Get float in HEX format

Question

I have the string char str [8]. It says a number in the HEX format. It is float and with a sign. How can I write it to a variable of type float?

For example:

char str[9] = "41700000\0";

I need get val from this: 15.0

`char str[8] = "41700000";` is not a string becaue it is not null-terminated. — MikeCAT, May 08 '20 at 17:40
It's hard to understand what you're asking. Are you saying, `"41700000"` is in hex. You want to convert it into a 32-bit `float`? — Fiddling Bits, May 08 '20 at 17:45
@FiddlingBits Yes. I made the opposite thing `float d = atof(str); printf("%X \n",*(unsigned int*)&d); ` and when i input 15.0 it gives me that number in hex. And I compared this to some site that gives you convert from Real2Hex. — Sam, May 08 '20 at 17:48
Perhaps with `printf("%.1f\n", ((union { int i; float f; }) { .i = atoi(str) }).f);`? — chux - Reinstate Monica, May 08 '20 at 17:53
@chux That's UB as well, since it's type punning a `float` with an `int` when there's no guarantee that `float` and `int` are the same size, or have the same endianness. — user3386109, May 08 '20 at 17:56
@user3386109 Fair enough - yet it is reversing the process that got OP there (should have used `unsigned, strtoul()` too). If it is UB that is reversing UB, does that make it BU? — chux - Reinstate Monica, May 08 '20 at 17:58

Boris Lipschitz · Answer 1 · 2020-05-08T18:39:35.130

1

This is a quick and dirty way to revert your 15 back from the hexadecimal 4 bytes representation. There is too much wrong with it to even start talking about it, though, and the right thing to do would be to ask yourself "why do i need this to begin from, and how can i do something better instead".

float hexStrToFloat(const char* str)
{
    union { unsigned int i; float f; } tmp;
    sscanf_s(str, "%x", &tmp.i);
    return(tmp.f);
}

Footnote: assumes little-endian, 32 bit or higher, machine.

edited May 08 '20 at 18:39

answered May 08 '20 at 17:59

Boris Lipschitz

1,514
8
12

I believe `*(float*)&tmp` is UB due to aliasing issues. – chux - Reinstate Monica May 08 '20 at 18:00
@ chux - Reinstate Monica, there are at least 4 bytes where that int is (assuming 32 bit machine or higher), C-style casting a pointer there and de-refing it as float pointer... don't think there is a (technical) problem with that in any way imaginable. But hey, taking whats pretty much is a binary dump of a floating point number and trying to convert it back to floating point by directly unpacking that binary representation back into a variable... thats just wrong on so many levels to begin from. – Boris Lipschitz May 08 '20 at 18:10
Suggest reviewing [What is the strict aliasing rule?](https://stackoverflow.com/a/98702/2410359). Aside: it is not a size issue. – chux - Reinstate Monica May 08 '20 at 18:19
I mean, can you imagine any situation, in which a compiler would optimize this in an awkward way and do something else other than putting tmp on stack? Anyway, modifying the answer to go without this cast, but do pretty much the same... – Boris Lipschitz May 08 '20 at 18:29
I can imagine a situation. Compilers today are _not your friend_ when they optimize and your code has UB. I doubt I can convince in comment space. If unconvinced, you may want to research or post a question. – chux - Reinstate Monica May 08 '20 at 18:33
Well, better be safe than sorry, using a union instead pretty much tells the compiles "hey, i WANT to overlay those types on top of each other, don't do anything funny with that", so i've edited the answer to just use union. – Boris Lipschitz May 08 '20 at 18:35
I truly hope we all agree its generally better to avoid this entire idea of taking hex dumps and putting them back into variables in general, though. – Boris Lipschitz May 08 '20 at 18:40
Agree - except through `unsigned char[]` is OK. Anything else has _issues_. – chux - Reinstate Monica May 08 '20 at 18:43

0___________ · Accepted Answer · 2020-05-08T18:26:44.650

1

You can pun the data:

Using memcpy

unsigned int u = 0x41700000;
float f;

memcpy(&f, &u, sizeof(f));
printf("%f\n", f);

Using union (IMO legal, many people have opposite opinion)

union un
{
    float f;
    unsigned u;
};

void foo(unsigned x)
{
    union un a = {.u = x};
    printf("%f\n", a.f);
}

I assume floats && integers have the same size.

Of course you will have to convert string from your question to the unsigned value - but it is relatively easy (scanf, atoi ....)

PS BTW many compilers will generate exactly the same code for both (without the memcpy call) https://godbolt.org/z/VaCcxS

edited May 08 '20 at 18:26

answered May 08 '20 at 18:13

0___________

60,014
4
34
74

May I ask another question, how to fill with leading zeros without for loop?. For example: `0x3f8` to `0x3f800000` – Sam May 08 '20 at 19:46
`0x3f8 << (4 * numberOfZeoes)` assuming 0x3f8 is a number not the string. – 0___________ May 08 '20 at 19:50

Get float in HEX format

2 Answers2