Is it possible to use float and double without setting FPU?

mrjbom · **Joined:** Sun Jul 21, 2019 7:34 am **Posts:** 300

This is a very stupid question, but I thought I'd ask it nonetheless.
Now in my early initialization code I need to use double for calculations, but I'm not sure I can perform it as it seems to require the FPU to be preconfigured, is this true?

iansjack · **Posted:** Wed Feb 28, 2024 3:55 pm

https://wiki.osdev.org/FPU

But why would you need to use floating-point calculations when booting an operating system?

eekee · **Posted:** Wed Feb 28, 2024 4:44 pm

We can help convert your calculations to integer ones, if you'd like.

Octocontrabass · **Joined:** Mon Mar 25, 2013 7:01 pm **Posts:** 5146

If you really need floating-point math, there are ways to do it without using the FPU.

But you probably don't need floating-point math.

mrjbom · **Joined:** Sun Jul 21, 2019 7:34 am **Posts:** 300

iansjack wrote:

But why would you need to use floating-point calculations when booting an operating system?

To calculate when I should remap the hash map.

Code:

m->remap = (uint32_t)(m->capacity * ((double)m->load_fac / 100));

For example, m->capacity is 8 and m->load_fac = 75, then, I will need to increase the size of the hash map when the number of elements in it reaches 6.
Specifically this calculation I could replace with something like:

Code:

m->remap = m->capacity / 4 * 3

But can it be done any better? So that I can have different load_fac values, for example?

iansjack · **Posted:** Thu Feb 29, 2024 2:30 am

The integer calculation

Code:

m->remap = (m->capacity * m->load_fac + 99) / 100

(if you want to round up) produces the same result. If you want to round down just use

Code:

m->remap = (m->capacity*m->load->fac) / 100

iProgramInCpp · **Joined:** Sun Apr 21, 2019 7:39 am **Posts:** 76

mrjbom wrote:

Specifically this calculation I could replace with something like:

Code:

m->remap = m->capacity / 4 * 3

This would work better as a replacement for what I quoted above.

Code:

m->remap = m->capacity * 3 / 4

mrjbom · **Joined:** Sun Jul 21, 2019 7:34 am **Posts:** 300

iProgramInCpp wrote:

mrjbom wrote:

Specifically this calculation I could replace with something like:

Code:

m->remap = m->capacity / 4 * 3

This would work better as a replacement for what I quoted above.

Code:

m->remap = m->capacity * 3 / 4

Why?

iansjack · **Posted:** Sat Mar 02, 2024 3:12 pm

You need to do the multiplication before the division (if using integer arithmetic) to get the correct result. And it’s best to use parentheses to ensure the correct order.

Code:

8 * (75 / 100) = 0

whereas

Code:

(8 * 75) / 100 = 6

The latter is obviously the result you want.

You’re looking for an integer result, so you don’t need floating-point calculations.

mrjbom · **Joined:** Sun Jul 21, 2019 7:34 am **Posts:** 300

iansjack wrote:

You need to do the multiplication before the division (if using integer arithmetic) to get the correct result. And it’s best to use parentheses to ensure the correct order.

Code:

8 * (75 / 100) = 0

whereas

Code:

(8 * 75) / 100 = 6

The latter is obviously the result you want.

You’re looking for an integer result, so you don’t need floating-point calculations.

I understand this.
He said that I should replace

Code:

m->remap = m->capacity / 4 * 3

with

Code:

m->remap = m->capacity * 3 * 4

which I did not understand.

thewrongchristian · **Joined:** Tue Apr 03, 2018 2:44 am **Posts:** 403

mrjbom wrote:

iansjack wrote:

You need to do the multiplication before the division (if using integer arithmetic) to get the correct result. And it’s best to use parentheses to ensure the correct order.

Code:

8 * (75 / 100) = 0

whereas

Code:

(8 * 75) / 100 = 6

The latter is obviously the result you want.

You’re looking for an integer result, so you don’t need floating-point calculations.

I understand this.
He said that I should replace

Code:

m->remap = m->capacity / 4 * 3

with

Code:

m->remap = m->capacity * 3 * 4

which I did not understand.

No.

The suggested code was:

Code:

m->remap = m->capacity * 3 / 4

The reason it was suggested is because doing the division by 4 first will lose you precision. If you multiply by 3 first, you lose no precision in the intermediate result before dividing by 4.

The other good thing about the code is that the optimiser will turn this all into shifts and adds, so it'll be blazing fast.

OSDev.org

Is it possible to use float and double without setting FPU?

Who is online