llvm dev - Jul 2010 - [LLVMdev] Win32 COFF Support patch 5 (the final patch in the saga!)

On Fri, Jul 30, 2010 at 2:34 PM, Daniel Dunbar <daniel at zuster.org>
wrote:
> On Fri, Jul 30, 2010 at 11:05 AM, Michael Spencer <bigcheesegs at
gmail.com> wrote:
>> I tried setting these to false, but the resulting executable (made
>> with link.exe) crashed. I looked at the differences in the object
>> file, but couldn't figure out what was causing it. I'm guessing
>> there's a bug in the path when these are false.
>>
>> Now that I know what these flags do, I'll try and figure out
exactly
>> what the problem is.
>
> If I had to guess, this is probably because the assembler isn't
> relaxing something that should be. Scattered symbols will hide this,
> because they will force most things to be relaxed, but I think we
> probably need a new bit to model the actual semantics in play. I've
> paged this stuff out, but I think the semantics the current
> implementation isn't modeling is the possible relocation of static
> symbols.
>
> If you try your example without the scattered symbols bit, but with
> -mc-relax-all, does the resulting binary work?
>
>  - Daniel
Can't do that with llc, but I'll just go change the code to force that
option in MC for this test.

Should the mc options be added to llc?

- Michael Spencer

On Fri, Jul 30, 2010 at 7:54 PM, Michael Spencer <bigcheesegs at
gmail.com> wrote:
> On Fri, Jul 30, 2010 at 2:34 PM, Daniel Dunbar <daniel at zuster.org>
wrote:
>> On Fri, Jul 30, 2010 at 11:05 AM, Michael Spencer <bigcheesegs at
gmail.com> wrote:
>>> I tried setting these to false, but the resulting executable (made
>>> with link.exe) crashed. I looked at the differences in the object
>>> file, but couldn't figure out what was causing it. I'm
guessing
>>> there's a bug in the path when these are false.
>>>
>>> Now that I know what these flags do, I'll try and figure out
exactly
>>> what the problem is.
>>
>> If I had to guess, this is probably because the assembler isn't
>> relaxing something that should be. Scattered symbols will hide this,
>> because they will force most things to be relaxed, but I think we
>> probably need a new bit to model the actual semantics in play. I've
>> paged this stuff out, but I think the semantics the current
>> implementation isn't modeling is the possible relocation of static
>> symbols.
>>
>> If you try your example without the scattered symbols bit, but with
>> -mc-relax-all, does the resulting binary work?
>>
>>  - Daniel
>
> Can't do that with llc, but I'll just go change the code to force
that
> option in MC for this test.
Err, I think it would be better to keep ScatteredSymbols=true than do
that. At least that is more likely to generate correct code. Adding a
FIXME above it is a fine start.
> Should the mc options be added to llc?
Sure, if you like.

 - Daniel
> - Michael Spencer
>

On Sat, Jul 31, 2010 at 12:37 AM, Daniel Dunbar <daniel at zuster.org>
wrote:
> On Fri, Jul 30, 2010 at 7:54 PM, Michael Spencer <bigcheesegs at
gmail.com> wrote:
>> On Fri, Jul 30, 2010 at 2:34 PM, Daniel Dunbar <daniel at
zuster.org> wrote:
>>> On Fri, Jul 30, 2010 at 11:05 AM, Michael Spencer <bigcheesegs
at gmail.com> wrote:
>>>> I tried setting these to false, but the resulting executable
(made
>>>> with link.exe) crashed. I looked at the differences in the
object
>>>> file, but couldn't figure out what was causing it. I'm
guessing
>>>> there's a bug in the path when these are false.
>>>>
>>>> Now that I know what these flags do, I'll try and figure
out exactly
>>>> what the problem is.
>>>
>>> If I had to guess, this is probably because the assembler isn't
>>> relaxing something that should be. Scattered symbols will hide
this,
>>> because they will force most things to be relaxed, but I think we
>>> probably need a new bit to model the actual semantics in play.
I've
>>> paged this stuff out, but I think the semantics the current
>>> implementation isn't modeling is the possible relocation of
static
>>> symbols.
>>>
>>> If you try your example without the scattered symbols bit, but with
>>> -mc-relax-all, does the resulting binary work?
>>>
>>>  - Daniel
>>
>> Can't do that with llc, but I'll just go change the code to
force that
>> option in MC for this test.
>
> Err, I think it would be better to keep ScatteredSymbols=true than do
> that. At least that is more likely to generate correct code. Adding a
> FIXME above it is a fine start.
I meant as a way to get the same behavior as -mc-relax-all on my local
tree, not to commit.
>> Should the mc options be added to llc?
>
> Sure, if you like.
Will do.

- Michael Spencer
>
>  - Daniel
>
>> - Michael Spencer
>>
>

On Sat, Jul 31, 2010 at 12:37 AM, Daniel Dunbar <daniel at zuster.org>
wrote:
> Err, I think it would be better to keep ScatteredSymbols=true than do
> that. At least that is more likely to generate correct code. Adding a
> FIXME above it is a fine start.
I tested it all out. Setting relax all to true doesn't fix the
problem, however, HasAbsolutizedSet was unneeded.

- Michael Spencer