Define function types for prototype declaration

[Kerilk]

This is pull request implements what I have been describing in a recent WG meeting: providing and using type function definitions to declare symbols and pointers. This yields several benefits:

• We stop code duplication. (It is less an issue nowadays since we generate the code, but still)
• Now that prototypes can be deactivated, this allows users declaring prototypes at a finer granularity. For instance, we could use this to declare loader extension prototypes by using the type declaration from the header. For now, the definitions cannot be validated by a prototype from the headers since CL_NO_NON_ICD_DISPATCH_EXTENSION_PROTOTYPES is activated: see: https://github.com/KhronosGroup/OpenCL-ICD-Loader/blob/ee329edc23258a53f9b5799446f7417d853ee143/loader/icd_dispatch.c#L26-L32. Instead, we could now write:

static clGetICDLoaderInfoOCLICD_t clGetICDLoaderInfoOCLICD;
cl_int
clGetICDLoaderInfoOCLICD(
   cl_icdl_info param_name,
   size_t       param_value_size,
   void *       param_value,
   size_t *     param_value_size_ret)
{

I kept the original exports as some users could be parsing the headers for the original declarations (We do in the CLHPP testing for instance)

I did cut this PR in 3 main patches, should we decide to partially apply these changes:

• Add prototypes to the extensions through the generator.
• Add prototype to the core functions in a new generated file cl_function_types
• Apply prototypes (manually) to cl_icd.h which benefits a lot from the change. Given how suffixes are not really used (and very inconsistently in cl_icd.h) we could also take it a step father and remove the pointer type definitions from this file and use the function types directly inside the dispatch structure. For now I decided to minimize the scope of the changes. I did remove duplicated prototype declarations in cl_icdl.h, and added a missing guard around cl_api_clCreateFromGLTexture which apparently only exist in OpenCL 1.2. Implemented the change as suggested.

I tested using the ICD-Loader and CLHPP tests. No issues to report.

[bashbaug]

LGTM! I have a few comments but it would be fine to address them in a subsequent PR (or not at all).

[bashbaug]

LGTM! I did some basic testing on Windows and everything seems to be in order.

[Kerilk]

LGTM! I did some basic testing on Windows and everything seems to be in order.

Thank for taking a look. Should we discuss this next week at the teleconference before merging?

[bashbaug]

Merging as discussed on the May 9th teleconference.