Hi all!

I started to work on cent/ucent support in LDC (and possible in upstream DMD). Here is the current state:

1) The pull request https://github.com/ldc-developers/ldc/pull/891/files implements cent/ucent based on the upcoming major LDC release (branch merge-2.067).
It is usable but relies on GCC because it uses the __int128 data type. My next step is to implement a synthetic int128_t type to enable support on all platforms.
Up to now I tested only on Linux/x86_64.

2) I already submitted some pull request for DMD, Druntime and Phobos to enable cent/ucent support.

How to continue?

As soon as I got it working on all compiler environment I like to create an upstream pull request. (May take some time because I am now busy with other tasks.)
But: I am not going to extend the DMD backend! This has 2 consequences. First, we need to decide how to integrate the code.
(Do we want to clutter the code with #if WANT_CENT as I currently do? Should we wait for DDMD?) Second, someone needs to work on the DMD backend if DMD should support cent/ucent, too.

If you like to help:
- clone & test
- Druntime/Phobos should support cent/ucent. I already updated/created some modules but more work is needed here
- add support to the DMD backend

Regards,
Kai

On Tuesday, 7 April 2015 at 15:55:24 UTC, Kai Nacke wrote:
> I started to work on cent/ucent support in LDC (and possible in upstream DMD). Here is the current state:

Hi, I appreciate all the work you're doing for LDC, but isn't ucent better implemented as a library type (e.g. as a FixNum struct for arbitrary fixed-size integers)? With the deprecation/removal of complex numbers, octal literals, and hex strings, I'm not sure if cent/ucent would pass the same inclusion requirements that those features didn't.

This doesn't necessarily obsolete your work - we could remove ucent from the language specification, but still expose the backend feature in some way that FixNum could use for added efficiency.

If there's already been a discussion about removing cent/ucent and I missed it, my apologies.

On Tuesday, 7 April 2015 at 18:43:06 UTC, Vladimir Panteleev wrote:
> On Tuesday, 7 April 2015 at 15:55:24 UTC, Kai Nacke wrote:
>> I started to work on cent/ucent support in LDC (and possible in upstream DMD). Here is the current state:
>
> Hi, I appreciate all the work you're doing for LDC, but isn't ucent better implemented as a library type (e.g. as a FixNum struct for arbitrary fixed-size integers)? With the deprecation/removal of complex numbers, octal literals, and hex strings, I'm not sure if cent/ucent would pass the same inclusion requirements that those features didn't.
>
> This doesn't necessarily obsolete your work - we could remove ucent from the language specification, but still expose the backend feature in some way that FixNum could use for added efficiency.
>
> If there's already been a discussion about removing cent/ucent and I missed it, my apologies.

Hi Vladimir,

for me this was a low-hanging fruit. In many cases I only had to add the special cases for max and min values. If it is not included then it is no disaster.

My motivation for the implementation is
- implementation of 128bit CAS
- evaluation of performance of bigintnoasm with 128bit vs. 64bit on X64

Regards,
Kai

Awesome !

"Kai Nacke"  wrote in message news:kxcbizohnxdtimjwlued@forum.dlang.org...

> But: I am not going to extend the DMD backend! This has 2 consequences. First, we need to decide how to integrate the code.
> (Do we want to clutter the code with #if WANT_CENT as I currently do? Should we wait for DDMD?) Second, someone needs to work on the DMD backend if DMD should support cent/ucent, too.

I'm happy to do the DMD backend, at least on x86_64.  I don't want WANT_CENT anywhere in there, it would be better to enable it through a runtime flag in Target in a similar way to the simd types.

On Tuesday, 7 April 2015 at 15:55:24 UTC, Kai Nacke wrote:
> Hi all!
>
> I started to work on cent/ucent support in LDC (and possible in upstream DMD). Here is the current state:

Hurray!
I missed that.

> If you like to help:
> - clone & test
> - Druntime/Phobos should support cent/ucent. I already updated/created some modules but more work is needed here
> - add support to the DMD backend

I once had added cent/ucent to std/traits.d

On Monday, 13 April 2015 at 09:00:30 UTC, Dominikus Dittes Scherkl wrote:
> I once had added cent/ucent to std/traits.d

TypeInfo for cent/ucent is now in druntime. std.traits has cent/ucent support.
I need to address some comments in std.format before it can be merged. Other modules (e.g. std.conv) still miss support.

Regards,
Kai

On Monday, 13 April 2015 at 20:06:11 UTC, Kai Nacke wrote:
> On Monday, 13 April 2015 at 09:00:30 UTC, Dominikus Dittes Scherkl wrote:
>> I once had added cent/ucent to std/traits.d
>
> TypeInfo for cent/ucent is now in druntime. std.traits has cent/ucent support.
Wow, cool.

> I need to address some comments in std.format before it can be merged. Other modules (e.g. std.conv) still miss support.
Hmm.
By the way: are also cent/ucent literals defined?
And if yes, what's the postfix for that? Was it "T"?

cent.max   = 170_141_183_460_469_231_731_637_303_715_884_105_727T
ucent.max  = 340_282_366_920_938_463_463_374_607_431_768_211_455UT

On Tuesday, 14 April 2015 at 10:42:31 UTC, Dominikus Dittes Scherkl wrote:
> By the way: are also cent/ucent literals defined?
> And if yes, what's the postfix for that? Was it "T"?

There is no postfix defined. If the value of an literal does not fit in long/ulong then the type is cent/ucent. I simply followed the rules of int/long.

Regards,
Kai