On 05/08/13 21:57, James Greenhalgh wrote:
This patch fixes up the vget_lane RTL patterns to better
exploit the behaviour of their target instructions, and
to allow variants keeping the result in the SIMD register file.
---
gcc/
2013-08-05 James Greenhalgh <james.greenha...@arm.com>
* config/aarch64/aarch64-simd-builtins.def (get_lane_signed): Remove.
(get_lane_unsigned): Likewise.
(dup_lane_scalar): Likewise.
(get_lane): enable for VALL.
* config/aarch64/aarch64-simd.md
(aarch64_dup_lane_scalar<mode>): Remove.
(aarch64_get_lane_signed<mode>): Likewise.
(aarch64_get_lane_unsigned<mode>): Likewise.
(aarch64_get_lane_extend<GPI:mode><VDQQH:mode>): New.
(aarch64_get_lane_zero_extendsi<mode>): Likewise.
(aarch64_get_lane<mode>): Enable for all vector modes.
(aarch64_get_lanedi): Remove misleading constraints.
* config/aarch64/arm_neon.h
(__aarch64_vget_lane_any): Define.
(__aarch64_vget<q>_lane_<fpsu><8,16,32,64>): Likewise.
(vget<q>_lane_<fpsu><8,16,32,64>): Use __aarch64_vget_lane macros.
(vdup<bhsd>_lane_<su><8,16,32,64>): Likewise.
* config/aarch64/iterators.md (VDQQH): New.
(VDQQHS): Likewise.
(vwcore): Likewise.
OK
/Marcus
