triton.extension.buffer.language.to_tensor

triton.extension.buffer.language.to_tensor(memref: buffer, writable: bool = True, target_shape=None, _semantic=None) → tensor

Create a tl.tensor from a bl.buffer.

参数:

• memref -- the input bl.buffer object.

• writable (bool) -- If set true, the resultant tensor is considered "writable" during bufferization.

Memref type:

bl.buffer

Example

@triton.jit
def kernel_func(XBLOCK: tl.constexpr):
    buffer1 = bl.alloc(tl.float32, [XBLOCK])
    buffer1.to_tensor(writable=True)
    buffer2 = bl.alloc(tl.float32, [XBLOCK])
    bl.to_tensor(buffer2, writable=True)

Special Restrictions

• Same type support constraints as alloc.