a for loop under vulkan doesn't work with count = count + k but ok with count += k, and Ok under cpu

[deMesnard]

A for loop in @ti.kernel and under acrh vulkan give incorrect result with count=count+k but is OK with count+= k and is also OK under arch cpu.

--coding:Latin-1 -

import taichi as ti
ti.init(arch=ti.vulkan)
@ti.kernel
def count_primes(n: int) -> int:
count = 0
for k in range(0, n):
count = count + k
#count += k
return count
print(count_primes(30))
#ti.profiler.print_kernel_profiler_info()

Result :
[Taichi] version 1.6.0, llvm 15.0.1, commit f1c6fbb, win, python 3.11.6
[Taichi] Starting on arch=vulkan
28

Process finished with exit code 0
28 is incorrect should be 435
With count+= k :

--coding:Latin-1 -

import taichi as ti
#ti.init(arch=ti.cpu, debug = True, verbose= True)
#ti.init(arch=ti.cpu)
#ti.init(arch=ti.vulkan, debug = True, verbose= True)
#ti.init(arch=ti.cuda)
#ti.init(arch=ti.opengl)
ti.init(arch=ti.vulkan)
#ti.init(kernel_profiler=True)

@ti.kernel
def count_primes(n: int) -> int:
count = 0
for k in range(0, n):
#count = count + k
count += k
return count
print(count_primes(30))
Result:
[Taichi] version 1.6.0, llvm 15.0.1, commit f1c6fbb, win, python 3.11.6
[Taichi] Starting on arch=vulkan
435

Process finished with exit code 0
is correct and work fine also in any cases under cpu instead vulkan

[deMesnard]

I am under Windows 11 PC : HP, CPU :AMD Ryzen 7 3700U with Radeon Vega Mobile Gfx 2.30 GHz
GPU : AMD radeon RX Vega 10, Vulkan : version up -to-date 31.0.21905.1001

[FieryAery]

count = count + k is not atomics
count += k is automatically atomics

[deMesnard]

OK thank you I found in the Taichi's guide the explanations about this question of "atomics", I understand now