Error when building a big 2D field.

[mushroomfire]

Describe the bug
The code will shut down when I try to allocate a big 2D field, maybe it's a bug.

To Reproduce

import taichi as ti
ti.init(arch=ti.cpu)

arr = ti.field(ti.f32, (40000000, 50))
#arr.fill(1.) 

@ti.kernel
def fill():
    for I in arr:
        arr[I] = 1.0
fill(arr)

Log/Screenshots

[Taichi] version 1.1.2, llvm 10.0.0, commit f25cf4a2, win, python 3.8.13
[Taichi] Starting on arch=x64
[W 10/09/22 19:49:27.691 22784] [D:/a/taichi/taichi/taichi/ir/snode.cpp:create_node@95] Snode index might be out of int32 boundary but int64 indexing is not supported yet.

But 2**31-1 > 40000000*50 is True!
BTW, arr.fill(1.0) also raise an error as below:

RuntimeError: [D:/a/taichi/taichi/taichi/transforms/demote_dense_struct_fors.cpp:convert_to_range_for@29] Assertion failure: total_bits <= 30

[FantasyVR]

Hi @mushroomfire. According to the doc:

By default, Taichi implicitly fits a field in a larger buffer with power-of-two dimensions.

Even though 40000000*50 is smaller than 2^31 - 1, it would still try to allocate 2^31 memory. You could init with packed mode to solve this problem ti.init(arch=ti.cpu, packed=True).

As for the errors when filling arr, it is because of the shape of your field is still too large.

[mushroomfire]

packed=True, that helps. Thanks a lot!