Make numba an optional dependency
#349
Conversation
| @@ -55,7 +103,7 @@ def _assemble_torsion( | |||
| return k_el, f_el, c_el | |||
|
|
|||
|
|
|||
| @njit(cache=True, nogil=True) # type: ignore | |||
There was a problem hiding this comment.
This impairs performance as compiled code is not cached. Why not directly define a njit decorator when njit cannot be imported. This solution appears to be overcomplex to me.
There was a problem hiding this comment.
Now that you mention this, that's an obvious solution thanks! I didn't realise this would impair the caching. If you have time to implement a fix I'd be happy to merge, otherwise I will get to this next week. Cheers!
There was a problem hiding this comment.
I mean, directly define the decorator on line 31:
def njit(**options):
def decorator(func):
def wrapper(*args, **kwargs):
return func(*args, **kwargs)
return wrapper
return decoratorThere was a problem hiding this comment.
The rest does not need to be changed, so cache will still work if numba is installed. Ppl download and run once, the functions will be compiled, cached, and reused. Without cache, those functions will be compiled every time when the analysis is performed. Since most ppl do not change the source code, caching could benefit most users I presume.
| @@ -648,7 +696,9 @@ def element_stress( | |||
|
|
|||
| # extrapolate results to nodes, ignore numba warnings about performance | |||
| with warnings.catch_warnings(): | |||
| warnings.simplefilter("ignore", category=NumbaPerformanceWarning) | |||
| if USE_NUMBA: | |||
| warnings.simplefilter("ignore", category=NumbaPerformanceWarning) | |||
There was a problem hiding this comment.
Change line 638 to:
sig_zxy_mzz_gp = np.zeros((n_points, 2), order='F')
sig_zxy_vx_gp = np.zeros((n_points, 2), order='F')
sig_zxy_vy_gp = np.zeros((n_points, 2), order='F')can suppress the warning so no need to import those extra utilities.
Fixes #348.