[AArch64] Rewrite the vdup_lane intrinsics in C

[AArch64] Rewrite the vdup_lane intrinsics in C

gcc/
	* config/aarch64/aarch64-simd-builtins.def
	(dup_lane_scalar): Remove.
	* config/aarch64/aarch64-simd.md
	(aarch64_simd_dup): Add 'w->w' alternative.
	(aarch64_dup_lane<mode>): Allow for VALL.
	(aarch64_dup_lane_scalar<mode>): Remove.
	(aarch64_dup_lane_<vswap_width_name><mode>): New.
	(aarch64_get_lane_signed<mode>): Add w->w altenative.
	(aarch64_get_lane_unsigned<mode>): Likewise.
	(aarch64_get_lane<mode>): Likewise.
	* config/aarch64/aarch64.c (aarch64_evpc_dup): New.
	(aarch64_expand_vec_perm_const_1): Use aarch64_evpc_dup.
	* config/aarch64/iterators.md (VSWAP_WIDTH): New.
	(VCON): Change container of V2SF.
	(vswap_width_name): Likewise.
	* config/aarch64/arm_neon.h
	(__aarch64_vdup_lane_any): New.
	(__aarch64_vdup<q>_lane<q>_<fpsu><8,16,32,64>): Likewise.
	(vdup<q>_n_<psuf><8,16,32,64>): Convert to C implementation.
	(vdup<q>_lane<q>_<fpsu><8,16,32,64>): Likewise.

gcc/testsuite/
	* gcc.target/aarch64/scalar_intrinsics.c
	(vdup<bhsd>_lane<su><8,16,32,64>): Force values to SIMD registers.

From-SVN: r202180