jax.jit(f)(x) != f(x)

[mariogeiger]

f(x) != jax.jit(f)(x)

code to reproduce

import jax
import jax.numpy as jnp
import jaxlib

print(jax.__version__, jaxlib.__version__)

def h(a, w):
    n = jnp.cross(w['y'], w['z'])
    return jnp.cross(a['x'], n / jnp.linalg.norm(n))

def f(state):
    return jax.vmap(jax.vmap(h, (None, 0), 0), (0, None), 0)(state['A'], state['B']), state

state = {
    'A': {
        'x': jnp.ones((5, 3))
    },
    'B': {
        'y': jnp.ones((3, 3)),
        'z': jnp.array([
            [1.0, 0.0, 0.0],
            [1.0, 1.0, 0.0],
            [1.0, 0.0, 1.0],
        ]),
    },
}

t1, _ = jax.jit(f)(state)
t2, _ = f(state)

assert jnp.all(t1 == t2), (t1, t2)

jax.__version__ == '0.2.17'
jaxlib.__version__ == '0.1.69'

[jakevdp]

Thanks for the question! I ran your code with jax 0.2.17 and jaxlib 0.1.59 on both CPU and GPU Colab backends, and the assertion passed in both cases. Can you say more about how you're running this?

[hawkinsp]

I can reproduce on my Mac with jaxlib 0.1.69. It might be CPU dependent.

[jakevdp]

For what it's worth, here's a Colab runtime where the issue does not reproduce: https://colab.research.google.com/drive/1tBX_ZGDm1X16WWm2ey_aY8a-wi4yJzyh

[mariogeiger]

I simplified the code a bit

[hawkinsp]

I think this is an LLVM bug. My best workaround for now is to set XLA_FLAGS="--xla_backend_extra_options=-vectorize-slp=false"

The bug only reproduces on machines without AVX512 support. This may be why the problem reproduces in some environments but not others. The output of lscpu on Linux will tell you whether your CPU supports AVX512.

[mariogeiger]

Here are the lscpu output for the two machine I tried on (both have the bug)

Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
Address sizes:       39 bits physical, 48 bits virtual
CPU(s):              8
On-line CPU(s) list: 0-7
Thread(s) per core:  2
Core(s) per socket:  4
Socket(s):           1
NUMA node(s):        1
Vendor ID:           GenuineIntel
CPU family:          6
Model:               142
Model name:          Intel(R) Core(TM) i7-8650U CPU @ 1.90GHz
Stepping:            10
CPU MHz:             876.022
CPU max MHz:         4200.0000
CPU min MHz:         400.0000
BogoMIPS:            4224.00
Virtualization:      VT-x
L1d cache:           32K
L1i cache:           32K
L2 cache:            256K
L3 cache:            8192K
NUMA node0 CPU(s):   0-7
Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1d

Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
CPU(s):              8
On-line CPU(s) list: 0-7
Thread(s) per core:  2
Core(s) per socket:  4
Socket(s):           1
NUMA node(s):        1
Vendor ID:           GenuineIntel
CPU family:          6
Model:               158
Model name:          Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz
Stepping:            9
CPU MHz:             800.052
CPU max MHz:         4200.0000
CPU min MHz:         800.0000
BogoMIPS:            7200.00
Virtualization:      VT-x
L1d cache:           32K
L1i cache:           32K
L2 cache:            256K
L3 cache:            8192K
NUMA node0 CPU(s):   0-7
Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1d