On 15/01/11 09:39, Walter Bright wrote:
> This is the 64 bit Linux version! Hopefully, the D2 64 bit alpha isn't too far behind.
>
> http://ftp.digitalmars.com/dmd1beta.zip
>
> The 64 bit support is ALPHA, so expect problems! Please post bugs here, not in bugzilla, until we do a real release.
>
>
> Known Problems
> --------------
>
> Code gen is not tuned for 64 bits. It's inefficient, especially with floating point.
>
> C ABI for struct and complex number pass/return is not followed.
>
> Dwarf symbolic debug info is generated, but not recognized by gdb. I have no idea what the problem is.
>
> std.zip is not working.
>
> std.boxer is not working.
>
>
>
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta

It seems the following doesn't work yet? (First issue I hit while trying to
run tango unittests - it compiles fine with the exception of the 64bit
vararg support for dmd which I'm working on).
test.d:
----
void foo(int, ...)
{
}
----
dmd -c -m64 test.d gives:

test.d(1): Error: '__va_argsave_t' is not defined, perhaps you need to
import std.c.stdarg; ?
test.d(1): Error: __va_argsave_t is used as a type
test.d(1): Error: variable test.foo.__va_argsave voids have no value

-- 
Robert
http://octarineparrot.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110115/85fa783d/attachment.html>

Having hacked around that, it seems not all x86_64 opcodes are supported? Not looked extensively, the first issue tango hits is the lack of pushq, no doubt there'll be others.

On 15 January 2011 12:19, Robert <robert at octarineparrot.com> wrote:

> On 15/01/11 09:39, Walter Bright wrote:
> > This is the 64 bit Linux version! Hopefully, the D2 64 bit alpha isn't too far behind.
> >
> > http://ftp.digitalmars.com/dmd1beta.zip
> >
> > The 64 bit support is ALPHA, so expect problems! Please post bugs here, not in bugzilla, until we do a real release.
> >
> >
> > Known Problems
> > --------------
> >
> > Code gen is not tuned for 64 bits. It's inefficient, especially with floating point.
> >
> > C ABI for struct and complex number pass/return is not followed.
> >
> > Dwarf symbolic debug info is generated, but not recognized by gdb. I have no idea what the problem is.
> >
> > std.zip is not working.
> >
> > std.boxer is not working.
> >
> >
> >
> > _______________________________________________
> > dmd-beta mailing list
> > dmd-beta at puremagic.com
> > http://lists.puremagic.com/mailman/listinfo/dmd-beta
>
> It seems the following doesn't work yet? (First issue I hit while trying to
> run tango unittests - it compiles fine with the exception of the 64bit
> vararg support for dmd which I'm working on).
> test.d:
> ----
> void foo(int, ...)
> {
> }
> ----
> dmd -c -m64 test.d gives:
>
> test.d(1): Error: '__va_argsave_t' is not defined, perhaps you need to
> import std.c.stdarg; ?
> test.d(1): Error: __va_argsave_t is used as a type
> test.d(1): Error: variable test.foo.__va_argsave voids have no value
>
> --
> Robert
> http://octarineparrot.com/
>



-- 
Robert
http://octarineparrot.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110115/b2b34583/attachment.html>

Robert wrote:
> Having hacked around that, it seems not all x86_64 opcodes are supported? Not looked extensively, the first issue tango hits is the lack of pushq, no doubt there'll be others.
>

According to my (AMD) instruction set reference, there is no pushq instruction. Just use push; they all push 8 bytes regardless of overrides.

On 15/01/11 19:48, Walter Bright wrote:
>
>
> Robert wrote:
>> Having hacked around that, it seems not all x86_64 opcodes are supported? Not looked extensively, the first issue tango hits is the lack of pushq, no doubt there'll be others.
>>
>
> According to my (AMD) instruction set reference, there is no pushq instruction. Just use push; they all push 8 bytes regardless of overrides.

Yes, I realised this shortly after I posted it, didn't get chance to reply though - ldc and gdc both accept it, just needed replacing to work with dmd.

-- 
Robert
http://octarineparrot.com/

Robert Clipsham wrote:
>
> Yes, I realised this shortly after I posted it, didn't get chance to reply though - ldc and gdc both accept it, just needed replacing to work with dmd.
>

I should add that dmd's 64 bit inline assembler has been tested only with Phobos' inline assembler usage.

obj2asm should work correctly with 64 bit elf files. As usual, I find obj2asm to be an invaluable tool in tracking down bad code gen.