[PATCH][AArch64] Add support for vget/vset_lane_f16 ACLE functions

[Chad Rosier]

Hi Tim,
> Hi Chad,
>
>> I'm checking the target triple to allow for half-float arguments/return
>> types in Sema.  However, I'm currently looking into defining the AArch64
>> PCS as an alternative solution.  I was hoping we could get these patches
>> into the 3.4 release, so I'm sending patches with the less invasive
>> change
>> for now.
>
> I don't think this is the right solution. ACLE itself says that __fp16
> (and by extension float16_t) should *not* be permitted as an argument
> or return type (section 4.1.2). That should make this patch rather
> easier, actually.

According to the latest documentation I have the ACLE prototypes for
vget/vset_lane_f16 are:

float16_t vget_lane_f16(float16x4_t v, const int lane)
float16_t vgetq_lane_f16(float16x8_t v, const int lane)

float16x4_t vset_lane_f16(float16_t a, float16x4_t v, const int lane)
float16x8_t vsetq_lane_f16(float16_t a, float16x8_t v, const int lane)

Perhaps, I'm misunderstanding your or you're using outdated documentation?

 Chad

> Cheers.
>
> Tim.
>


-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation