triton.language.tensor

class triton.language.tensor(handle, type: dtype)

Represents an N-dimensional array of values or pointers.

tensor is the fundamental data structure in Triton programs. Most functions in triton.language operate on and return tensors.

Most of the named member functions here are duplicates of the free functions in triton.language. For example, triton.language.sqrt(x) is equivalent to x.sqrt().

tensor also defines most of the magic/dunder methods, so you can write x+y, x << 2, etc.

Constructors

__init__(handle, type: dtype): Not called by user code.

Methods

`__init__`(handle, type)	Not called by user code.
`abs`([_semantic])	Forwards to `abs()` free function
`advance`(offsets[, _semantic])	Forwards to `advance()` free function
`associative_scan`(axis, combine_fn[, ...])	Forwards to `associative_scan()` free function
`atomic_add`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_add()` free function
`atomic_and`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_and()` free function
`atomic_cas`(cmp, val[, sem, scope, _semantic])	Forwards to `atomic_cas()` free function
`atomic_max`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_max()` free function
`atomic_min`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_min()` free function
`atomic_or`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_or()` free function
`atomic_xchg`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_xchg()` free function
`atomic_xor`(val[, mask, sem, scope, _semantic])	Forwards to `atomic_xor()` free function
`broadcast_to`(*shape[, _semantic])	Forwards to `broadcast_to()` free function
`cast`(dtype[, fp_downcast_rounding, bitcast, ...])	Forwards to `cast()` free function
`ceil`([_semantic])	Forwards to `ceil()` free function
`cos`([_semantic])	Forwards to `cos()` free function
`erf`([_semantic])	Forwards to `erf()` free function
`exp`([_semantic])	Forwards to `exp()` free function
`exp2`([_semantic])	Forwards to `exp2()` free function
`expand_dims`(axis[, _semantic])	Forwards to `expand_dims()` free function
`extract_slice`(offsets, sizes, strides[, ...])	Forwards to `extract_slice()` free function
`floor`([_semantic])	Forwards to `floor()` free function
`gather`(index, axis[, _semantic])	Forwards to `gather()` free function
`gather_out_to_ub`(index, index_boundary, dim, ...)	Forwards to `gather_out_to_ub()` free function
`get_element`(indice[, _semantic, _generator])	Forwards to `get_element()` free function
`histogram`(num_bins[, mask, _semantic, ...])	Forwards to `histogram()` free function
`index_put`(index, value, dim, index_boundary, ...)	Forwards to `index_put()` free function
`index_select_simd`(dim, index, src_shape, ...)	Forwards to `index_select_simd()` free function
`insert_slice`(sub, offsets, sizes, strides[, ...])	Forwards to `insert_slice()` free function
`item`([_semantic, _generator])	Forwards to `item()` free function
`log`([_semantic])	Forwards to `log()` free function
`log2`([_semantic])	Forwards to `log2()` free function
`logical_and`(other[, _semantic])
`logical_or`(other[, _semantic])
`permute`(*dims[, _semantic])	Forwards to `permute()` free function
`reduce`(axis, combine_fn[, keep_dims, ...])	Forwards to `reduce()` free function
`reshape`(*shape[, can_reorder, _semantic, ...])	Forwards to `reshape()` free function
`rsqrt`([_semantic])	Forwards to `rsqrt()` free function
`scatter_ub_to_out`(value, index, ...[, _semantic])	Forwards to `scatter_ub_to_out()` free function
`sin`([_semantic])	Forwards to `sin()` free function
`sort`([dim, descending])
`split`([_semantic, _generator])	Forwards to `split()` free function
`sqrt`([_semantic])	Forwards to `sqrt()` free function
`sqrt_rn`([_semantic])	Forwards to `sqrt_rn()` free function
`store`(value[, mask, boundary_check, ...])	Forwards to `store()` free function
`sync_block_all`(event_id[, _semantic])	Forwards to `sync_block_all()` free function
`sync_block_set`(receiver, event_id[, _semantic])	Forwards to `sync_block_set()` free function
`sync_block_wait`(receiver, event_id[, _semantic])	Forwards to `sync_block_wait()` free function
`to`(dtype[, fp_downcast_rounding, bitcast, ...])	Alias for `tensor.cast()`.
`trans`(*dims[, _semantic])	Forwards to `trans()` free function
`view`(*shape[, _semantic])	Forwards to `view()` free function

Attributes

`T`	Transposes a 2D tensor.
`argmax`
`argmin`
`atan2`
`cdiv`
`cumprod`
`cumsum`
`finitef`
`flip`
`isfinited`
`max`
`min`
`ravel`
`reduce_or`
`sigmoid`
`softmax`
`sum`
`xor_sum`
`type`