Re: [PATCH 1/4] Aarch64: Func to detect args passed in float regs

[Simon Marchi <simon.marchi@polymtl.ca>]

On 2018-08-28 11:49, Alan Hayward wrote:
>> On 28 Aug 2018, at 16:43, Simon Marchi <simon.marchi@polymtl.ca> 
>> wrote:
>> 
>> On 2018-08-20 05:29, Alan Hayward wrote:
>>> aapcs_is_vfp_call_or_return_candidate is as an eventual replacement
>>> for is_hfa_or_hva.
>>> This function is based on the GCC code
>>> gcc/config/aarch64/aarch64.c:aarch64_vfp_is_call_or_return_candidate 
>>> ()
>>> 2018-08-20  Alan Hayward  <alan.hayward@arm.com>
>>> 	* aarch64-tdep.c (HA_MAX_NUM_FLDS): New macro.
>>> 	(aapcs_is_vfp_call_or_return_candidate_1): New function.
>>> 	(aapcs_is_vfp_call_or_return_candidate): Likewise.
>> 
>> I'm not an AArch64 expert, but I didn't spot anything suspicious.  The 
>> documentation
>> helps a lot to understand, thanks for that.
>> 
>> 
>>> +/* Return true if an argument, whose type is described by TYPE, can
>>> be passed or
>>> +   returned in simd/fp registers, providing enough parameter passing 
>>> registers
>>> +   are available.  This is as described in the AAPCS64.
>>> +
>>> +   Upon successful return, *COUNT returns the number of needed 
>>> registers,
>>> +   *FUNDAMENTAL_TYPE contains the type of those registers.
>>> +
>>> +   Candidate as per the AAPCS64 5.4.2.C is either a:
>>> +   - float.
>>> +   - short-vector.
>>> +   - HFA (Homogeneous Floating-point Aggregate, 4.3.5.1). A 
>>> Composite
>>> type where
>>> +     all the members are floats and has at most 4 members.
>>> +   - HVA (Homogeneous Short-vector Aggregate, 4.3.5.2). A Composite 
>>> type where
>>> +     all the members are short vectors and has at most 4 members.
>>> +   - Complex (7.1.1)
>>> +
>>> +   Note that HFAs and HVAs can include nested structures and arrays. 
>>>  */
>>> +
>>> +bool
>> 
>> static?
>> 
> 
> Yes, it should be static. But, in this patch nothing calls the 
> function, which
> causes a compile failure. I make it static in patch 2.
> A little awkward, but seemed the best way to do it.
> 
> 
> Alan.

Ah, I see.  Another option would be to put __attribute__((unused)) 
temporarily.  But I'm fine with what you've done too.

Simon