consistent TFloat prototypes

[GerHobbelt]

Tfloat patch 1 - TFloat used consistently in prototypes tesseract-ocr#3492

Just saw these this morning; anyway: question is: do we want the various specific implementations to have TFloat in their function definition (correctly matching the typedef) or do we want the compiler to match up float <-> TFloat and double <-> TFloat instead?

(if you decide on the latter direction, that would be using the C++ compiler and removing the FAST_TFLOAT #ifdef/#else conditions in the AVX/SSE codes, but it would also mean slightly more work for the compiler, so slightly longer compile times. 🤔 Just thinking out loud and wondering what your preferred coding style for these beasts would be, in the end.)

[stweil]

I prefer using float and double explicitly in the implementations (instead of TFloat) because that's required if we want to have Tesseract binaries which support both variants at the same time. As long as the decision is made at build time, the #ifdefs should stay in the source files to minimize the generated code.

[GerHobbelt]

Okay, that would nil pullreq #5.

Question: what do you feel about having that stuff (same code for float and double) done as C++ template instantiations? (one for float, one for double). While I'm not a big fan of templates, this would be a good use IMO as it would take a single written codebase and have the compiler do the work of duplication, thus killing the risk of subtle bugs which occur when someone fixes one version, but is in a bit of a hurry and misses to fix the other equivalent code as well? --> that would mean #5 would be rewritten with a bit of templating on top to accomplish this: write once, have two.

[stweil]

Thanks.

[stweil]

I reverted this commit in my branch because I prefer to use double explicitly.

[GerHobbelt]

👍 got it.