Floating point is useless. You don't need anything more than either Q16.16 or Q32.32 fixed point. Prove me wrong

Floating point is useless. You don't need anything more than either Q16.16 or Q32.32 fixed point. Prove me wrong.

Attached: 能代.jpg (540x763, 71K)

but how mak game witout flating paint?

I tend to agree with you, but I'm a brainlet

Original PlayStation handled it just fine

Clearly you've never run regression on a cost function in your life.

Attached: Regression_pic_assymetrique.gif (610x460, 22K)

What supports Q?

MY DICK AAAAAAAAAAHHHH

I like to use modern floating point as the "minimum step" for optimization problems (especially machine learning).

I would imagine that floating point is pretty important for scientific research as well.

It's handy when you've got an FPGA and don't want to waste the fabric with unnecessary floating point hardware

Yes, it is
but 16 bit FP useful for sound and video

Holy shit OP here, my buzz has died down now, can't believe I made this retarded thread.

32bit FP is god-tier for HDR and/or dealing with extremely faint signals

Everything, it's an abstraction. Math on them is just normal math.

When was the last time you took a numerical derivative? Numerical derivatives instantly rob you of half of your digits of precision. If you're taking derivatives all the time -- and scientific researchers are -- then you need to start with all the digits you can get.

Bring on the quad- and oct-precision routines!

apparently canucks are too brain addled to understand what floating point is.

how would you represent the number 4294967296? as 1.0? why? are you dumb?

for real live aplication of this alweys posible select fixed point value

>how would you represent the number 4294967296? as 1.0? why? are you dumb?
easy

>just fine
No...

>decide to use fixed point
>waste valuable development time implementing basic arithmetic operations
>it's buggier than floats because you didn't bother to handle all the edge cases
>it's slower than floats because the logic is in software and uses branches

Testing
onions basedboy onions
ONIONS Onions onions onions

onions

I'd like to float her point, if you get what I mean.

Go home nigger

I'd delete the thread if I could friendo.

>apparently canucks are too brain addled to understand what floating point is.
Right, so when my function optimizes to an arbitrarily small fractional value from a large value I'll just change data types on the fly. Truly fucking genius.

yes it did. inb4 muh texture warping that wasn't even related to fixed points

>Prove me wrong.
Division is futile. You will be approximated...

Floating point is an approximation too

That's OK, we knew you were a faggot anyway.

You seem pretty divided on the subject

Attached: 1511803411002.png (557x576, 425K)

99% aplication integer point is enought
FP gust brainlet version of integer