[llvm] [AArch64][VecLib] Add vector function mappings for the modf, sincos, sincospi vector intrinsics (PR #175098)

[Benjamin Maxwell via llvm-commits]

================
@@ -1548,6 +1548,11 @@ TLI_DEFINE_VECFUNC("modff", "armpl_vmodfq_f32", FIXED(4), NOMASK, "_ZGV_LLVM_N4v
 TLI_DEFINE_VECFUNC("modf", "armpl_svmodf_f64_x", SCALABLE(2), MASKED, "_ZGVsMxvl8", NOCC)
 TLI_DEFINE_VECFUNC("modff", "armpl_svmodf_f32_x", SCALABLE(4), MASKED, "_ZGVsMxvl4", NOCC)
 
+TLI_DEFINE_VECFUNC("llvm.modf.f64", "armpl_vmodfq_f64_stret", FIXED(2), NOMASK, "_ZGV_LLVM_N2v", CallingConv::AArch64_VectorCall)
+TLI_DEFINE_VECFUNC("llvm.modf.f32", "armpl_vmodfq_f32_stret", FIXED(4), NOMASK, "_ZGV_LLVM_N4v", CallingConv::AArch64_VectorCall)
+TLI_DEFINE_VECFUNC("llvm.modf.f64", "armpl_svmodf_f64_stret_x", SCALABLE(2), MASKED, "_ZGVsMxv", NOCC)
+TLI_DEFINE_VECFUNC("llvm.modf.f32", "armpl_svmodf_f32_stret_x", SCALABLE(4), MASKED, "_ZGVsMxv", NOCC)
----------------
MacDue wrote:

I don't know if this is really an issue in practice, but the AArch64 VFABI does not support this kind of mapping. If you look at the RGB example: https://github.com/ARM-software/abi-aa/blob/main/vfabia64/vfabia64.rst#rgb-example

You see if we were to follow the VFABI the struct result should be returned using an output parameter (which is a vector of pointers). Obviously, this is not what we want, but it's something we noticed when looking at this in the past, and thought might need amending before libraries provided these functions. 

https://github.com/llvm/llvm-project/pull/175098