Code breaks on vulkan backend with v1.1.2

[neozhaoliang]

This code crashes on intel GT2 card, the commit is e2fb83bdc4f3b4cc6cc86cca4ea35433883a49e6:

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

d = 3
num_round = 1000000
max_steps = 1000000

ivec = ti.types.vector(d, int)
origin = ivec([0, 0, 0]) # ivec([0] * d)

dirs = ti.Vector.field(d, int, shape=2*d)
for k in range(d):
    dirs[2 *k][k] = 1
    dirs[2 * k + 1][k] = -1

print(dirs)

@ti.func
def choose_random_direction():
    ind = int(ti.random() * 2 * d)
    return dirs[ind]

@ti.kernel
def walk() -> float:
    success = 0
    for _ in range(num_round):
        pos = origin
        for step in range(max_steps):
            pos += choose_random_direction()
            if all(pos == origin):
                success += 1
                break

    return success / num_round

print(walk())

Error message:

[E 11/21/22 12:57:53.167 1630179] [vulkan_device.cpp:submit@1708] Vulkan Error : -4 : failed to submit command buffer


Traceback (most recent call last):
  File "/home/zhao/China_R_Meeting/return_probability_Z^d.py", line 36, in <module>
    print(do_experiments())
  File "/home/zhao/.local/lib/python3.10/site-packages/taichi/lang/kernel_impl.py", line 942, in wrapped
    return primal(*args, **kwargs)
  File "/home/zhao/.local/lib/python3.10/site-packages/taichi/lang/kernel_impl.py", line 869, in __call__
    return self.runtime.compiled_functions[key](*args)
  File "/home/zhao/.local/lib/python3.10/site-packages/taichi/lang/kernel_impl.py", line 792, in func__
    runtime_ops.sync()
  File "/home/zhao/.local/lib/python3.10/site-packages/taichi/lang/runtime_ops.py", line 8, in sync
    impl.get_runtime().sync()
  File "/home/zhao/.local/lib/python3.10/site-packages/taichi/lang/impl.py", line 453, in sync
    self.prog.synchronize()
RuntimeError: [vulkan_device.cpp:submit@1708] Vulkan Error : -4 : failed to submit command buffer
[E 11/21/22 12:57:53.209 1630179] [vulkan_device.cpp:submit@1708] Vulkan Error : -4 : failed to submit command buffer


[E 11/21/22 12:57:53.210 1630179] [vulkan_device.cpp:submit@1708] Vulkan Error : -4 : failed to submit command buffer


terminate called after throwing an instance of 'std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >'
已放弃

Though the parent commits run without raising errors, they all give the wrong result 0.0 on both the Vulkan and OpenGL backends. But the code works fine on Cuda +v1.0.

Info: the program works for num_rounds = 100K but fails for num_rounds=1000K.

[ailzhang]

@neozhaoliang hmmm I've run the script and it works fine on master branch. Does this repro consistently on your end?

[ailzhang]

@neozhaoliang can you still repro this issue?

[neozhaoliang]

Using current master branch [Taichi] version 1.5.0, llvm 14.0.0, commit 4af54d49, linux, python 3.10.6

The program runs but gives the wrong error 0.0

[bobcao3]

RTX 4090 on windows produced this result:

[[ 1  0  0]
 [-1  0  0]
 [ 0  1  0]
 [ 0 -1  0]
 [ 0  0  1]
 [ 0  0 -1]]
0.34026598930358887