parameter types and bounds

[ntessore]

This PR introduces the possibility to give types and bounds to the individual parameters in an object definition. So far, the parameter types only serve to give a number of default bounds on parameters (see below), but they will be more useful in the future.

An example from the Sérsic source is

params
{
    { "x",   POSITION_X },
    { "y",   POSITION_Y },
    { "r",   RADIUS     },
    { "mag", MAGNITUDE  },
    { "n",   PARAMETER, { 0.5f, 8.0f } },
    { "q",   AXIS_RATIO },
    { "pa",  POS_ANGLE  }
};

The second (optional) argument is the parameter type.
There are a number of common, predefined parameter types: POSITION_X, POSITION_Y, RADIUS, MAGNITUDE, AXIS_RATIO, POS_ANGLE.
The generic type for parameters without special, pre-defined meaning is PARAMETER, which is the default.

The third (optional) argument is the definition of hard bounds for the parameter. This serves to limit the possible range of parameter values, for example to exclude unphysical quantities. Lensed will check whether a prior agrees with the bounds, and display a warning if the prior is partially out of the range, or an error if it is fully out of range.

warning: source.n: prior partially outside parameter bounds [0.5, 8]
Part of the prior lies outside of the allowed range of parameter values. Values 
will be drawn from the prior until a valid one is found. If the bounds of prior 
and parameter do not overlap significantly, this can be slow. You might want to 
change the prior accordingly.

error: source.q: prior does not include parameter bounds [0, 1]
The prior does not include the allowed range of values for the parameter. It is 
impossible to draw a valid parameter value. Please fix the prior to include the 
range of values indicated above.

We can also define bounds to be set by default for certain parameter types. So far, these are

• RADIUS has bounds [0, inf)
• AXIS_RATIO has bounds [0, 1]

[ntessore]

cc @fabiobg83 @rbmetcalf

[ntessore]

This needs some fixup to make the tests pass on Travis again before it can be merged.

[ntessore]

This is actually quite hard, as there is inconsistency on what doesn't work when one has strings inside a struct initialiser.
On some OpenCL implementations, it seems that initialising a fixed-size array with a string literal is broken:

char name[16] = "test"; // won't work for some implementations

Unfortunately, on other platforms (e.g. Apple) it appears that assigning a string literal to a constant char* is broken instead:

constant char* str = "test"; // does not work sometimes

The above are just examples to showcase what does not work; in the real case, the strings are members of a struct that is initialised. I'm investigating.

[ntessore]

Taking a closer look at what the compiler puts into memory in the first case above reveals the buggy behaviour.

This is what the sie parameter definition should look like, in the format name[16] | type | bounds:

78 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 | 01 00 00 00 | 0.000000 0.000000
79 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 | 02 00 00 00 | 0.000000 0.000000
72 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 | 03 00 00 00 | 0.000000 0.000000
71 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 | 05 00 00 00 | 0.000000 0.000000
70 61 00 00 00 00 00 00 00 00 00 00 00 00 00 00 | 06 00 00 00 | 0.000000 0.000000

And here is what it actually looks like using the offending compiler:

78 00 01 00 00 00 00 00 00 00 00 00 00 00 00 00 | 01 00 00 00 | 0.000000 0.000000
79 00 02 00 00 00 00 00 00 00 00 00 00 00 00 00 | 02 00 00 00 | 0.000000 0.000000
72 00 03 00 00 00 00 00 00 00 00 00 00 00 00 00 | 03 00 00 00 | 0.000000 0.000000
71 00 05 00 00 00 00 00 00 00 00 00 00 00 00 00 | 05 00 00 00 | 0.000000 0.000000
70 61 00 06 00 00 00 00 00 00 00 00 00 00 00 00 | 00 06 00 00 | 0.000000 0.000000

It seems like the string is not correctly padded to a char[16] array, and the struct alignment is subsequently taken from the first item of the array.

[ntessore]

As this is a problem with some OpenCL implementations, I will merge this and open a separate issue.