Taichi if statement bug that causes performance issue

[FieryAery]

Describe the bug
The idea is to bypass particle that has 0 mass using if statement, but it turns out that the code will still evaluate the content inside the if statement even when mass is 0. Also, different content inside if statement is generating different execution time when it should have been bypassed by the false condition.

To Reproduce

import taichi as ti
from taichi.math import vec3,vec2
import time
ti.init(arch = ti.gpu, debug=False)
Particle = ti.types.struct(mass=ti.uint32,pos=vec3,vel=vec3,force=vec3,density=float)
bufferP0,bufferP1 = Particle.field(shape=(128,64,32)),Particle.field(shape=(128,64,32))

@ti.kernel
def test_Main_exec():
    for pos in ti.grouped(ti.ndrange(128,64,32)):
        p0,p1 = Particle(),Particle()
        for offset in ti.grouped(ti.ndrange((-1,10),(-1,10),(-1,10))):
            NPos = pos + offset
            incoming0,incoming1 = bufferP0[NPos],bufferP1[NPos]
            if incoming0.mass > 0: #deleting content inside this statement generate different execution time result
                p0.pos = p0.pos*p0.mass + incoming0.pos*incoming0.mass
                p0.mass = incoming0.mass
            if incoming1.mass > 0: #same here
                p1.pos = p1.pos*p1.mass + incoming1.pos*incoming1.mass
                p1.mass = incoming1.mass
        bufferP0[pos],bufferP1[pos]=p0,p1

if __name__ == "__main__":
    start_time = time.time()
    prev_time = 0
    while True:
        test_Main_exec()
        ti.sync()
        end_time = time.time()
        execution_time = (end_time - start_time)*1000
        print(f"Execution time: {execution_time-prev_time:.1f} ms per frame")
        prev_time = execution_time

deleting

p0.mass = incoming0.mass
p1.mass = incoming1.mass

will generate different execution time (60ms to 8ms), which means the content inside if statement is not bypassed. Also 8ms to 60ms is waaaay too much for just a simple p0.mass = incoming0.mass and p1.mass = incoming1.mass.