets.

Attributes