Codebase Browser
llvm
Go to App

llvm/llvm/test/CodeGen/ARM/big-endian-vector-callee.ll 

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple armeb-eabi -mattr v7,neon -float-abi soft %s -o - | FileCheck %s -check-prefix SOFT
; RUN: llc -mtriple armeb-eabi -mattr v7,neon -float-abi hard %s -o - | FileCheck %s -check-prefix HARD

define i64 @test_i64_f64(double %p) {
; SOFT-LABEL: test_i64_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define i64 @test_i64_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_i64_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define i64 @test_i64_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_i64_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define i64 @test_i64_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_i64_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define i64 @test_i64_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_i64_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define i64 @test_i64_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_i64_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r0, r2, d16
; SOFT-NEXT:    adds r1, r0, r0
; SOFT-NEXT:    adc r0, r2, r2
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_i64_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vmov r0, r2, d16
; HARD-NEXT:    adds r1, r0, r0
; HARD-NEXT:    adc r0, r2, r2
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to i64
    %3 = add i64 %2, %2
    ret i64 %3
}

define double @test_f64_i64(i64 %p) {
; SOFT-LABEL: test_f64_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define double @test_f64_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_f64_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define double @test_f64_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_f64_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define double @test_f64_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_f64_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define double @test_f64_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_f64_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define double @test_f64_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_f64_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_f64_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to double
    %3 = fadd double %2, %2
    ret double %3
}

define <1 x i64> @test_v1i64_i64(i64 %p) {
; SOFT-LABEL: test_v1i64_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <1 x i64> @test_v1i64_f64(double %p) {
; SOFT-LABEL: test_v1i64_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <1 x i64> @test_v1i64_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_v1i64_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <1 x i64> @test_v1i64_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_v1i64_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <1 x i64> @test_v1i64_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_v1i64_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <1 x i64> @test_v1i64_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_v1i64_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v1i64_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vadd.i64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to <1 x i64>
    %3 = add <1 x i64> %2, %2
    ret <1 x i64> %3
}

define <2 x float> @test_v2f32_i64(i64 %p) {
; SOFT-LABEL: test_v2f32_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x float> @test_v2f32_f64(double %p) {
; SOFT-LABEL: test_v2f32_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x float> @test_v2f32_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_v2f32_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x float> @test_v2f32_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_v2f32_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x float> @test_v2f32_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_v2f32_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev32.16 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev32.16 d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x float> @test_v2f32_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_v2f32_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev32.8 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f32_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev32.8 d16, d16
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to <2 x float>
    %3 = fadd <2 x float> %2, %2
    ret <2 x float> %3
}

define <2 x i32> @test_v2i32_i64(i64 %p) {
; SOFT-LABEL: test_v2i32_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <2 x i32> @test_v2i32_f64(double %p) {
; SOFT-LABEL: test_v2i32_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <2 x i32> @test_v2i32_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_v2i32_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vrev64.32 d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <2 x i32> @test_v2i32_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_v2i32_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <2 x i32> @test_v2i32_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_v2i32_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev32.16 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev32.16 d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <2 x i32> @test_v2i32_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_v2i32_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev32.8 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i32_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev32.8 d16, d16
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev64.32 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to <2 x i32>
    %3 = add <2 x i32> %2, %2
    ret <2 x i32> %3
}

define <4 x i16> @test_v4i16_i64(i64 %p) {
; SOFT-LABEL: test_v4i16_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <4 x i16> @test_v4i16_f64(double %p) {
; SOFT-LABEL: test_v4i16_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <4 x i16> @test_v4i16_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_v4i16_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vrev64.16 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <4 x i16> @test_v4i16_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_v4i16_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev32.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev32.16 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <4 x i16> @test_v4i16_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_v4i16_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev32.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev32.16 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <4 x i16> @test_v4i16_v8i8(<8 x i8> %p) {
; SOFT-LABEL: test_v4i16_v8i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev16.8 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i16_v8i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 d16, d0
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev16.8 d16, d16
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev64.16 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i8> %p, %p
    %2 = bitcast <8 x i8> %1 to <4 x i16>
    %3 = add <4 x i16> %2, %2
    ret <4 x i16> %3
}

define <8 x i8> @test_v8i8_i64(i64 %p) {
; SOFT-LABEL: test_v8i8_i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    adds r1, r1, r1
; SOFT-NEXT:    adc r0, r0, r0
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    adds r1, r1, r1
; HARD-NEXT:    adc r0, r0, r0
; HARD-NEXT:    vmov d16, r1, r0
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = add i64 %p, %p
    %2 = bitcast i64 %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define <8 x i8> @test_v8i8_f64(double %p) {
; SOFT-LABEL: test_v8i8_f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.f64 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd double %p, %p
    %2 = bitcast double %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define <8 x i8> @test_v8i8_v1i64(<1 x i64> %p) {
; SOFT-LABEL: test_v8i8_v1i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_v1i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 d16, d0, d0
; HARD-NEXT:    vrev64.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <1 x i64> %p, %p
    %2 = bitcast <1 x i64> %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define <8 x i8> @test_v8i8_v2f32(<2 x float> %p) {
; SOFT-LABEL: test_v8i8_v2f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.f32 d16, d16, d16
; SOFT-NEXT:    vrev32.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_v2f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.f32 d16, d16, d16
; HARD-NEXT:    vrev32.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = fadd <2 x float> %p, %p
    %2 = bitcast <2 x float> %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define <8 x i8> @test_v8i8_v2i32(<2 x i32> %p) {
; SOFT-LABEL: test_v8i8_v2i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 d16, d16
; SOFT-NEXT:    vadd.i32 d16, d16, d16
; SOFT-NEXT:    vrev32.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_v2i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 d16, d0
; HARD-NEXT:    vadd.i32 d16, d16, d16
; HARD-NEXT:    vrev32.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i32> %p, %p
    %2 = bitcast <2 x i32> %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define <8 x i8> @test_v8i8_v4i16(<4 x i16> %p) {
; SOFT-LABEL: test_v8i8_v4i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 d16, d16
; SOFT-NEXT:    vadd.i16 d16, d16, d16
; SOFT-NEXT:    vrev16.8 d16, d16
; SOFT-NEXT:    vadd.i8 d16, d16, d16
; SOFT-NEXT:    vrev64.8 d16, d16
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i8_v4i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 d16, d0
; HARD-NEXT:    vadd.i16 d16, d16, d16
; HARD-NEXT:    vrev16.8 d16, d16
; HARD-NEXT:    vadd.i8 d16, d16, d16
; HARD-NEXT:    vrev64.8 d0, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i16> %p, %p
    %2 = bitcast <4 x i16> %1 to <8 x i8>
    %3 = add <8 x i8> %2, %2
    ret <8 x i8> %3
}

define fp128 @test_f128_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_f128_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vrev64.32 q8, q9
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define fp128 @test_f128_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_f128_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define fp128 @test_f128_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_f128_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define fp128 @test_f128_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_f128_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define fp128 @test_f128_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_f128_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define fp128 @test_f128_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_f128_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vmov r2, r3, d17
; SOFT-NEXT:    vmov r0, r1, d16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_f128_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vmov r2, r3, d17
; HARD-NEXT:    vmov r0, r1, d16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to fp128
    %3 = fadd fp128 %2, %2
    ret fp128 %3
}

define <2 x double> @test_v2f64_f128(fp128 %p) {
; SOFT-LABEL: test_v2f64_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v2f64_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x double> @test_v2f64_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_v2f64_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f64_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x double> @test_v2f64_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_v2f64_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f64_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x double> @test_v2f64_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_v2f64_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f64_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x double> @test_v2f64_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_v2f64_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f64_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q8, q8
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x double> @test_v2f64_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_v2f64_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.f64 d18, d16, d16
; SOFT-NEXT:    vadd.f64 d16, d17, d17
; SOFT-NEXT:    vmov r1, r0, d18
; SOFT-NEXT:    vmov r3, r2, d16
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2f64_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q8, q8
; HARD-NEXT:    vadd.f64 d1, d17, d17
; HARD-NEXT:    vadd.f64 d0, d16, d16
; HARD-NEXT:    bx lr
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to <2 x double>
    %3 = fadd <2 x double> %2, %2
    ret <2 x double> %3
}

define <2 x i64> @test_v2i64_f128(fp128 %p) {
; SOFT-LABEL: test_v2i64_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v2i64_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <2 x i64> @test_v2i64_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_v2i64_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vadd.i64 q8, q9, q9
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i64_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    bx lr
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <2 x i64> @test_v2i64_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_v2i64_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i64_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    bx lr
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <2 x i64> @test_v2i64_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_v2i64_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i64_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    bx lr
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <2 x i64> @test_v2i64_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_v2i64_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i64_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q8, q8
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    bx lr
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <2 x i64> @test_v2i64_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_v2i64_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v2i64_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q8, q8
; HARD-NEXT:    vadd.i64 q0, q8, q8
; HARD-NEXT:    bx lr
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to <2 x i64>
    %3 = add <2 x i64> %2, %2
    ret <2 x i64> %3
}

define <4 x float> @test_v4f32_f128(fp128 %p) {
; SOFT-LABEL: test_v4f32_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v4f32_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x float> @test_v4f32_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_v4f32_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vrev64.32 q8, q9
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4f32_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x float> @test_v4f32_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_v4f32_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4f32_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x float> @test_v4f32_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_v4f32_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4f32_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x float> @test_v4f32_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_v4f32_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4f32_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x float> @test_v4f32_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_v4f32_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4f32_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to <4 x float>
    %3 = fadd <4 x float> %2, %2
    ret <4 x float> %3
}

define <4 x i32> @test_v4i32_f128(fp128 %p) {
; SOFT-LABEL: test_v4i32_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v4i32_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <4 x i32> @test_v4i32_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_v4i32_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vrev64.32 q8, q9
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i32_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <4 x i32> @test_v4i32_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_v4i32_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i32_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vrev64.32 q8, q8
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <4 x i32> @test_v4i32_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_v4i32_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i32_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <4 x i32> @test_v4i32_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_v4i32_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i32_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <4 x i32> @test_v4i32_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_v4i32_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v4i32_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev64.32 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to <4 x i32>
    %3 = add <4 x i32> %2, %2
    ret <4 x i32> %3
}

define <8 x i16> @test_v8i16_f128(fp128 %p) {
; SOFT-LABEL: test_v8i16_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v8i16_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <8 x i16> @test_v8i16_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_v8i16_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vrev64.16 q8, q9
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i16_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.16 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <8 x i16> @test_v8i16_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_v8i16_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i16_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vrev64.16 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <8 x i16> @test_v8i16_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_v8i16_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i16_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <8 x i16> @test_v8i16_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_v8i16_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev32.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i16_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev32.16 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <8 x i16> @test_v8i16_v16i8(<16 x i8> %p) {
; SOFT-LABEL: test_v8i16_v16i8:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev16.8 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v8i16_v16i8:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.8 q8, q0
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev16.8 q8, q8
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev64.16 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <16 x i8> %p, %p
    %2 = bitcast <16 x i8> %1 to <8 x i16>
    %3 = add <8 x i16> %2, %2
    ret <8 x i16> %3
}

define <16 x i8> @test_v16i8_f128(fp128 %p) {
; SOFT-LABEL: test_v16i8_f128:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    .save {r11, lr}
; SOFT-NEXT:    push {r11, lr}
; SOFT-NEXT:    .pad #16
; SOFT-NEXT:    sub sp, sp, #16
; SOFT-NEXT:    stm sp, {r0, r1, r2, r3}
; SOFT-NEXT:    bl __addtf3
; SOFT-NEXT:    vmov.32 d17[0], r2
; SOFT-NEXT:    vmov.32 d16[0], r0
; SOFT-NEXT:    vmov.32 d17[1], r3
; SOFT-NEXT:    vmov.32 d16[1], r1
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    add sp, sp, #16
; SOFT-NEXT:    pop {r11, pc}
;
; HARD-LABEL: test_v16i8_f128:
; HARD:       @ %bb.0:
; HARD-NEXT:    .save {r11, lr}
; HARD-NEXT:    push {r11, lr}
; HARD-NEXT:    .pad #16
; HARD-NEXT:    sub sp, sp, #16
; HARD-NEXT:    stm sp, {r0, r1, r2, r3}
; HARD-NEXT:    bl __addtf3
; HARD-NEXT:    vmov.32 d17[0], r2
; HARD-NEXT:    vmov.32 d16[0], r0
; HARD-NEXT:    vmov.32 d17[1], r3
; HARD-NEXT:    vmov.32 d16[1], r1
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    add sp, sp, #16
; HARD-NEXT:    pop {r11, pc}
    %1 = fadd fp128 %p, %p
    %2 = bitcast fp128 %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}

define <16 x i8> @test_v16i8_v2f64(<2 x double> %p) {
; SOFT-LABEL: test_v16i8_v2f64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d16, r3, r2
; SOFT-NEXT:    vmov d17, r1, r0
; SOFT-NEXT:    vadd.f64 d19, d16, d16
; SOFT-NEXT:    vadd.f64 d18, d17, d17
; SOFT-NEXT:    vrev64.8 q8, q9
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v16i8_v2f64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.f64 d17, d1, d1
; HARD-NEXT:    vadd.f64 d16, d0, d0
; HARD-NEXT:    vrev64.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <2 x double> %p, %p
    %2 = bitcast <2 x double> %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}

define <16 x i8> @test_v16i8_v2i64(<2 x i64> %p) {
; SOFT-LABEL: test_v16i8_v2i64:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vadd.i64 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v16i8_v2i64:
; HARD:       @ %bb.0:
; HARD-NEXT:    vadd.i64 q8, q0, q0
; HARD-NEXT:    vrev64.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <2 x i64> %p, %p
    %2 = bitcast <2 x i64> %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}

define <16 x i8> @test_v16i8_v4f32(<4 x float> %p) {
; SOFT-LABEL: test_v16i8_v4f32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.f32 q8, q8, q8
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v16i8_v4f32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.f32 q8, q8, q8
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    bx lr
    %1 = fadd <4 x float> %p, %p
    %2 = bitcast <4 x float> %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}

define <16 x i8> @test_v16i8_v4i32(<4 x i32> %p) {
; SOFT-LABEL: test_v16i8_v4i32:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.32 q8, q8
; SOFT-NEXT:    vadd.i32 q8, q8, q8
; SOFT-NEXT:    vrev32.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v16i8_v4i32:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.32 q8, q0
; HARD-NEXT:    vadd.i32 q8, q8, q8
; HARD-NEXT:    vrev32.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <4 x i32> %p, %p
    %2 = bitcast <4 x i32> %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}

define <16 x i8> @test_v16i8_v8i16(<8 x i16> %p) {
; SOFT-LABEL: test_v16i8_v8i16:
; SOFT:       @ %bb.0:
; SOFT-NEXT:    vmov d17, r3, r2
; SOFT-NEXT:    vmov d16, r1, r0
; SOFT-NEXT:    vrev64.16 q8, q8
; SOFT-NEXT:    vadd.i16 q8, q8, q8
; SOFT-NEXT:    vrev16.8 q8, q8
; SOFT-NEXT:    vadd.i8 q8, q8, q8
; SOFT-NEXT:    vrev64.8 q8, q8
; SOFT-NEXT:    vmov r1, r0, d16
; SOFT-NEXT:    vmov r3, r2, d17
; SOFT-NEXT:    bx lr
;
; HARD-LABEL: test_v16i8_v8i16:
; HARD:       @ %bb.0:
; HARD-NEXT:    vrev64.16 q8, q0
; HARD-NEXT:    vadd.i16 q8, q8, q8
; HARD-NEXT:    vrev16.8 q8, q8
; HARD-NEXT:    vadd.i8 q8, q8, q8
; HARD-NEXT:    vrev64.8 q0, q8
; HARD-NEXT:    bx lr
    %1 = add <8 x i16> %p, %p
    %2 = bitcast <8 x i16> %1 to <16 x i8>
    %3 = add <16 x i8> %2, %2
    ret <16 x i8> %3
}