[cfe-dev] [PATCH] Add preliminary Alpha support

[Eli Friedman]

On Fri, May 13, 2011 at 5:57 AM, Matt Turner <mattst88 at gmail.com> wrote:
> On Wed, May 11, 2011 at 4:48 PM, Andrew Lenharth <andrewl at lenharth.org> wrote:
>> On Wed, May 11, 2011 at 3:40 PM, Matt Turner <mattst88 at gmail.com> wrote:
>>> Signed-off-by: Matt Turner <mattst88 at gmail.com>
>>> ---
>>> So LLVM's Alpha backend works, somewhat. It generates code that segfaults at
>>> any optimization level except -O0, but this is a start at reviving it.
>>
>> The patch looks good, but I haven't looked at clang stuff for quite a
>> while.  Others can comment more here.
>>
>>> Things I'm unsure about:
>>>        DescriptionString - I think everything is natually aligned, but the MIPS
>>>        desc string I looked at specified 32-bit alignment for 8 and 16-bit units.
>>>
>>>        getDefaultFeatures - This seems to be where ISA additions should be created,
>>>        or something, but I see 'FIXME: This should not be here.' in other arch's
>>>        code. So, what goes here?
>>
>> If your assumption is correct, then probably nothing.  I took 21164
>> support out a few years ago, so only 21264 features are supported.
>>
>>>        getTargetDefines - I don't know what this function does. Sets what the
>>>        preprocessor is supposed to define? gcc defines __alpha__ and some other
>>>        things dependent on features supported. Should these things be defined here?
>>>
>>>        getTargetBuiltins - No idea what this is.
>>
>> I would assume things like:
>> __builtin_alpha_mskwh
>> __builtin_alpha_minub8
>>
>>>        validateAsmConstraint - I think this is for validating inline assembly inputs
>>>        and outputs? If so, this shouldn't be a problem.
>>
>> Yes.  I think you should only need to check INT/FP ness to validate most asm.
>>
>>>        getClobbers - Must be for inline assembly. The only relevant comments I see
>>>        in other code is '// FIXME: Is this really right?', so not terribly helpful.
>>
>> There shouldn't be hidden clobbers, so this doesn't seem too bad.
>
> Thanks for the review. Since there don't seem to be any NAKs, can we
> commit this and I'll send follow-up patches to fill out things like
> validateAsmConstraint?

Feel free to commit this.

-Eli