cuda.tile.min

cuda.tile.min(x, /, axis=None, *, keepdims=False, flush_to_zero=False)

Performs min reduction on tile along the axis.

Parameters:

• x (Tile) – input tile.

• axis (None | const int | tuple[const int,...]) – the axis for reduction. The default, axis=None, will reduce all of the elements. For argmin and argmax, tuple of axis is not supported.

• keepdims (const bool) – If true, preserves the number of dimension from the input tile. flush_to_zero (const bool): If True, flushes subnormal inputs and results to sign-preserving zero, default is False.

Returns:

Examples

Reduce all axes.

SnippetComplete Example

tx = ct.arange(8, dtype=ct.int32).reshape((2, 4))
print("input:", tx)
print("reduced:", ct.min(tx, None))

Copy

import cuda.tile as ct
import torch

@ct.kernel
def kernel():
    tx = ct.arange(8, dtype=ct.int32).reshape((2, 4))
    print("input:", tx)
    print("reduced:", ct.min(tx, None))


torch.cuda.init()
ct.launch(torch.cuda.current_stream(), (1,), kernel, ())
torch.cuda.synchronize()

Output

input: [[0, 1, 2, 3], [4, 5, 6, 7]]
reduced: 0

Reduce axis 1 and keepdims.

SnippetComplete Example

tx = ct.arange(8, dtype=ct.int32).reshape((2, 4))
print("input:", tx)
print("reduced:", ct.min(tx, 1, keepdims=True))

Copy

import cuda.tile as ct
import torch

@ct.kernel
def kernel():
    tx = ct.arange(8, dtype=ct.int32).reshape((2, 4))
    print("input:", tx)
    print("reduced:", ct.min(tx, 1, keepdims=True))


torch.cuda.init()
ct.launch(torch.cuda.current_stream(), (1,), kernel, ())
torch.cuda.synchronize()

Output

input: [[0, 1, 2, 3], [4, 5, 6, 7]]
reduced: [[0], [4]]

Reduce axes (1, 2).

SnippetComplete Example

tx = ct.arange(8, dtype=ct.int32).reshape((2, 2, 2))
print("input:", tx)
print("reduced:", ct.min(tx, (1, 2)))

Copy

import cuda.tile as ct
import torch

@ct.kernel
def kernel():
    tx = ct.arange(8, dtype=ct.int32).reshape((2, 2, 2))
    print("input:", tx)
    print("reduced:", ct.min(tx, (1, 2)))


torch.cuda.init()
ct.launch(torch.cuda.current_stream(), (1,), kernel, ())
torch.cuda.synchronize()

Output

input: [[[0, 1], [2, 3]], [[4, 5], [6, 7]]]
reduced: [0, 4]

Return type:

Tile