[llvm] [TLI] Use AArch64 vector calling convention for ArmPL routines (PR #135790)

[David Sherwood via llvm-commits]

================
@@ -218,21 +218,21 @@ declare float @llvm.log10.f32(float) #0
 ; SLEEFGNUABI_RISCV: declare <vscale x 2 x double> @Sleef_sindx_u10rvvm2(<vscale x 2 x double>)
 ; SLEEFGNUABI_RISCV: declare <vscale x 4 x float> @Sleef_log10fx_u10rvvm2(<vscale x 4 x float>)
 
-; ARMPL: declare <2 x double> @armpl_vmodfq_f64(<2 x double>, ptr)
+; ARMPL: declare aarch64_vector_pcs <2 x double> @armpl_vmodfq_f64(<2 x double>, ptr)
 ; ARMPL: declare <vscale x 2 x double> @armpl_svmodf_f64_x(<vscale x 2 x double>, ptr, <vscale x 2 x i1>)
----------------
david-arm wrote:

Yeah I think so, which is why the attribute isn't added in the amath.h header shipped with ArmPL. I thought it's best to simply mirror what we do for the declarations in that header.

https://github.com/llvm/llvm-project/pull/135790