[BUG] Vector field assignment error in taichi v1.7

[Ives0721]

Description

When do assignment to a VectorField f = ti.Vector.field(2, float, shape=(2, 1)) (i.e. f[i,j][k] = 2.0), f[0,0][0] cannot be assigned right value.

To Reproduce
Here's a sample_code.py to reproduce the bug.

import taichi as ti

ti.init(arch=ti.gpu)

f = ti.Vector.field(2, float, shape=(2, 1))
feq = ti.Vector.field(2, float, shape=(2, 1))

@ti.kernel
def foo():
    f[0, 0][0] = 1.0            # The bug at here
    f[0, 0][1] = 2.0
    f[1, 0][0] = 3.0
    f[1, 0][1] = 4.0

    print('f[0, 0] =', f[0, 0])
    print('f[1, 0] =', f[1, 0])

    feq[0, 0] = f[0, 0]
    feq[1, 0] = f[1, 0]

    print('feq[0, 0] =', feq[0, 0])
    print('feq[1, 0] =', feq[1, 0])

    f[0, 0][1] = 5.0

    print('f[0, 0] =', f[0, 0])
    print('feq[0, 0] =', feq[0, 0])


foo()

Log

When in TaiChi v.1.7, wrong result shows like this.

$ python sample_code.py
[Taichi] version 1.7.1, llvm 15.0.1, commit 0f143b2f, win, python 3.10.14
[Taichi] Starting on arch=cuda
f[0, 0] = [0.000000, 2.000000]  
f[1, 0] = [3.000000, 4.000000]  
feq[0, 0] = [0.000000, 2.000000]
feq[1, 0] = [3.000000, 4.000000]
f[0, 0] = [0.000000, 5.000000]  
feq[0, 0] = [0.000000, 2.000000]

When in TaiChi v.1.6, correct result shows like this:

$ python sample_code.py
[Taichi] version 1.6.0, llvm 15.0.1, commit f1c6fbbd, win, python 3.9.16
[Taichi] Starting on arch=cuda
f[0, 0] = [1.000000, 2.000000]  
f[1, 0] = [3.000000, 4.000000]  
feq[0, 0] = [1.000000, 2.000000]
feq[1, 0] = [3.000000, 4.000000]
f[0, 0] = [1.000000, 5.000000]  
feq[0, 0] = [1.000000, 2.000000]