Debugger: sceKernelPrintf improvement, QOL adjustments

[Nemoumbra]

1. Now sceKernelPrintf can print %f floats. Note: it doesn't use the floating registers (if anyone has some evidence of a real PSP game that prepares f0 or something before the call, please do show).
   Example with a1 equal to 0x3f8ccccd:
   

2. The "Copy Float (32 bit)" menu item has been moved to its own section because I suspect some people might have memorized the layout and thus experience slight inconveniences due to it now being the last element. Well, that's no longer a problem. We may also later add other floating types here.

3. This commit does not fix the recently added menu item so I decided to do that.
   I have also renamed it to have consistent terms across the code:
   

4. Lastly, check the notes here: https://en.cppreference.com/w/cpp/io/basic_stringstream/str

[hrydgard]

Yeah seems likely that printf (varargs) parameters go through regular registers. I do seem to recall that floating point arguments to regular functions can go through fp registers, but didn't check now.

[hrydgard]

To clarify my previous comment about this, here's what I would have done here:

char temp[64];
snprintf(temp, sizeof(temp), "%f", f_arg);
result += temp;

Though, to be fair, yours is better "C++", mine is just reflective of my somewhat excessive obsession with avoiding allocations. Doing that is important when running PPSSPP in debug mode for things that happen a lot, but since this shouldn't be very spammy, doing it the C++ way is fine and I'm not even sure in reality that there's more than one excessive allocation happening here (the stringstream has to allocate a buffer, while my way allocates it on the stack which is cheaper).

[fp64]

The raw "%f" (which both operator<< and std::to_string use under the hood) in general gives you "wrong" float representation (produces different bitpattern when fed to e.g. strtof). The simplest version that guarantees (assuming round-to-nearest rounding mode) float->string->float round-trip is "%.9g". Not sure what the desired behaviour for sceKernelPrintf is, but for right-click menu I would expect "value-preserving" representation.

As for stringstream... why not use aforementioned std::to_string?

[hrydgard]

That's a good point about copy - roundtrippable would be good there.

[Nemoumbra]

#17203 (comment)
Here I explained my choice with using the stringstream and not std::to_string()

[fp64]

Here I explained my choice with using the stringstream and not std::to_string()

Oh, yes, I was wrong: operator<< uses (by default) "%g" (see https://en.cppreference.com/w/cpp/locale/num_put/put for details) and std::to_string uses "%f".

Still, "%g" is "%.6g", which is not enough for round-trip.

[Nemoumbra]

Of course, we're not done with sceKernelPrintf. I only added the %f flag, but no precision params are available as of now. I guess, this is a PoC.
This is what I think we can do next:

1. Add support for things like "%.2f". I think this will downgrade the current precision we have for %f, but that's how the PSP works and... we want PPSSPP to emulate it, don't we?
2. If sceKernelPrintf supports this %g, it should also be added.
3. Finally, if the precision can be improved without breaking the correctness, improve the "Copy Float (32 bit)" option.

[fp64]

float->string->float round-trip is "%.9g"

For those curious: the formula is $1 + \lceil p \times \log_{10} 2 \rceil$ (and $\lfloor (p - 1) \times \log_{10} 2 \rfloor$ for longest safe string->float->string round-trip, which is 6 for float, which has $p = 24$). See e.g. this.