[22/22] arm: [MVE intrinsics] rework vhaddq vhsubq vmulhq vqaddq vqsubq vqdmulhq vrhaddq vrmulhq

Message ID 20230418134608.244751-23-christophe.lyon@arm.com
State Accepted
Headers
Series arm: New framework for MVE intrinsics |

Checks

Context Check Description
snail/gcc-patch-check success Github commit url

Commit Message

Christophe Lyon April 18, 2023, 1:46 p.m. UTC
  Implement vhaddq, vhsubq, vmulhq, vqaddq, vqsubq, vqdmulhq, vrhaddq, vrmulhq using the new MVE builtins framework.

2022-09-08  Christophe Lyon <christophe.lyon@arm.com>

	gcc/
	* config/arm/arm-mve-builtins-base.cc (FUNCTION_WITH_M_N_NO_F)
	(FUNCTION_WITHOUT_N_NO_F, FUNCTION_WITH_M_N_NO_U_F): New.
	(vhaddq, vhsubq, vmulhq, vqaddq, vqsubq, vqdmulhq, vrhaddq)
	(vrmulhq): New.
	* config/arm/arm-mve-builtins-base.def (vhaddq, vhsubq, vmulhq)
	(vqaddq, vqsubq, vqdmulhq, vrhaddq, vrmulhq): New.
	* config/arm/arm-mve-builtins-base.h (vhaddq, vhsubq, vmulhq)
	(vqaddq, vqsubq, vqdmulhq, vrhaddq, vrmulhq): New.
	* config/arm/arm_mve.h (vhsubq): Remove.
	(vhaddq): Remove.
	(vhaddq_m): Remove.
	(vhsubq_m): Remove.
	(vhaddq_x): Remove.
	(vhsubq_x): Remove.
	(vhsubq_u8): Remove.
	(vhsubq_n_u8): Remove.
	(vhaddq_u8): Remove.
	(vhaddq_n_u8): Remove.
	(vhsubq_s8): Remove.
	(vhsubq_n_s8): Remove.
	(vhaddq_s8): Remove.
	(vhaddq_n_s8): Remove.
	(vhsubq_u16): Remove.
	(vhsubq_n_u16): Remove.
	(vhaddq_u16): Remove.
	(vhaddq_n_u16): Remove.
	(vhsubq_s16): Remove.
	(vhsubq_n_s16): Remove.
	(vhaddq_s16): Remove.
	(vhaddq_n_s16): Remove.
	(vhsubq_u32): Remove.
	(vhsubq_n_u32): Remove.
	(vhaddq_u32): Remove.
	(vhaddq_n_u32): Remove.
	(vhsubq_s32): Remove.
	(vhsubq_n_s32): Remove.
	(vhaddq_s32): Remove.
	(vhaddq_n_s32): Remove.
	(vhaddq_m_n_s8): Remove.
	(vhaddq_m_n_s32): Remove.
	(vhaddq_m_n_s16): Remove.
	(vhaddq_m_n_u8): Remove.
	(vhaddq_m_n_u32): Remove.
	(vhaddq_m_n_u16): Remove.
	(vhaddq_m_s8): Remove.
	(vhaddq_m_s32): Remove.
	(vhaddq_m_s16): Remove.
	(vhaddq_m_u8): Remove.
	(vhaddq_m_u32): Remove.
	(vhaddq_m_u16): Remove.
	(vhsubq_m_n_s8): Remove.
	(vhsubq_m_n_s32): Remove.
	(vhsubq_m_n_s16): Remove.
	(vhsubq_m_n_u8): Remove.
	(vhsubq_m_n_u32): Remove.
	(vhsubq_m_n_u16): Remove.
	(vhsubq_m_s8): Remove.
	(vhsubq_m_s32): Remove.
	(vhsubq_m_s16): Remove.
	(vhsubq_m_u8): Remove.
	(vhsubq_m_u32): Remove.
	(vhsubq_m_u16): Remove.
	(vhaddq_x_n_s8): Remove.
	(vhaddq_x_n_s16): Remove.
	(vhaddq_x_n_s32): Remove.
	(vhaddq_x_n_u8): Remove.
	(vhaddq_x_n_u16): Remove.
	(vhaddq_x_n_u32): Remove.
	(vhaddq_x_s8): Remove.
	(vhaddq_x_s16): Remove.
	(vhaddq_x_s32): Remove.
	(vhaddq_x_u8): Remove.
	(vhaddq_x_u16): Remove.
	(vhaddq_x_u32): Remove.
	(vhsubq_x_n_s8): Remove.
	(vhsubq_x_n_s16): Remove.
	(vhsubq_x_n_s32): Remove.
	(vhsubq_x_n_u8): Remove.
	(vhsubq_x_n_u16): Remove.
	(vhsubq_x_n_u32): Remove.
	(vhsubq_x_s8): Remove.
	(vhsubq_x_s16): Remove.
	(vhsubq_x_s32): Remove.
	(vhsubq_x_u8): Remove.
	(vhsubq_x_u16): Remove.
	(vhsubq_x_u32): Remove.
	(__arm_vhsubq_u8): Remove.
	(__arm_vhsubq_n_u8): Remove.
	(__arm_vhaddq_u8): Remove.
	(__arm_vhaddq_n_u8): Remove.
	(__arm_vhsubq_s8): Remove.
	(__arm_vhsubq_n_s8): Remove.
	(__arm_vhaddq_s8): Remove.
	(__arm_vhaddq_n_s8): Remove.
	(__arm_vhsubq_u16): Remove.
	(__arm_vhsubq_n_u16): Remove.
	(__arm_vhaddq_u16): Remove.
	(__arm_vhaddq_n_u16): Remove.
	(__arm_vhsubq_s16): Remove.
	(__arm_vhsubq_n_s16): Remove.
	(__arm_vhaddq_s16): Remove.
	(__arm_vhaddq_n_s16): Remove.
	(__arm_vhsubq_u32): Remove.
	(__arm_vhsubq_n_u32): Remove.
	(__arm_vhaddq_u32): Remove.
	(__arm_vhaddq_n_u32): Remove.
	(__arm_vhsubq_s32): Remove.
	(__arm_vhsubq_n_s32): Remove.
	(__arm_vhaddq_s32): Remove.
	(__arm_vhaddq_n_s32): Remove.
	(__arm_vhaddq_m_n_s8): Remove.
	(__arm_vhaddq_m_n_s32): Remove.
	(__arm_vhaddq_m_n_s16): Remove.
	(__arm_vhaddq_m_n_u8): Remove.
	(__arm_vhaddq_m_n_u32): Remove.
	(__arm_vhaddq_m_n_u16): Remove.
	(__arm_vhaddq_m_s8): Remove.
	(__arm_vhaddq_m_s32): Remove.
	(__arm_vhaddq_m_s16): Remove.
	(__arm_vhaddq_m_u8): Remove.
	(__arm_vhaddq_m_u32): Remove.
	(__arm_vhaddq_m_u16): Remove.
	(__arm_vhsubq_m_n_s8): Remove.
	(__arm_vhsubq_m_n_s32): Remove.
	(__arm_vhsubq_m_n_s16): Remove.
	(__arm_vhsubq_m_n_u8): Remove.
	(__arm_vhsubq_m_n_u32): Remove.
	(__arm_vhsubq_m_n_u16): Remove.
	(__arm_vhsubq_m_s8): Remove.
	(__arm_vhsubq_m_s32): Remove.
	(__arm_vhsubq_m_s16): Remove.
	(__arm_vhsubq_m_u8): Remove.
	(__arm_vhsubq_m_u32): Remove.
	(__arm_vhsubq_m_u16): Remove.
	(__arm_vhaddq_x_n_s8): Remove.
	(__arm_vhaddq_x_n_s16): Remove.
	(__arm_vhaddq_x_n_s32): Remove.
	(__arm_vhaddq_x_n_u8): Remove.
	(__arm_vhaddq_x_n_u16): Remove.
	(__arm_vhaddq_x_n_u32): Remove.
	(__arm_vhaddq_x_s8): Remove.
	(__arm_vhaddq_x_s16): Remove.
	(__arm_vhaddq_x_s32): Remove.
	(__arm_vhaddq_x_u8): Remove.
	(__arm_vhaddq_x_u16): Remove.
	(__arm_vhaddq_x_u32): Remove.
	(__arm_vhsubq_x_n_s8): Remove.
	(__arm_vhsubq_x_n_s16): Remove.
	(__arm_vhsubq_x_n_s32): Remove.
	(__arm_vhsubq_x_n_u8): Remove.
	(__arm_vhsubq_x_n_u16): Remove.
	(__arm_vhsubq_x_n_u32): Remove.
	(__arm_vhsubq_x_s8): Remove.
	(__arm_vhsubq_x_s16): Remove.
	(__arm_vhsubq_x_s32): Remove.
	(__arm_vhsubq_x_u8): Remove.
	(__arm_vhsubq_x_u16): Remove.
	(__arm_vhsubq_x_u32): Remove.
	(__arm_vhsubq): Remove.
	(__arm_vhaddq): Remove.
	(__arm_vhaddq_m): Remove.
	(__arm_vhsubq_m): Remove.
	(__arm_vhaddq_x): Remove.
	(__arm_vhsubq_x): Remove.
	(vmulhq): Remove.
	(vmulhq_m): Remove.
	(vmulhq_x): Remove.
	(vmulhq_u8): Remove.
	(vmulhq_s8): Remove.
	(vmulhq_u16): Remove.
	(vmulhq_s16): Remove.
	(vmulhq_u32): Remove.
	(vmulhq_s32): Remove.
	(vmulhq_m_s8): Remove.
	(vmulhq_m_s32): Remove.
	(vmulhq_m_s16): Remove.
	(vmulhq_m_u8): Remove.
	(vmulhq_m_u32): Remove.
	(vmulhq_m_u16): Remove.
	(vmulhq_x_s8): Remove.
	(vmulhq_x_s16): Remove.
	(vmulhq_x_s32): Remove.
	(vmulhq_x_u8): Remove.
	(vmulhq_x_u16): Remove.
	(vmulhq_x_u32): Remove.
	(__arm_vmulhq_u8): Remove.
	(__arm_vmulhq_s8): Remove.
	(__arm_vmulhq_u16): Remove.
	(__arm_vmulhq_s16): Remove.
	(__arm_vmulhq_u32): Remove.
	(__arm_vmulhq_s32): Remove.
	(__arm_vmulhq_m_s8): Remove.
	(__arm_vmulhq_m_s32): Remove.
	(__arm_vmulhq_m_s16): Remove.
	(__arm_vmulhq_m_u8): Remove.
	(__arm_vmulhq_m_u32): Remove.
	(__arm_vmulhq_m_u16): Remove.
	(__arm_vmulhq_x_s8): Remove.
	(__arm_vmulhq_x_s16): Remove.
	(__arm_vmulhq_x_s32): Remove.
	(__arm_vmulhq_x_u8): Remove.
	(__arm_vmulhq_x_u16): Remove.
	(__arm_vmulhq_x_u32): Remove.
	(__arm_vmulhq): Remove.
	(__arm_vmulhq_m): Remove.
	(__arm_vmulhq_x): Remove.
	(vqsubq): Remove.
	(vqaddq): Remove.
	(vqaddq_m): Remove.
	(vqsubq_m): Remove.
	(vqsubq_u8): Remove.
	(vqsubq_n_u8): Remove.
	(vqaddq_u8): Remove.
	(vqaddq_n_u8): Remove.
	(vqsubq_s8): Remove.
	(vqsubq_n_s8): Remove.
	(vqaddq_s8): Remove.
	(vqaddq_n_s8): Remove.
	(vqsubq_u16): Remove.
	(vqsubq_n_u16): Remove.
	(vqaddq_u16): Remove.
	(vqaddq_n_u16): Remove.
	(vqsubq_s16): Remove.
	(vqsubq_n_s16): Remove.
	(vqaddq_s16): Remove.
	(vqaddq_n_s16): Remove.
	(vqsubq_u32): Remove.
	(vqsubq_n_u32): Remove.
	(vqaddq_u32): Remove.
	(vqaddq_n_u32): Remove.
	(vqsubq_s32): Remove.
	(vqsubq_n_s32): Remove.
	(vqaddq_s32): Remove.
	(vqaddq_n_s32): Remove.
	(vqaddq_m_n_s8): Remove.
	(vqaddq_m_n_s32): Remove.
	(vqaddq_m_n_s16): Remove.
	(vqaddq_m_n_u8): Remove.
	(vqaddq_m_n_u32): Remove.
	(vqaddq_m_n_u16): Remove.
	(vqaddq_m_s8): Remove.
	(vqaddq_m_s32): Remove.
	(vqaddq_m_s16): Remove.
	(vqaddq_m_u8): Remove.
	(vqaddq_m_u32): Remove.
	(vqaddq_m_u16): Remove.
	(vqsubq_m_n_s8): Remove.
	(vqsubq_m_n_s32): Remove.
	(vqsubq_m_n_s16): Remove.
	(vqsubq_m_n_u8): Remove.
	(vqsubq_m_n_u32): Remove.
	(vqsubq_m_n_u16): Remove.
	(vqsubq_m_s8): Remove.
	(vqsubq_m_s32): Remove.
	(vqsubq_m_s16): Remove.
	(vqsubq_m_u8): Remove.
	(vqsubq_m_u32): Remove.
	(vqsubq_m_u16): Remove.
	(__arm_vqsubq_u8): Remove.
	(__arm_vqsubq_n_u8): Remove.
	(__arm_vqaddq_u8): Remove.
	(__arm_vqaddq_n_u8): Remove.
	(__arm_vqsubq_s8): Remove.
	(__arm_vqsubq_n_s8): Remove.
	(__arm_vqaddq_s8): Remove.
	(__arm_vqaddq_n_s8): Remove.
	(__arm_vqsubq_u16): Remove.
	(__arm_vqsubq_n_u16): Remove.
	(__arm_vqaddq_u16): Remove.
	(__arm_vqaddq_n_u16): Remove.
	(__arm_vqsubq_s16): Remove.
	(__arm_vqsubq_n_s16): Remove.
	(__arm_vqaddq_s16): Remove.
	(__arm_vqaddq_n_s16): Remove.
	(__arm_vqsubq_u32): Remove.
	(__arm_vqsubq_n_u32): Remove.
	(__arm_vqaddq_u32): Remove.
	(__arm_vqaddq_n_u32): Remove.
	(__arm_vqsubq_s32): Remove.
	(__arm_vqsubq_n_s32): Remove.
	(__arm_vqaddq_s32): Remove.
	(__arm_vqaddq_n_s32): Remove.
	(__arm_vqaddq_m_n_s8): Remove.
	(__arm_vqaddq_m_n_s32): Remove.
	(__arm_vqaddq_m_n_s16): Remove.
	(__arm_vqaddq_m_n_u8): Remove.
	(__arm_vqaddq_m_n_u32): Remove.
	(__arm_vqaddq_m_n_u16): Remove.
	(__arm_vqaddq_m_s8): Remove.
	(__arm_vqaddq_m_s32): Remove.
	(__arm_vqaddq_m_s16): Remove.
	(__arm_vqaddq_m_u8): Remove.
	(__arm_vqaddq_m_u32): Remove.
	(__arm_vqaddq_m_u16): Remove.
	(__arm_vqsubq_m_n_s8): Remove.
	(__arm_vqsubq_m_n_s32): Remove.
	(__arm_vqsubq_m_n_s16): Remove.
	(__arm_vqsubq_m_n_u8): Remove.
	(__arm_vqsubq_m_n_u32): Remove.
	(__arm_vqsubq_m_n_u16): Remove.
	(__arm_vqsubq_m_s8): Remove.
	(__arm_vqsubq_m_s32): Remove.
	(__arm_vqsubq_m_s16): Remove.
	(__arm_vqsubq_m_u8): Remove.
	(__arm_vqsubq_m_u32): Remove.
	(__arm_vqsubq_m_u16): Remove.
	(__arm_vqsubq): Remove.
	(__arm_vqaddq): Remove.
	(__arm_vqaddq_m): Remove.
	(__arm_vqsubq_m): Remove.
	(vqdmulhq): Remove.
	(vqdmulhq_m): Remove.
	(vqdmulhq_s8): Remove.
	(vqdmulhq_n_s8): Remove.
	(vqdmulhq_s16): Remove.
	(vqdmulhq_n_s16): Remove.
	(vqdmulhq_s32): Remove.
	(vqdmulhq_n_s32): Remove.
	(vqdmulhq_m_n_s8): Remove.
	(vqdmulhq_m_n_s32): Remove.
	(vqdmulhq_m_n_s16): Remove.
	(vqdmulhq_m_s8): Remove.
	(vqdmulhq_m_s32): Remove.
	(vqdmulhq_m_s16): Remove.
	(__arm_vqdmulhq_s8): Remove.
	(__arm_vqdmulhq_n_s8): Remove.
	(__arm_vqdmulhq_s16): Remove.
	(__arm_vqdmulhq_n_s16): Remove.
	(__arm_vqdmulhq_s32): Remove.
	(__arm_vqdmulhq_n_s32): Remove.
	(__arm_vqdmulhq_m_n_s8): Remove.
	(__arm_vqdmulhq_m_n_s32): Remove.
	(__arm_vqdmulhq_m_n_s16): Remove.
	(__arm_vqdmulhq_m_s8): Remove.
	(__arm_vqdmulhq_m_s32): Remove.
	(__arm_vqdmulhq_m_s16): Remove.
	(__arm_vqdmulhq): Remove.
	(__arm_vqdmulhq_m): Remove.
	(vrhaddq): Remove.
	(vrhaddq_m): Remove.
	(vrhaddq_x): Remove.
	(vrhaddq_u8): Remove.
	(vrhaddq_s8): Remove.
	(vrhaddq_u16): Remove.
	(vrhaddq_s16): Remove.
	(vrhaddq_u32): Remove.
	(vrhaddq_s32): Remove.
	(vrhaddq_m_s8): Remove.
	(vrhaddq_m_s32): Remove.
	(vrhaddq_m_s16): Remove.
	(vrhaddq_m_u8): Remove.
	(vrhaddq_m_u32): Remove.
	(vrhaddq_m_u16): Remove.
	(vrhaddq_x_s8): Remove.
	(vrhaddq_x_s16): Remove.
	(vrhaddq_x_s32): Remove.
	(vrhaddq_x_u8): Remove.
	(vrhaddq_x_u16): Remove.
	(vrhaddq_x_u32): Remove.
	(__arm_vrhaddq_u8): Remove.
	(__arm_vrhaddq_s8): Remove.
	(__arm_vrhaddq_u16): Remove.
	(__arm_vrhaddq_s16): Remove.
	(__arm_vrhaddq_u32): Remove.
	(__arm_vrhaddq_s32): Remove.
	(__arm_vrhaddq_m_s8): Remove.
	(__arm_vrhaddq_m_s32): Remove.
	(__arm_vrhaddq_m_s16): Remove.
	(__arm_vrhaddq_m_u8): Remove.
	(__arm_vrhaddq_m_u32): Remove.
	(__arm_vrhaddq_m_u16): Remove.
	(__arm_vrhaddq_x_s8): Remove.
	(__arm_vrhaddq_x_s16): Remove.
	(__arm_vrhaddq_x_s32): Remove.
	(__arm_vrhaddq_x_u8): Remove.
	(__arm_vrhaddq_x_u16): Remove.
	(__arm_vrhaddq_x_u32): Remove.
	(__arm_vrhaddq): Remove.
	(__arm_vrhaddq_m): Remove.
	(__arm_vrhaddq_x): Remove.
	(vrmulhq): Remove.
	(vrmulhq_m): Remove.
	(vrmulhq_x): Remove.
	(vrmulhq_u8): Remove.
	(vrmulhq_s8): Remove.
	(vrmulhq_u16): Remove.
	(vrmulhq_s16): Remove.
	(vrmulhq_u32): Remove.
	(vrmulhq_s32): Remove.
	(vrmulhq_m_s8): Remove.
	(vrmulhq_m_s32): Remove.
	(vrmulhq_m_s16): Remove.
	(vrmulhq_m_u8): Remove.
	(vrmulhq_m_u32): Remove.
	(vrmulhq_m_u16): Remove.
	(vrmulhq_x_s8): Remove.
	(vrmulhq_x_s16): Remove.
	(vrmulhq_x_s32): Remove.
	(vrmulhq_x_u8): Remove.
	(vrmulhq_x_u16): Remove.
	(vrmulhq_x_u32): Remove.
	(__arm_vrmulhq_u8): Remove.
	(__arm_vrmulhq_s8): Remove.
	(__arm_vrmulhq_u16): Remove.
	(__arm_vrmulhq_s16): Remove.
	(__arm_vrmulhq_u32): Remove.
	(__arm_vrmulhq_s32): Remove.
	(__arm_vrmulhq_m_s8): Remove.
	(__arm_vrmulhq_m_s32): Remove.
	(__arm_vrmulhq_m_s16): Remove.
	(__arm_vrmulhq_m_u8): Remove.
	(__arm_vrmulhq_m_u32): Remove.
	(__arm_vrmulhq_m_u16): Remove.
	(__arm_vrmulhq_x_s8): Remove.
	(__arm_vrmulhq_x_s16): Remove.
	(__arm_vrmulhq_x_s32): Remove.
	(__arm_vrmulhq_x_u8): Remove.
	(__arm_vrmulhq_x_u16): Remove.
	(__arm_vrmulhq_x_u32): Remove.
	(__arm_vrmulhq): Remove.
	(__arm_vrmulhq_m): Remove.
	(__arm_vrmulhq_x): Remove.
---
 gcc/config/arm/arm-mve-builtins-base.cc  |   35 +
 gcc/config/arm/arm-mve-builtins-base.def |    8 +
 gcc/config/arm/arm-mve-builtins-base.h   |    8 +
 gcc/config/arm/arm_mve.h                 | 3203 ----------------------
 4 files changed, 51 insertions(+), 3203 deletions(-)
  

Comments

Kyrylo Tkachov May 3, 2023, 8:51 a.m. UTC | #1
> -----Original Message-----
> From: Christophe Lyon <christophe.lyon@arm.com>
> Sent: Tuesday, April 18, 2023 2:46 PM
> To: gcc-patches@gcc.gnu.org; Kyrylo Tkachov <Kyrylo.Tkachov@arm.com>;
> Richard Earnshaw <Richard.Earnshaw@arm.com>; Richard Sandiford
> <Richard.Sandiford@arm.com>
> Cc: Christophe Lyon <Christophe.Lyon@arm.com>
> Subject: [PATCH 22/22] arm: [MVE intrinsics] rework vhaddq vhsubq vmulhq
> vqaddq vqsubq vqdmulhq vrhaddq vrmulhq
> 
> Implement vhaddq, vhsubq, vmulhq, vqaddq, vqsubq, vqdmulhq, vrhaddq,
> vrmulhq using the new MVE builtins framework.

Ok.
Thanks,
Kyrill

> 
> 2022-09-08  Christophe Lyon <christophe.lyon@arm.com>
> 
> 	gcc/
> 	* config/arm/arm-mve-builtins-base.cc
> (FUNCTION_WITH_M_N_NO_F)
> 	(FUNCTION_WITHOUT_N_NO_F, FUNCTION_WITH_M_N_NO_U_F):
> New.
> 	(vhaddq, vhsubq, vmulhq, vqaddq, vqsubq, vqdmulhq, vrhaddq)
> 	(vrmulhq): New.
> 	* config/arm/arm-mve-builtins-base.def (vhaddq, vhsubq, vmulhq)
> 	(vqaddq, vqsubq, vqdmulhq, vrhaddq, vrmulhq): New.
> 	* config/arm/arm-mve-builtins-base.h (vhaddq, vhsubq, vmulhq)
> 	(vqaddq, vqsubq, vqdmulhq, vrhaddq, vrmulhq): New.
> 	* config/arm/arm_mve.h (vhsubq): Remove.
> 	(vhaddq): Remove.
> 	(vhaddq_m): Remove.
> 	(vhsubq_m): Remove.
> 	(vhaddq_x): Remove.
> 	(vhsubq_x): Remove.
> 	(vhsubq_u8): Remove.
> 	(vhsubq_n_u8): Remove.
> 	(vhaddq_u8): Remove.
> 	(vhaddq_n_u8): Remove.
> 	(vhsubq_s8): Remove.
> 	(vhsubq_n_s8): Remove.
> 	(vhaddq_s8): Remove.
> 	(vhaddq_n_s8): Remove.
> 	(vhsubq_u16): Remove.
> 	(vhsubq_n_u16): Remove.
> 	(vhaddq_u16): Remove.
> 	(vhaddq_n_u16): Remove.
> 	(vhsubq_s16): Remove.
> 	(vhsubq_n_s16): Remove.
> 	(vhaddq_s16): Remove.
> 	(vhaddq_n_s16): Remove.
> 	(vhsubq_u32): Remove.
> 	(vhsubq_n_u32): Remove.
> 	(vhaddq_u32): Remove.
> 	(vhaddq_n_u32): Remove.
> 	(vhsubq_s32): Remove.
> 	(vhsubq_n_s32): Remove.
> 	(vhaddq_s32): Remove.
> 	(vhaddq_n_s32): Remove.
> 	(vhaddq_m_n_s8): Remove.
> 	(vhaddq_m_n_s32): Remove.
> 	(vhaddq_m_n_s16): Remove.
> 	(vhaddq_m_n_u8): Remove.
> 	(vhaddq_m_n_u32): Remove.
> 	(vhaddq_m_n_u16): Remove.
> 	(vhaddq_m_s8): Remove.
> 	(vhaddq_m_s32): Remove.
> 	(vhaddq_m_s16): Remove.
> 	(vhaddq_m_u8): Remove.
> 	(vhaddq_m_u32): Remove.
> 	(vhaddq_m_u16): Remove.
> 	(vhsubq_m_n_s8): Remove.
> 	(vhsubq_m_n_s32): Remove.
> 	(vhsubq_m_n_s16): Remove.
> 	(vhsubq_m_n_u8): Remove.
> 	(vhsubq_m_n_u32): Remove.
> 	(vhsubq_m_n_u16): Remove.
> 	(vhsubq_m_s8): Remove.
> 	(vhsubq_m_s32): Remove.
> 	(vhsubq_m_s16): Remove.
> 	(vhsubq_m_u8): Remove.
> 	(vhsubq_m_u32): Remove.
> 	(vhsubq_m_u16): Remove.
> 	(vhaddq_x_n_s8): Remove.
> 	(vhaddq_x_n_s16): Remove.
> 	(vhaddq_x_n_s32): Remove.
> 	(vhaddq_x_n_u8): Remove.
> 	(vhaddq_x_n_u16): Remove.
> 	(vhaddq_x_n_u32): Remove.
> 	(vhaddq_x_s8): Remove.
> 	(vhaddq_x_s16): Remove.
> 	(vhaddq_x_s32): Remove.
> 	(vhaddq_x_u8): Remove.
> 	(vhaddq_x_u16): Remove.
> 	(vhaddq_x_u32): Remove.
> 	(vhsubq_x_n_s8): Remove.
> 	(vhsubq_x_n_s16): Remove.
> 	(vhsubq_x_n_s32): Remove.
> 	(vhsubq_x_n_u8): Remove.
> 	(vhsubq_x_n_u16): Remove.
> 	(vhsubq_x_n_u32): Remove.
> 	(vhsubq_x_s8): Remove.
> 	(vhsubq_x_s16): Remove.
> 	(vhsubq_x_s32): Remove.
> 	(vhsubq_x_u8): Remove.
> 	(vhsubq_x_u16): Remove.
> 	(vhsubq_x_u32): Remove.
> 	(__arm_vhsubq_u8): Remove.
> 	(__arm_vhsubq_n_u8): Remove.
> 	(__arm_vhaddq_u8): Remove.
> 	(__arm_vhaddq_n_u8): Remove.
> 	(__arm_vhsubq_s8): Remove.
> 	(__arm_vhsubq_n_s8): Remove.
> 	(__arm_vhaddq_s8): Remove.
> 	(__arm_vhaddq_n_s8): Remove.
> 	(__arm_vhsubq_u16): Remove.
> 	(__arm_vhsubq_n_u16): Remove.
> 	(__arm_vhaddq_u16): Remove.
> 	(__arm_vhaddq_n_u16): Remove.
> 	(__arm_vhsubq_s16): Remove.
> 	(__arm_vhsubq_n_s16): Remove.
> 	(__arm_vhaddq_s16): Remove.
> 	(__arm_vhaddq_n_s16): Remove.
> 	(__arm_vhsubq_u32): Remove.
> 	(__arm_vhsubq_n_u32): Remove.
> 	(__arm_vhaddq_u32): Remove.
> 	(__arm_vhaddq_n_u32): Remove.
> 	(__arm_vhsubq_s32): Remove.
> 	(__arm_vhsubq_n_s32): Remove.
> 	(__arm_vhaddq_s32): Remove.
> 	(__arm_vhaddq_n_s32): Remove.
> 	(__arm_vhaddq_m_n_s8): Remove.
> 	(__arm_vhaddq_m_n_s32): Remove.
> 	(__arm_vhaddq_m_n_s16): Remove.
> 	(__arm_vhaddq_m_n_u8): Remove.
> 	(__arm_vhaddq_m_n_u32): Remove.
> 	(__arm_vhaddq_m_n_u16): Remove.
> 	(__arm_vhaddq_m_s8): Remove.
> 	(__arm_vhaddq_m_s32): Remove.
> 	(__arm_vhaddq_m_s16): Remove.
> 	(__arm_vhaddq_m_u8): Remove.
> 	(__arm_vhaddq_m_u32): Remove.
> 	(__arm_vhaddq_m_u16): Remove.
> 	(__arm_vhsubq_m_n_s8): Remove.
> 	(__arm_vhsubq_m_n_s32): Remove.
> 	(__arm_vhsubq_m_n_s16): Remove.
> 	(__arm_vhsubq_m_n_u8): Remove.
> 	(__arm_vhsubq_m_n_u32): Remove.
> 	(__arm_vhsubq_m_n_u16): Remove.
> 	(__arm_vhsubq_m_s8): Remove.
> 	(__arm_vhsubq_m_s32): Remove.
> 	(__arm_vhsubq_m_s16): Remove.
> 	(__arm_vhsubq_m_u8): Remove.
> 	(__arm_vhsubq_m_u32): Remove.
> 	(__arm_vhsubq_m_u16): Remove.
> 	(__arm_vhaddq_x_n_s8): Remove.
> 	(__arm_vhaddq_x_n_s16): Remove.
> 	(__arm_vhaddq_x_n_s32): Remove.
> 	(__arm_vhaddq_x_n_u8): Remove.
> 	(__arm_vhaddq_x_n_u16): Remove.
> 	(__arm_vhaddq_x_n_u32): Remove.
> 	(__arm_vhaddq_x_s8): Remove.
> 	(__arm_vhaddq_x_s16): Remove.
> 	(__arm_vhaddq_x_s32): Remove.
> 	(__arm_vhaddq_x_u8): Remove.
> 	(__arm_vhaddq_x_u16): Remove.
> 	(__arm_vhaddq_x_u32): Remove.
> 	(__arm_vhsubq_x_n_s8): Remove.
> 	(__arm_vhsubq_x_n_s16): Remove.
> 	(__arm_vhsubq_x_n_s32): Remove.
> 	(__arm_vhsubq_x_n_u8): Remove.
> 	(__arm_vhsubq_x_n_u16): Remove.
> 	(__arm_vhsubq_x_n_u32): Remove.
> 	(__arm_vhsubq_x_s8): Remove.
> 	(__arm_vhsubq_x_s16): Remove.
> 	(__arm_vhsubq_x_s32): Remove.
> 	(__arm_vhsubq_x_u8): Remove.
> 	(__arm_vhsubq_x_u16): Remove.
> 	(__arm_vhsubq_x_u32): Remove.
> 	(__arm_vhsubq): Remove.
> 	(__arm_vhaddq): Remove.
> 	(__arm_vhaddq_m): Remove.
> 	(__arm_vhsubq_m): Remove.
> 	(__arm_vhaddq_x): Remove.
> 	(__arm_vhsubq_x): Remove.
> 	(vmulhq): Remove.
> 	(vmulhq_m): Remove.
> 	(vmulhq_x): Remove.
> 	(vmulhq_u8): Remove.
> 	(vmulhq_s8): Remove.
> 	(vmulhq_u16): Remove.
> 	(vmulhq_s16): Remove.
> 	(vmulhq_u32): Remove.
> 	(vmulhq_s32): Remove.
> 	(vmulhq_m_s8): Remove.
> 	(vmulhq_m_s32): Remove.
> 	(vmulhq_m_s16): Remove.
> 	(vmulhq_m_u8): Remove.
> 	(vmulhq_m_u32): Remove.
> 	(vmulhq_m_u16): Remove.
> 	(vmulhq_x_s8): Remove.
> 	(vmulhq_x_s16): Remove.
> 	(vmulhq_x_s32): Remove.
> 	(vmulhq_x_u8): Remove.
> 	(vmulhq_x_u16): Remove.
> 	(vmulhq_x_u32): Remove.
> 	(__arm_vmulhq_u8): Remove.
> 	(__arm_vmulhq_s8): Remove.
> 	(__arm_vmulhq_u16): Remove.
> 	(__arm_vmulhq_s16): Remove.
> 	(__arm_vmulhq_u32): Remove.
> 	(__arm_vmulhq_s32): Remove.
> 	(__arm_vmulhq_m_s8): Remove.
> 	(__arm_vmulhq_m_s32): Remove.
> 	(__arm_vmulhq_m_s16): Remove.
> 	(__arm_vmulhq_m_u8): Remove.
> 	(__arm_vmulhq_m_u32): Remove.
> 	(__arm_vmulhq_m_u16): Remove.
> 	(__arm_vmulhq_x_s8): Remove.
> 	(__arm_vmulhq_x_s16): Remove.
> 	(__arm_vmulhq_x_s32): Remove.
> 	(__arm_vmulhq_x_u8): Remove.
> 	(__arm_vmulhq_x_u16): Remove.
> 	(__arm_vmulhq_x_u32): Remove.
> 	(__arm_vmulhq): Remove.
> 	(__arm_vmulhq_m): Remove.
> 	(__arm_vmulhq_x): Remove.
> 	(vqsubq): Remove.
> 	(vqaddq): Remove.
> 	(vqaddq_m): Remove.
> 	(vqsubq_m): Remove.
> 	(vqsubq_u8): Remove.
> 	(vqsubq_n_u8): Remove.
> 	(vqaddq_u8): Remove.
> 	(vqaddq_n_u8): Remove.
> 	(vqsubq_s8): Remove.
> 	(vqsubq_n_s8): Remove.
> 	(vqaddq_s8): Remove.
> 	(vqaddq_n_s8): Remove.
> 	(vqsubq_u16): Remove.
> 	(vqsubq_n_u16): Remove.
> 	(vqaddq_u16): Remove.
> 	(vqaddq_n_u16): Remove.
> 	(vqsubq_s16): Remove.
> 	(vqsubq_n_s16): Remove.
> 	(vqaddq_s16): Remove.
> 	(vqaddq_n_s16): Remove.
> 	(vqsubq_u32): Remove.
> 	(vqsubq_n_u32): Remove.
> 	(vqaddq_u32): Remove.
> 	(vqaddq_n_u32): Remove.
> 	(vqsubq_s32): Remove.
> 	(vqsubq_n_s32): Remove.
> 	(vqaddq_s32): Remove.
> 	(vqaddq_n_s32): Remove.
> 	(vqaddq_m_n_s8): Remove.
> 	(vqaddq_m_n_s32): Remove.
> 	(vqaddq_m_n_s16): Remove.
> 	(vqaddq_m_n_u8): Remove.
> 	(vqaddq_m_n_u32): Remove.
> 	(vqaddq_m_n_u16): Remove.
> 	(vqaddq_m_s8): Remove.
> 	(vqaddq_m_s32): Remove.
> 	(vqaddq_m_s16): Remove.
> 	(vqaddq_m_u8): Remove.
> 	(vqaddq_m_u32): Remove.
> 	(vqaddq_m_u16): Remove.
> 	(vqsubq_m_n_s8): Remove.
> 	(vqsubq_m_n_s32): Remove.
> 	(vqsubq_m_n_s16): Remove.
> 	(vqsubq_m_n_u8): Remove.
> 	(vqsubq_m_n_u32): Remove.
> 	(vqsubq_m_n_u16): Remove.
> 	(vqsubq_m_s8): Remove.
> 	(vqsubq_m_s32): Remove.
> 	(vqsubq_m_s16): Remove.
> 	(vqsubq_m_u8): Remove.
> 	(vqsubq_m_u32): Remove.
> 	(vqsubq_m_u16): Remove.
> 	(__arm_vqsubq_u8): Remove.
> 	(__arm_vqsubq_n_u8): Remove.
> 	(__arm_vqaddq_u8): Remove.
> 	(__arm_vqaddq_n_u8): Remove.
> 	(__arm_vqsubq_s8): Remove.
> 	(__arm_vqsubq_n_s8): Remove.
> 	(__arm_vqaddq_s8): Remove.
> 	(__arm_vqaddq_n_s8): Remove.
> 	(__arm_vqsubq_u16): Remove.
> 	(__arm_vqsubq_n_u16): Remove.
> 	(__arm_vqaddq_u16): Remove.
> 	(__arm_vqaddq_n_u16): Remove.
> 	(__arm_vqsubq_s16): Remove.
> 	(__arm_vqsubq_n_s16): Remove.
> 	(__arm_vqaddq_s16): Remove.
> 	(__arm_vqaddq_n_s16): Remove.
> 	(__arm_vqsubq_u32): Remove.
> 	(__arm_vqsubq_n_u32): Remove.
> 	(__arm_vqaddq_u32): Remove.
> 	(__arm_vqaddq_n_u32): Remove.
> 	(__arm_vqsubq_s32): Remove.
> 	(__arm_vqsubq_n_s32): Remove.
> 	(__arm_vqaddq_s32): Remove.
> 	(__arm_vqaddq_n_s32): Remove.
> 	(__arm_vqaddq_m_n_s8): Remove.
> 	(__arm_vqaddq_m_n_s32): Remove.
> 	(__arm_vqaddq_m_n_s16): Remove.
> 	(__arm_vqaddq_m_n_u8): Remove.
> 	(__arm_vqaddq_m_n_u32): Remove.
> 	(__arm_vqaddq_m_n_u16): Remove.
> 	(__arm_vqaddq_m_s8): Remove.
> 	(__arm_vqaddq_m_s32): Remove.
> 	(__arm_vqaddq_m_s16): Remove.
> 	(__arm_vqaddq_m_u8): Remove.
> 	(__arm_vqaddq_m_u32): Remove.
> 	(__arm_vqaddq_m_u16): Remove.
> 	(__arm_vqsubq_m_n_s8): Remove.
> 	(__arm_vqsubq_m_n_s32): Remove.
> 	(__arm_vqsubq_m_n_s16): Remove.
> 	(__arm_vqsubq_m_n_u8): Remove.
> 	(__arm_vqsubq_m_n_u32): Remove.
> 	(__arm_vqsubq_m_n_u16): Remove.
> 	(__arm_vqsubq_m_s8): Remove.
> 	(__arm_vqsubq_m_s32): Remove.
> 	(__arm_vqsubq_m_s16): Remove.
> 	(__arm_vqsubq_m_u8): Remove.
> 	(__arm_vqsubq_m_u32): Remove.
> 	(__arm_vqsubq_m_u16): Remove.
> 	(__arm_vqsubq): Remove.
> 	(__arm_vqaddq): Remove.
> 	(__arm_vqaddq_m): Remove.
> 	(__arm_vqsubq_m): Remove.
> 	(vqdmulhq): Remove.
> 	(vqdmulhq_m): Remove.
> 	(vqdmulhq_s8): Remove.
> 	(vqdmulhq_n_s8): Remove.
> 	(vqdmulhq_s16): Remove.
> 	(vqdmulhq_n_s16): Remove.
> 	(vqdmulhq_s32): Remove.
> 	(vqdmulhq_n_s32): Remove.
> 	(vqdmulhq_m_n_s8): Remove.
> 	(vqdmulhq_m_n_s32): Remove.
> 	(vqdmulhq_m_n_s16): Remove.
> 	(vqdmulhq_m_s8): Remove.
> 	(vqdmulhq_m_s32): Remove.
> 	(vqdmulhq_m_s16): Remove.
> 	(__arm_vqdmulhq_s8): Remove.
> 	(__arm_vqdmulhq_n_s8): Remove.
> 	(__arm_vqdmulhq_s16): Remove.
> 	(__arm_vqdmulhq_n_s16): Remove.
> 	(__arm_vqdmulhq_s32): Remove.
> 	(__arm_vqdmulhq_n_s32): Remove.
> 	(__arm_vqdmulhq_m_n_s8): Remove.
> 	(__arm_vqdmulhq_m_n_s32): Remove.
> 	(__arm_vqdmulhq_m_n_s16): Remove.
> 	(__arm_vqdmulhq_m_s8): Remove.
> 	(__arm_vqdmulhq_m_s32): Remove.
> 	(__arm_vqdmulhq_m_s16): Remove.
> 	(__arm_vqdmulhq): Remove.
> 	(__arm_vqdmulhq_m): Remove.
> 	(vrhaddq): Remove.
> 	(vrhaddq_m): Remove.
> 	(vrhaddq_x): Remove.
> 	(vrhaddq_u8): Remove.
> 	(vrhaddq_s8): Remove.
> 	(vrhaddq_u16): Remove.
> 	(vrhaddq_s16): Remove.
> 	(vrhaddq_u32): Remove.
> 	(vrhaddq_s32): Remove.
> 	(vrhaddq_m_s8): Remove.
> 	(vrhaddq_m_s32): Remove.
> 	(vrhaddq_m_s16): Remove.
> 	(vrhaddq_m_u8): Remove.
> 	(vrhaddq_m_u32): Remove.
> 	(vrhaddq_m_u16): Remove.
> 	(vrhaddq_x_s8): Remove.
> 	(vrhaddq_x_s16): Remove.
> 	(vrhaddq_x_s32): Remove.
> 	(vrhaddq_x_u8): Remove.
> 	(vrhaddq_x_u16): Remove.
> 	(vrhaddq_x_u32): Remove.
> 	(__arm_vrhaddq_u8): Remove.
> 	(__arm_vrhaddq_s8): Remove.
> 	(__arm_vrhaddq_u16): Remove.
> 	(__arm_vrhaddq_s16): Remove.
> 	(__arm_vrhaddq_u32): Remove.
> 	(__arm_vrhaddq_s32): Remove.
> 	(__arm_vrhaddq_m_s8): Remove.
> 	(__arm_vrhaddq_m_s32): Remove.
> 	(__arm_vrhaddq_m_s16): Remove.
> 	(__arm_vrhaddq_m_u8): Remove.
> 	(__arm_vrhaddq_m_u32): Remove.
> 	(__arm_vrhaddq_m_u16): Remove.
> 	(__arm_vrhaddq_x_s8): Remove.
> 	(__arm_vrhaddq_x_s16): Remove.
> 	(__arm_vrhaddq_x_s32): Remove.
> 	(__arm_vrhaddq_x_u8): Remove.
> 	(__arm_vrhaddq_x_u16): Remove.
> 	(__arm_vrhaddq_x_u32): Remove.
> 	(__arm_vrhaddq): Remove.
> 	(__arm_vrhaddq_m): Remove.
> 	(__arm_vrhaddq_x): Remove.
> 	(vrmulhq): Remove.
> 	(vrmulhq_m): Remove.
> 	(vrmulhq_x): Remove.
> 	(vrmulhq_u8): Remove.
> 	(vrmulhq_s8): Remove.
> 	(vrmulhq_u16): Remove.
> 	(vrmulhq_s16): Remove.
> 	(vrmulhq_u32): Remove.
> 	(vrmulhq_s32): Remove.
> 	(vrmulhq_m_s8): Remove.
> 	(vrmulhq_m_s32): Remove.
> 	(vrmulhq_m_s16): Remove.
> 	(vrmulhq_m_u8): Remove.
> 	(vrmulhq_m_u32): Remove.
> 	(vrmulhq_m_u16): Remove.
> 	(vrmulhq_x_s8): Remove.
> 	(vrmulhq_x_s16): Remove.
> 	(vrmulhq_x_s32): Remove.
> 	(vrmulhq_x_u8): Remove.
> 	(vrmulhq_x_u16): Remove.
> 	(vrmulhq_x_u32): Remove.
> 	(__arm_vrmulhq_u8): Remove.
> 	(__arm_vrmulhq_s8): Remove.
> 	(__arm_vrmulhq_u16): Remove.
> 	(__arm_vrmulhq_s16): Remove.
> 	(__arm_vrmulhq_u32): Remove.
> 	(__arm_vrmulhq_s32): Remove.
> 	(__arm_vrmulhq_m_s8): Remove.
> 	(__arm_vrmulhq_m_s32): Remove.
> 	(__arm_vrmulhq_m_s16): Remove.
> 	(__arm_vrmulhq_m_u8): Remove.
> 	(__arm_vrmulhq_m_u32): Remove.
> 	(__arm_vrmulhq_m_u16): Remove.
> 	(__arm_vrmulhq_x_s8): Remove.
> 	(__arm_vrmulhq_x_s16): Remove.
> 	(__arm_vrmulhq_x_s32): Remove.
> 	(__arm_vrmulhq_x_u8): Remove.
> 	(__arm_vrmulhq_x_u16): Remove.
> 	(__arm_vrmulhq_x_u32): Remove.
> 	(__arm_vrmulhq): Remove.
> 	(__arm_vrmulhq_m): Remove.
> 	(__arm_vrmulhq_x): Remove.
> ---
>  gcc/config/arm/arm-mve-builtins-base.cc  |   35 +
>  gcc/config/arm/arm-mve-builtins-base.def |    8 +
>  gcc/config/arm/arm-mve-builtins-base.h   |    8 +
>  gcc/config/arm/arm_mve.h                 | 3203 ----------------------
>  4 files changed, 51 insertions(+), 3203 deletions(-)
> 
> diff --git a/gcc/config/arm/arm-mve-builtins-base.cc b/gcc/config/arm/arm-
> mve-builtins-base.cc
> index 9722c861faf..668f1fe9cda 100644
> --- a/gcc/config/arm/arm-mve-builtins-base.cc
> +++ b/gcc/config/arm/arm-mve-builtins-base.cc
> @@ -115,13 +115,48 @@ namespace arm_mve {
>      -1, -1, -1,								\
>      -1, -1, -1))
> 
> +  /* Helper for builtins with only unspec codes, _m predicated and _n
> +     overrides, but no floating-point version.  */
> +#define FUNCTION_WITH_M_N_NO_F(NAME, UNSPEC) FUNCTION
> 		\
> +  (NAME, unspec_mve_function_exact_insn,				\
> +   (UNSPEC##_S, UNSPEC##_U, -1,
> 	\
> +    UNSPEC##_N_S, UNSPEC##_N_U, -1,					\
> +    UNSPEC##_M_S, UNSPEC##_M_U, -1,
> 	\
> +    UNSPEC##_M_N_S, UNSPEC##_M_N_U, -1))
> +
> +  /* Helper for builtins with only unspec codes, _m predicated
> +     overrides, no _n and no floating-point version.  */
> +#define FUNCTION_WITHOUT_N_NO_F(NAME, UNSPEC) FUNCTION
> 		\
> +  (NAME, unspec_mve_function_exact_insn,				\
> +   (UNSPEC##_S, UNSPEC##_U, -1,
> 	\
> +    -1, -1, -1,								\
> +    UNSPEC##_M_S, UNSPEC##_M_U, -1,
> 	\
> +    -1, -1, -1))
> +
> +  /* Helper for builtins with only unspec codes, _m predicated and _n
> +     overrides, but no unsigned and floating-point versions.  */
> +#define FUNCTION_WITH_M_N_NO_U_F(NAME, UNSPEC) FUNCTION
> 		\
> +  (NAME, unspec_mve_function_exact_insn,				\
> +   (UNSPEC##_S, -1, -1,							\
> +    UNSPEC##_N_S, -1, -1,						\
> +    UNSPEC##_M_S, -1, -1,						\
> +    UNSPEC##_M_N_S, -1, -1))
> +
>  FUNCTION_WITH_RTX_M_N (vaddq, PLUS, VADDQ)
>  FUNCTION_WITH_RTX_M (vandq, AND, VANDQ)
>  FUNCTION_WITHOUT_M_N (vcreateq, VCREATEQ)
>  FUNCTION_WITH_RTX_M (veorq, XOR, VEORQ)
> +FUNCTION_WITH_M_N_NO_F (vhaddq, VHADDQ)
> +FUNCTION_WITH_M_N_NO_F (vhsubq, VHSUBQ)
> +FUNCTION_WITHOUT_N_NO_F (vmulhq, VMULHQ)
>  FUNCTION_WITH_RTX_M_N (vmulq, MULT, VMULQ)
>  FUNCTION_WITH_RTX_M_N_NO_N_F (vorrq, IOR, VORRQ)
> +FUNCTION_WITH_M_N_NO_F (vqaddq, VQADDQ)
> +FUNCTION_WITH_M_N_NO_U_F (vqdmulhq, VQDMULHQ)
> +FUNCTION_WITH_M_N_NO_F (vqsubq, VQSUBQ)
>  FUNCTION (vreinterpretq, vreinterpretq_impl,)
> +FUNCTION_WITHOUT_N_NO_F (vrhaddq, VRHADDQ)
> +FUNCTION_WITHOUT_N_NO_F (vrmulhq, VRMULHQ)
>  FUNCTION_WITH_RTX_M_N (vsubq, MINUS, VSUBQ)
>  FUNCTION (vuninitializedq, vuninitializedq_impl,)
> 
> diff --git a/gcc/config/arm/arm-mve-builtins-base.def b/gcc/config/arm/arm-
> mve-builtins-base.def
> index 1bfd15f973c..d256f3ebb2d 100644
> --- a/gcc/config/arm/arm-mve-builtins-base.def
> +++ b/gcc/config/arm/arm-mve-builtins-base.def
> @@ -22,9 +22,17 @@ DEF_MVE_FUNCTION (vaddq, binary_opt_n,
> all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vandq, binary, all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vcreateq, create, all_integer_with_64, none)
>  DEF_MVE_FUNCTION (veorq, binary, all_integer, mx_or_none)
> +DEF_MVE_FUNCTION (vhaddq, binary_opt_n, all_integer, mx_or_none)
> +DEF_MVE_FUNCTION (vhsubq, binary_opt_n, all_integer, mx_or_none)
> +DEF_MVE_FUNCTION (vmulhq, binary, all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vmulq, binary_opt_n, all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vorrq, binary_orrq, all_integer, mx_or_none)
> +DEF_MVE_FUNCTION (vqaddq, binary_opt_n, all_integer, m_or_none)
> +DEF_MVE_FUNCTION (vqdmulhq, binary_opt_n, all_signed, m_or_none)
> +DEF_MVE_FUNCTION (vqsubq, binary_opt_n, all_integer, m_or_none)
>  DEF_MVE_FUNCTION (vreinterpretq, unary_convert, reinterpret_integer,
> none)
> +DEF_MVE_FUNCTION (vrhaddq, binary, all_integer, mx_or_none)
> +DEF_MVE_FUNCTION (vrmulhq, binary, all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vsubq, binary_opt_n, all_integer, mx_or_none)
>  DEF_MVE_FUNCTION (vuninitializedq, inherent, all_integer_with_64, none)
>  #undef REQUIRES_FLOAT
> diff --git a/gcc/config/arm/arm-mve-builtins-base.h b/gcc/config/arm/arm-
> mve-builtins-base.h
> index 8dd6bff01bf..d64cb5e1dec 100644
> --- a/gcc/config/arm/arm-mve-builtins-base.h
> +++ b/gcc/config/arm/arm-mve-builtins-base.h
> @@ -27,9 +27,17 @@ extern const function_base *const vaddq;
>  extern const function_base *const vandq;
>  extern const function_base *const vcreateq;
>  extern const function_base *const veorq;
> +extern const function_base *const vhaddq;
> +extern const function_base *const vhsubq;
> +extern const function_base *const vmulhq;
>  extern const function_base *const vmulq;
>  extern const function_base *const vorrq;
> +extern const function_base *const vqaddq;
> +extern const function_base *const vqdmulhq;
> +extern const function_base *const vqsubq;
>  extern const function_base *const vreinterpretq;
> +extern const function_base *const vrhaddq;
> +extern const function_base *const vrmulhq;
>  extern const function_base *const vsubq;
>  extern const function_base *const vuninitializedq;
> 
> diff --git a/gcc/config/arm/arm_mve.h b/gcc/config/arm/arm_mve.h
> index 4810e2977d3..9c5d14794a1 100644
> --- a/gcc/config/arm/arm_mve.h
> +++ b/gcc/config/arm/arm_mve.h
> @@ -61,21 +61,14 @@
>  #define vaddlvq_p(__a, __p) __arm_vaddlvq_p(__a, __p)
>  #define vcmpneq(__a, __b) __arm_vcmpneq(__a, __b)
>  #define vshlq(__a, __b) __arm_vshlq(__a, __b)
> -#define vrmulhq(__a, __b) __arm_vrmulhq(__a, __b)
> -#define vrhaddq(__a, __b) __arm_vrhaddq(__a, __b)
> -#define vqsubq(__a, __b) __arm_vqsubq(__a, __b)
> -#define vqaddq(__a, __b) __arm_vqaddq(__a, __b)
>  #define vornq(__a, __b) __arm_vornq(__a, __b)
>  #define vmulltq_int(__a, __b) __arm_vmulltq_int(__a, __b)
>  #define vmullbq_int(__a, __b) __arm_vmullbq_int(__a, __b)
> -#define vmulhq(__a, __b) __arm_vmulhq(__a, __b)
>  #define vmladavq(__a, __b) __arm_vmladavq(__a, __b)
>  #define vminvq(__a, __b) __arm_vminvq(__a, __b)
>  #define vminq(__a, __b) __arm_vminq(__a, __b)
>  #define vmaxvq(__a, __b) __arm_vmaxvq(__a, __b)
>  #define vmaxq(__a, __b) __arm_vmaxq(__a, __b)
> -#define vhsubq(__a, __b) __arm_vhsubq(__a, __b)
> -#define vhaddq(__a, __b) __arm_vhaddq(__a, __b)
>  #define vcmphiq(__a, __b) __arm_vcmphiq(__a, __b)
>  #define vcmpeqq(__a, __b) __arm_vcmpeqq(__a, __b)
>  #define vcmpcsq(__a, __b) __arm_vcmpcsq(__a, __b)
> @@ -104,7 +97,6 @@
>  #define vcmpgeq(__a, __b) __arm_vcmpgeq(__a, __b)
>  #define vqshluq(__a, __imm) __arm_vqshluq(__a, __imm)
>  #define vqrdmulhq(__a, __b) __arm_vqrdmulhq(__a, __b)
> -#define vqdmulhq(__a, __b) __arm_vqdmulhq(__a, __b)
>  #define vmlsdavxq(__a, __b) __arm_vmlsdavxq(__a, __b)
>  #define vmlsdavq(__a, __b) __arm_vmlsdavq(__a, __b)
>  #define vmladavxq(__a, __b) __arm_vmladavxq(__a, __b)
> @@ -236,10 +228,8 @@
>  #define vbrsrq_m(__inactive, __a, __b, __p) __arm_vbrsrq_m(__inactive,
> __a, __b, __p)
>  #define vcaddq_rot270_m(__inactive, __a, __b, __p)
> __arm_vcaddq_rot270_m(__inactive, __a, __b, __p)
>  #define vcaddq_rot90_m(__inactive, __a, __b, __p)
> __arm_vcaddq_rot90_m(__inactive, __a, __b, __p)
> -#define vhaddq_m(__inactive, __a, __b, __p) __arm_vhaddq_m(__inactive,
> __a, __b, __p)
>  #define vhcaddq_rot270_m(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot270_m(__inactive, __a, __b, __p)
>  #define vhcaddq_rot90_m(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot90_m(__inactive, __a, __b, __p)
> -#define vhsubq_m(__inactive, __a, __b, __p) __arm_vhsubq_m(__inactive,
> __a, __b, __p)
>  #define vmaxq_m(__inactive, __a, __b, __p) __arm_vmaxq_m(__inactive,
> __a, __b, __p)
>  #define vminq_m(__inactive, __a, __b, __p) __arm_vminq_m(__inactive,
> __a, __b, __p)
>  #define vmladavaq_p(__a, __b, __c, __p) __arm_vmladavaq_p(__a, __b, __c,
> __p)
> @@ -248,18 +238,15 @@
>  #define vmlasq_m(__a, __b, __c, __p) __arm_vmlasq_m(__a, __b, __c, __p)
>  #define vmlsdavaq_p(__a, __b, __c, __p) __arm_vmlsdavaq_p(__a, __b, __c,
> __p)
>  #define vmlsdavaxq_p(__a, __b, __c, __p) __arm_vmlsdavaxq_p(__a, __b,
> __c, __p)
> -#define vmulhq_m(__inactive, __a, __b, __p) __arm_vmulhq_m(__inactive,
> __a, __b, __p)
>  #define vmullbq_int_m(__inactive, __a, __b, __p)
> __arm_vmullbq_int_m(__inactive, __a, __b, __p)
>  #define vmulltq_int_m(__inactive, __a, __b, __p)
> __arm_vmulltq_int_m(__inactive, __a, __b, __p)
>  #define vornq_m(__inactive, __a, __b, __p) __arm_vornq_m(__inactive, __a,
> __b, __p)
> -#define vqaddq_m(__inactive, __a, __b, __p) __arm_vqaddq_m(__inactive,
> __a, __b, __p)
>  #define vqdmladhq_m(__inactive, __a, __b, __p)
> __arm_vqdmladhq_m(__inactive, __a, __b, __p)
>  #define vqdmlashq_m(__a, __b, __c, __p) __arm_vqdmlashq_m(__a, __b,
> __c, __p)
>  #define vqdmladhxq_m(__inactive, __a, __b, __p)
> __arm_vqdmladhxq_m(__inactive, __a, __b, __p)
>  #define vqdmlahq_m(__a, __b, __c, __p) __arm_vqdmlahq_m(__a, __b, __c,
> __p)
>  #define vqdmlsdhq_m(__inactive, __a, __b, __p)
> __arm_vqdmlsdhq_m(__inactive, __a, __b, __p)
>  #define vqdmlsdhxq_m(__inactive, __a, __b, __p)
> __arm_vqdmlsdhxq_m(__inactive, __a, __b, __p)
> -#define vqdmulhq_m(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m(__inactive, __a, __b, __p)
>  #define vqrdmladhq_m(__inactive, __a, __b, __p)
> __arm_vqrdmladhq_m(__inactive, __a, __b, __p)
>  #define vqrdmladhxq_m(__inactive, __a, __b, __p)
> __arm_vqrdmladhxq_m(__inactive, __a, __b, __p)
>  #define vqrdmlahq_m(__a, __b, __c, __p) __arm_vqrdmlahq_m(__a, __b,
> __c, __p)
> @@ -270,9 +257,6 @@
>  #define vqrshlq_m(__inactive, __a, __b, __p) __arm_vqrshlq_m(__inactive,
> __a, __b, __p)
>  #define vqshlq_m_n(__inactive, __a, __imm, __p)
> __arm_vqshlq_m_n(__inactive, __a, __imm, __p)
>  #define vqshlq_m(__inactive, __a, __b, __p) __arm_vqshlq_m(__inactive,
> __a, __b, __p)
> -#define vqsubq_m(__inactive, __a, __b, __p) __arm_vqsubq_m(__inactive,
> __a, __b, __p)
> -#define vrhaddq_m(__inactive, __a, __b, __p)
> __arm_vrhaddq_m(__inactive, __a, __b, __p)
> -#define vrmulhq_m(__inactive, __a, __b, __p)
> __arm_vrmulhq_m(__inactive, __a, __b, __p)
>  #define vrshlq_m(__inactive, __a, __b, __p) __arm_vrshlq_m(__inactive,
> __a, __b, __p)
>  #define vrshrq_m(__inactive, __a, __imm, __p) __arm_vrshrq_m(__inactive,
> __a, __imm, __p)
>  #define vshlq_m_n(__inactive, __a, __imm, __p)
> __arm_vshlq_m_n(__inactive, __a, __imm, __p)
> @@ -384,19 +368,14 @@
>  #define vclsq_x(__a, __p) __arm_vclsq_x(__a, __p)
>  #define vclzq_x(__a, __p) __arm_vclzq_x(__a, __p)
>  #define vnegq_x(__a, __p) __arm_vnegq_x(__a, __p)
> -#define vmulhq_x(__a, __b, __p) __arm_vmulhq_x(__a, __b, __p)
>  #define vmullbq_poly_x(__a, __b, __p) __arm_vmullbq_poly_x(__a, __b,
> __p)
>  #define vmullbq_int_x(__a, __b, __p) __arm_vmullbq_int_x(__a, __b, __p)
>  #define vmulltq_poly_x(__a, __b, __p) __arm_vmulltq_poly_x(__a, __b, __p)
>  #define vmulltq_int_x(__a, __b, __p) __arm_vmulltq_int_x(__a, __b, __p)
>  #define vcaddq_rot90_x(__a, __b, __p) __arm_vcaddq_rot90_x(__a, __b,
> __p)
>  #define vcaddq_rot270_x(__a, __b, __p) __arm_vcaddq_rot270_x(__a, __b,
> __p)
> -#define vhaddq_x(__a, __b, __p) __arm_vhaddq_x(__a, __b, __p)
>  #define vhcaddq_rot90_x(__a, __b, __p) __arm_vhcaddq_rot90_x(__a, __b,
> __p)
>  #define vhcaddq_rot270_x(__a, __b, __p) __arm_vhcaddq_rot270_x(__a,
> __b, __p)
> -#define vhsubq_x(__a, __b, __p) __arm_vhsubq_x(__a, __b, __p)
> -#define vrhaddq_x(__a, __b, __p) __arm_vrhaddq_x(__a, __b, __p)
> -#define vrmulhq_x(__a, __b, __p) __arm_vrmulhq_x(__a, __b, __p)
>  #define vbicq_x(__a, __b, __p) __arm_vbicq_x(__a, __b, __p)
>  #define vbrsrq_x(__a, __b, __p) __arm_vbrsrq_x(__a, __b, __p)
>  #define vmovlbq_x(__a, __p) __arm_vmovlbq_x(__a, __p)
> @@ -662,25 +641,14 @@
>  #define vshlq_u8(__a, __b) __arm_vshlq_u8(__a, __b)
>  #define vshlq_u16(__a, __b) __arm_vshlq_u16(__a, __b)
>  #define vshlq_u32(__a, __b) __arm_vshlq_u32(__a, __b)
> -#define vrmulhq_u8(__a, __b) __arm_vrmulhq_u8(__a, __b)
> -#define vrhaddq_u8(__a, __b) __arm_vrhaddq_u8(__a, __b)
> -#define vqsubq_u8(__a, __b) __arm_vqsubq_u8(__a, __b)
> -#define vqsubq_n_u8(__a, __b) __arm_vqsubq_n_u8(__a, __b)
> -#define vqaddq_u8(__a, __b) __arm_vqaddq_u8(__a, __b)
> -#define vqaddq_n_u8(__a, __b) __arm_vqaddq_n_u8(__a, __b)
>  #define vornq_u8(__a, __b) __arm_vornq_u8(__a, __b)
>  #define vmulltq_int_u8(__a, __b) __arm_vmulltq_int_u8(__a, __b)
>  #define vmullbq_int_u8(__a, __b) __arm_vmullbq_int_u8(__a, __b)
> -#define vmulhq_u8(__a, __b) __arm_vmulhq_u8(__a, __b)
>  #define vmladavq_u8(__a, __b) __arm_vmladavq_u8(__a, __b)
>  #define vminvq_u8(__a, __b) __arm_vminvq_u8(__a, __b)
>  #define vminq_u8(__a, __b) __arm_vminq_u8(__a, __b)
>  #define vmaxvq_u8(__a, __b) __arm_vmaxvq_u8(__a, __b)
>  #define vmaxq_u8(__a, __b) __arm_vmaxq_u8(__a, __b)
> -#define vhsubq_u8(__a, __b) __arm_vhsubq_u8(__a, __b)
> -#define vhsubq_n_u8(__a, __b) __arm_vhsubq_n_u8(__a, __b)
> -#define vhaddq_u8(__a, __b) __arm_vhaddq_u8(__a, __b)
> -#define vhaddq_n_u8(__a, __b) __arm_vhaddq_n_u8(__a, __b)
>  #define vcmpneq_n_u8(__a, __b) __arm_vcmpneq_n_u8(__a, __b)
>  #define vcmphiq_u8(__a, __b) __arm_vcmphiq_u8(__a, __b)
>  #define vcmphiq_n_u8(__a, __b) __arm_vcmphiq_n_u8(__a, __b)
> @@ -725,24 +693,15 @@
>  #define vshlq_r_s8(__a, __b) __arm_vshlq_r_s8(__a, __b)
>  #define vrshlq_s8(__a, __b) __arm_vrshlq_s8(__a, __b)
>  #define vrshlq_n_s8(__a, __b) __arm_vrshlq_n_s8(__a, __b)
> -#define vrmulhq_s8(__a, __b) __arm_vrmulhq_s8(__a, __b)
> -#define vrhaddq_s8(__a, __b) __arm_vrhaddq_s8(__a, __b)
> -#define vqsubq_s8(__a, __b) __arm_vqsubq_s8(__a, __b)
> -#define vqsubq_n_s8(__a, __b) __arm_vqsubq_n_s8(__a, __b)
>  #define vqshlq_s8(__a, __b) __arm_vqshlq_s8(__a, __b)
>  #define vqshlq_r_s8(__a, __b) __arm_vqshlq_r_s8(__a, __b)
>  #define vqrshlq_s8(__a, __b) __arm_vqrshlq_s8(__a, __b)
>  #define vqrshlq_n_s8(__a, __b) __arm_vqrshlq_n_s8(__a, __b)
>  #define vqrdmulhq_s8(__a, __b) __arm_vqrdmulhq_s8(__a, __b)
>  #define vqrdmulhq_n_s8(__a, __b) __arm_vqrdmulhq_n_s8(__a, __b)
> -#define vqdmulhq_s8(__a, __b) __arm_vqdmulhq_s8(__a, __b)
> -#define vqdmulhq_n_s8(__a, __b) __arm_vqdmulhq_n_s8(__a, __b)
> -#define vqaddq_s8(__a, __b) __arm_vqaddq_s8(__a, __b)
> -#define vqaddq_n_s8(__a, __b) __arm_vqaddq_n_s8(__a, __b)
>  #define vornq_s8(__a, __b) __arm_vornq_s8(__a, __b)
>  #define vmulltq_int_s8(__a, __b) __arm_vmulltq_int_s8(__a, __b)
>  #define vmullbq_int_s8(__a, __b) __arm_vmullbq_int_s8(__a, __b)
> -#define vmulhq_s8(__a, __b) __arm_vmulhq_s8(__a, __b)
>  #define vmlsdavxq_s8(__a, __b) __arm_vmlsdavxq_s8(__a, __b)
>  #define vmlsdavq_s8(__a, __b) __arm_vmlsdavq_s8(__a, __b)
>  #define vmladavxq_s8(__a, __b) __arm_vmladavxq_s8(__a, __b)
> @@ -751,12 +710,8 @@
>  #define vminq_s8(__a, __b) __arm_vminq_s8(__a, __b)
>  #define vmaxvq_s8(__a, __b) __arm_vmaxvq_s8(__a, __b)
>  #define vmaxq_s8(__a, __b) __arm_vmaxq_s8(__a, __b)
> -#define vhsubq_s8(__a, __b) __arm_vhsubq_s8(__a, __b)
> -#define vhsubq_n_s8(__a, __b) __arm_vhsubq_n_s8(__a, __b)
>  #define vhcaddq_rot90_s8(__a, __b) __arm_vhcaddq_rot90_s8(__a, __b)
>  #define vhcaddq_rot270_s8(__a, __b) __arm_vhcaddq_rot270_s8(__a, __b)
> -#define vhaddq_s8(__a, __b) __arm_vhaddq_s8(__a, __b)
> -#define vhaddq_n_s8(__a, __b) __arm_vhaddq_n_s8(__a, __b)
>  #define vcaddq_rot90_s8(__a, __b) __arm_vcaddq_rot90_s8(__a, __b)
>  #define vcaddq_rot270_s8(__a, __b) __arm_vcaddq_rot270_s8(__a, __b)
>  #define vbrsrq_n_s8(__a, __b) __arm_vbrsrq_n_s8(__a, __b)
> @@ -766,25 +721,14 @@
>  #define vshlq_n_s8(__a,  __imm) __arm_vshlq_n_s8(__a,  __imm)
>  #define vrshrq_n_s8(__a,  __imm) __arm_vrshrq_n_s8(__a,  __imm)
>  #define vqshlq_n_s8(__a,  __imm) __arm_vqshlq_n_s8(__a,  __imm)
> -#define vrmulhq_u16(__a, __b) __arm_vrmulhq_u16(__a, __b)
> -#define vrhaddq_u16(__a, __b) __arm_vrhaddq_u16(__a, __b)
> -#define vqsubq_u16(__a, __b) __arm_vqsubq_u16(__a, __b)
> -#define vqsubq_n_u16(__a, __b) __arm_vqsubq_n_u16(__a, __b)
> -#define vqaddq_u16(__a, __b) __arm_vqaddq_u16(__a, __b)
> -#define vqaddq_n_u16(__a, __b) __arm_vqaddq_n_u16(__a, __b)
>  #define vornq_u16(__a, __b) __arm_vornq_u16(__a, __b)
>  #define vmulltq_int_u16(__a, __b) __arm_vmulltq_int_u16(__a, __b)
>  #define vmullbq_int_u16(__a, __b) __arm_vmullbq_int_u16(__a, __b)
> -#define vmulhq_u16(__a, __b) __arm_vmulhq_u16(__a, __b)
>  #define vmladavq_u16(__a, __b) __arm_vmladavq_u16(__a, __b)
>  #define vminvq_u16(__a, __b) __arm_vminvq_u16(__a, __b)
>  #define vminq_u16(__a, __b) __arm_vminq_u16(__a, __b)
>  #define vmaxvq_u16(__a, __b) __arm_vmaxvq_u16(__a, __b)
>  #define vmaxq_u16(__a, __b) __arm_vmaxq_u16(__a, __b)
> -#define vhsubq_u16(__a, __b) __arm_vhsubq_u16(__a, __b)
> -#define vhsubq_n_u16(__a, __b) __arm_vhsubq_n_u16(__a, __b)
> -#define vhaddq_u16(__a, __b) __arm_vhaddq_u16(__a, __b)
> -#define vhaddq_n_u16(__a, __b) __arm_vhaddq_n_u16(__a, __b)
>  #define vcmpneq_n_u16(__a, __b) __arm_vcmpneq_n_u16(__a, __b)
>  #define vcmphiq_u16(__a, __b) __arm_vcmphiq_u16(__a, __b)
>  #define vcmphiq_n_u16(__a, __b) __arm_vcmphiq_n_u16(__a, __b)
> @@ -829,24 +773,15 @@
>  #define vshlq_r_s16(__a, __b) __arm_vshlq_r_s16(__a, __b)
>  #define vrshlq_s16(__a, __b) __arm_vrshlq_s16(__a, __b)
>  #define vrshlq_n_s16(__a, __b) __arm_vrshlq_n_s16(__a, __b)
> -#define vrmulhq_s16(__a, __b) __arm_vrmulhq_s16(__a, __b)
> -#define vrhaddq_s16(__a, __b) __arm_vrhaddq_s16(__a, __b)
> -#define vqsubq_s16(__a, __b) __arm_vqsubq_s16(__a, __b)
> -#define vqsubq_n_s16(__a, __b) __arm_vqsubq_n_s16(__a, __b)
>  #define vqshlq_s16(__a, __b) __arm_vqshlq_s16(__a, __b)
>  #define vqshlq_r_s16(__a, __b) __arm_vqshlq_r_s16(__a, __b)
>  #define vqrshlq_s16(__a, __b) __arm_vqrshlq_s16(__a, __b)
>  #define vqrshlq_n_s16(__a, __b) __arm_vqrshlq_n_s16(__a, __b)
>  #define vqrdmulhq_s16(__a, __b) __arm_vqrdmulhq_s16(__a, __b)
>  #define vqrdmulhq_n_s16(__a, __b) __arm_vqrdmulhq_n_s16(__a, __b)
> -#define vqdmulhq_s16(__a, __b) __arm_vqdmulhq_s16(__a, __b)
> -#define vqdmulhq_n_s16(__a, __b) __arm_vqdmulhq_n_s16(__a, __b)
> -#define vqaddq_s16(__a, __b) __arm_vqaddq_s16(__a, __b)
> -#define vqaddq_n_s16(__a, __b) __arm_vqaddq_n_s16(__a, __b)
>  #define vornq_s16(__a, __b) __arm_vornq_s16(__a, __b)
>  #define vmulltq_int_s16(__a, __b) __arm_vmulltq_int_s16(__a, __b)
>  #define vmullbq_int_s16(__a, __b) __arm_vmullbq_int_s16(__a, __b)
> -#define vmulhq_s16(__a, __b) __arm_vmulhq_s16(__a, __b)
>  #define vmlsdavxq_s16(__a, __b) __arm_vmlsdavxq_s16(__a, __b)
>  #define vmlsdavq_s16(__a, __b) __arm_vmlsdavq_s16(__a, __b)
>  #define vmladavxq_s16(__a, __b) __arm_vmladavxq_s16(__a, __b)
> @@ -855,12 +790,8 @@
>  #define vminq_s16(__a, __b) __arm_vminq_s16(__a, __b)
>  #define vmaxvq_s16(__a, __b) __arm_vmaxvq_s16(__a, __b)
>  #define vmaxq_s16(__a, __b) __arm_vmaxq_s16(__a, __b)
> -#define vhsubq_s16(__a, __b) __arm_vhsubq_s16(__a, __b)
> -#define vhsubq_n_s16(__a, __b) __arm_vhsubq_n_s16(__a, __b)
>  #define vhcaddq_rot90_s16(__a, __b) __arm_vhcaddq_rot90_s16(__a, __b)
>  #define vhcaddq_rot270_s16(__a, __b) __arm_vhcaddq_rot270_s16(__a,
> __b)
> -#define vhaddq_s16(__a, __b) __arm_vhaddq_s16(__a, __b)
> -#define vhaddq_n_s16(__a, __b) __arm_vhaddq_n_s16(__a, __b)
>  #define vcaddq_rot90_s16(__a, __b) __arm_vcaddq_rot90_s16(__a, __b)
>  #define vcaddq_rot270_s16(__a, __b) __arm_vcaddq_rot270_s16(__a, __b)
>  #define vbrsrq_n_s16(__a, __b) __arm_vbrsrq_n_s16(__a, __b)
> @@ -870,25 +801,14 @@
>  #define vshlq_n_s16(__a,  __imm) __arm_vshlq_n_s16(__a,  __imm)
>  #define vrshrq_n_s16(__a,  __imm) __arm_vrshrq_n_s16(__a,  __imm)
>  #define vqshlq_n_s16(__a,  __imm) __arm_vqshlq_n_s16(__a,  __imm)
> -#define vrmulhq_u32(__a, __b) __arm_vrmulhq_u32(__a, __b)
> -#define vrhaddq_u32(__a, __b) __arm_vrhaddq_u32(__a, __b)
> -#define vqsubq_u32(__a, __b) __arm_vqsubq_u32(__a, __b)
> -#define vqsubq_n_u32(__a, __b) __arm_vqsubq_n_u32(__a, __b)
> -#define vqaddq_u32(__a, __b) __arm_vqaddq_u32(__a, __b)
> -#define vqaddq_n_u32(__a, __b) __arm_vqaddq_n_u32(__a, __b)
>  #define vornq_u32(__a, __b) __arm_vornq_u32(__a, __b)
>  #define vmulltq_int_u32(__a, __b) __arm_vmulltq_int_u32(__a, __b)
>  #define vmullbq_int_u32(__a, __b) __arm_vmullbq_int_u32(__a, __b)
> -#define vmulhq_u32(__a, __b) __arm_vmulhq_u32(__a, __b)
>  #define vmladavq_u32(__a, __b) __arm_vmladavq_u32(__a, __b)
>  #define vminvq_u32(__a, __b) __arm_vminvq_u32(__a, __b)
>  #define vminq_u32(__a, __b) __arm_vminq_u32(__a, __b)
>  #define vmaxvq_u32(__a, __b) __arm_vmaxvq_u32(__a, __b)
>  #define vmaxq_u32(__a, __b) __arm_vmaxq_u32(__a, __b)
> -#define vhsubq_u32(__a, __b) __arm_vhsubq_u32(__a, __b)
> -#define vhsubq_n_u32(__a, __b) __arm_vhsubq_n_u32(__a, __b)
> -#define vhaddq_u32(__a, __b) __arm_vhaddq_u32(__a, __b)
> -#define vhaddq_n_u32(__a, __b) __arm_vhaddq_n_u32(__a, __b)
>  #define vcmpneq_n_u32(__a, __b) __arm_vcmpneq_n_u32(__a, __b)
>  #define vcmphiq_u32(__a, __b) __arm_vcmphiq_u32(__a, __b)
>  #define vcmphiq_n_u32(__a, __b) __arm_vcmphiq_n_u32(__a, __b)
> @@ -933,24 +853,15 @@
>  #define vshlq_r_s32(__a, __b) __arm_vshlq_r_s32(__a, __b)
>  #define vrshlq_s32(__a, __b) __arm_vrshlq_s32(__a, __b)
>  #define vrshlq_n_s32(__a, __b) __arm_vrshlq_n_s32(__a, __b)
> -#define vrmulhq_s32(__a, __b) __arm_vrmulhq_s32(__a, __b)
> -#define vrhaddq_s32(__a, __b) __arm_vrhaddq_s32(__a, __b)
> -#define vqsubq_s32(__a, __b) __arm_vqsubq_s32(__a, __b)
> -#define vqsubq_n_s32(__a, __b) __arm_vqsubq_n_s32(__a, __b)
>  #define vqshlq_s32(__a, __b) __arm_vqshlq_s32(__a, __b)
>  #define vqshlq_r_s32(__a, __b) __arm_vqshlq_r_s32(__a, __b)
>  #define vqrshlq_s32(__a, __b) __arm_vqrshlq_s32(__a, __b)
>  #define vqrshlq_n_s32(__a, __b) __arm_vqrshlq_n_s32(__a, __b)
>  #define vqrdmulhq_s32(__a, __b) __arm_vqrdmulhq_s32(__a, __b)
>  #define vqrdmulhq_n_s32(__a, __b) __arm_vqrdmulhq_n_s32(__a, __b)
> -#define vqdmulhq_s32(__a, __b) __arm_vqdmulhq_s32(__a, __b)
> -#define vqdmulhq_n_s32(__a, __b) __arm_vqdmulhq_n_s32(__a, __b)
> -#define vqaddq_s32(__a, __b) __arm_vqaddq_s32(__a, __b)
> -#define vqaddq_n_s32(__a, __b) __arm_vqaddq_n_s32(__a, __b)
>  #define vornq_s32(__a, __b) __arm_vornq_s32(__a, __b)
>  #define vmulltq_int_s32(__a, __b) __arm_vmulltq_int_s32(__a, __b)
>  #define vmullbq_int_s32(__a, __b) __arm_vmullbq_int_s32(__a, __b)
> -#define vmulhq_s32(__a, __b) __arm_vmulhq_s32(__a, __b)
>  #define vmlsdavxq_s32(__a, __b) __arm_vmlsdavxq_s32(__a, __b)
>  #define vmlsdavq_s32(__a, __b) __arm_vmlsdavq_s32(__a, __b)
>  #define vmladavxq_s32(__a, __b) __arm_vmladavxq_s32(__a, __b)
> @@ -959,12 +870,8 @@
>  #define vminq_s32(__a, __b) __arm_vminq_s32(__a, __b)
>  #define vmaxvq_s32(__a, __b) __arm_vmaxvq_s32(__a, __b)
>  #define vmaxq_s32(__a, __b) __arm_vmaxq_s32(__a, __b)
> -#define vhsubq_s32(__a, __b) __arm_vhsubq_s32(__a, __b)
> -#define vhsubq_n_s32(__a, __b) __arm_vhsubq_n_s32(__a, __b)
>  #define vhcaddq_rot90_s32(__a, __b) __arm_vhcaddq_rot90_s32(__a, __b)
>  #define vhcaddq_rot270_s32(__a, __b) __arm_vhcaddq_rot270_s32(__a,
> __b)
> -#define vhaddq_s32(__a, __b) __arm_vhaddq_s32(__a, __b)
> -#define vhaddq_n_s32(__a, __b) __arm_vhaddq_n_s32(__a, __b)
>  #define vcaddq_rot90_s32(__a, __b) __arm_vcaddq_rot90_s32(__a, __b)
>  #define vcaddq_rot270_s32(__a, __b) __arm_vcaddq_rot270_s32(__a, __b)
>  #define vbrsrq_n_s32(__a, __b) __arm_vbrsrq_n_s32(__a, __b)
> @@ -1634,36 +1541,12 @@
>  #define vcaddq_rot90_m_u8(__inactive, __a, __b, __p)
> __arm_vcaddq_rot90_m_u8(__inactive, __a, __b, __p)
>  #define vcaddq_rot90_m_u32(__inactive, __a, __b, __p)
> __arm_vcaddq_rot90_m_u32(__inactive, __a, __b, __p)
>  #define vcaddq_rot90_m_u16(__inactive, __a, __b, __p)
> __arm_vcaddq_rot90_m_u16(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_s8(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_s8(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_s32(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_s32(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_s16(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_s16(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_u8(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_u8(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_u32(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_u32(__inactive, __a, __b, __p)
> -#define vhaddq_m_n_u16(__inactive, __a, __b, __p)
> __arm_vhaddq_m_n_u16(__inactive, __a, __b, __p)
> -#define vhaddq_m_s8(__inactive, __a, __b, __p)
> __arm_vhaddq_m_s8(__inactive, __a, __b, __p)
> -#define vhaddq_m_s32(__inactive, __a, __b, __p)
> __arm_vhaddq_m_s32(__inactive, __a, __b, __p)
> -#define vhaddq_m_s16(__inactive, __a, __b, __p)
> __arm_vhaddq_m_s16(__inactive, __a, __b, __p)
> -#define vhaddq_m_u8(__inactive, __a, __b, __p)
> __arm_vhaddq_m_u8(__inactive, __a, __b, __p)
> -#define vhaddq_m_u32(__inactive, __a, __b, __p)
> __arm_vhaddq_m_u32(__inactive, __a, __b, __p)
> -#define vhaddq_m_u16(__inactive, __a, __b, __p)
> __arm_vhaddq_m_u16(__inactive, __a, __b, __p)
>  #define vhcaddq_rot270_m_s8(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot270_m_s8(__inactive, __a, __b, __p)
>  #define vhcaddq_rot270_m_s32(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot270_m_s32(__inactive, __a, __b, __p)
>  #define vhcaddq_rot270_m_s16(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot270_m_s16(__inactive, __a, __b, __p)
>  #define vhcaddq_rot90_m_s8(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot90_m_s8(__inactive, __a, __b, __p)
>  #define vhcaddq_rot90_m_s32(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot90_m_s32(__inactive, __a, __b, __p)
>  #define vhcaddq_rot90_m_s16(__inactive, __a, __b, __p)
> __arm_vhcaddq_rot90_m_s16(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_s8(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_s8(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_s32(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_s32(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_s16(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_s16(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_u8(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_u8(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_u32(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_u32(__inactive, __a, __b, __p)
> -#define vhsubq_m_n_u16(__inactive, __a, __b, __p)
> __arm_vhsubq_m_n_u16(__inactive, __a, __b, __p)
> -#define vhsubq_m_s8(__inactive, __a, __b, __p)
> __arm_vhsubq_m_s8(__inactive, __a, __b, __p)
> -#define vhsubq_m_s32(__inactive, __a, __b, __p)
> __arm_vhsubq_m_s32(__inactive, __a, __b, __p)
> -#define vhsubq_m_s16(__inactive, __a, __b, __p)
> __arm_vhsubq_m_s16(__inactive, __a, __b, __p)
> -#define vhsubq_m_u8(__inactive, __a, __b, __p)
> __arm_vhsubq_m_u8(__inactive, __a, __b, __p)
> -#define vhsubq_m_u32(__inactive, __a, __b, __p)
> __arm_vhsubq_m_u32(__inactive, __a, __b, __p)
> -#define vhsubq_m_u16(__inactive, __a, __b, __p)
> __arm_vhsubq_m_u16(__inactive, __a, __b, __p)
>  #define vmaxq_m_s8(__inactive, __a, __b, __p)
> __arm_vmaxq_m_s8(__inactive, __a, __b, __p)
>  #define vmaxq_m_s32(__inactive, __a, __b, __p)
> __arm_vmaxq_m_s32(__inactive, __a, __b, __p)
>  #define vmaxq_m_s16(__inactive, __a, __b, __p)
> __arm_vmaxq_m_s16(__inactive, __a, __b, __p)
> @@ -1703,12 +1586,6 @@
>  #define vmlsdavaxq_p_s8(__a, __b, __c, __p) __arm_vmlsdavaxq_p_s8(__a,
> __b, __c, __p)
>  #define vmlsdavaxq_p_s32(__a, __b, __c, __p)
> __arm_vmlsdavaxq_p_s32(__a, __b, __c, __p)
>  #define vmlsdavaxq_p_s16(__a, __b, __c, __p)
> __arm_vmlsdavaxq_p_s16(__a, __b, __c, __p)
> -#define vmulhq_m_s8(__inactive, __a, __b, __p)
> __arm_vmulhq_m_s8(__inactive, __a, __b, __p)
> -#define vmulhq_m_s32(__inactive, __a, __b, __p)
> __arm_vmulhq_m_s32(__inactive, __a, __b, __p)
> -#define vmulhq_m_s16(__inactive, __a, __b, __p)
> __arm_vmulhq_m_s16(__inactive, __a, __b, __p)
> -#define vmulhq_m_u8(__inactive, __a, __b, __p)
> __arm_vmulhq_m_u8(__inactive, __a, __b, __p)
> -#define vmulhq_m_u32(__inactive, __a, __b, __p)
> __arm_vmulhq_m_u32(__inactive, __a, __b, __p)
> -#define vmulhq_m_u16(__inactive, __a, __b, __p)
> __arm_vmulhq_m_u16(__inactive, __a, __b, __p)
>  #define vmullbq_int_m_s8(__inactive, __a, __b, __p)
> __arm_vmullbq_int_m_s8(__inactive, __a, __b, __p)
>  #define vmullbq_int_m_s32(__inactive, __a, __b, __p)
> __arm_vmullbq_int_m_s32(__inactive, __a, __b, __p)
>  #define vmullbq_int_m_s16(__inactive, __a, __b, __p)
> __arm_vmullbq_int_m_s16(__inactive, __a, __b, __p)
> @@ -1727,18 +1604,6 @@
>  #define vornq_m_u8(__inactive, __a, __b, __p)
> __arm_vornq_m_u8(__inactive, __a, __b, __p)
>  #define vornq_m_u32(__inactive, __a, __b, __p)
> __arm_vornq_m_u32(__inactive, __a, __b, __p)
>  #define vornq_m_u16(__inactive, __a, __b, __p)
> __arm_vornq_m_u16(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_s8(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_s8(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_s32(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_s32(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_s16(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_s16(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_u8(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_u8(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_u32(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_u32(__inactive, __a, __b, __p)
> -#define vqaddq_m_n_u16(__inactive, __a, __b, __p)
> __arm_vqaddq_m_n_u16(__inactive, __a, __b, __p)
> -#define vqaddq_m_s8(__inactive, __a, __b, __p)
> __arm_vqaddq_m_s8(__inactive, __a, __b, __p)
> -#define vqaddq_m_s32(__inactive, __a, __b, __p)
> __arm_vqaddq_m_s32(__inactive, __a, __b, __p)
> -#define vqaddq_m_s16(__inactive, __a, __b, __p)
> __arm_vqaddq_m_s16(__inactive, __a, __b, __p)
> -#define vqaddq_m_u8(__inactive, __a, __b, __p)
> __arm_vqaddq_m_u8(__inactive, __a, __b, __p)
> -#define vqaddq_m_u32(__inactive, __a, __b, __p)
> __arm_vqaddq_m_u32(__inactive, __a, __b, __p)
> -#define vqaddq_m_u16(__inactive, __a, __b, __p)
> __arm_vqaddq_m_u16(__inactive, __a, __b, __p)
>  #define vqdmladhq_m_s8(__inactive, __a, __b, __p)
> __arm_vqdmladhq_m_s8(__inactive, __a, __b, __p)
>  #define vqdmladhq_m_s32(__inactive, __a, __b, __p)
> __arm_vqdmladhq_m_s32(__inactive, __a, __b, __p)
>  #define vqdmladhq_m_s16(__inactive, __a, __b, __p)
> __arm_vqdmladhq_m_s16(__inactive, __a, __b, __p)
> @@ -1757,12 +1622,6 @@
>  #define vqdmlsdhxq_m_s8(__inactive, __a, __b, __p)
> __arm_vqdmlsdhxq_m_s8(__inactive, __a, __b, __p)
>  #define vqdmlsdhxq_m_s32(__inactive, __a, __b, __p)
> __arm_vqdmlsdhxq_m_s32(__inactive, __a, __b, __p)
>  #define vqdmlsdhxq_m_s16(__inactive, __a, __b, __p)
> __arm_vqdmlsdhxq_m_s16(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_n_s8(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_n_s8(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_n_s32(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_n_s32(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_n_s16(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_n_s16(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_s8(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_s8(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_s32(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_s32(__inactive, __a, __b, __p)
> -#define vqdmulhq_m_s16(__inactive, __a, __b, __p)
> __arm_vqdmulhq_m_s16(__inactive, __a, __b, __p)
>  #define vqrdmladhq_m_s8(__inactive, __a, __b, __p)
> __arm_vqrdmladhq_m_s8(__inactive, __a, __b, __p)
>  #define vqrdmladhq_m_s32(__inactive, __a, __b, __p)
> __arm_vqrdmladhq_m_s32(__inactive, __a, __b, __p)
>  #define vqrdmladhq_m_s16(__inactive, __a, __b, __p)
> __arm_vqrdmladhq_m_s16(__inactive, __a, __b, __p)
> @@ -1805,30 +1664,6 @@
>  #define vqshlq_m_u8(__inactive, __a, __b, __p)
> __arm_vqshlq_m_u8(__inactive, __a, __b, __p)
>  #define vqshlq_m_u32(__inactive, __a, __b, __p)
> __arm_vqshlq_m_u32(__inactive, __a, __b, __p)
>  #define vqshlq_m_u16(__inactive, __a, __b, __p)
> __arm_vqshlq_m_u16(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_s8(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_s8(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_s32(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_s32(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_s16(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_s16(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_u8(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_u8(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_u32(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_u32(__inactive, __a, __b, __p)
> -#define vqsubq_m_n_u16(__inactive, __a, __b, __p)
> __arm_vqsubq_m_n_u16(__inactive, __a, __b, __p)
> -#define vqsubq_m_s8(__inactive, __a, __b, __p)
> __arm_vqsubq_m_s8(__inactive, __a, __b, __p)
> -#define vqsubq_m_s32(__inactive, __a, __b, __p)
> __arm_vqsubq_m_s32(__inactive, __a, __b, __p)
> -#define vqsubq_m_s16(__inactive, __a, __b, __p)
> __arm_vqsubq_m_s16(__inactive, __a, __b, __p)
> -#define vqsubq_m_u8(__inactive, __a, __b, __p)
> __arm_vqsubq_m_u8(__inactive, __a, __b, __p)
> -#define vqsubq_m_u32(__inactive, __a, __b, __p)
> __arm_vqsubq_m_u32(__inactive, __a, __b, __p)
> -#define vqsubq_m_u16(__inactive, __a, __b, __p)
> __arm_vqsubq_m_u16(__inactive, __a, __b, __p)
> -#define vrhaddq_m_s8(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_s8(__inactive, __a, __b, __p)
> -#define vrhaddq_m_s32(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_s32(__inactive, __a, __b, __p)
> -#define vrhaddq_m_s16(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_s16(__inactive, __a, __b, __p)
> -#define vrhaddq_m_u8(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_u8(__inactive, __a, __b, __p)
> -#define vrhaddq_m_u32(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_u32(__inactive, __a, __b, __p)
> -#define vrhaddq_m_u16(__inactive, __a, __b, __p)
> __arm_vrhaddq_m_u16(__inactive, __a, __b, __p)
> -#define vrmulhq_m_s8(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_s8(__inactive, __a, __b, __p)
> -#define vrmulhq_m_s32(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_s32(__inactive, __a, __b, __p)
> -#define vrmulhq_m_s16(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_s16(__inactive, __a, __b, __p)
> -#define vrmulhq_m_u8(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_u8(__inactive, __a, __b, __p)
> -#define vrmulhq_m_u32(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_u32(__inactive, __a, __b, __p)
> -#define vrmulhq_m_u16(__inactive, __a, __b, __p)
> __arm_vrmulhq_m_u16(__inactive, __a, __b, __p)
>  #define vrshlq_m_s8(__inactive, __a, __b, __p)
> __arm_vrshlq_m_s8(__inactive, __a, __b, __p)
>  #define vrshlq_m_s32(__inactive, __a, __b, __p)
> __arm_vrshlq_m_s32(__inactive, __a, __b, __p)
>  #define vrshlq_m_s16(__inactive, __a, __b, __p)
> __arm_vrshlq_m_s16(__inactive, __a, __b, __p)
> @@ -2315,12 +2150,6 @@
>  #define vnegq_x_s8(__a, __p) __arm_vnegq_x_s8(__a, __p)
>  #define vnegq_x_s16(__a, __p) __arm_vnegq_x_s16(__a, __p)
>  #define vnegq_x_s32(__a, __p) __arm_vnegq_x_s32(__a, __p)
> -#define vmulhq_x_s8(__a, __b, __p) __arm_vmulhq_x_s8(__a, __b, __p)
> -#define vmulhq_x_s16(__a, __b, __p) __arm_vmulhq_x_s16(__a, __b, __p)
> -#define vmulhq_x_s32(__a, __b, __p) __arm_vmulhq_x_s32(__a, __b, __p)
> -#define vmulhq_x_u8(__a, __b, __p) __arm_vmulhq_x_u8(__a, __b, __p)
> -#define vmulhq_x_u16(__a, __b, __p) __arm_vmulhq_x_u16(__a, __b, __p)
> -#define vmulhq_x_u32(__a, __b, __p) __arm_vmulhq_x_u32(__a, __b, __p)
>  #define vmullbq_poly_x_p8(__a, __b, __p) __arm_vmullbq_poly_x_p8(__a,
> __b, __p)
>  #define vmullbq_poly_x_p16(__a, __b, __p)
> __arm_vmullbq_poly_x_p16(__a, __b, __p)
>  #define vmullbq_int_x_s8(__a, __b, __p) __arm_vmullbq_int_x_s8(__a, __b,
> __p)
> @@ -2349,48 +2178,12 @@
>  #define vcaddq_rot270_x_u8(__a, __b, __p)
> __arm_vcaddq_rot270_x_u8(__a, __b, __p)
>  #define vcaddq_rot270_x_u16(__a, __b, __p)
> __arm_vcaddq_rot270_x_u16(__a, __b, __p)
>  #define vcaddq_rot270_x_u32(__a, __b, __p)
> __arm_vcaddq_rot270_x_u32(__a, __b, __p)
> -#define vhaddq_x_n_s8(__a, __b, __p) __arm_vhaddq_x_n_s8(__a, __b,
> __p)
> -#define vhaddq_x_n_s16(__a, __b, __p) __arm_vhaddq_x_n_s16(__a, __b,
> __p)
> -#define vhaddq_x_n_s32(__a, __b, __p) __arm_vhaddq_x_n_s32(__a, __b,
> __p)
> -#define vhaddq_x_n_u8(__a, __b, __p) __arm_vhaddq_x_n_u8(__a, __b,
> __p)
> -#define vhaddq_x_n_u16(__a, __b, __p) __arm_vhaddq_x_n_u16(__a, __b,
> __p)
> -#define vhaddq_x_n_u32(__a, __b, __p) __arm_vhaddq_x_n_u32(__a, __b,
> __p)
> -#define vhaddq_x_s8(__a, __b, __p) __arm_vhaddq_x_s8(__a, __b, __p)
> -#define vhaddq_x_s16(__a, __b, __p) __arm_vhaddq_x_s16(__a, __b, __p)
> -#define vhaddq_x_s32(__a, __b, __p) __arm_vhaddq_x_s32(__a, __b, __p)
> -#define vhaddq_x_u8(__a, __b, __p) __arm_vhaddq_x_u8(__a, __b, __p)
> -#define vhaddq_x_u16(__a, __b, __p) __arm_vhaddq_x_u16(__a, __b, __p)
> -#define vhaddq_x_u32(__a, __b, __p) __arm_vhaddq_x_u32(__a, __b, __p)
>  #define vhcaddq_rot90_x_s8(__a, __b, __p)
> __arm_vhcaddq_rot90_x_s8(__a, __b, __p)
>  #define vhcaddq_rot90_x_s16(__a, __b, __p)
> __arm_vhcaddq_rot90_x_s16(__a, __b, __p)
>  #define vhcaddq_rot90_x_s32(__a, __b, __p)
> __arm_vhcaddq_rot90_x_s32(__a, __b, __p)
>  #define vhcaddq_rot270_x_s8(__a, __b, __p)
> __arm_vhcaddq_rot270_x_s8(__a, __b, __p)
>  #define vhcaddq_rot270_x_s16(__a, __b, __p)
> __arm_vhcaddq_rot270_x_s16(__a, __b, __p)
>  #define vhcaddq_rot270_x_s32(__a, __b, __p)
> __arm_vhcaddq_rot270_x_s32(__a, __b, __p)
> -#define vhsubq_x_n_s8(__a, __b, __p) __arm_vhsubq_x_n_s8(__a, __b,
> __p)
> -#define vhsubq_x_n_s16(__a, __b, __p) __arm_vhsubq_x_n_s16(__a, __b,
> __p)
> -#define vhsubq_x_n_s32(__a, __b, __p) __arm_vhsubq_x_n_s32(__a, __b,
> __p)
> -#define vhsubq_x_n_u8(__a, __b, __p) __arm_vhsubq_x_n_u8(__a, __b,
> __p)
> -#define vhsubq_x_n_u16(__a, __b, __p) __arm_vhsubq_x_n_u16(__a, __b,
> __p)
> -#define vhsubq_x_n_u32(__a, __b, __p) __arm_vhsubq_x_n_u32(__a, __b,
> __p)
> -#define vhsubq_x_s8(__a, __b, __p) __arm_vhsubq_x_s8(__a, __b, __p)
> -#define vhsubq_x_s16(__a, __b, __p) __arm_vhsubq_x_s16(__a, __b, __p)
> -#define vhsubq_x_s32(__a, __b, __p) __arm_vhsubq_x_s32(__a, __b, __p)
> -#define vhsubq_x_u8(__a, __b, __p) __arm_vhsubq_x_u8(__a, __b, __p)
> -#define vhsubq_x_u16(__a, __b, __p) __arm_vhsubq_x_u16(__a, __b, __p)
> -#define vhsubq_x_u32(__a, __b, __p) __arm_vhsubq_x_u32(__a, __b, __p)
> -#define vrhaddq_x_s8(__a, __b, __p) __arm_vrhaddq_x_s8(__a, __b, __p)
> -#define vrhaddq_x_s16(__a, __b, __p) __arm_vrhaddq_x_s16(__a, __b,
> __p)
> -#define vrhaddq_x_s32(__a, __b, __p) __arm_vrhaddq_x_s32(__a, __b,
> __p)
> -#define vrhaddq_x_u8(__a, __b, __p) __arm_vrhaddq_x_u8(__a, __b, __p)
> -#define vrhaddq_x_u16(__a, __b, __p) __arm_vrhaddq_x_u16(__a, __b,
> __p)
> -#define vrhaddq_x_u32(__a, __b, __p) __arm_vrhaddq_x_u32(__a, __b,
> __p)
> -#define vrmulhq_x_s8(__a, __b, __p) __arm_vrmulhq_x_s8(__a, __b, __p)
> -#define vrmulhq_x_s16(__a, __b, __p) __arm_vrmulhq_x_s16(__a, __b,
> __p)
> -#define vrmulhq_x_s32(__a, __b, __p) __arm_vrmulhq_x_s32(__a, __b,
> __p)
> -#define vrmulhq_x_u8(__a, __b, __p) __arm_vrmulhq_x_u8(__a, __b, __p)
> -#define vrmulhq_x_u16(__a, __b, __p) __arm_vrmulhq_x_u16(__a, __b,
> __p)
> -#define vrmulhq_x_u32(__a, __b, __p) __arm_vrmulhq_x_u32(__a, __b,
> __p)
>  #define vbicq_x_s8(__a, __b, __p) __arm_vbicq_x_s8(__a, __b, __p)
>  #define vbicq_x_s16(__a, __b, __p) __arm_vbicq_x_s16(__a, __b, __p)
>  #define vbicq_x_s32(__a, __b, __p) __arm_vbicq_x_s32(__a, __b, __p)
> @@ -3351,48 +3144,6 @@ __arm_vshlq_u32 (uint32x4_t __a, int32x4_t __b)
>    return __builtin_mve_vshlq_uv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vrmulhq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vrhaddq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vqsubq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_u8 (uint8x16_t __a, uint8_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vqaddq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_u8 (uint8x16_t __a, uint8_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_uv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline uint8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_u8 (uint8x16_t __a, uint8x16_t __b)
> @@ -3414,13 +3165,6 @@ __arm_vmullbq_int_u8 (uint8x16_t __a,
> uint8x16_t __b)
>    return __builtin_mve_vmullbq_int_uv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vmulhq_uv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq_u8 (uint8x16_t __a, uint8x16_t __b)
> @@ -3456,34 +3200,6 @@ __arm_vmaxq_u8 (uint8x16_t __a, uint8x16_t
> __b)
>    return __builtin_mve_vmaxq_uv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vhsubq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_u8 (uint8x16_t __a, uint8_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
> -{
> -  return __builtin_mve_vhaddq_uv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_u8 (uint8x16_t __a, uint8_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_uv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq_n_u8 (uint8x16_t __a, uint8_t __b)
> @@ -3794,34 +3510,6 @@ __arm_vrshlq_n_s8 (int8x16_t __a, int32_t __b)
>    return __builtin_mve_vrshlq_n_sv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vrmulhq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vrhaddq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vqsubq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_s8 (int8x16_t __a, int8_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_sv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq_s8 (int8x16_t __a, int8x16_t __b)
> @@ -3864,34 +3552,6 @@ __arm_vqrdmulhq_n_s8 (int8x16_t __a, int8_t
> __b)
>    return __builtin_mve_vqrdmulhq_n_sv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_n_s8 (int8x16_t __a, int8_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_n_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vqaddq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_s8 (int8x16_t __a, int8_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_sv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_s8 (int8x16_t __a, int8x16_t __b)
> @@ -3913,13 +3573,6 @@ __arm_vmullbq_int_s8 (int8x16_t __a, int8x16_t
> __b)
>    return __builtin_mve_vmullbq_int_sv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vmulhq_sv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq_s8 (int8x16_t __a, int8x16_t __b)
> @@ -3976,20 +3629,6 @@ __arm_vmaxq_s8 (int8x16_t __a, int8x16_t __b)
>    return __builtin_mve_vmaxq_sv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vhsubq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_s8 (int8x16_t __a, int8_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_sv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90_s8 (int8x16_t __a, int8x16_t __b)
> @@ -4004,20 +3643,6 @@ __arm_vhcaddq_rot270_s8 (int8x16_t __a,
> int8x16_t __b)
>    return __builtin_mve_vhcaddq_rot270_sv16qi (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_s8 (int8x16_t __a, int8x16_t __b)
> -{
> -  return __builtin_mve_vhaddq_sv16qi (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_s8 (int8x16_t __a, int8_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_sv16qi (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90_s8 (int8x16_t __a, int8x16_t __b)
> @@ -4081,48 +3706,6 @@ __arm_vqshlq_n_s8 (int8x16_t __a, const int
> __imm)
>    return __builtin_mve_vqshlq_n_sv16qi (__a, __imm);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vrmulhq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vrhaddq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vqsubq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_u16 (uint16x8_t __a, uint16_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vqaddq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_u16 (uint16x8_t __a, uint16_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_uv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline uint16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_u16 (uint16x8_t __a, uint16x8_t __b)
> @@ -4144,13 +3727,6 @@ __arm_vmullbq_int_u16 (uint16x8_t __a,
> uint16x8_t __b)
>    return __builtin_mve_vmullbq_int_uv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vmulhq_uv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq_u16 (uint16x8_t __a, uint16x8_t __b)
> @@ -4186,34 +3762,6 @@ __arm_vmaxq_u16 (uint16x8_t __a, uint16x8_t
> __b)
>    return __builtin_mve_vmaxq_uv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vhsubq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_u16 (uint16x8_t __a, uint16_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
> -{
> -  return __builtin_mve_vhaddq_uv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_u16 (uint16x8_t __a, uint16_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_uv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq_n_u16 (uint16x8_t __a, uint16_t __b)
> @@ -4524,34 +4072,6 @@ __arm_vrshlq_n_s16 (int16x8_t __a, int32_t __b)
>    return __builtin_mve_vrshlq_n_sv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vrmulhq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vrhaddq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vqsubq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_s16 (int16x8_t __a, int16_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_sv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq_s16 (int16x8_t __a, int16x8_t __b)
> @@ -4594,34 +4114,6 @@ __arm_vqrdmulhq_n_s16 (int16x8_t __a, int16_t
> __b)
>    return __builtin_mve_vqrdmulhq_n_sv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_n_s16 (int16x8_t __a, int16_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_n_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vqaddq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_s16 (int16x8_t __a, int16_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_sv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_s16 (int16x8_t __a, int16x8_t __b)
> @@ -4643,13 +4135,6 @@ __arm_vmullbq_int_s16 (int16x8_t __a, int16x8_t
> __b)
>    return __builtin_mve_vmullbq_int_sv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vmulhq_sv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq_s16 (int16x8_t __a, int16x8_t __b)
> @@ -4706,20 +4191,6 @@ __arm_vmaxq_s16 (int16x8_t __a, int16x8_t __b)
>    return __builtin_mve_vmaxq_sv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vhsubq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_s16 (int16x8_t __a, int16_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_sv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90_s16 (int16x8_t __a, int16x8_t __b)
> @@ -4734,20 +4205,6 @@ __arm_vhcaddq_rot270_s16 (int16x8_t __a,
> int16x8_t __b)
>    return __builtin_mve_vhcaddq_rot270_sv8hi (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_s16 (int16x8_t __a, int16x8_t __b)
> -{
> -  return __builtin_mve_vhaddq_sv8hi (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_s16 (int16x8_t __a, int16_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_sv8hi (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90_s16 (int16x8_t __a, int16x8_t __b)
> @@ -4811,48 +4268,6 @@ __arm_vqshlq_n_s16 (int16x8_t __a, const int
> __imm)
>    return __builtin_mve_vqshlq_n_sv8hi (__a, __imm);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vrmulhq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vrhaddq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vqsubq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_u32 (uint32x4_t __a, uint32_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vqaddq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_u32 (uint32x4_t __a, uint32_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_uv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_u32 (uint32x4_t __a, uint32x4_t __b)
> @@ -4874,13 +4289,6 @@ __arm_vmullbq_int_u32 (uint32x4_t __a,
> uint32x4_t __b)
>    return __builtin_mve_vmullbq_int_uv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vmulhq_uv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq_u32 (uint32x4_t __a, uint32x4_t __b)
> @@ -4916,34 +4324,6 @@ __arm_vmaxq_u32 (uint32x4_t __a, uint32x4_t
> __b)
>    return __builtin_mve_vmaxq_uv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vhsubq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_u32 (uint32x4_t __a, uint32_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
> -{
> -  return __builtin_mve_vhaddq_uv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_u32 (uint32x4_t __a, uint32_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_uv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq_n_u32 (uint32x4_t __a, uint32_t __b)
> @@ -5254,34 +4634,6 @@ __arm_vrshlq_n_s32 (int32x4_t __a, int32_t __b)
>    return __builtin_mve_vrshlq_n_sv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vrmulhq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vrhaddq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vqsubq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_n_s32 (int32x4_t __a, int32_t __b)
> -{
> -  return __builtin_mve_vqsubq_n_sv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq_s32 (int32x4_t __a, int32x4_t __b)
> @@ -5324,34 +4676,6 @@ __arm_vqrdmulhq_n_s32 (int32x4_t __a, int32_t
> __b)
>    return __builtin_mve_vqrdmulhq_n_sv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_n_s32 (int32x4_t __a, int32_t __b)
> -{
> -  return __builtin_mve_vqdmulhq_n_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vqaddq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_n_s32 (int32x4_t __a, int32_t __b)
> -{
> -  return __builtin_mve_vqaddq_n_sv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq_s32 (int32x4_t __a, int32x4_t __b)
> @@ -5373,13 +4697,6 @@ __arm_vmullbq_int_s32 (int32x4_t __a, int32x4_t
> __b)
>    return __builtin_mve_vmullbq_int_sv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vmulhq_sv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq_s32 (int32x4_t __a, int32x4_t __b)
> @@ -5436,20 +4753,6 @@ __arm_vmaxq_s32 (int32x4_t __a, int32x4_t __b)
>    return __builtin_mve_vmaxq_sv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vhsubq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_n_s32 (int32x4_t __a, int32_t __b)
> -{
> -  return __builtin_mve_vhsubq_n_sv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90_s32 (int32x4_t __a, int32x4_t __b)
> @@ -5464,20 +4767,6 @@ __arm_vhcaddq_rot270_s32 (int32x4_t __a,
> int32x4_t __b)
>    return __builtin_mve_vhcaddq_rot270_sv4si (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_s32 (int32x4_t __a, int32x4_t __b)
> -{
> -  return __builtin_mve_vhaddq_sv4si (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_n_s32 (int32x4_t __a, int32_t __b)
> -{
> -  return __builtin_mve_vhaddq_n_sv4si (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90_s32 (int32x4_t __a, int32x4_t __b)
> @@ -9005,90 +8294,6 @@ __arm_vcaddq_rot90_m_u16 (uint16x8_t
> __inactive, uint16x8_t __a, uint16x8_t __b,
>    return __builtin_mve_vcaddq_rot90_m_uv8hi (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot270_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t
> __b, mve_pred16_t __p)
> @@ -9131,90 +8336,6 @@ __arm_vhcaddq_rot90_m_s16 (int16x8_t
> __inactive, int16x8_t __a, int16x8_t __b, m
>    return __builtin_mve_vhcaddq_rot90_m_sv8hi (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmaxq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -9488,48 +8609,6 @@ __arm_vmlsdavaxq_p_s16 (int32_t __a, int16x8_t
> __b, int16x8_t __c, mve_pred16_t
>    return __builtin_mve_vmlsdavaxq_p_sv8hi (__a, __b, __c, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmullbq_int_m_s8 (int16x8_t __inactive, int8x16_t __a, int8x16_t
> __b, mve_pred16_t __p)
> @@ -9656,90 +8735,6 @@ __arm_vornq_m_u16 (uint16x8_t __inactive,
> uint16x8_t __a, uint16x8_t __b, mve_pr
>    return __builtin_mve_vornq_m_uv8hi (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_n_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqaddq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqdmladhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -9845,48 +8840,6 @@ __arm_vqdmlsdhxq_m_s16 (int16x8_t __inactive,
> int16x8_t __a, int16x8_t __b, mve_
>    return __builtin_mve_vqdmlsdhxq_m_sv8hi (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_n_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_n_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_n_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqdmulhq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqrdmladhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t
> __b, mve_pred16_t __p)
> @@ -10202,174 +9155,6 @@ __arm_vqshlq_m_u16 (uint16x8_t __inactive,
> uint16x8_t __a, int16x8_t __b, mve_pr
>    return __builtin_mve_vqshlq_m_uv8hi (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_n_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vqsubq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv8hi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv16qi (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv4si (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t
> __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv8hi (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vrshlq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -13289,48 +12074,6 @@ __arm_vnegq_x_s32 (int32x4_t __a,
> mve_pred16_t __p)
>    return __builtin_mve_vnegq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv16qi (__arm_vuninitializedq_s8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv8hi (__arm_vuninitializedq_s16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv16qi (__arm_vuninitializedq_u8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv8hi (__arm_vuninitializedq_u16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vmulhq_m_uv4si (__arm_vuninitializedq_u32 (), __a,
> __b, __p);
> -}
> -
>  __extension__ extern __inline uint16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmullbq_poly_x_p8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t
> __p)
> @@ -13527,90 +12270,6 @@ __arm_vcaddq_rot270_x_u32 (uint32x4_t __a,
> uint32x4_t __b, mve_pred16_t __p)
>    return __builtin_mve_vcaddq_rot270_m_uv4si (__arm_vuninitializedq_u32
> (), __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_s8 (int8x16_t __a, int8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv16qi (__arm_vuninitializedq_s8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_s16 (int16x8_t __a, int16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv8hi (__arm_vuninitializedq_s16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_s32 (int32x4_t __a, int32_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_sv4si (__arm_vuninitializedq_s32 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_u8 (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv16qi (__arm_vuninitializedq_u8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_u16 (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv8hi (__arm_vuninitializedq_u16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_n_u32 (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_n_uv4si (__arm_vuninitializedq_u32 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv16qi (__arm_vuninitializedq_s8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv8hi (__arm_vuninitializedq_s16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv16qi (__arm_vuninitializedq_u8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv8hi (__arm_vuninitializedq_u16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhaddq_m_uv4si (__arm_vuninitializedq_u32 (), __a,
> __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t
> __p)
> @@ -13653,174 +12312,6 @@ __arm_vhcaddq_rot270_x_s32 (int32x4_t __a,
> int32x4_t __b, mve_pred16_t __p)
>    return __builtin_mve_vhcaddq_rot270_m_sv4si (__arm_vuninitializedq_s32
> (), __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_s8 (int8x16_t __a, int8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv16qi (__arm_vuninitializedq_s8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_s16 (int16x8_t __a, int16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv8hi (__arm_vuninitializedq_s16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_s32 (int32x4_t __a, int32_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_sv4si (__arm_vuninitializedq_s32 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_u8 (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv16qi (__arm_vuninitializedq_u8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_u16 (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv8hi (__arm_vuninitializedq_u16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_n_u32 (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_n_uv4si (__arm_vuninitializedq_u32 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv16qi (__arm_vuninitializedq_s8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv8hi (__arm_vuninitializedq_s16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv16qi (__arm_vuninitializedq_u8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv8hi (__arm_vuninitializedq_u16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vhsubq_m_uv4si (__arm_vuninitializedq_u32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv16qi (__arm_vuninitializedq_s8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv8hi (__arm_vuninitializedq_s16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv16qi (__arm_vuninitializedq_u8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv8hi (__arm_vuninitializedq_u16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrhaddq_m_uv4si (__arm_vuninitializedq_u32 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv16qi (__arm_vuninitializedq_s8 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv8hi (__arm_vuninitializedq_s16 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_sv4si (__arm_vuninitializedq_s32 (), __a,
> __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv16qi (__arm_vuninitializedq_u8 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv8hi (__arm_vuninitializedq_u16 (),
> __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> -  return __builtin_mve_vrmulhq_m_uv4si (__arm_vuninitializedq_u32 (),
> __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vbicq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> @@ -18558,48 +17049,6 @@ __arm_vshlq (uint32x4_t __a, int32x4_t __b)
>   return __arm_vshlq_u32 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vrmulhq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vrhaddq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vqsubq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint8x16_t __a, uint8_t __b)
> -{
> - return __arm_vqsubq_n_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vqaddq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint8x16_t __a, uint8_t __b)
> -{
> - return __arm_vqaddq_n_u8 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (uint8x16_t __a, uint8x16_t __b)
> @@ -18621,13 +17070,6 @@ __arm_vmullbq_int (uint8x16_t __a, uint8x16_t
> __b)
>   return __arm_vmullbq_int_u8 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vmulhq_u8 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq (uint8x16_t __a, uint8x16_t __b)
> @@ -18663,34 +17105,6 @@ __arm_vmaxq (uint8x16_t __a, uint8x16_t __b)
>   return __arm_vmaxq_u8 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vhsubq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint8x16_t __a, uint8_t __b)
> -{
> - return __arm_vhsubq_n_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint8x16_t __a, uint8x16_t __b)
> -{
> - return __arm_vhaddq_u8 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint8x16_t __a, uint8_t __b)
> -{
> - return __arm_vhaddq_n_u8 (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq (uint8x16_t __a, uint8_t __b)
> @@ -18999,34 +17413,6 @@ __arm_vrshlq (int8x16_t __a, int32_t __b)
>   return __arm_vrshlq_n_s8 (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vrmulhq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vrhaddq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vqsubq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int8x16_t __a, int8_t __b)
> -{
> - return __arm_vqsubq_n_s8 (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq (int8x16_t __a, int8x16_t __b)
> @@ -19069,34 +17455,6 @@ __arm_vqrdmulhq (int8x16_t __a, int8_t __b)
>   return __arm_vqrdmulhq_n_s8 (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vqdmulhq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int8x16_t __a, int8_t __b)
> -{
> - return __arm_vqdmulhq_n_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vqaddq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int8x16_t __a, int8_t __b)
> -{
> - return __arm_vqaddq_n_s8 (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (int8x16_t __a, int8x16_t __b)
> @@ -19118,13 +17476,6 @@ __arm_vmullbq_int (int8x16_t __a, int8x16_t
> __b)
>   return __arm_vmullbq_int_s8 (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vmulhq_s8 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq (int8x16_t __a, int8x16_t __b)
> @@ -19181,20 +17532,6 @@ __arm_vmaxq (int8x16_t __a, int8x16_t __b)
>   return __arm_vmaxq_s8 (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vhsubq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int8x16_t __a, int8_t __b)
> -{
> - return __arm_vhsubq_n_s8 (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90 (int8x16_t __a, int8x16_t __b)
> @@ -19209,20 +17546,6 @@ __arm_vhcaddq_rot270 (int8x16_t __a,
> int8x16_t __b)
>   return __arm_vhcaddq_rot270_s8 (__a, __b);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int8x16_t __a, int8x16_t __b)
> -{
> - return __arm_vhaddq_s8 (__a, __b);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int8x16_t __a, int8_t __b)
> -{
> - return __arm_vhaddq_n_s8 (__a, __b);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90 (int8x16_t __a, int8x16_t __b)
> @@ -19286,48 +17609,6 @@ __arm_vqshlq_n (int8x16_t __a, const int
> __imm)
>   return __arm_vqshlq_n_s8 (__a, __imm);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vrmulhq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vrhaddq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vqsubq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint16x8_t __a, uint16_t __b)
> -{
> - return __arm_vqsubq_n_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vqaddq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint16x8_t __a, uint16_t __b)
> -{
> - return __arm_vqaddq_n_u16 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (uint16x8_t __a, uint16x8_t __b)
> @@ -19349,13 +17630,6 @@ __arm_vmullbq_int (uint16x8_t __a, uint16x8_t
> __b)
>   return __arm_vmullbq_int_u16 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vmulhq_u16 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq (uint16x8_t __a, uint16x8_t __b)
> @@ -19391,34 +17665,6 @@ __arm_vmaxq (uint16x8_t __a, uint16x8_t __b)
>   return __arm_vmaxq_u16 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vhsubq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint16x8_t __a, uint16_t __b)
> -{
> - return __arm_vhsubq_n_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint16x8_t __a, uint16x8_t __b)
> -{
> - return __arm_vhaddq_u16 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint16x8_t __a, uint16_t __b)
> -{
> - return __arm_vhaddq_n_u16 (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq (uint16x8_t __a, uint16_t __b)
> @@ -19727,34 +17973,6 @@ __arm_vrshlq (int16x8_t __a, int32_t __b)
>   return __arm_vrshlq_n_s16 (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vrmulhq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vrhaddq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vqsubq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int16x8_t __a, int16_t __b)
> -{
> - return __arm_vqsubq_n_s16 (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq (int16x8_t __a, int16x8_t __b)
> @@ -19797,34 +18015,6 @@ __arm_vqrdmulhq (int16x8_t __a, int16_t __b)
>   return __arm_vqrdmulhq_n_s16 (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vqdmulhq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int16x8_t __a, int16_t __b)
> -{
> - return __arm_vqdmulhq_n_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vqaddq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int16x8_t __a, int16_t __b)
> -{
> - return __arm_vqaddq_n_s16 (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (int16x8_t __a, int16x8_t __b)
> @@ -19846,13 +18036,6 @@ __arm_vmullbq_int (int16x8_t __a, int16x8_t
> __b)
>   return __arm_vmullbq_int_s16 (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vmulhq_s16 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq (int16x8_t __a, int16x8_t __b)
> @@ -19909,20 +18092,6 @@ __arm_vmaxq (int16x8_t __a, int16x8_t __b)
>   return __arm_vmaxq_s16 (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vhsubq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int16x8_t __a, int16_t __b)
> -{
> - return __arm_vhsubq_n_s16 (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90 (int16x8_t __a, int16x8_t __b)
> @@ -19937,20 +18106,6 @@ __arm_vhcaddq_rot270 (int16x8_t __a,
> int16x8_t __b)
>   return __arm_vhcaddq_rot270_s16 (__a, __b);
>  }
> 
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int16x8_t __a, int16x8_t __b)
> -{
> - return __arm_vhaddq_s16 (__a, __b);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int16x8_t __a, int16_t __b)
> -{
> - return __arm_vhaddq_n_s16 (__a, __b);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90 (int16x8_t __a, int16x8_t __b)
> @@ -20014,48 +18169,6 @@ __arm_vqshlq_n (int16x8_t __a, const int
> __imm)
>   return __arm_vqshlq_n_s16 (__a, __imm);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vrmulhq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vrhaddq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vqsubq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (uint32x4_t __a, uint32_t __b)
> -{
> - return __arm_vqsubq_n_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vqaddq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (uint32x4_t __a, uint32_t __b)
> -{
> - return __arm_vqaddq_n_u32 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (uint32x4_t __a, uint32x4_t __b)
> @@ -20077,13 +18190,6 @@ __arm_vmullbq_int (uint32x4_t __a, uint32x4_t
> __b)
>   return __arm_vmullbq_int_u32 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vmulhq_u32 (__a, __b);
> -}
> -
>  __extension__ extern __inline uint32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmladavq (uint32x4_t __a, uint32x4_t __b)
> @@ -20119,34 +18225,6 @@ __arm_vmaxq (uint32x4_t __a, uint32x4_t __b)
>   return __arm_vmaxq_u32 (__a, __b);
>  }
> 
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vhsubq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (uint32x4_t __a, uint32_t __b)
> -{
> - return __arm_vhsubq_n_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint32x4_t __a, uint32x4_t __b)
> -{
> - return __arm_vhaddq_u32 (__a, __b);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (uint32x4_t __a, uint32_t __b)
> -{
> - return __arm_vhaddq_n_u32 (__a, __b);
> -}
> -
>  __extension__ extern __inline mve_pred16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcmpneq (uint32x4_t __a, uint32_t __b)
> @@ -20455,34 +18533,6 @@ __arm_vrshlq (int32x4_t __a, int32_t __b)
>   return __arm_vrshlq_n_s32 (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vrmulhq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vrhaddq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vqsubq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq (int32x4_t __a, int32_t __b)
> -{
> - return __arm_vqsubq_n_s32 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqshlq (int32x4_t __a, int32x4_t __b)
> @@ -20525,34 +18575,6 @@ __arm_vqrdmulhq (int32x4_t __a, int32_t __b)
>   return __arm_vqrdmulhq_n_s32 (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vqdmulhq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq (int32x4_t __a, int32_t __b)
> -{
> - return __arm_vqdmulhq_n_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vqaddq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq (int32x4_t __a, int32_t __b)
> -{
> - return __arm_vqaddq_n_s32 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vornq (int32x4_t __a, int32x4_t __b)
> @@ -20574,13 +18596,6 @@ __arm_vmullbq_int (int32x4_t __a, int32x4_t
> __b)
>   return __arm_vmullbq_int_s32 (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vmulhq_s32 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmlsdavxq (int32x4_t __a, int32x4_t __b)
> @@ -20637,20 +18652,6 @@ __arm_vmaxq (int32x4_t __a, int32x4_t __b)
>   return __arm_vmaxq_s32 (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vhsubq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq (int32x4_t __a, int32_t __b)
> -{
> - return __arm_vhsubq_n_s32 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90 (int32x4_t __a, int32x4_t __b)
> @@ -20665,20 +18666,6 @@ __arm_vhcaddq_rot270 (int32x4_t __a,
> int32x4_t __b)
>   return __arm_vhcaddq_rot270_s32 (__a, __b);
>  }
> 
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int32x4_t __a, int32x4_t __b)
> -{
> - return __arm_vhaddq_s32 (__a, __b);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq (int32x4_t __a, int32_t __b)
> -{
> - return __arm_vhaddq_n_s32 (__a, __b);
> -}
> -
>  __extension__ extern __inline int32x4_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vcaddq_rot90 (int32x4_t __a, int32x4_t __b)
> @@ -24165,90 +22152,6 @@ __arm_vcaddq_rot90_m (uint16x8_t __inactive,
> uint16x8_t __a, uint16x8_t __b, mve
>   return __arm_vcaddq_rot90_m_u16 (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_n_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhaddq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot270_m (int8x16_t __inactive, int8x16_t __a, int8x16_t
> __b, mve_pred16_t __p)
> @@ -24291,90 +22194,6 @@ __arm_vhcaddq_rot90_m (int16x8_t __inactive,
> int16x8_t __a, int16x8_t __b, mve_p
>   return __arm_vhcaddq_rot90_m_s16 (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_n_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vhsubq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmaxq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -24648,48 +22467,6 @@ __arm_vmlsdavaxq_p (int32_t __a, int16x8_t
> __b, int16x8_t __c, mve_pred16_t __p)
>   return __arm_vmlsdavaxq_p_s16 (__a, __b, __c, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vmulhq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmullbq_int_m (int16x8_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -24816,90 +22593,6 @@ __arm_vornq_m (uint16x8_t __inactive,
> uint16x8_t __a, uint16x8_t __b, mve_pred16
>   return __arm_vornq_m_u16 (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_n_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqaddq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqdmladhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -25005,48 +22698,6 @@ __arm_vqdmlsdhxq_m (int16x8_t __inactive,
> int16x8_t __a, int16x8_t __b, mve_pred
>   return __arm_vqdmlsdhxq_m_s16 (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_n_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_n_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_n_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqdmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqdmulhq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vqrdmladhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -25362,174 +23013,6 @@ __arm_vqshlq_m (uint16x8_t __inactive,
> uint16x8_t __a, int16x8_t __b, mve_pred16
>   return __arm_vqshlq_m_u16 (__inactive, __a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_n_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vqsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vqsubq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_s8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_s32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_s16 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_u8 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_u32 (__inactive, __a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
> mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_m_u16 (__inactive, __a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vrshlq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b,
> mve_pred16_t __p)
> @@ -27980,48 +25463,6 @@ __arm_vnegq_x (int32x4_t __a, mve_pred16_t
> __p)
>   return __arm_vnegq_x_s32 (__a, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vmulhq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vmulhq_x_u32 (__a, __b, __p);
> -}
> -
>  __extension__ extern __inline uint16x8_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vmullbq_poly_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> @@ -28218,90 +25659,6 @@ __arm_vcaddq_rot270_x (uint32x4_t __a,
> uint32x4_t __b, mve_pred16_t __p)
>   return __arm_vcaddq_rot270_x_u32 (__a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int8x16_t __a, int8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int16x8_t __a, int16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int32x4_t __a, int32_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_n_u32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhaddq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhaddq_x_u32 (__a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vhcaddq_rot90_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> @@ -28344,174 +25701,6 @@ __arm_vhcaddq_rot270_x (int32x4_t __a,
> int32x4_t __b, mve_pred16_t __p)
>   return __arm_vhcaddq_rot270_x_s32 (__a, __b, __p);
>  }
> 
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int8x16_t __a, int8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int16x8_t __a, int16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int32x4_t __a, int32_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_n_u32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vhsubq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vhsubq_x_u32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrhaddq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrhaddq_x_u32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_s8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_s16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline int32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_s32 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint8x16_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_u8 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint16x8_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_u16 (__a, __b, __p);
> -}
> -
> -__extension__ extern __inline uint32x4_t
> -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> -__arm_vrmulhq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
> -{
> - return __arm_vrmulhq_x_u32 (__a, __b, __p);
> -}
> -
>  __extension__ extern __inline int8x16_t
>  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>  __arm_vbicq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
> @@ -32685,42 +29874,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrshlq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrshlq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vrmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
> -#define __arm_vrhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
> -#define __arm_vqsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vqsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vqsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vqsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vqshluq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
>    int (*)[__ARM_mve_type_int8x16_t]: __arm_vqshluq_n_s8
> (__ARM_mve_coerce(__p0, int8x16_t), p1), \
> @@ -32831,32 +29984,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqdmullbq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqdmullbq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vqdmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqdmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqdmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqdmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> -
> -#define __arm_vqaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vqaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vqaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vqaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vmulltq_poly(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -32879,22 +30006,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmulltq_int_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmulltq_int_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> 
> -#define __arm_vhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vhcaddq_rot270(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -32909,22 +30020,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhcaddq_rot90_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhcaddq_rot90_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vhsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vminq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -32975,16 +30070,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmovnbq_u16 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmovnbq_u32 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> 
> -#define __arm_vmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vmullbq_int(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -34580,42 +31665,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrshlq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrshlq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vrmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
> -#define __arm_vrhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
> -#define __arm_vqsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vqsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vqsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vqsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vqsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vqshlq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -34694,32 +31743,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqrdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
>    int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqrdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)));})
> 
> -#define __arm_vqdmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqdmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqdmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqdmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> -
> -#define __arm_vqaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vqaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vqaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vqaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vqaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vqaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vqaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vqaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vornq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -34750,16 +31773,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmullbq_int_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmullbq_int_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> 
> -#define __arm_vmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vminq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -34794,22 +31807,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vmaxaq_s16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vmaxaq_s32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vhsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vhcaddq_rot90(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -34824,22 +31821,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhcaddq_rot270_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
>    int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhcaddq_rot270_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)));})
> 
> -#define __arm_vhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce3(p1, int)), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t)), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t),
> __ARM_mve_coerce(__p1, int16x8_t)), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t),
> __ARM_mve_coerce(__p1, int32x4_t)), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t),
> __ARM_mve_coerce(__p1, uint8x16_t)), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t)), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t)));})
> -
>  #define __arm_vcaddq_rot90(p0,p1) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0,
> \
> @@ -35910,16 +32891,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint8x16_t]: __arm_vmovltq_x_u8
> (__ARM_mve_coerce(__p1, uint8x16_t), p2), \
>    int (*)[__ARM_mve_type_uint16x8_t]: __arm_vmovltq_x_u16
> (__ARM_mve_coerce(__p1, uint16x8_t), p2));})
> 
> -#define __arm_vmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vmullbq_int_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> @@ -36095,16 +33066,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int8x16_t]: __arm_vrev16q_x_s8
> (__ARM_mve_coerce(__p1, int8x16_t), p2), \
>    int (*)[__ARM_mve_type_uint8x16_t]: __arm_vrev16q_x_u8
> (__ARM_mve_coerce(__p1, uint8x16_t), p2));})
> 
> -#define __arm_vrhaddq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrhaddq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrhaddq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrhaddq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrhaddq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrhaddq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrhaddq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vshlq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> @@ -36115,16 +33076,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vshlq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vshlq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3));})
> 
> -#define __arm_vrmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vrmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vrmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vrmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vrmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vrmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vrmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vrshlq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> @@ -36236,22 +33187,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t]: __arm_vshrq_x_n_u16
> (__ARM_mve_coerce(__p1, uint16x8_t), p2, p3), \
>    int (*)[__ARM_mve_type_uint32x4_t]: __arm_vshrq_x_n_u32
> (__ARM_mve_coerce(__p1, uint32x4_t), p2, p3));})
> 
> -#define __arm_vhaddq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_u8( __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_u16( __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhaddq_x_n_u32( __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhaddq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhaddq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhaddq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhaddq_x_u8( __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhaddq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhaddq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vhcaddq_rot270_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> @@ -36266,22 +33201,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhcaddq_rot90_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
>    int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhcaddq_rot90_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3));})
> 
> -#define __arm_vhsubq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]:
> __arm_vhsubq_x_n_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce3(p2, int), p3), \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vhsubq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vhsubq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vhsubq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vhsubq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vhsubq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vhsubq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vclsq_x(p1,p2) ({ __typeof(p1) __p1 = (p1); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)])0, \
>    int (*)[__ARM_mve_type_int8x16_t]: __arm_vclsq_x_s8
> (__ARM_mve_coerce(__p1, int8x16_t), p2), \
> @@ -36446,28 +33365,6 @@ extern void *__ARM_undef;
>    int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int16x8_t]: __arm_vqshlq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
>    int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int32x4_t]: __arm_vqshlq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3));})
> 
> -#define __arm_vrhaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vrhaddq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vrhaddq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vrhaddq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vrhaddq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vrhaddq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vrhaddq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
> -#define __arm_vrmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vrmulhq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vrmulhq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vrmulhq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vrmulhq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vrmulhq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vrmulhq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vrshlq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36509,23 +33406,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vsliq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t),  p2, p3), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vsliq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t),  p2, p3));})
> 
> -#define __arm_vqsubq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int_n]: __arm_vqsubq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vqsubq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vqsubq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_int_n]: __arm_vqsubq_m_n_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int_n]: __arm_vqsubq_m_n_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int_n]: __arm_vqsubq_m_n_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vqsubq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vqsubq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vqsubq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vqsubq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vqsubq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vqsubq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vqrdmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36799,23 +33679,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vsriq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t),
> __ARM_mve_coerce(__p1, uint16x8_t), p2, p3), \
>    int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vsriq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t),
> __ARM_mve_coerce(__p1, uint32x4_t), p2, p3));})
> 
> -#define __arm_vhaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int_n]: __arm_vhaddq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vhaddq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vhaddq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_int_n]: __arm_vhaddq_m_n_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int_n]: __arm_vhaddq_m_n_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int_n]: __arm_vhaddq_m_n_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vhaddq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vhaddq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vhaddq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vhaddq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vhaddq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vhaddq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vhcaddq_rot270_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0);
> \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36832,23 +33695,6 @@ extern void *__ARM_undef;
>    int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vhcaddq_rot90_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
>    int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vhcaddq_rot90_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3));})
> 
> -#define __arm_vhsubq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vhsubq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vhsubq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vhsubq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vhsubq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vhsubq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vhsubq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int_n]: __arm_vhsubq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vhsubq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vhsubq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_int_n]: __arm_vhsubq_m_n_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int_n]: __arm_vhsubq_m_n_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int_n]: __arm_vhsubq_m_n_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2,
> int), p3));})
> -
>  #define __arm_vmaxq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36893,17 +33739,6 @@ extern void *__ARM_undef;
>    int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int_n]: __arm_vmlasq_m_n_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
>    int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int_n]: __arm_vmlasq_m_n_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2,
> int), p3));})
> 
> -#define __arm_vmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vmulhq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vmulhq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vmulhq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vmulhq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vmulhq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vmulhq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vmullbq_int_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36933,23 +33768,6 @@ extern void *__ARM_undef;
>    int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vmulltq_poly_m_p8 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
>    int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vmulltq_poly_m_p16
> (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1,
> uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3));})
> 
> -#define __arm_vqaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int_n]: __arm_vqaddq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t),
> __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vqaddq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vqaddq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_int_n]: __arm_vqaddq_m_n_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_int_n]: __arm_vqaddq_m_n_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_int_n]: __arm_vqaddq_m_n_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vqaddq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vqaddq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vqaddq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_m
> ve_type_uint8x16_t]: __arm_vqaddq_m_u8 (__ARM_mve_coerce(__p0,
> uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_m
> ve_type_uint16x8_t]: __arm_vqaddq_m_u16 (__ARM_mve_coerce(__p0,
> uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_m
> ve_type_uint32x4_t]: __arm_vqaddq_m_u32 (__ARM_mve_coerce(__p0,
> uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vqdmlahq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -36958,17 +33776,6 @@ extern void *__ARM_undef;
>    int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vqdmlahq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
>    int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vqdmlahq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3));})
> 
> -#define __arm_vqdmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
> -  __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int
> (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typ
> eid(__p2)])0, \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int_n]: __arm_vqdmulhq_m_n_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int_n]: __arm_vqdmulhq_m_n_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int_n]: __arm_vqdmulhq_m_n_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2,
> int), p3), \
> -  int
> (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve
> _type_int8x16_t]: __arm_vqdmulhq_m_s8 (__ARM_mve_coerce(__p0,
> int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2,
> int8x16_t), p3), \
> -  int
> (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve
> _type_int16x8_t]: __arm_vqdmulhq_m_s16 (__ARM_mve_coerce(__p0,
> int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2,
> int16x8_t), p3), \
> -  int
> (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve
> _type_int32x4_t]: __arm_vqdmulhq_m_s32 (__ARM_mve_coerce(__p0,
> int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2,
> int32x4_t), p3));})
> -
>  #define __arm_vqdmullbq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
>    __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
> @@ -37562,16 +34369,6 @@ extern void *__ARM_undef;
>    int (*)[__ARM_mve_type_uint8x16_t]: __arm_vmovltq_x_u8
> (__ARM_mve_coerce(__p1, uint8x16_t), p2), \
>    int (*)[__ARM_mve_type_uint16x8_t]: __arm_vmovltq_x_u16
> (__ARM_mve_coerce(__p1, uint16x8_t), p2));})
> 
> -#define __arm_vmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
> -  __typeof(p2) __p2 = (p2); \
> -  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> -  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]:
> __arm_vmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t),
> __ARM_mve_coerce(__p2, int8x16_t), p3), \
> -  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]:
> __arm_vmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t),
> __ARM_mve_coerce(__p2, int16x8_t), p3), \
> -  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]:
> __arm_vmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t),
> __ARM_mve_coerce(__p2, int32x4_t), p3), \
> -  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]:
> __arm_vmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t),
> __ARM_mve_coerce(__p2, uint8x16_t), p3), \
> -  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]:
> __arm_vmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t),
> __ARM_mve_coerce(__p2, uint16x8_t), p3), \
> -  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]:
> __arm_vmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t),
> __ARM_mve_coerce(__p2, uint32x4_t), p3));})
> -
>  #define __arm_vmullbq_int_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
>    __typeof(p2) __p2 = (p2); \
>    _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0,
> \
> --
> 2.34.1
  

Patch

diff --git a/gcc/config/arm/arm-mve-builtins-base.cc b/gcc/config/arm/arm-mve-builtins-base.cc
index 9722c861faf..668f1fe9cda 100644
--- a/gcc/config/arm/arm-mve-builtins-base.cc
+++ b/gcc/config/arm/arm-mve-builtins-base.cc
@@ -115,13 +115,48 @@  namespace arm_mve {
     -1, -1, -1,								\
     -1, -1, -1))
 
+  /* Helper for builtins with only unspec codes, _m predicated and _n
+     overrides, but no floating-point version.  */
+#define FUNCTION_WITH_M_N_NO_F(NAME, UNSPEC) FUNCTION			\
+  (NAME, unspec_mve_function_exact_insn,				\
+   (UNSPEC##_S, UNSPEC##_U, -1,						\
+    UNSPEC##_N_S, UNSPEC##_N_U, -1,					\
+    UNSPEC##_M_S, UNSPEC##_M_U, -1,					\
+    UNSPEC##_M_N_S, UNSPEC##_M_N_U, -1))
+
+  /* Helper for builtins with only unspec codes, _m predicated
+     overrides, no _n and no floating-point version.  */
+#define FUNCTION_WITHOUT_N_NO_F(NAME, UNSPEC) FUNCTION			\
+  (NAME, unspec_mve_function_exact_insn,				\
+   (UNSPEC##_S, UNSPEC##_U, -1,						\
+    -1, -1, -1,								\
+    UNSPEC##_M_S, UNSPEC##_M_U, -1,					\
+    -1, -1, -1))
+
+  /* Helper for builtins with only unspec codes, _m predicated and _n
+     overrides, but no unsigned and floating-point versions.  */
+#define FUNCTION_WITH_M_N_NO_U_F(NAME, UNSPEC) FUNCTION			\
+  (NAME, unspec_mve_function_exact_insn,				\
+   (UNSPEC##_S, -1, -1,							\
+    UNSPEC##_N_S, -1, -1,						\
+    UNSPEC##_M_S, -1, -1,						\
+    UNSPEC##_M_N_S, -1, -1))
+
 FUNCTION_WITH_RTX_M_N (vaddq, PLUS, VADDQ)
 FUNCTION_WITH_RTX_M (vandq, AND, VANDQ)
 FUNCTION_WITHOUT_M_N (vcreateq, VCREATEQ)
 FUNCTION_WITH_RTX_M (veorq, XOR, VEORQ)
+FUNCTION_WITH_M_N_NO_F (vhaddq, VHADDQ)
+FUNCTION_WITH_M_N_NO_F (vhsubq, VHSUBQ)
+FUNCTION_WITHOUT_N_NO_F (vmulhq, VMULHQ)
 FUNCTION_WITH_RTX_M_N (vmulq, MULT, VMULQ)
 FUNCTION_WITH_RTX_M_N_NO_N_F (vorrq, IOR, VORRQ)
+FUNCTION_WITH_M_N_NO_F (vqaddq, VQADDQ)
+FUNCTION_WITH_M_N_NO_U_F (vqdmulhq, VQDMULHQ)
+FUNCTION_WITH_M_N_NO_F (vqsubq, VQSUBQ)
 FUNCTION (vreinterpretq, vreinterpretq_impl,)
+FUNCTION_WITHOUT_N_NO_F (vrhaddq, VRHADDQ)
+FUNCTION_WITHOUT_N_NO_F (vrmulhq, VRMULHQ)
 FUNCTION_WITH_RTX_M_N (vsubq, MINUS, VSUBQ)
 FUNCTION (vuninitializedq, vuninitializedq_impl,)
 
diff --git a/gcc/config/arm/arm-mve-builtins-base.def b/gcc/config/arm/arm-mve-builtins-base.def
index 1bfd15f973c..d256f3ebb2d 100644
--- a/gcc/config/arm/arm-mve-builtins-base.def
+++ b/gcc/config/arm/arm-mve-builtins-base.def
@@ -22,9 +22,17 @@  DEF_MVE_FUNCTION (vaddq, binary_opt_n, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vandq, binary, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vcreateq, create, all_integer_with_64, none)
 DEF_MVE_FUNCTION (veorq, binary, all_integer, mx_or_none)
+DEF_MVE_FUNCTION (vhaddq, binary_opt_n, all_integer, mx_or_none)
+DEF_MVE_FUNCTION (vhsubq, binary_opt_n, all_integer, mx_or_none)
+DEF_MVE_FUNCTION (vmulhq, binary, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vmulq, binary_opt_n, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vorrq, binary_orrq, all_integer, mx_or_none)
+DEF_MVE_FUNCTION (vqaddq, binary_opt_n, all_integer, m_or_none)
+DEF_MVE_FUNCTION (vqdmulhq, binary_opt_n, all_signed, m_or_none)
+DEF_MVE_FUNCTION (vqsubq, binary_opt_n, all_integer, m_or_none)
 DEF_MVE_FUNCTION (vreinterpretq, unary_convert, reinterpret_integer, none)
+DEF_MVE_FUNCTION (vrhaddq, binary, all_integer, mx_or_none)
+DEF_MVE_FUNCTION (vrmulhq, binary, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vsubq, binary_opt_n, all_integer, mx_or_none)
 DEF_MVE_FUNCTION (vuninitializedq, inherent, all_integer_with_64, none)
 #undef REQUIRES_FLOAT
diff --git a/gcc/config/arm/arm-mve-builtins-base.h b/gcc/config/arm/arm-mve-builtins-base.h
index 8dd6bff01bf..d64cb5e1dec 100644
--- a/gcc/config/arm/arm-mve-builtins-base.h
+++ b/gcc/config/arm/arm-mve-builtins-base.h
@@ -27,9 +27,17 @@  extern const function_base *const vaddq;
 extern const function_base *const vandq;
 extern const function_base *const vcreateq;
 extern const function_base *const veorq;
+extern const function_base *const vhaddq;
+extern const function_base *const vhsubq;
+extern const function_base *const vmulhq;
 extern const function_base *const vmulq;
 extern const function_base *const vorrq;
+extern const function_base *const vqaddq;
+extern const function_base *const vqdmulhq;
+extern const function_base *const vqsubq;
 extern const function_base *const vreinterpretq;
+extern const function_base *const vrhaddq;
+extern const function_base *const vrmulhq;
 extern const function_base *const vsubq;
 extern const function_base *const vuninitializedq;
 
diff --git a/gcc/config/arm/arm_mve.h b/gcc/config/arm/arm_mve.h
index 4810e2977d3..9c5d14794a1 100644
--- a/gcc/config/arm/arm_mve.h
+++ b/gcc/config/arm/arm_mve.h
@@ -61,21 +61,14 @@ 
 #define vaddlvq_p(__a, __p) __arm_vaddlvq_p(__a, __p)
 #define vcmpneq(__a, __b) __arm_vcmpneq(__a, __b)
 #define vshlq(__a, __b) __arm_vshlq(__a, __b)
-#define vrmulhq(__a, __b) __arm_vrmulhq(__a, __b)
-#define vrhaddq(__a, __b) __arm_vrhaddq(__a, __b)
-#define vqsubq(__a, __b) __arm_vqsubq(__a, __b)
-#define vqaddq(__a, __b) __arm_vqaddq(__a, __b)
 #define vornq(__a, __b) __arm_vornq(__a, __b)
 #define vmulltq_int(__a, __b) __arm_vmulltq_int(__a, __b)
 #define vmullbq_int(__a, __b) __arm_vmullbq_int(__a, __b)
-#define vmulhq(__a, __b) __arm_vmulhq(__a, __b)
 #define vmladavq(__a, __b) __arm_vmladavq(__a, __b)
 #define vminvq(__a, __b) __arm_vminvq(__a, __b)
 #define vminq(__a, __b) __arm_vminq(__a, __b)
 #define vmaxvq(__a, __b) __arm_vmaxvq(__a, __b)
 #define vmaxq(__a, __b) __arm_vmaxq(__a, __b)
-#define vhsubq(__a, __b) __arm_vhsubq(__a, __b)
-#define vhaddq(__a, __b) __arm_vhaddq(__a, __b)
 #define vcmphiq(__a, __b) __arm_vcmphiq(__a, __b)
 #define vcmpeqq(__a, __b) __arm_vcmpeqq(__a, __b)
 #define vcmpcsq(__a, __b) __arm_vcmpcsq(__a, __b)
@@ -104,7 +97,6 @@ 
 #define vcmpgeq(__a, __b) __arm_vcmpgeq(__a, __b)
 #define vqshluq(__a, __imm) __arm_vqshluq(__a, __imm)
 #define vqrdmulhq(__a, __b) __arm_vqrdmulhq(__a, __b)
-#define vqdmulhq(__a, __b) __arm_vqdmulhq(__a, __b)
 #define vmlsdavxq(__a, __b) __arm_vmlsdavxq(__a, __b)
 #define vmlsdavq(__a, __b) __arm_vmlsdavq(__a, __b)
 #define vmladavxq(__a, __b) __arm_vmladavxq(__a, __b)
@@ -236,10 +228,8 @@ 
 #define vbrsrq_m(__inactive, __a, __b, __p) __arm_vbrsrq_m(__inactive, __a, __b, __p)
 #define vcaddq_rot270_m(__inactive, __a, __b, __p) __arm_vcaddq_rot270_m(__inactive, __a, __b, __p)
 #define vcaddq_rot90_m(__inactive, __a, __b, __p) __arm_vcaddq_rot90_m(__inactive, __a, __b, __p)
-#define vhaddq_m(__inactive, __a, __b, __p) __arm_vhaddq_m(__inactive, __a, __b, __p)
 #define vhcaddq_rot270_m(__inactive, __a, __b, __p) __arm_vhcaddq_rot270_m(__inactive, __a, __b, __p)
 #define vhcaddq_rot90_m(__inactive, __a, __b, __p) __arm_vhcaddq_rot90_m(__inactive, __a, __b, __p)
-#define vhsubq_m(__inactive, __a, __b, __p) __arm_vhsubq_m(__inactive, __a, __b, __p)
 #define vmaxq_m(__inactive, __a, __b, __p) __arm_vmaxq_m(__inactive, __a, __b, __p)
 #define vminq_m(__inactive, __a, __b, __p) __arm_vminq_m(__inactive, __a, __b, __p)
 #define vmladavaq_p(__a, __b, __c, __p) __arm_vmladavaq_p(__a, __b, __c, __p)
@@ -248,18 +238,15 @@ 
 #define vmlasq_m(__a, __b, __c, __p) __arm_vmlasq_m(__a, __b, __c, __p)
 #define vmlsdavaq_p(__a, __b, __c, __p) __arm_vmlsdavaq_p(__a, __b, __c, __p)
 #define vmlsdavaxq_p(__a, __b, __c, __p) __arm_vmlsdavaxq_p(__a, __b, __c, __p)
-#define vmulhq_m(__inactive, __a, __b, __p) __arm_vmulhq_m(__inactive, __a, __b, __p)
 #define vmullbq_int_m(__inactive, __a, __b, __p) __arm_vmullbq_int_m(__inactive, __a, __b, __p)
 #define vmulltq_int_m(__inactive, __a, __b, __p) __arm_vmulltq_int_m(__inactive, __a, __b, __p)
 #define vornq_m(__inactive, __a, __b, __p) __arm_vornq_m(__inactive, __a, __b, __p)
-#define vqaddq_m(__inactive, __a, __b, __p) __arm_vqaddq_m(__inactive, __a, __b, __p)
 #define vqdmladhq_m(__inactive, __a, __b, __p) __arm_vqdmladhq_m(__inactive, __a, __b, __p)
 #define vqdmlashq_m(__a, __b, __c, __p) __arm_vqdmlashq_m(__a, __b, __c, __p)
 #define vqdmladhxq_m(__inactive, __a, __b, __p) __arm_vqdmladhxq_m(__inactive, __a, __b, __p)
 #define vqdmlahq_m(__a, __b, __c, __p) __arm_vqdmlahq_m(__a, __b, __c, __p)
 #define vqdmlsdhq_m(__inactive, __a, __b, __p) __arm_vqdmlsdhq_m(__inactive, __a, __b, __p)
 #define vqdmlsdhxq_m(__inactive, __a, __b, __p) __arm_vqdmlsdhxq_m(__inactive, __a, __b, __p)
-#define vqdmulhq_m(__inactive, __a, __b, __p) __arm_vqdmulhq_m(__inactive, __a, __b, __p)
 #define vqrdmladhq_m(__inactive, __a, __b, __p) __arm_vqrdmladhq_m(__inactive, __a, __b, __p)
 #define vqrdmladhxq_m(__inactive, __a, __b, __p) __arm_vqrdmladhxq_m(__inactive, __a, __b, __p)
 #define vqrdmlahq_m(__a, __b, __c, __p) __arm_vqrdmlahq_m(__a, __b, __c, __p)
@@ -270,9 +257,6 @@ 
 #define vqrshlq_m(__inactive, __a, __b, __p) __arm_vqrshlq_m(__inactive, __a, __b, __p)
 #define vqshlq_m_n(__inactive, __a, __imm, __p) __arm_vqshlq_m_n(__inactive, __a, __imm, __p)
 #define vqshlq_m(__inactive, __a, __b, __p) __arm_vqshlq_m(__inactive, __a, __b, __p)
-#define vqsubq_m(__inactive, __a, __b, __p) __arm_vqsubq_m(__inactive, __a, __b, __p)
-#define vrhaddq_m(__inactive, __a, __b, __p) __arm_vrhaddq_m(__inactive, __a, __b, __p)
-#define vrmulhq_m(__inactive, __a, __b, __p) __arm_vrmulhq_m(__inactive, __a, __b, __p)
 #define vrshlq_m(__inactive, __a, __b, __p) __arm_vrshlq_m(__inactive, __a, __b, __p)
 #define vrshrq_m(__inactive, __a, __imm, __p) __arm_vrshrq_m(__inactive, __a, __imm, __p)
 #define vshlq_m_n(__inactive, __a, __imm, __p) __arm_vshlq_m_n(__inactive, __a, __imm, __p)
@@ -384,19 +368,14 @@ 
 #define vclsq_x(__a, __p) __arm_vclsq_x(__a, __p)
 #define vclzq_x(__a, __p) __arm_vclzq_x(__a, __p)
 #define vnegq_x(__a, __p) __arm_vnegq_x(__a, __p)
-#define vmulhq_x(__a, __b, __p) __arm_vmulhq_x(__a, __b, __p)
 #define vmullbq_poly_x(__a, __b, __p) __arm_vmullbq_poly_x(__a, __b, __p)
 #define vmullbq_int_x(__a, __b, __p) __arm_vmullbq_int_x(__a, __b, __p)
 #define vmulltq_poly_x(__a, __b, __p) __arm_vmulltq_poly_x(__a, __b, __p)
 #define vmulltq_int_x(__a, __b, __p) __arm_vmulltq_int_x(__a, __b, __p)
 #define vcaddq_rot90_x(__a, __b, __p) __arm_vcaddq_rot90_x(__a, __b, __p)
 #define vcaddq_rot270_x(__a, __b, __p) __arm_vcaddq_rot270_x(__a, __b, __p)
-#define vhaddq_x(__a, __b, __p) __arm_vhaddq_x(__a, __b, __p)
 #define vhcaddq_rot90_x(__a, __b, __p) __arm_vhcaddq_rot90_x(__a, __b, __p)
 #define vhcaddq_rot270_x(__a, __b, __p) __arm_vhcaddq_rot270_x(__a, __b, __p)
-#define vhsubq_x(__a, __b, __p) __arm_vhsubq_x(__a, __b, __p)
-#define vrhaddq_x(__a, __b, __p) __arm_vrhaddq_x(__a, __b, __p)
-#define vrmulhq_x(__a, __b, __p) __arm_vrmulhq_x(__a, __b, __p)
 #define vbicq_x(__a, __b, __p) __arm_vbicq_x(__a, __b, __p)
 #define vbrsrq_x(__a, __b, __p) __arm_vbrsrq_x(__a, __b, __p)
 #define vmovlbq_x(__a, __p) __arm_vmovlbq_x(__a, __p)
@@ -662,25 +641,14 @@ 
 #define vshlq_u8(__a, __b) __arm_vshlq_u8(__a, __b)
 #define vshlq_u16(__a, __b) __arm_vshlq_u16(__a, __b)
 #define vshlq_u32(__a, __b) __arm_vshlq_u32(__a, __b)
-#define vrmulhq_u8(__a, __b) __arm_vrmulhq_u8(__a, __b)
-#define vrhaddq_u8(__a, __b) __arm_vrhaddq_u8(__a, __b)
-#define vqsubq_u8(__a, __b) __arm_vqsubq_u8(__a, __b)
-#define vqsubq_n_u8(__a, __b) __arm_vqsubq_n_u8(__a, __b)
-#define vqaddq_u8(__a, __b) __arm_vqaddq_u8(__a, __b)
-#define vqaddq_n_u8(__a, __b) __arm_vqaddq_n_u8(__a, __b)
 #define vornq_u8(__a, __b) __arm_vornq_u8(__a, __b)
 #define vmulltq_int_u8(__a, __b) __arm_vmulltq_int_u8(__a, __b)
 #define vmullbq_int_u8(__a, __b) __arm_vmullbq_int_u8(__a, __b)
-#define vmulhq_u8(__a, __b) __arm_vmulhq_u8(__a, __b)
 #define vmladavq_u8(__a, __b) __arm_vmladavq_u8(__a, __b)
 #define vminvq_u8(__a, __b) __arm_vminvq_u8(__a, __b)
 #define vminq_u8(__a, __b) __arm_vminq_u8(__a, __b)
 #define vmaxvq_u8(__a, __b) __arm_vmaxvq_u8(__a, __b)
 #define vmaxq_u8(__a, __b) __arm_vmaxq_u8(__a, __b)
-#define vhsubq_u8(__a, __b) __arm_vhsubq_u8(__a, __b)
-#define vhsubq_n_u8(__a, __b) __arm_vhsubq_n_u8(__a, __b)
-#define vhaddq_u8(__a, __b) __arm_vhaddq_u8(__a, __b)
-#define vhaddq_n_u8(__a, __b) __arm_vhaddq_n_u8(__a, __b)
 #define vcmpneq_n_u8(__a, __b) __arm_vcmpneq_n_u8(__a, __b)
 #define vcmphiq_u8(__a, __b) __arm_vcmphiq_u8(__a, __b)
 #define vcmphiq_n_u8(__a, __b) __arm_vcmphiq_n_u8(__a, __b)
@@ -725,24 +693,15 @@ 
 #define vshlq_r_s8(__a, __b) __arm_vshlq_r_s8(__a, __b)
 #define vrshlq_s8(__a, __b) __arm_vrshlq_s8(__a, __b)
 #define vrshlq_n_s8(__a, __b) __arm_vrshlq_n_s8(__a, __b)
-#define vrmulhq_s8(__a, __b) __arm_vrmulhq_s8(__a, __b)
-#define vrhaddq_s8(__a, __b) __arm_vrhaddq_s8(__a, __b)
-#define vqsubq_s8(__a, __b) __arm_vqsubq_s8(__a, __b)
-#define vqsubq_n_s8(__a, __b) __arm_vqsubq_n_s8(__a, __b)
 #define vqshlq_s8(__a, __b) __arm_vqshlq_s8(__a, __b)
 #define vqshlq_r_s8(__a, __b) __arm_vqshlq_r_s8(__a, __b)
 #define vqrshlq_s8(__a, __b) __arm_vqrshlq_s8(__a, __b)
 #define vqrshlq_n_s8(__a, __b) __arm_vqrshlq_n_s8(__a, __b)
 #define vqrdmulhq_s8(__a, __b) __arm_vqrdmulhq_s8(__a, __b)
 #define vqrdmulhq_n_s8(__a, __b) __arm_vqrdmulhq_n_s8(__a, __b)
-#define vqdmulhq_s8(__a, __b) __arm_vqdmulhq_s8(__a, __b)
-#define vqdmulhq_n_s8(__a, __b) __arm_vqdmulhq_n_s8(__a, __b)
-#define vqaddq_s8(__a, __b) __arm_vqaddq_s8(__a, __b)
-#define vqaddq_n_s8(__a, __b) __arm_vqaddq_n_s8(__a, __b)
 #define vornq_s8(__a, __b) __arm_vornq_s8(__a, __b)
 #define vmulltq_int_s8(__a, __b) __arm_vmulltq_int_s8(__a, __b)
 #define vmullbq_int_s8(__a, __b) __arm_vmullbq_int_s8(__a, __b)
-#define vmulhq_s8(__a, __b) __arm_vmulhq_s8(__a, __b)
 #define vmlsdavxq_s8(__a, __b) __arm_vmlsdavxq_s8(__a, __b)
 #define vmlsdavq_s8(__a, __b) __arm_vmlsdavq_s8(__a, __b)
 #define vmladavxq_s8(__a, __b) __arm_vmladavxq_s8(__a, __b)
@@ -751,12 +710,8 @@ 
 #define vminq_s8(__a, __b) __arm_vminq_s8(__a, __b)
 #define vmaxvq_s8(__a, __b) __arm_vmaxvq_s8(__a, __b)
 #define vmaxq_s8(__a, __b) __arm_vmaxq_s8(__a, __b)
-#define vhsubq_s8(__a, __b) __arm_vhsubq_s8(__a, __b)
-#define vhsubq_n_s8(__a, __b) __arm_vhsubq_n_s8(__a, __b)
 #define vhcaddq_rot90_s8(__a, __b) __arm_vhcaddq_rot90_s8(__a, __b)
 #define vhcaddq_rot270_s8(__a, __b) __arm_vhcaddq_rot270_s8(__a, __b)
-#define vhaddq_s8(__a, __b) __arm_vhaddq_s8(__a, __b)
-#define vhaddq_n_s8(__a, __b) __arm_vhaddq_n_s8(__a, __b)
 #define vcaddq_rot90_s8(__a, __b) __arm_vcaddq_rot90_s8(__a, __b)
 #define vcaddq_rot270_s8(__a, __b) __arm_vcaddq_rot270_s8(__a, __b)
 #define vbrsrq_n_s8(__a, __b) __arm_vbrsrq_n_s8(__a, __b)
@@ -766,25 +721,14 @@ 
 #define vshlq_n_s8(__a,  __imm) __arm_vshlq_n_s8(__a,  __imm)
 #define vrshrq_n_s8(__a,  __imm) __arm_vrshrq_n_s8(__a,  __imm)
 #define vqshlq_n_s8(__a,  __imm) __arm_vqshlq_n_s8(__a,  __imm)
-#define vrmulhq_u16(__a, __b) __arm_vrmulhq_u16(__a, __b)
-#define vrhaddq_u16(__a, __b) __arm_vrhaddq_u16(__a, __b)
-#define vqsubq_u16(__a, __b) __arm_vqsubq_u16(__a, __b)
-#define vqsubq_n_u16(__a, __b) __arm_vqsubq_n_u16(__a, __b)
-#define vqaddq_u16(__a, __b) __arm_vqaddq_u16(__a, __b)
-#define vqaddq_n_u16(__a, __b) __arm_vqaddq_n_u16(__a, __b)
 #define vornq_u16(__a, __b) __arm_vornq_u16(__a, __b)
 #define vmulltq_int_u16(__a, __b) __arm_vmulltq_int_u16(__a, __b)
 #define vmullbq_int_u16(__a, __b) __arm_vmullbq_int_u16(__a, __b)
-#define vmulhq_u16(__a, __b) __arm_vmulhq_u16(__a, __b)
 #define vmladavq_u16(__a, __b) __arm_vmladavq_u16(__a, __b)
 #define vminvq_u16(__a, __b) __arm_vminvq_u16(__a, __b)
 #define vminq_u16(__a, __b) __arm_vminq_u16(__a, __b)
 #define vmaxvq_u16(__a, __b) __arm_vmaxvq_u16(__a, __b)
 #define vmaxq_u16(__a, __b) __arm_vmaxq_u16(__a, __b)
-#define vhsubq_u16(__a, __b) __arm_vhsubq_u16(__a, __b)
-#define vhsubq_n_u16(__a, __b) __arm_vhsubq_n_u16(__a, __b)
-#define vhaddq_u16(__a, __b) __arm_vhaddq_u16(__a, __b)
-#define vhaddq_n_u16(__a, __b) __arm_vhaddq_n_u16(__a, __b)
 #define vcmpneq_n_u16(__a, __b) __arm_vcmpneq_n_u16(__a, __b)
 #define vcmphiq_u16(__a, __b) __arm_vcmphiq_u16(__a, __b)
 #define vcmphiq_n_u16(__a, __b) __arm_vcmphiq_n_u16(__a, __b)
@@ -829,24 +773,15 @@ 
 #define vshlq_r_s16(__a, __b) __arm_vshlq_r_s16(__a, __b)
 #define vrshlq_s16(__a, __b) __arm_vrshlq_s16(__a, __b)
 #define vrshlq_n_s16(__a, __b) __arm_vrshlq_n_s16(__a, __b)
-#define vrmulhq_s16(__a, __b) __arm_vrmulhq_s16(__a, __b)
-#define vrhaddq_s16(__a, __b) __arm_vrhaddq_s16(__a, __b)
-#define vqsubq_s16(__a, __b) __arm_vqsubq_s16(__a, __b)
-#define vqsubq_n_s16(__a, __b) __arm_vqsubq_n_s16(__a, __b)
 #define vqshlq_s16(__a, __b) __arm_vqshlq_s16(__a, __b)
 #define vqshlq_r_s16(__a, __b) __arm_vqshlq_r_s16(__a, __b)
 #define vqrshlq_s16(__a, __b) __arm_vqrshlq_s16(__a, __b)
 #define vqrshlq_n_s16(__a, __b) __arm_vqrshlq_n_s16(__a, __b)
 #define vqrdmulhq_s16(__a, __b) __arm_vqrdmulhq_s16(__a, __b)
 #define vqrdmulhq_n_s16(__a, __b) __arm_vqrdmulhq_n_s16(__a, __b)
-#define vqdmulhq_s16(__a, __b) __arm_vqdmulhq_s16(__a, __b)
-#define vqdmulhq_n_s16(__a, __b) __arm_vqdmulhq_n_s16(__a, __b)
-#define vqaddq_s16(__a, __b) __arm_vqaddq_s16(__a, __b)
-#define vqaddq_n_s16(__a, __b) __arm_vqaddq_n_s16(__a, __b)
 #define vornq_s16(__a, __b) __arm_vornq_s16(__a, __b)
 #define vmulltq_int_s16(__a, __b) __arm_vmulltq_int_s16(__a, __b)
 #define vmullbq_int_s16(__a, __b) __arm_vmullbq_int_s16(__a, __b)
-#define vmulhq_s16(__a, __b) __arm_vmulhq_s16(__a, __b)
 #define vmlsdavxq_s16(__a, __b) __arm_vmlsdavxq_s16(__a, __b)
 #define vmlsdavq_s16(__a, __b) __arm_vmlsdavq_s16(__a, __b)
 #define vmladavxq_s16(__a, __b) __arm_vmladavxq_s16(__a, __b)
@@ -855,12 +790,8 @@ 
 #define vminq_s16(__a, __b) __arm_vminq_s16(__a, __b)
 #define vmaxvq_s16(__a, __b) __arm_vmaxvq_s16(__a, __b)
 #define vmaxq_s16(__a, __b) __arm_vmaxq_s16(__a, __b)
-#define vhsubq_s16(__a, __b) __arm_vhsubq_s16(__a, __b)
-#define vhsubq_n_s16(__a, __b) __arm_vhsubq_n_s16(__a, __b)
 #define vhcaddq_rot90_s16(__a, __b) __arm_vhcaddq_rot90_s16(__a, __b)
 #define vhcaddq_rot270_s16(__a, __b) __arm_vhcaddq_rot270_s16(__a, __b)
-#define vhaddq_s16(__a, __b) __arm_vhaddq_s16(__a, __b)
-#define vhaddq_n_s16(__a, __b) __arm_vhaddq_n_s16(__a, __b)
 #define vcaddq_rot90_s16(__a, __b) __arm_vcaddq_rot90_s16(__a, __b)
 #define vcaddq_rot270_s16(__a, __b) __arm_vcaddq_rot270_s16(__a, __b)
 #define vbrsrq_n_s16(__a, __b) __arm_vbrsrq_n_s16(__a, __b)
@@ -870,25 +801,14 @@ 
 #define vshlq_n_s16(__a,  __imm) __arm_vshlq_n_s16(__a,  __imm)
 #define vrshrq_n_s16(__a,  __imm) __arm_vrshrq_n_s16(__a,  __imm)
 #define vqshlq_n_s16(__a,  __imm) __arm_vqshlq_n_s16(__a,  __imm)
-#define vrmulhq_u32(__a, __b) __arm_vrmulhq_u32(__a, __b)
-#define vrhaddq_u32(__a, __b) __arm_vrhaddq_u32(__a, __b)
-#define vqsubq_u32(__a, __b) __arm_vqsubq_u32(__a, __b)
-#define vqsubq_n_u32(__a, __b) __arm_vqsubq_n_u32(__a, __b)
-#define vqaddq_u32(__a, __b) __arm_vqaddq_u32(__a, __b)
-#define vqaddq_n_u32(__a, __b) __arm_vqaddq_n_u32(__a, __b)
 #define vornq_u32(__a, __b) __arm_vornq_u32(__a, __b)
 #define vmulltq_int_u32(__a, __b) __arm_vmulltq_int_u32(__a, __b)
 #define vmullbq_int_u32(__a, __b) __arm_vmullbq_int_u32(__a, __b)
-#define vmulhq_u32(__a, __b) __arm_vmulhq_u32(__a, __b)
 #define vmladavq_u32(__a, __b) __arm_vmladavq_u32(__a, __b)
 #define vminvq_u32(__a, __b) __arm_vminvq_u32(__a, __b)
 #define vminq_u32(__a, __b) __arm_vminq_u32(__a, __b)
 #define vmaxvq_u32(__a, __b) __arm_vmaxvq_u32(__a, __b)
 #define vmaxq_u32(__a, __b) __arm_vmaxq_u32(__a, __b)
-#define vhsubq_u32(__a, __b) __arm_vhsubq_u32(__a, __b)
-#define vhsubq_n_u32(__a, __b) __arm_vhsubq_n_u32(__a, __b)
-#define vhaddq_u32(__a, __b) __arm_vhaddq_u32(__a, __b)
-#define vhaddq_n_u32(__a, __b) __arm_vhaddq_n_u32(__a, __b)
 #define vcmpneq_n_u32(__a, __b) __arm_vcmpneq_n_u32(__a, __b)
 #define vcmphiq_u32(__a, __b) __arm_vcmphiq_u32(__a, __b)
 #define vcmphiq_n_u32(__a, __b) __arm_vcmphiq_n_u32(__a, __b)
@@ -933,24 +853,15 @@ 
 #define vshlq_r_s32(__a, __b) __arm_vshlq_r_s32(__a, __b)
 #define vrshlq_s32(__a, __b) __arm_vrshlq_s32(__a, __b)
 #define vrshlq_n_s32(__a, __b) __arm_vrshlq_n_s32(__a, __b)
-#define vrmulhq_s32(__a, __b) __arm_vrmulhq_s32(__a, __b)
-#define vrhaddq_s32(__a, __b) __arm_vrhaddq_s32(__a, __b)
-#define vqsubq_s32(__a, __b) __arm_vqsubq_s32(__a, __b)
-#define vqsubq_n_s32(__a, __b) __arm_vqsubq_n_s32(__a, __b)
 #define vqshlq_s32(__a, __b) __arm_vqshlq_s32(__a, __b)
 #define vqshlq_r_s32(__a, __b) __arm_vqshlq_r_s32(__a, __b)
 #define vqrshlq_s32(__a, __b) __arm_vqrshlq_s32(__a, __b)
 #define vqrshlq_n_s32(__a, __b) __arm_vqrshlq_n_s32(__a, __b)
 #define vqrdmulhq_s32(__a, __b) __arm_vqrdmulhq_s32(__a, __b)
 #define vqrdmulhq_n_s32(__a, __b) __arm_vqrdmulhq_n_s32(__a, __b)
-#define vqdmulhq_s32(__a, __b) __arm_vqdmulhq_s32(__a, __b)
-#define vqdmulhq_n_s32(__a, __b) __arm_vqdmulhq_n_s32(__a, __b)
-#define vqaddq_s32(__a, __b) __arm_vqaddq_s32(__a, __b)
-#define vqaddq_n_s32(__a, __b) __arm_vqaddq_n_s32(__a, __b)
 #define vornq_s32(__a, __b) __arm_vornq_s32(__a, __b)
 #define vmulltq_int_s32(__a, __b) __arm_vmulltq_int_s32(__a, __b)
 #define vmullbq_int_s32(__a, __b) __arm_vmullbq_int_s32(__a, __b)
-#define vmulhq_s32(__a, __b) __arm_vmulhq_s32(__a, __b)
 #define vmlsdavxq_s32(__a, __b) __arm_vmlsdavxq_s32(__a, __b)
 #define vmlsdavq_s32(__a, __b) __arm_vmlsdavq_s32(__a, __b)
 #define vmladavxq_s32(__a, __b) __arm_vmladavxq_s32(__a, __b)
@@ -959,12 +870,8 @@ 
 #define vminq_s32(__a, __b) __arm_vminq_s32(__a, __b)
 #define vmaxvq_s32(__a, __b) __arm_vmaxvq_s32(__a, __b)
 #define vmaxq_s32(__a, __b) __arm_vmaxq_s32(__a, __b)
-#define vhsubq_s32(__a, __b) __arm_vhsubq_s32(__a, __b)
-#define vhsubq_n_s32(__a, __b) __arm_vhsubq_n_s32(__a, __b)
 #define vhcaddq_rot90_s32(__a, __b) __arm_vhcaddq_rot90_s32(__a, __b)
 #define vhcaddq_rot270_s32(__a, __b) __arm_vhcaddq_rot270_s32(__a, __b)
-#define vhaddq_s32(__a, __b) __arm_vhaddq_s32(__a, __b)
-#define vhaddq_n_s32(__a, __b) __arm_vhaddq_n_s32(__a, __b)
 #define vcaddq_rot90_s32(__a, __b) __arm_vcaddq_rot90_s32(__a, __b)
 #define vcaddq_rot270_s32(__a, __b) __arm_vcaddq_rot270_s32(__a, __b)
 #define vbrsrq_n_s32(__a, __b) __arm_vbrsrq_n_s32(__a, __b)
@@ -1634,36 +1541,12 @@ 
 #define vcaddq_rot90_m_u8(__inactive, __a, __b, __p) __arm_vcaddq_rot90_m_u8(__inactive, __a, __b, __p)
 #define vcaddq_rot90_m_u32(__inactive, __a, __b, __p) __arm_vcaddq_rot90_m_u32(__inactive, __a, __b, __p)
 #define vcaddq_rot90_m_u16(__inactive, __a, __b, __p) __arm_vcaddq_rot90_m_u16(__inactive, __a, __b, __p)
-#define vhaddq_m_n_s8(__inactive, __a, __b, __p) __arm_vhaddq_m_n_s8(__inactive, __a, __b, __p)
-#define vhaddq_m_n_s32(__inactive, __a, __b, __p) __arm_vhaddq_m_n_s32(__inactive, __a, __b, __p)
-#define vhaddq_m_n_s16(__inactive, __a, __b, __p) __arm_vhaddq_m_n_s16(__inactive, __a, __b, __p)
-#define vhaddq_m_n_u8(__inactive, __a, __b, __p) __arm_vhaddq_m_n_u8(__inactive, __a, __b, __p)
-#define vhaddq_m_n_u32(__inactive, __a, __b, __p) __arm_vhaddq_m_n_u32(__inactive, __a, __b, __p)
-#define vhaddq_m_n_u16(__inactive, __a, __b, __p) __arm_vhaddq_m_n_u16(__inactive, __a, __b, __p)
-#define vhaddq_m_s8(__inactive, __a, __b, __p) __arm_vhaddq_m_s8(__inactive, __a, __b, __p)
-#define vhaddq_m_s32(__inactive, __a, __b, __p) __arm_vhaddq_m_s32(__inactive, __a, __b, __p)
-#define vhaddq_m_s16(__inactive, __a, __b, __p) __arm_vhaddq_m_s16(__inactive, __a, __b, __p)
-#define vhaddq_m_u8(__inactive, __a, __b, __p) __arm_vhaddq_m_u8(__inactive, __a, __b, __p)
-#define vhaddq_m_u32(__inactive, __a, __b, __p) __arm_vhaddq_m_u32(__inactive, __a, __b, __p)
-#define vhaddq_m_u16(__inactive, __a, __b, __p) __arm_vhaddq_m_u16(__inactive, __a, __b, __p)
 #define vhcaddq_rot270_m_s8(__inactive, __a, __b, __p) __arm_vhcaddq_rot270_m_s8(__inactive, __a, __b, __p)
 #define vhcaddq_rot270_m_s32(__inactive, __a, __b, __p) __arm_vhcaddq_rot270_m_s32(__inactive, __a, __b, __p)
 #define vhcaddq_rot270_m_s16(__inactive, __a, __b, __p) __arm_vhcaddq_rot270_m_s16(__inactive, __a, __b, __p)
 #define vhcaddq_rot90_m_s8(__inactive, __a, __b, __p) __arm_vhcaddq_rot90_m_s8(__inactive, __a, __b, __p)
 #define vhcaddq_rot90_m_s32(__inactive, __a, __b, __p) __arm_vhcaddq_rot90_m_s32(__inactive, __a, __b, __p)
 #define vhcaddq_rot90_m_s16(__inactive, __a, __b, __p) __arm_vhcaddq_rot90_m_s16(__inactive, __a, __b, __p)
-#define vhsubq_m_n_s8(__inactive, __a, __b, __p) __arm_vhsubq_m_n_s8(__inactive, __a, __b, __p)
-#define vhsubq_m_n_s32(__inactive, __a, __b, __p) __arm_vhsubq_m_n_s32(__inactive, __a, __b, __p)
-#define vhsubq_m_n_s16(__inactive, __a, __b, __p) __arm_vhsubq_m_n_s16(__inactive, __a, __b, __p)
-#define vhsubq_m_n_u8(__inactive, __a, __b, __p) __arm_vhsubq_m_n_u8(__inactive, __a, __b, __p)
-#define vhsubq_m_n_u32(__inactive, __a, __b, __p) __arm_vhsubq_m_n_u32(__inactive, __a, __b, __p)
-#define vhsubq_m_n_u16(__inactive, __a, __b, __p) __arm_vhsubq_m_n_u16(__inactive, __a, __b, __p)
-#define vhsubq_m_s8(__inactive, __a, __b, __p) __arm_vhsubq_m_s8(__inactive, __a, __b, __p)
-#define vhsubq_m_s32(__inactive, __a, __b, __p) __arm_vhsubq_m_s32(__inactive, __a, __b, __p)
-#define vhsubq_m_s16(__inactive, __a, __b, __p) __arm_vhsubq_m_s16(__inactive, __a, __b, __p)
-#define vhsubq_m_u8(__inactive, __a, __b, __p) __arm_vhsubq_m_u8(__inactive, __a, __b, __p)
-#define vhsubq_m_u32(__inactive, __a, __b, __p) __arm_vhsubq_m_u32(__inactive, __a, __b, __p)
-#define vhsubq_m_u16(__inactive, __a, __b, __p) __arm_vhsubq_m_u16(__inactive, __a, __b, __p)
 #define vmaxq_m_s8(__inactive, __a, __b, __p) __arm_vmaxq_m_s8(__inactive, __a, __b, __p)
 #define vmaxq_m_s32(__inactive, __a, __b, __p) __arm_vmaxq_m_s32(__inactive, __a, __b, __p)
 #define vmaxq_m_s16(__inactive, __a, __b, __p) __arm_vmaxq_m_s16(__inactive, __a, __b, __p)
@@ -1703,12 +1586,6 @@ 
 #define vmlsdavaxq_p_s8(__a, __b, __c, __p) __arm_vmlsdavaxq_p_s8(__a, __b, __c, __p)
 #define vmlsdavaxq_p_s32(__a, __b, __c, __p) __arm_vmlsdavaxq_p_s32(__a, __b, __c, __p)
 #define vmlsdavaxq_p_s16(__a, __b, __c, __p) __arm_vmlsdavaxq_p_s16(__a, __b, __c, __p)
-#define vmulhq_m_s8(__inactive, __a, __b, __p) __arm_vmulhq_m_s8(__inactive, __a, __b, __p)
-#define vmulhq_m_s32(__inactive, __a, __b, __p) __arm_vmulhq_m_s32(__inactive, __a, __b, __p)
-#define vmulhq_m_s16(__inactive, __a, __b, __p) __arm_vmulhq_m_s16(__inactive, __a, __b, __p)
-#define vmulhq_m_u8(__inactive, __a, __b, __p) __arm_vmulhq_m_u8(__inactive, __a, __b, __p)
-#define vmulhq_m_u32(__inactive, __a, __b, __p) __arm_vmulhq_m_u32(__inactive, __a, __b, __p)
-#define vmulhq_m_u16(__inactive, __a, __b, __p) __arm_vmulhq_m_u16(__inactive, __a, __b, __p)
 #define vmullbq_int_m_s8(__inactive, __a, __b, __p) __arm_vmullbq_int_m_s8(__inactive, __a, __b, __p)
 #define vmullbq_int_m_s32(__inactive, __a, __b, __p) __arm_vmullbq_int_m_s32(__inactive, __a, __b, __p)
 #define vmullbq_int_m_s16(__inactive, __a, __b, __p) __arm_vmullbq_int_m_s16(__inactive, __a, __b, __p)
@@ -1727,18 +1604,6 @@ 
 #define vornq_m_u8(__inactive, __a, __b, __p) __arm_vornq_m_u8(__inactive, __a, __b, __p)
 #define vornq_m_u32(__inactive, __a, __b, __p) __arm_vornq_m_u32(__inactive, __a, __b, __p)
 #define vornq_m_u16(__inactive, __a, __b, __p) __arm_vornq_m_u16(__inactive, __a, __b, __p)
-#define vqaddq_m_n_s8(__inactive, __a, __b, __p) __arm_vqaddq_m_n_s8(__inactive, __a, __b, __p)
-#define vqaddq_m_n_s32(__inactive, __a, __b, __p) __arm_vqaddq_m_n_s32(__inactive, __a, __b, __p)
-#define vqaddq_m_n_s16(__inactive, __a, __b, __p) __arm_vqaddq_m_n_s16(__inactive, __a, __b, __p)
-#define vqaddq_m_n_u8(__inactive, __a, __b, __p) __arm_vqaddq_m_n_u8(__inactive, __a, __b, __p)
-#define vqaddq_m_n_u32(__inactive, __a, __b, __p) __arm_vqaddq_m_n_u32(__inactive, __a, __b, __p)
-#define vqaddq_m_n_u16(__inactive, __a, __b, __p) __arm_vqaddq_m_n_u16(__inactive, __a, __b, __p)
-#define vqaddq_m_s8(__inactive, __a, __b, __p) __arm_vqaddq_m_s8(__inactive, __a, __b, __p)
-#define vqaddq_m_s32(__inactive, __a, __b, __p) __arm_vqaddq_m_s32(__inactive, __a, __b, __p)
-#define vqaddq_m_s16(__inactive, __a, __b, __p) __arm_vqaddq_m_s16(__inactive, __a, __b, __p)
-#define vqaddq_m_u8(__inactive, __a, __b, __p) __arm_vqaddq_m_u8(__inactive, __a, __b, __p)
-#define vqaddq_m_u32(__inactive, __a, __b, __p) __arm_vqaddq_m_u32(__inactive, __a, __b, __p)
-#define vqaddq_m_u16(__inactive, __a, __b, __p) __arm_vqaddq_m_u16(__inactive, __a, __b, __p)
 #define vqdmladhq_m_s8(__inactive, __a, __b, __p) __arm_vqdmladhq_m_s8(__inactive, __a, __b, __p)
 #define vqdmladhq_m_s32(__inactive, __a, __b, __p) __arm_vqdmladhq_m_s32(__inactive, __a, __b, __p)
 #define vqdmladhq_m_s16(__inactive, __a, __b, __p) __arm_vqdmladhq_m_s16(__inactive, __a, __b, __p)
@@ -1757,12 +1622,6 @@ 
 #define vqdmlsdhxq_m_s8(__inactive, __a, __b, __p) __arm_vqdmlsdhxq_m_s8(__inactive, __a, __b, __p)
 #define vqdmlsdhxq_m_s32(__inactive, __a, __b, __p) __arm_vqdmlsdhxq_m_s32(__inactive, __a, __b, __p)
 #define vqdmlsdhxq_m_s16(__inactive, __a, __b, __p) __arm_vqdmlsdhxq_m_s16(__inactive, __a, __b, __p)
-#define vqdmulhq_m_n_s8(__inactive, __a, __b, __p) __arm_vqdmulhq_m_n_s8(__inactive, __a, __b, __p)
-#define vqdmulhq_m_n_s32(__inactive, __a, __b, __p) __arm_vqdmulhq_m_n_s32(__inactive, __a, __b, __p)
-#define vqdmulhq_m_n_s16(__inactive, __a, __b, __p) __arm_vqdmulhq_m_n_s16(__inactive, __a, __b, __p)
-#define vqdmulhq_m_s8(__inactive, __a, __b, __p) __arm_vqdmulhq_m_s8(__inactive, __a, __b, __p)
-#define vqdmulhq_m_s32(__inactive, __a, __b, __p) __arm_vqdmulhq_m_s32(__inactive, __a, __b, __p)
-#define vqdmulhq_m_s16(__inactive, __a, __b, __p) __arm_vqdmulhq_m_s16(__inactive, __a, __b, __p)
 #define vqrdmladhq_m_s8(__inactive, __a, __b, __p) __arm_vqrdmladhq_m_s8(__inactive, __a, __b, __p)
 #define vqrdmladhq_m_s32(__inactive, __a, __b, __p) __arm_vqrdmladhq_m_s32(__inactive, __a, __b, __p)
 #define vqrdmladhq_m_s16(__inactive, __a, __b, __p) __arm_vqrdmladhq_m_s16(__inactive, __a, __b, __p)
@@ -1805,30 +1664,6 @@ 
 #define vqshlq_m_u8(__inactive, __a, __b, __p) __arm_vqshlq_m_u8(__inactive, __a, __b, __p)
 #define vqshlq_m_u32(__inactive, __a, __b, __p) __arm_vqshlq_m_u32(__inactive, __a, __b, __p)
 #define vqshlq_m_u16(__inactive, __a, __b, __p) __arm_vqshlq_m_u16(__inactive, __a, __b, __p)
-#define vqsubq_m_n_s8(__inactive, __a, __b, __p) __arm_vqsubq_m_n_s8(__inactive, __a, __b, __p)
-#define vqsubq_m_n_s32(__inactive, __a, __b, __p) __arm_vqsubq_m_n_s32(__inactive, __a, __b, __p)
-#define vqsubq_m_n_s16(__inactive, __a, __b, __p) __arm_vqsubq_m_n_s16(__inactive, __a, __b, __p)
-#define vqsubq_m_n_u8(__inactive, __a, __b, __p) __arm_vqsubq_m_n_u8(__inactive, __a, __b, __p)
-#define vqsubq_m_n_u32(__inactive, __a, __b, __p) __arm_vqsubq_m_n_u32(__inactive, __a, __b, __p)
-#define vqsubq_m_n_u16(__inactive, __a, __b, __p) __arm_vqsubq_m_n_u16(__inactive, __a, __b, __p)
-#define vqsubq_m_s8(__inactive, __a, __b, __p) __arm_vqsubq_m_s8(__inactive, __a, __b, __p)
-#define vqsubq_m_s32(__inactive, __a, __b, __p) __arm_vqsubq_m_s32(__inactive, __a, __b, __p)
-#define vqsubq_m_s16(__inactive, __a, __b, __p) __arm_vqsubq_m_s16(__inactive, __a, __b, __p)
-#define vqsubq_m_u8(__inactive, __a, __b, __p) __arm_vqsubq_m_u8(__inactive, __a, __b, __p)
-#define vqsubq_m_u32(__inactive, __a, __b, __p) __arm_vqsubq_m_u32(__inactive, __a, __b, __p)
-#define vqsubq_m_u16(__inactive, __a, __b, __p) __arm_vqsubq_m_u16(__inactive, __a, __b, __p)
-#define vrhaddq_m_s8(__inactive, __a, __b, __p) __arm_vrhaddq_m_s8(__inactive, __a, __b, __p)
-#define vrhaddq_m_s32(__inactive, __a, __b, __p) __arm_vrhaddq_m_s32(__inactive, __a, __b, __p)
-#define vrhaddq_m_s16(__inactive, __a, __b, __p) __arm_vrhaddq_m_s16(__inactive, __a, __b, __p)
-#define vrhaddq_m_u8(__inactive, __a, __b, __p) __arm_vrhaddq_m_u8(__inactive, __a, __b, __p)
-#define vrhaddq_m_u32(__inactive, __a, __b, __p) __arm_vrhaddq_m_u32(__inactive, __a, __b, __p)
-#define vrhaddq_m_u16(__inactive, __a, __b, __p) __arm_vrhaddq_m_u16(__inactive, __a, __b, __p)
-#define vrmulhq_m_s8(__inactive, __a, __b, __p) __arm_vrmulhq_m_s8(__inactive, __a, __b, __p)
-#define vrmulhq_m_s32(__inactive, __a, __b, __p) __arm_vrmulhq_m_s32(__inactive, __a, __b, __p)
-#define vrmulhq_m_s16(__inactive, __a, __b, __p) __arm_vrmulhq_m_s16(__inactive, __a, __b, __p)
-#define vrmulhq_m_u8(__inactive, __a, __b, __p) __arm_vrmulhq_m_u8(__inactive, __a, __b, __p)
-#define vrmulhq_m_u32(__inactive, __a, __b, __p) __arm_vrmulhq_m_u32(__inactive, __a, __b, __p)
-#define vrmulhq_m_u16(__inactive, __a, __b, __p) __arm_vrmulhq_m_u16(__inactive, __a, __b, __p)
 #define vrshlq_m_s8(__inactive, __a, __b, __p) __arm_vrshlq_m_s8(__inactive, __a, __b, __p)
 #define vrshlq_m_s32(__inactive, __a, __b, __p) __arm_vrshlq_m_s32(__inactive, __a, __b, __p)
 #define vrshlq_m_s16(__inactive, __a, __b, __p) __arm_vrshlq_m_s16(__inactive, __a, __b, __p)
@@ -2315,12 +2150,6 @@ 
 #define vnegq_x_s8(__a, __p) __arm_vnegq_x_s8(__a, __p)
 #define vnegq_x_s16(__a, __p) __arm_vnegq_x_s16(__a, __p)
 #define vnegq_x_s32(__a, __p) __arm_vnegq_x_s32(__a, __p)
-#define vmulhq_x_s8(__a, __b, __p) __arm_vmulhq_x_s8(__a, __b, __p)
-#define vmulhq_x_s16(__a, __b, __p) __arm_vmulhq_x_s16(__a, __b, __p)
-#define vmulhq_x_s32(__a, __b, __p) __arm_vmulhq_x_s32(__a, __b, __p)
-#define vmulhq_x_u8(__a, __b, __p) __arm_vmulhq_x_u8(__a, __b, __p)
-#define vmulhq_x_u16(__a, __b, __p) __arm_vmulhq_x_u16(__a, __b, __p)
-#define vmulhq_x_u32(__a, __b, __p) __arm_vmulhq_x_u32(__a, __b, __p)
 #define vmullbq_poly_x_p8(__a, __b, __p) __arm_vmullbq_poly_x_p8(__a, __b, __p)
 #define vmullbq_poly_x_p16(__a, __b, __p) __arm_vmullbq_poly_x_p16(__a, __b, __p)
 #define vmullbq_int_x_s8(__a, __b, __p) __arm_vmullbq_int_x_s8(__a, __b, __p)
@@ -2349,48 +2178,12 @@ 
 #define vcaddq_rot270_x_u8(__a, __b, __p) __arm_vcaddq_rot270_x_u8(__a, __b, __p)
 #define vcaddq_rot270_x_u16(__a, __b, __p) __arm_vcaddq_rot270_x_u16(__a, __b, __p)
 #define vcaddq_rot270_x_u32(__a, __b, __p) __arm_vcaddq_rot270_x_u32(__a, __b, __p)
-#define vhaddq_x_n_s8(__a, __b, __p) __arm_vhaddq_x_n_s8(__a, __b, __p)
-#define vhaddq_x_n_s16(__a, __b, __p) __arm_vhaddq_x_n_s16(__a, __b, __p)
-#define vhaddq_x_n_s32(__a, __b, __p) __arm_vhaddq_x_n_s32(__a, __b, __p)
-#define vhaddq_x_n_u8(__a, __b, __p) __arm_vhaddq_x_n_u8(__a, __b, __p)
-#define vhaddq_x_n_u16(__a, __b, __p) __arm_vhaddq_x_n_u16(__a, __b, __p)
-#define vhaddq_x_n_u32(__a, __b, __p) __arm_vhaddq_x_n_u32(__a, __b, __p)
-#define vhaddq_x_s8(__a, __b, __p) __arm_vhaddq_x_s8(__a, __b, __p)
-#define vhaddq_x_s16(__a, __b, __p) __arm_vhaddq_x_s16(__a, __b, __p)
-#define vhaddq_x_s32(__a, __b, __p) __arm_vhaddq_x_s32(__a, __b, __p)
-#define vhaddq_x_u8(__a, __b, __p) __arm_vhaddq_x_u8(__a, __b, __p)
-#define vhaddq_x_u16(__a, __b, __p) __arm_vhaddq_x_u16(__a, __b, __p)
-#define vhaddq_x_u32(__a, __b, __p) __arm_vhaddq_x_u32(__a, __b, __p)
 #define vhcaddq_rot90_x_s8(__a, __b, __p) __arm_vhcaddq_rot90_x_s8(__a, __b, __p)
 #define vhcaddq_rot90_x_s16(__a, __b, __p) __arm_vhcaddq_rot90_x_s16(__a, __b, __p)
 #define vhcaddq_rot90_x_s32(__a, __b, __p) __arm_vhcaddq_rot90_x_s32(__a, __b, __p)
 #define vhcaddq_rot270_x_s8(__a, __b, __p) __arm_vhcaddq_rot270_x_s8(__a, __b, __p)
 #define vhcaddq_rot270_x_s16(__a, __b, __p) __arm_vhcaddq_rot270_x_s16(__a, __b, __p)
 #define vhcaddq_rot270_x_s32(__a, __b, __p) __arm_vhcaddq_rot270_x_s32(__a, __b, __p)
-#define vhsubq_x_n_s8(__a, __b, __p) __arm_vhsubq_x_n_s8(__a, __b, __p)
-#define vhsubq_x_n_s16(__a, __b, __p) __arm_vhsubq_x_n_s16(__a, __b, __p)
-#define vhsubq_x_n_s32(__a, __b, __p) __arm_vhsubq_x_n_s32(__a, __b, __p)
-#define vhsubq_x_n_u8(__a, __b, __p) __arm_vhsubq_x_n_u8(__a, __b, __p)
-#define vhsubq_x_n_u16(__a, __b, __p) __arm_vhsubq_x_n_u16(__a, __b, __p)
-#define vhsubq_x_n_u32(__a, __b, __p) __arm_vhsubq_x_n_u32(__a, __b, __p)
-#define vhsubq_x_s8(__a, __b, __p) __arm_vhsubq_x_s8(__a, __b, __p)
-#define vhsubq_x_s16(__a, __b, __p) __arm_vhsubq_x_s16(__a, __b, __p)
-#define vhsubq_x_s32(__a, __b, __p) __arm_vhsubq_x_s32(__a, __b, __p)
-#define vhsubq_x_u8(__a, __b, __p) __arm_vhsubq_x_u8(__a, __b, __p)
-#define vhsubq_x_u16(__a, __b, __p) __arm_vhsubq_x_u16(__a, __b, __p)
-#define vhsubq_x_u32(__a, __b, __p) __arm_vhsubq_x_u32(__a, __b, __p)
-#define vrhaddq_x_s8(__a, __b, __p) __arm_vrhaddq_x_s8(__a, __b, __p)
-#define vrhaddq_x_s16(__a, __b, __p) __arm_vrhaddq_x_s16(__a, __b, __p)
-#define vrhaddq_x_s32(__a, __b, __p) __arm_vrhaddq_x_s32(__a, __b, __p)
-#define vrhaddq_x_u8(__a, __b, __p) __arm_vrhaddq_x_u8(__a, __b, __p)
-#define vrhaddq_x_u16(__a, __b, __p) __arm_vrhaddq_x_u16(__a, __b, __p)
-#define vrhaddq_x_u32(__a, __b, __p) __arm_vrhaddq_x_u32(__a, __b, __p)
-#define vrmulhq_x_s8(__a, __b, __p) __arm_vrmulhq_x_s8(__a, __b, __p)
-#define vrmulhq_x_s16(__a, __b, __p) __arm_vrmulhq_x_s16(__a, __b, __p)
-#define vrmulhq_x_s32(__a, __b, __p) __arm_vrmulhq_x_s32(__a, __b, __p)
-#define vrmulhq_x_u8(__a, __b, __p) __arm_vrmulhq_x_u8(__a, __b, __p)
-#define vrmulhq_x_u16(__a, __b, __p) __arm_vrmulhq_x_u16(__a, __b, __p)
-#define vrmulhq_x_u32(__a, __b, __p) __arm_vrmulhq_x_u32(__a, __b, __p)
 #define vbicq_x_s8(__a, __b, __p) __arm_vbicq_x_s8(__a, __b, __p)
 #define vbicq_x_s16(__a, __b, __p) __arm_vbicq_x_s16(__a, __b, __p)
 #define vbicq_x_s32(__a, __b, __p) __arm_vbicq_x_s32(__a, __b, __p)
@@ -3351,48 +3144,6 @@  __arm_vshlq_u32 (uint32x4_t __a, int32x4_t __b)
   return __builtin_mve_vshlq_uv4si (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vrmulhq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vrhaddq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vqsubq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_u8 (uint8x16_t __a, uint8_t __b)
-{
-  return __builtin_mve_vqsubq_n_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vqaddq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_u8 (uint8x16_t __a, uint8_t __b)
-{
-  return __builtin_mve_vqaddq_n_uv16qi (__a, __b);
-}
-
 __extension__ extern __inline uint8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_u8 (uint8x16_t __a, uint8x16_t __b)
@@ -3414,13 +3165,6 @@  __arm_vmullbq_int_u8 (uint8x16_t __a, uint8x16_t __b)
   return __builtin_mve_vmullbq_int_uv16qi (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vmulhq_uv16qi (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq_u8 (uint8x16_t __a, uint8x16_t __b)
@@ -3456,34 +3200,6 @@  __arm_vmaxq_u8 (uint8x16_t __a, uint8x16_t __b)
   return __builtin_mve_vmaxq_uv16qi (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vhsubq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_u8 (uint8x16_t __a, uint8_t __b)
-{
-  return __builtin_mve_vhsubq_n_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
-{
-  return __builtin_mve_vhaddq_uv16qi (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_u8 (uint8x16_t __a, uint8_t __b)
-{
-  return __builtin_mve_vhaddq_n_uv16qi (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq_n_u8 (uint8x16_t __a, uint8_t __b)
@@ -3794,34 +3510,6 @@  __arm_vrshlq_n_s8 (int8x16_t __a, int32_t __b)
   return __builtin_mve_vrshlq_n_sv16qi (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vrmulhq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vrhaddq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vqsubq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_s8 (int8x16_t __a, int8_t __b)
-{
-  return __builtin_mve_vqsubq_n_sv16qi (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq_s8 (int8x16_t __a, int8x16_t __b)
@@ -3864,34 +3552,6 @@  __arm_vqrdmulhq_n_s8 (int8x16_t __a, int8_t __b)
   return __builtin_mve_vqrdmulhq_n_sv16qi (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vqdmulhq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_n_s8 (int8x16_t __a, int8_t __b)
-{
-  return __builtin_mve_vqdmulhq_n_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vqaddq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_s8 (int8x16_t __a, int8_t __b)
-{
-  return __builtin_mve_vqaddq_n_sv16qi (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_s8 (int8x16_t __a, int8x16_t __b)
@@ -3913,13 +3573,6 @@  __arm_vmullbq_int_s8 (int8x16_t __a, int8x16_t __b)
   return __builtin_mve_vmullbq_int_sv16qi (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vmulhq_sv16qi (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq_s8 (int8x16_t __a, int8x16_t __b)
@@ -3976,20 +3629,6 @@  __arm_vmaxq_s8 (int8x16_t __a, int8x16_t __b)
   return __builtin_mve_vmaxq_sv16qi (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vhsubq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_s8 (int8x16_t __a, int8_t __b)
-{
-  return __builtin_mve_vhsubq_n_sv16qi (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90_s8 (int8x16_t __a, int8x16_t __b)
@@ -4004,20 +3643,6 @@  __arm_vhcaddq_rot270_s8 (int8x16_t __a, int8x16_t __b)
   return __builtin_mve_vhcaddq_rot270_sv16qi (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_s8 (int8x16_t __a, int8x16_t __b)
-{
-  return __builtin_mve_vhaddq_sv16qi (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_s8 (int8x16_t __a, int8_t __b)
-{
-  return __builtin_mve_vhaddq_n_sv16qi (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90_s8 (int8x16_t __a, int8x16_t __b)
@@ -4081,48 +3706,6 @@  __arm_vqshlq_n_s8 (int8x16_t __a, const int __imm)
   return __builtin_mve_vqshlq_n_sv16qi (__a, __imm);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vrmulhq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vrhaddq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vqsubq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_u16 (uint16x8_t __a, uint16_t __b)
-{
-  return __builtin_mve_vqsubq_n_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vqaddq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_u16 (uint16x8_t __a, uint16_t __b)
-{
-  return __builtin_mve_vqaddq_n_uv8hi (__a, __b);
-}
-
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_u16 (uint16x8_t __a, uint16x8_t __b)
@@ -4144,13 +3727,6 @@  __arm_vmullbq_int_u16 (uint16x8_t __a, uint16x8_t __b)
   return __builtin_mve_vmullbq_int_uv8hi (__a, __b);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vmulhq_uv8hi (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq_u16 (uint16x8_t __a, uint16x8_t __b)
@@ -4186,34 +3762,6 @@  __arm_vmaxq_u16 (uint16x8_t __a, uint16x8_t __b)
   return __builtin_mve_vmaxq_uv8hi (__a, __b);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vhsubq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_u16 (uint16x8_t __a, uint16_t __b)
-{
-  return __builtin_mve_vhsubq_n_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
-{
-  return __builtin_mve_vhaddq_uv8hi (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_u16 (uint16x8_t __a, uint16_t __b)
-{
-  return __builtin_mve_vhaddq_n_uv8hi (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq_n_u16 (uint16x8_t __a, uint16_t __b)
@@ -4524,34 +4072,6 @@  __arm_vrshlq_n_s16 (int16x8_t __a, int32_t __b)
   return __builtin_mve_vrshlq_n_sv8hi (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vrmulhq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vrhaddq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vqsubq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_s16 (int16x8_t __a, int16_t __b)
-{
-  return __builtin_mve_vqsubq_n_sv8hi (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq_s16 (int16x8_t __a, int16x8_t __b)
@@ -4594,34 +4114,6 @@  __arm_vqrdmulhq_n_s16 (int16x8_t __a, int16_t __b)
   return __builtin_mve_vqrdmulhq_n_sv8hi (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vqdmulhq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_n_s16 (int16x8_t __a, int16_t __b)
-{
-  return __builtin_mve_vqdmulhq_n_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vqaddq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_s16 (int16x8_t __a, int16_t __b)
-{
-  return __builtin_mve_vqaddq_n_sv8hi (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_s16 (int16x8_t __a, int16x8_t __b)
@@ -4643,13 +4135,6 @@  __arm_vmullbq_int_s16 (int16x8_t __a, int16x8_t __b)
   return __builtin_mve_vmullbq_int_sv8hi (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vmulhq_sv8hi (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq_s16 (int16x8_t __a, int16x8_t __b)
@@ -4706,20 +4191,6 @@  __arm_vmaxq_s16 (int16x8_t __a, int16x8_t __b)
   return __builtin_mve_vmaxq_sv8hi (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vhsubq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_s16 (int16x8_t __a, int16_t __b)
-{
-  return __builtin_mve_vhsubq_n_sv8hi (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90_s16 (int16x8_t __a, int16x8_t __b)
@@ -4734,20 +4205,6 @@  __arm_vhcaddq_rot270_s16 (int16x8_t __a, int16x8_t __b)
   return __builtin_mve_vhcaddq_rot270_sv8hi (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_s16 (int16x8_t __a, int16x8_t __b)
-{
-  return __builtin_mve_vhaddq_sv8hi (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_s16 (int16x8_t __a, int16_t __b)
-{
-  return __builtin_mve_vhaddq_n_sv8hi (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90_s16 (int16x8_t __a, int16x8_t __b)
@@ -4811,48 +4268,6 @@  __arm_vqshlq_n_s16 (int16x8_t __a, const int __imm)
   return __builtin_mve_vqshlq_n_sv8hi (__a, __imm);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vrmulhq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vrhaddq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vqsubq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_u32 (uint32x4_t __a, uint32_t __b)
-{
-  return __builtin_mve_vqsubq_n_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vqaddq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_u32 (uint32x4_t __a, uint32_t __b)
-{
-  return __builtin_mve_vqaddq_n_uv4si (__a, __b);
-}
-
 __extension__ extern __inline uint32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_u32 (uint32x4_t __a, uint32x4_t __b)
@@ -4874,13 +4289,6 @@  __arm_vmullbq_int_u32 (uint32x4_t __a, uint32x4_t __b)
   return __builtin_mve_vmullbq_int_uv4si (__a, __b);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vmulhq_uv4si (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq_u32 (uint32x4_t __a, uint32x4_t __b)
@@ -4916,34 +4324,6 @@  __arm_vmaxq_u32 (uint32x4_t __a, uint32x4_t __b)
   return __builtin_mve_vmaxq_uv4si (__a, __b);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vhsubq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_u32 (uint32x4_t __a, uint32_t __b)
-{
-  return __builtin_mve_vhsubq_n_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
-{
-  return __builtin_mve_vhaddq_uv4si (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_u32 (uint32x4_t __a, uint32_t __b)
-{
-  return __builtin_mve_vhaddq_n_uv4si (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq_n_u32 (uint32x4_t __a, uint32_t __b)
@@ -5254,34 +4634,6 @@  __arm_vrshlq_n_s32 (int32x4_t __a, int32_t __b)
   return __builtin_mve_vrshlq_n_sv4si (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vrmulhq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vrhaddq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vqsubq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_n_s32 (int32x4_t __a, int32_t __b)
-{
-  return __builtin_mve_vqsubq_n_sv4si (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq_s32 (int32x4_t __a, int32x4_t __b)
@@ -5324,34 +4676,6 @@  __arm_vqrdmulhq_n_s32 (int32x4_t __a, int32_t __b)
   return __builtin_mve_vqrdmulhq_n_sv4si (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vqdmulhq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_n_s32 (int32x4_t __a, int32_t __b)
-{
-  return __builtin_mve_vqdmulhq_n_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vqaddq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_n_s32 (int32x4_t __a, int32_t __b)
-{
-  return __builtin_mve_vqaddq_n_sv4si (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq_s32 (int32x4_t __a, int32x4_t __b)
@@ -5373,13 +4697,6 @@  __arm_vmullbq_int_s32 (int32x4_t __a, int32x4_t __b)
   return __builtin_mve_vmullbq_int_sv4si (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vmulhq_sv4si (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq_s32 (int32x4_t __a, int32x4_t __b)
@@ -5436,20 +4753,6 @@  __arm_vmaxq_s32 (int32x4_t __a, int32x4_t __b)
   return __builtin_mve_vmaxq_sv4si (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vhsubq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_n_s32 (int32x4_t __a, int32_t __b)
-{
-  return __builtin_mve_vhsubq_n_sv4si (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90_s32 (int32x4_t __a, int32x4_t __b)
@@ -5464,20 +4767,6 @@  __arm_vhcaddq_rot270_s32 (int32x4_t __a, int32x4_t __b)
   return __builtin_mve_vhcaddq_rot270_sv4si (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_s32 (int32x4_t __a, int32x4_t __b)
-{
-  return __builtin_mve_vhaddq_sv4si (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_n_s32 (int32x4_t __a, int32_t __b)
-{
-  return __builtin_mve_vhaddq_n_sv4si (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90_s32 (int32x4_t __a, int32x4_t __b)
@@ -9005,90 +8294,6 @@  __arm_vcaddq_rot90_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b,
   return __builtin_mve_vcaddq_rot90_m_uv8hi (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot270_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -9131,90 +8336,6 @@  __arm_vhcaddq_rot90_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, m
   return __builtin_mve_vhcaddq_rot90_m_sv8hi (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmaxq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -9488,48 +8609,6 @@  __arm_vmlsdavaxq_p_s16 (int32_t __a, int16x8_t __b, int16x8_t __c, mve_pred16_t
   return __builtin_mve_vmlsdavaxq_p_sv8hi (__a, __b, __c, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmullbq_int_m_s8 (int16x8_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -9656,90 +8735,6 @@  __arm_vornq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pr
   return __builtin_mve_vornq_m_uv8hi (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_n_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqaddq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqdmladhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -9845,48 +8840,6 @@  __arm_vqdmlsdhxq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_
   return __builtin_mve_vqdmlsdhxq_m_sv8hi (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_n_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_n_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_n_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqdmulhq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqrdmladhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -10202,174 +9155,6 @@  __arm_vqshlq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, int16x8_t __b, mve_pr
   return __builtin_mve_vqshlq_m_uv8hi (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_s8 (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_s32 (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_s16 (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_n_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_n_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vqsubq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_s32 (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_s16 (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv8hi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_u8 (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv16qi (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_u32 (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv4si (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m_u16 (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv8hi (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vrshlq_m_s8 (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -13289,48 +12074,6 @@  __arm_vnegq_x_s32 (int32x4_t __a, mve_pred16_t __p)
   return __builtin_mve_vnegq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vmulhq_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmullbq_poly_x_p8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
@@ -13527,90 +12270,6 @@  __arm_vcaddq_rot270_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
   return __builtin_mve_vcaddq_rot270_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_s8 (int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_s16 (int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_s32 (int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_u8 (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_u16 (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_n_u32 (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_n_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhaddq_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -13653,174 +12312,6 @@  __arm_vhcaddq_rot270_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
   return __builtin_mve_vhcaddq_rot270_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_s8 (int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_s16 (int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_s32 (int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_u8 (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_u16 (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_n_u32 (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_n_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vhsubq_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrhaddq_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv16qi (__arm_vuninitializedq_s8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_s16 (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv8hi (__arm_vuninitializedq_s16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_s32 (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_sv4si (__arm_vuninitializedq_s32 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_u8 (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv16qi (__arm_vuninitializedq_u8 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_u16 (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv8hi (__arm_vuninitializedq_u16 (), __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x_u32 (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
-  return __builtin_mve_vrmulhq_m_uv4si (__arm_vuninitializedq_u32 (), __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vbicq_x_s8 (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -18558,48 +17049,6 @@  __arm_vshlq (uint32x4_t __a, int32x4_t __b)
  return __arm_vshlq_u32 (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vrmulhq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vrhaddq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vqsubq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint8x16_t __a, uint8_t __b)
-{
- return __arm_vqsubq_n_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vqaddq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint8x16_t __a, uint8_t __b)
-{
- return __arm_vqaddq_n_u8 (__a, __b);
-}
-
 __extension__ extern __inline uint8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (uint8x16_t __a, uint8x16_t __b)
@@ -18621,13 +17070,6 @@  __arm_vmullbq_int (uint8x16_t __a, uint8x16_t __b)
  return __arm_vmullbq_int_u8 (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vmulhq_u8 (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq (uint8x16_t __a, uint8x16_t __b)
@@ -18663,34 +17105,6 @@  __arm_vmaxq (uint8x16_t __a, uint8x16_t __b)
  return __arm_vmaxq_u8 (__a, __b);
 }
 
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vhsubq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint8x16_t __a, uint8_t __b)
-{
- return __arm_vhsubq_n_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint8x16_t __a, uint8x16_t __b)
-{
- return __arm_vhaddq_u8 (__a, __b);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint8x16_t __a, uint8_t __b)
-{
- return __arm_vhaddq_n_u8 (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq (uint8x16_t __a, uint8_t __b)
@@ -18999,34 +17413,6 @@  __arm_vrshlq (int8x16_t __a, int32_t __b)
  return __arm_vrshlq_n_s8 (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vrmulhq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vrhaddq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vqsubq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int8x16_t __a, int8_t __b)
-{
- return __arm_vqsubq_n_s8 (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq (int8x16_t __a, int8x16_t __b)
@@ -19069,34 +17455,6 @@  __arm_vqrdmulhq (int8x16_t __a, int8_t __b)
  return __arm_vqrdmulhq_n_s8 (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vqdmulhq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int8x16_t __a, int8_t __b)
-{
- return __arm_vqdmulhq_n_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vqaddq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int8x16_t __a, int8_t __b)
-{
- return __arm_vqaddq_n_s8 (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (int8x16_t __a, int8x16_t __b)
@@ -19118,13 +17476,6 @@  __arm_vmullbq_int (int8x16_t __a, int8x16_t __b)
  return __arm_vmullbq_int_s8 (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vmulhq_s8 (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq (int8x16_t __a, int8x16_t __b)
@@ -19181,20 +17532,6 @@  __arm_vmaxq (int8x16_t __a, int8x16_t __b)
  return __arm_vmaxq_s8 (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vhsubq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int8x16_t __a, int8_t __b)
-{
- return __arm_vhsubq_n_s8 (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90 (int8x16_t __a, int8x16_t __b)
@@ -19209,20 +17546,6 @@  __arm_vhcaddq_rot270 (int8x16_t __a, int8x16_t __b)
  return __arm_vhcaddq_rot270_s8 (__a, __b);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int8x16_t __a, int8x16_t __b)
-{
- return __arm_vhaddq_s8 (__a, __b);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int8x16_t __a, int8_t __b)
-{
- return __arm_vhaddq_n_s8 (__a, __b);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90 (int8x16_t __a, int8x16_t __b)
@@ -19286,48 +17609,6 @@  __arm_vqshlq_n (int8x16_t __a, const int __imm)
  return __arm_vqshlq_n_s8 (__a, __imm);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vrmulhq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vrhaddq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vqsubq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint16x8_t __a, uint16_t __b)
-{
- return __arm_vqsubq_n_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vqaddq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint16x8_t __a, uint16_t __b)
-{
- return __arm_vqaddq_n_u16 (__a, __b);
-}
-
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (uint16x8_t __a, uint16x8_t __b)
@@ -19349,13 +17630,6 @@  __arm_vmullbq_int (uint16x8_t __a, uint16x8_t __b)
  return __arm_vmullbq_int_u16 (__a, __b);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vmulhq_u16 (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq (uint16x8_t __a, uint16x8_t __b)
@@ -19391,34 +17665,6 @@  __arm_vmaxq (uint16x8_t __a, uint16x8_t __b)
  return __arm_vmaxq_u16 (__a, __b);
 }
 
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vhsubq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint16x8_t __a, uint16_t __b)
-{
- return __arm_vhsubq_n_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint16x8_t __a, uint16x8_t __b)
-{
- return __arm_vhaddq_u16 (__a, __b);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint16x8_t __a, uint16_t __b)
-{
- return __arm_vhaddq_n_u16 (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq (uint16x8_t __a, uint16_t __b)
@@ -19727,34 +17973,6 @@  __arm_vrshlq (int16x8_t __a, int32_t __b)
  return __arm_vrshlq_n_s16 (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vrmulhq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vrhaddq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vqsubq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int16x8_t __a, int16_t __b)
-{
- return __arm_vqsubq_n_s16 (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq (int16x8_t __a, int16x8_t __b)
@@ -19797,34 +18015,6 @@  __arm_vqrdmulhq (int16x8_t __a, int16_t __b)
  return __arm_vqrdmulhq_n_s16 (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vqdmulhq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int16x8_t __a, int16_t __b)
-{
- return __arm_vqdmulhq_n_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vqaddq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int16x8_t __a, int16_t __b)
-{
- return __arm_vqaddq_n_s16 (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (int16x8_t __a, int16x8_t __b)
@@ -19846,13 +18036,6 @@  __arm_vmullbq_int (int16x8_t __a, int16x8_t __b)
  return __arm_vmullbq_int_s16 (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vmulhq_s16 (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq (int16x8_t __a, int16x8_t __b)
@@ -19909,20 +18092,6 @@  __arm_vmaxq (int16x8_t __a, int16x8_t __b)
  return __arm_vmaxq_s16 (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vhsubq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int16x8_t __a, int16_t __b)
-{
- return __arm_vhsubq_n_s16 (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90 (int16x8_t __a, int16x8_t __b)
@@ -19937,20 +18106,6 @@  __arm_vhcaddq_rot270 (int16x8_t __a, int16x8_t __b)
  return __arm_vhcaddq_rot270_s16 (__a, __b);
 }
 
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int16x8_t __a, int16x8_t __b)
-{
- return __arm_vhaddq_s16 (__a, __b);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int16x8_t __a, int16_t __b)
-{
- return __arm_vhaddq_n_s16 (__a, __b);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90 (int16x8_t __a, int16x8_t __b)
@@ -20014,48 +18169,6 @@  __arm_vqshlq_n (int16x8_t __a, const int __imm)
  return __arm_vqshlq_n_s16 (__a, __imm);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vrmulhq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vrhaddq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vqsubq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (uint32x4_t __a, uint32_t __b)
-{
- return __arm_vqsubq_n_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vqaddq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (uint32x4_t __a, uint32_t __b)
-{
- return __arm_vqaddq_n_u32 (__a, __b);
-}
-
 __extension__ extern __inline uint32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (uint32x4_t __a, uint32x4_t __b)
@@ -20077,13 +18190,6 @@  __arm_vmullbq_int (uint32x4_t __a, uint32x4_t __b)
  return __arm_vmullbq_int_u32 (__a, __b);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vmulhq_u32 (__a, __b);
-}
-
 __extension__ extern __inline uint32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmladavq (uint32x4_t __a, uint32x4_t __b)
@@ -20119,34 +18225,6 @@  __arm_vmaxq (uint32x4_t __a, uint32x4_t __b)
  return __arm_vmaxq_u32 (__a, __b);
 }
 
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vhsubq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (uint32x4_t __a, uint32_t __b)
-{
- return __arm_vhsubq_n_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint32x4_t __a, uint32x4_t __b)
-{
- return __arm_vhaddq_u32 (__a, __b);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (uint32x4_t __a, uint32_t __b)
-{
- return __arm_vhaddq_n_u32 (__a, __b);
-}
-
 __extension__ extern __inline mve_pred16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcmpneq (uint32x4_t __a, uint32_t __b)
@@ -20455,34 +18533,6 @@  __arm_vrshlq (int32x4_t __a, int32_t __b)
  return __arm_vrshlq_n_s32 (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vrmulhq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vrhaddq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vqsubq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq (int32x4_t __a, int32_t __b)
-{
- return __arm_vqsubq_n_s32 (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqshlq (int32x4_t __a, int32x4_t __b)
@@ -20525,34 +18575,6 @@  __arm_vqrdmulhq (int32x4_t __a, int32_t __b)
  return __arm_vqrdmulhq_n_s32 (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vqdmulhq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq (int32x4_t __a, int32_t __b)
-{
- return __arm_vqdmulhq_n_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vqaddq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq (int32x4_t __a, int32_t __b)
-{
- return __arm_vqaddq_n_s32 (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vornq (int32x4_t __a, int32x4_t __b)
@@ -20574,13 +18596,6 @@  __arm_vmullbq_int (int32x4_t __a, int32x4_t __b)
  return __arm_vmullbq_int_s32 (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vmulhq_s32 (__a, __b);
-}
-
 __extension__ extern __inline int32_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmlsdavxq (int32x4_t __a, int32x4_t __b)
@@ -20637,20 +18652,6 @@  __arm_vmaxq (int32x4_t __a, int32x4_t __b)
  return __arm_vmaxq_s32 (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vhsubq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq (int32x4_t __a, int32_t __b)
-{
- return __arm_vhsubq_n_s32 (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90 (int32x4_t __a, int32x4_t __b)
@@ -20665,20 +18666,6 @@  __arm_vhcaddq_rot270 (int32x4_t __a, int32x4_t __b)
  return __arm_vhcaddq_rot270_s32 (__a, __b);
 }
 
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int32x4_t __a, int32x4_t __b)
-{
- return __arm_vhaddq_s32 (__a, __b);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq (int32x4_t __a, int32_t __b)
-{
- return __arm_vhaddq_n_s32 (__a, __b);
-}
-
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vcaddq_rot90 (int32x4_t __a, int32x4_t __b)
@@ -24165,90 +22152,6 @@  __arm_vcaddq_rot90_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve
  return __arm_vcaddq_rot90_m_u16 (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_n_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_m_u16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot270_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -24291,90 +22194,6 @@  __arm_vhcaddq_rot90_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_p
  return __arm_vhcaddq_rot90_m_s16 (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_n_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_m_u16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmaxq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -24648,48 +22467,6 @@  __arm_vmlsdavaxq_p (int32_t __a, int16x8_t __b, int16x8_t __c, mve_pred16_t __p)
  return __arm_vmlsdavaxq_p_s16 (__a, __b, __c, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_m_u16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmullbq_int_m (int16x8_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -24816,90 +22593,6 @@  __arm_vornq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16
  return __arm_vornq_m_u16 (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_n_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vqaddq_m_u16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqdmladhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -25005,48 +22698,6 @@  __arm_vqdmlsdhxq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred
  return __arm_vqdmlsdhxq_m_s16 (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_n_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_n_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_n_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqdmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vqdmulhq_m_s16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vqrdmladhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -25362,174 +23013,6 @@  __arm_vqshlq_m (uint16x8_t __inactive, uint16x8_t __a, int16x8_t __b, mve_pred16
  return __arm_vqshlq_m_u16 (__inactive, __a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int8x16_t __inactive, int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int32x4_t __inactive, int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int16x8_t __inactive, int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_n_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vqsubq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vqsubq_m_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_m_u16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_s8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (int32x4_t __inactive, int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_s32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (int16x8_t __inactive, int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_s16 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (uint8x16_t __inactive, uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_u8 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (uint32x4_t __inactive, uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_u32 (__inactive, __a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_m (uint16x8_t __inactive, uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_m_u16 (__inactive, __a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vrshlq_m (int8x16_t __inactive, int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -27980,48 +25463,6 @@  __arm_vnegq_x (int32x4_t __a, mve_pred16_t __p)
  return __arm_vnegq_x_s32 (__a, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vmulhq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vmulhq_x_u32 (__a, __b, __p);
-}
-
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vmullbq_poly_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
@@ -28218,90 +25659,6 @@  __arm_vcaddq_rot270_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
  return __arm_vcaddq_rot270_x_u32 (__a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_n_u32 (__a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhaddq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhaddq_x_u32 (__a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vhcaddq_rot90_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -28344,174 +25701,6 @@  __arm_vhcaddq_rot270_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
  return __arm_vhcaddq_rot270_x_s32 (__a, __b, __p);
 }
 
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int8x16_t __a, int8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int16x8_t __a, int16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int32x4_t __a, int32_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint8x16_t __a, uint8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint16x8_t __a, uint16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint32x4_t __a, uint32_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_n_u32 (__a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vhsubq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vhsubq_x_u32 (__a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrhaddq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrhaddq_x_u32 (__a, __b, __p);
-}
-
-__extension__ extern __inline int8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_s8 (__a, __b, __p);
-}
-
-__extension__ extern __inline int16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (int16x8_t __a, int16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_s16 (__a, __b, __p);
-}
-
-__extension__ extern __inline int32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (int32x4_t __a, int32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_s32 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint8x16_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (uint8x16_t __a, uint8x16_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_u8 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint16x8_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (uint16x8_t __a, uint16x8_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_u16 (__a, __b, __p);
-}
-
-__extension__ extern __inline uint32x4_t
-__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-__arm_vrmulhq_x (uint32x4_t __a, uint32x4_t __b, mve_pred16_t __p)
-{
- return __arm_vrmulhq_x_u32 (__a, __b, __p);
-}
-
 __extension__ extern __inline int8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 __arm_vbicq_x (int8x16_t __a, int8x16_t __b, mve_pred16_t __p)
@@ -32685,42 +29874,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrshlq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrshlq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vrmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
-#define __arm_vrhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
-#define __arm_vqsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vqshluq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
   int (*)[__ARM_mve_type_int8x16_t]: __arm_vqshluq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), p1), \
@@ -32831,32 +29984,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqdmullbq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqdmullbq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vqdmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqdmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqdmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqdmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
-
-#define __arm_vqaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vmulltq_poly(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -32879,22 +30006,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulltq_int_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulltq_int_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
 
-#define __arm_vhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vhcaddq_rot270(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -32909,22 +30020,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhcaddq_rot90_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhcaddq_rot90_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vhsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vminq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -32975,16 +30070,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint16x8_t]: __arm_vmovnbq_u16 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32x4_t]: __arm_vmovnbq_u32 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
 
-#define __arm_vmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vmullbq_int(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -34580,42 +31665,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrshlq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrshlq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vrmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
-#define __arm_vrhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
-#define __arm_vqsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vqshlq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -34694,32 +31743,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqrdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqrdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)));})
 
-#define __arm_vqdmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqdmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqdmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqdmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
-
-#define __arm_vqaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vornq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -34750,16 +31773,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmullbq_int_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmullbq_int_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
 
-#define __arm_vmulhq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vmulhq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmulhq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmulhq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulhq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulhq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulhq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vminq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -34794,22 +31807,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmaxaq_s16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmaxaq_s32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vhsubq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhsubq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhsubq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhsubq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhsubq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhsubq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhsubq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vhcaddq_rot90(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -34824,22 +31821,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhcaddq_rot270_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhcaddq_rot270_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)));})
 
-#define __arm_vhaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce3(p1, int)), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
 #define __arm_vcaddq_rot90(p0,p1) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
@@ -35910,16 +32891,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint8x16_t]: __arm_vmovltq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), p2), \
   int (*)[__ARM_mve_type_uint16x8_t]: __arm_vmovltq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), p2));})
 
-#define __arm_vmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vmullbq_int_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
@@ -36095,16 +33066,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int8x16_t]: __arm_vrev16q_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), p2), \
   int (*)[__ARM_mve_type_uint8x16_t]: __arm_vrev16q_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), p2));})
 
-#define __arm_vrhaddq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrhaddq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrhaddq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrhaddq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrhaddq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrhaddq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrhaddq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vshlq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
@@ -36115,16 +33076,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]: __arm_vshlq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]: __arm_vshlq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3));})
 
-#define __arm_vrmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vrshlq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
@@ -36236,22 +33187,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t]: __arm_vshrq_x_n_u16 (__ARM_mve_coerce(__p1, uint16x8_t), p2, p3), \
   int (*)[__ARM_mve_type_uint32x4_t]: __arm_vshrq_x_n_u32 (__ARM_mve_coerce(__p1, uint32x4_t), p2, p3));})
 
-#define __arm_vhaddq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_u8( __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_u16( __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_x_n_u32( __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhaddq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhaddq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhaddq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhaddq_x_u8( __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhaddq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhaddq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vhcaddq_rot270_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
@@ -36266,22 +33201,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhcaddq_rot90_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhcaddq_rot90_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3));})
 
-#define __arm_vhsubq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_x_n_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhsubq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhsubq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhsubq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhsubq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhsubq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhsubq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vclsq_x(p1,p2) ({ __typeof(p1) __p1 = (p1); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)])0, \
   int (*)[__ARM_mve_type_int8x16_t]: __arm_vclsq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), p2), \
@@ -36446,28 +33365,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqshlq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqshlq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3));})
 
-#define __arm_vrhaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrhaddq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrhaddq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrhaddq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrhaddq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrhaddq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrhaddq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
-#define __arm_vrmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vrmulhq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vrmulhq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vrmulhq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vrmulhq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vrmulhq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vrmulhq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vrshlq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36509,23 +33406,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vsliq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t),  p2, p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vsliq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t),  p2, p3));})
 
-#define __arm_vqsubq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqsubq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqsubq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqsubq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqsubq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqsubq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqsubq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqsubq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vqrdmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36799,23 +33679,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vsriq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), p2, p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vsriq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), p2, p3));})
 
-#define __arm_vhaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhaddq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhaddq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhaddq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhaddq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhaddq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhaddq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhaddq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vhcaddq_rot270_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36832,23 +33695,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhcaddq_rot90_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhcaddq_rot90_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3));})
 
-#define __arm_vhsubq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vhsubq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vhsubq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vhsubq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vhsubq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vhsubq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vhsubq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vhsubq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3));})
-
 #define __arm_vmaxq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36893,17 +33739,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vmlasq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vmlasq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3));})
 
-#define __arm_vmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vmulhq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmulhq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmulhq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulhq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulhq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulhq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vmullbq_int_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36933,23 +33768,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulltq_poly_m_p8 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
   int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulltq_poly_m_p16 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3));})
 
-#define __arm_vqaddq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_int_n]: __arm_vqaddq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqaddq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqaddq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqaddq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vqaddq_m_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vqaddq_m_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vqaddq_m_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vqdmlahq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -36958,17 +33776,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqdmlahq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
   int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqdmlahq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3));})
 
-#define __arm_vqdmulhq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
-  __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_m_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_m_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int_n]: __arm_vqdmulhq_m_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce3(p2, int), p3), \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vqdmulhq_m_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vqdmulhq_m_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vqdmulhq_m_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3));})
-
 #define __arm_vqdmullbq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
   __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
@@ -37562,16 +34369,6 @@  extern void *__ARM_undef;
   int (*)[__ARM_mve_type_uint8x16_t]: __arm_vmovltq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), p2), \
   int (*)[__ARM_mve_type_uint16x8_t]: __arm_vmovltq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), p2));})
 
-#define __arm_vmulhq_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
-  __typeof(p2) __p2 = (p2); \
-  _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \
-  int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vmulhq_x_s8 (__ARM_mve_coerce(__p1, int8x16_t), __ARM_mve_coerce(__p2, int8x16_t), p3), \
-  int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vmulhq_x_s16 (__ARM_mve_coerce(__p1, int16x8_t), __ARM_mve_coerce(__p2, int16x8_t), p3), \
-  int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vmulhq_x_s32 (__ARM_mve_coerce(__p1, int32x4_t), __ARM_mve_coerce(__p2, int32x4_t), p3), \
-  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vmulhq_x_u8 (__ARM_mve_coerce(__p1, uint8x16_t), __ARM_mve_coerce(__p2, uint8x16_t), p3), \
-  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vmulhq_x_u16 (__ARM_mve_coerce(__p1, uint16x8_t), __ARM_mve_coerce(__p2, uint16x8_t), p3), \
-  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vmulhq_x_u32 (__ARM_mve_coerce(__p1, uint32x4_t), __ARM_mve_coerce(__p2, uint32x4_t), p3));})
-
 #define __arm_vmullbq_int_x(p1,p2,p3) ({ __typeof(p1) __p1 = (p1); \
   __typeof(p2) __p2 = (p2); \
   _Generic( (int (*)[__ARM_mve_typeid(__p1)][__ARM_mve_typeid(__p2)])0, \