[Bug] Can't deduce copy extents from args

[LyricZhao]

import torch
import tilelang
from tilelang import language as T


@tilelang.jit()
def get_buggy_kernel():
    num_tokens = T.symbolic('num_tokens')

    @T.prim_func
    def buggy_kernel(indices: T.Tensor[(num_tokens, ), 'int'],
                     x: T.Tensor[(num_tokens, ), 'float']):
        with T.Kernel(num_tokens, threads=32) as pid:
            idx = T.alloc_local([1], 'int')
            T.copy(indices[pid], idx[0])
            x[idx[0]] = x[idx[0]] + 1

    return buggy_kernel


if __name__ == '__main__':
    kernel = get_buggy_kernel()
    print(kernel.get_kernel_source())

    i = torch.arange(128, dtype=torch.int, device='cuda')
    x = torch.randn((128, ), dtype=torch.float, device='cuda')
    kernel(i, x)

error: Can't deduce copy extents from args, the error is raised by T.copy(indices[pid], idx[0]).

[LeiWang1999]

closed as resolved via pr #946