Auto vectorizing of ARM NEON integer operation in gcc arm-linux-gnueabihf-gcc -mfpu=neon -O2 -S int_average.c arm-linux-gnueabihf-gcc -mfpu=neon -O3 -S int_average.c

int_average.c

int int_average(int* array, int size)
{
	int i;
	long long total = 0;

	if (size <= 0) {
		return 0;
	}
	for (i = 0; i < size; i++) {
		total += array[i];
	}
	return total / size;
}

int_average.s.O2

	.syntax unified
	.arch armv7-a
	.eabi_attribute 27, 3
	.eabi_attribute 28, 1
	.fpu neon
	.eabi_attribute 20, 1
	.eabi_attribute 21, 1
	.eabi_attribute 23, 3
	.eabi_attribute 24, 1
	.eabi_attribute 25, 1
	.eabi_attribute 26, 2
	.eabi_attribute 30, 2
	.eabi_attribute 34, 1
	.eabi_attribute 18, 4
	.thumb
	.file	"int_average.c"
	.global	__aeabi_ldivmod
	.text
	.align	2
	.global	int_average
	.thumb
	.thumb_func
	.type	int_average, %function
int_average:
	@ args = 0, pretend = 0, frame = 0
	@ frame_needed = 0, uses_anonymous_args = 0
	push	{r3, r4, r5, lr}
	subs	r3, r1, #0
	ble	.L4
	subs	r2, r0, #4
	movs	r1, #0
	movs	r0, #0
	movs	r4, #0
.L3:
	ldr	r5, [r2, #4]!
	adds	r4, r4, #1
	adds	r0, r0, r5
	adc	r1, r1, r5, asr #31
	cmp	r4, r3
	bne	.L3
	mov	r2, r4
	asrs	r3, r4, #31
	bl	__aeabi_ldivmod
	pop	{r3, r4, r5, pc}
.L4:
	movs	r0, #0
	pop	{r3, r4, r5, pc}
	.size	int_average, .-int_average
	.ident	"GCC: (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3"
	.section	.note.GNU-stack,"",%progbits

int_average.s.O3

	.syntax unified
	.arch armv7-a
	.eabi_attribute 27, 3
	.eabi_attribute 28, 1
	.fpu neon
	.eabi_attribute 20, 1
	.eabi_attribute 21, 1
	.eabi_attribute 23, 3
	.eabi_attribute 24, 1
	.eabi_attribute 25, 1
	.eabi_attribute 26, 2
	.eabi_attribute 30, 2
	.eabi_attribute 34, 1
	.eabi_attribute 18, 4
	.thumb
	.file	"int_average.c"
	.global	__aeabi_ldivmod
	.text
	.align	2
	.global	int_average
	.thumb
	.thumb_func
	.type	int_average, %function
int_average:
	@ args = 0, pretend = 0, frame = 0
	@ frame_needed = 0, uses_anonymous_args = 0
	cmp	r1, #0
	push	{r3, r4, r5, r6, r7, lr}
	ble	.L10
	ubfx	r2, r0, #2, #2
	negs	r2, r2
	and	r2, r2, #3
	cmp	r2, r1
	it	cs
	movcs	r2, r1
	cmp	r2, #0
	beq	.L11
	subs	r3, r0, #4
	movs	r4, #0
	movs	r5, #0
	movs	r6, #0
.L4:
	ldr	r7, [r3, #4]!
	adds	r6, r6, #1
	adds	r4, r4, r7
	adc	r5, r5, r7, asr #31
	cmp	r6, r2
	bcc	.L4
	cmp	r1, r2
	beq	.L5
.L3:
	rsb	lr, r2, r1
	lsr	ip, lr, #2
	lsl	r7, ip, #2
	cmp	ip, #0
	beq	.L6
	vmov.i32	q9, #0  @ v2di
	add	r2, r0, r2, lsl #2
	movs	r3, #0
.L7:
	vldmia	r2!, {d16-d17}
	vmovl.s32 q10, d16
	adds	r3, r3, #1
	vmovl.s32 q8, d17
	vadd.i64	q9, q10, q9
	cmp	r3, ip
	vadd.i64	q9, q8, q9
	bcc	.L7
	vadd.i64	d18, d18, d19
	adds	r6, r6, r7
	vmov	r2, r3, d18  @ v2di
	adds	r4, r4, r2
	adc	r5, r5, r3
	cmp	lr, r7
	beq	.L5
.L6:
	add	r0, r0, r6, lsl #2
.L9:
	ldr	r3, [r0], #4
	adds	r6, r6, #1
	adds	r4, r4, r3
	adc	r5, r5, r3, asr #31
	cmp	r1, r6
	bgt	.L9
.L5:
	mov	r2, r1
	mov	r0, r4
	mov	r1, r5
	asrs	r3, r2, #31
	bl	__aeabi_ldivmod
	pop	{r3, r4, r5, r6, r7, pc}
.L10:
	movs	r0, #0
	pop	{r3, r4, r5, r6, r7, pc}
.L11:
	movs	r4, #0
	movs	r5, #0
	mov	r6, r2
	b	.L3
	.size	int_average, .-int_average
	.ident	"GCC: (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3"
	.section	.note.GNU-stack,"",%progbits