amdgcn: Extend reductions to all types

amdgcn: Extend reductions to all types

Add support for V64DFmode addition, and V64DImode min, max.  There's no
direct hardware support for these, so we use regular vector instructions
and separate lane shift instructions.

Also add support for V64QI and V64HI reductions. Some of these require
additional extends and truncates, because AMD GCN has 32-bit vector lanes.

2020-03-02  Andrew Stubbs  <ams@codesourcery.com>

	gcc/
	* config/gcn/gcn-valu.md (dpp_move<mode>): New.
	(reduc_insn): Use 'U' and 'B' operand codes.
	(reduc_<reduc_op>_scal_<mode>): Allow all types.
	(reduc_<reduc_op>_scal_v64di): Delete.
	(*<reduc_op>_dpp_shr_<mode>): Allow all 1reg types.
	(*plus_carry_dpp_shr_v64si): Change to ...
	(*plus_carry_dpp_shr_<mode>): ... this and allow all 1reg int types.
	(mov_from_lane63_v64di): Change to ...
	(mov_from_lane63_<mode>): ... this, and allow all 64-bit modes.
	* config/gcn/gcn.c (gcn_expand_dpp_shr_insn): Increase buffer size.
	Support UNSPEC_MOV_DPP_SHR output formats.
	(gcn_expand_reduc_scalar): Add "use_moves" reductions.
	Add "use_extends" reductions.
	(print_operand_address): Add 'I' and 'U' codes.
	* config/gcn/gcn.md (unspec): Add UNSPEC_MOV_DPP_SHR.