From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: 
X-Spam-Status: No, score=-1.9 required=5.0 tests=BAYES_00,FREEMAIL_FROM
	autolearn=unavailable autolearn_force=no version=3.4.4
X-Received: by 10.157.41.167 with SMTP id n36mr4023682otb.106.1493049970775;
        Mon, 24 Apr 2017 09:06:10 -0700 (PDT)
X-Received: by 10.157.37.213 with SMTP id q79mr223367ota.11.1493049970739;
 Mon, 24 Apr 2017 09:06:10 -0700 (PDT)
Path: 
 eternal-september.org!reader01.eternal-september.org!reader02.eternal-september.org!news.eternal-september.org!news.eternal-september.org!feeder.eternal-september.org!news.glorb.com!c26no1032249itd.0!news-out.google.com!v18ni3241ita.0!nntp.google.com!c26no1032247itd.0!postnews.google.com!glegroupsg2000goo.googlegroups.com!not-for-mail
Newsgroups: comp.lang.ada
Date: Mon, 24 Apr 2017 09:06:10 -0700 (PDT)
Complaints-To: groups-abuse@google.com
Injection-Info: glegroupsg2000goo.googlegroups.com;
 posting-host=173.71.201.205;
 posting-account=QF6XPQoAAABce2NyPxxDAaKdAkN6RgAf
NNTP-Posting-Host: 173.71.201.205
User-Agent: G2/1.0
MIME-Version: 1.0
Message-ID: <c298b2db-ecfe-4597-8eec-7b69650dcc85@googlegroups.com>
Subject: avoiding builtin memset
From: Jere <jhb.chat@gmail.com>
Injection-Date: Mon, 24 Apr 2017 16:06:10 +0000
Content-Type: text/plain; charset=UTF-8
Xref: news.eternal-september.org comp.lang.ada:46619
Date: 2017-04-24T09:06:10-07:00
List-Id: <comp.lang.ada>

GNAT GPL 2016
Windows 10
Cross compiled to arm cortex m0+
Full Optimization

With a small runtime that I am modifying, I am not linking in the standard c
libraries.  This means whenever I do an array initialize, gcc tries to link 
in a non existent memset call.  I started working on an Ada version of 
memset which I export out.  The problem comes when memset tries to 
recursively call itself.  The compiler is too smart for me.

At the end of my memset I have a simple loop:
      while Current_Address < End_Address loop
         Convert_8.To_Pointer (Current_Address).all := Uint8_Value;
         Current_Address := Current_Address + 1;
      end loop;

It serves two purposes:
1.  It finishes up any leftover bytes on an unaligned array
2.  If the data set is small enough (<= 16), the function skips immediately
    to this loop and just does a byte copy on the whole thing rather
    than try and do all the extra logic for an aligned copy

However GNAT tries to convert that to another memset call, which doesn't
work well, since I am in memset.

My immediate workaround is to make the loop more complex:
      Count := Count * 2; --  To avod recursive memset call
      while Count > 0 loop
         Convert_8.To_Pointer (Current_Address).all := Uint8_Value;
         Current_Address := Current_Address + 1;
         Count := Count - 2;
      end loop;

But I don't really like this as it adds unnecessary overhead.

I looked around for gcc switches to inhibit calls to memset, but the only
one I found ( -fno-builtins ) only works on C files.  I really don't want to 
write it in C (or  assembly for that matter).

I think I can also do ASM statements, but I would hate to have to do ASM if
there is an Ada solution.  I also don't know if GCC would just optimize
the ASM to a memset call anyways.

Do I have any other options that would let me do it in Ada?  I didn't see
pragmas that jumped out at me.