Arguments for apply()

[mloubout]

So there is a little "bug" in operator line 152.

        for argname in o_vals.keys():
            if not arguments[argname].shape == o_vals[argname].shape:
                raise InvalidArgument("Shapes must match")

The little problem with this is, any passed argument needs to have the exact same shape, including number of time steps. I kinda of defeat the purpose of generic dimensions if there sizes can't be changed.

AN example would be

u = TimeData(name='u', shape=(100, ), time_order=1, space_order=2, time_dim=100)
# Derive the stencil according to devito conventions
eqn = Eq(u.dt, u.dx)
stencil = solve(eqn, u.forward)[0]
op = Operator([Eq(u.forward, stencil)], subs={h: dx, s: dt})

# Execute the generated Devito stencil operator
op.apply()

the doing

u2 = TimeData(name='u', shape=(100, ), time_order=1, space_order=2, time_dim=200)
op.apply(u=u2)

produces

InvalidArgumentTraceback (most recent call last)<ipython-input-7-3e325d1513f1> in <module>()
      6 # stencil = solve(eqn, u2.forward)[0]
      7 # op = Operator([Eq(u2.forward, stencil)], subs={h: dx, s: dt})
----> 8 op.apply(u=u2)
/Users/mloubout/Dropbox/London/CodeGen/devito/devito/operator.pyc in apply(self, *args, **kwargs)
    478         """Apply the stencil kernel to a set of data objects"""
    479         # Build the arguments list to invoke the kernel function
--> 480         arguments, dim_sizes = self.arguments(*args, **kwargs)
    481 
    482         # Invoke kernel function with args
/Users/mloubout/Dropbox/London/CodeGen/devito/devito/operator.pyc in arguments(self, *args, **kwargs)
    152         for argname in o_vals.keys():
    153             if not arguments[argname].shape == o_vals[argname].shape:
--> 154                 raise InvalidArgument("Shapes must match")
    155 
    156             arguments[argname] = o_vals[argname]
InvalidArgument: Shapes must match

[FabioLuporini]

I wonder if this our usual caching @mlange05

[mloubout]

Well actually, thinking of it, it would be odd and error prone to pass arguments with differet sizes to apply as there is other objects ,m, src,rec that may not be compatible with this size leading to segfaults or errors.

[FabioLuporini]

Can this be closed @mlange05 ?

[mloubout]

This should be fixed by last week PRs

[FabioLuporini]

yes, but there's a new (smaller) issue now that is waiting for #273 ...

[FabioLuporini]

@mlange05 we can probably close this now?

[mlange05]

Yes, I think the relevant parts have been addressed. Closing.