neon.s

	.arch armv7-a
	.eabi_attribute 27, 3
	.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, 6
	.eabi_attribute 18, 2
	.file	"neon.c"
	.section	.rodata
	.align	2
.LC0:
	.ascii	"Time for compute : %d microseconds\012\000"
	.text
	.align	2
	.global	main
	.type	main, %function
main:
	.fnstart
.LFB1870:
	@ args = 0, pretend = 0, frame = 16024
	@ frame_needed = 1, uses_anonymous_args = 0
	stmfd	sp!, {fp, lr}
	.save {fp, lr}
	.setfp fp, sp, #4
	add	fp, sp, #4
	.pad #16000
	sub	sp, sp, #16000
	.pad #24
	sub	sp, sp, #24
	sub	r3, fp, #16000
	sub	r3, r3, #4
	sub	r3, r3, #24
	str	r3, [fp, #-8]
	mov	r3, #0
	str	r3, [fp, #-12]
	b	.L2
.L3:
	ldr	r1, [fp, #-12]
	ldr	r3, [fp, #-12]
	add	r2, r3, #1
	movw	r3, #:lower16:-16024
	movt	r3, #:upper16:-16024
	mov	r1, r1, asl #2
	sub	r0, fp, #4
	add	r1, r1, r0
	add	r3, r1, r3
	str	r2, [r3, #0]
	ldr	r3, [fp, #-12]
	add	r3, r3, #1
	str	r3, [fp, #-12]
.L2:
	ldr	r2, [fp, #-12]
	movw	r3, #3999
	cmp	r2, r3
	bls	.L3
	sub	r3, fp, #20
	mov	r0, r3
	mov	r1, #0
	bl	gettimeofday
	sub	r3, fp, #16000
	sub	r3, r3, #4
	sub	r3, r3, #24
	mov	r0, r3
	mov	r1, #4000
	bl	double_elements
	sub	r3, fp, #28
	mov	r0, r3
	mov	r1, #0
	bl	gettimeofday
	mov	r0, #10
	bl	putchar
	movw	r3, #:lower16:.LC0
	movt	r3, #:upper16:.LC0
	ldr	r1, [fp, #-24]
	ldr	r2, [fp, #-16]
	rsb	r1, r2, r1
	ldr	r0, [fp, #-28]
	ldr	r2, [fp, #-20]
	rsb	r0, r2, r0
	movw	r2, #:lower16:1000000
	movt	r2, #:upper16:1000000
	mul	r2, r2, r0
	add	r2, r1, r2
	mov	r0, r3
	mov	r1, r2
	bl	printf
	mov	r3, #0
	mov	r0, r3
	sub	sp, fp, #4
	ldmfd	sp!, {fp, pc}
	.fnend
	.size	main, .-main
	.align	2
	.global	double_elements
	.type	double_elements, %function
double_elements:
	.fnstart
.LFB1871:
	@ args = 0, pretend = 0, frame = 96
	@ frame_needed = 1, uses_anonymous_args = 0
	@ link register save eliminated.
	str	fp, [sp, #-4]!
	.save {fp}
	.setfp fp, sp, #0
	add	fp, sp, #0
	.pad #100
	sub	sp, sp, #100
	str	r0, [fp, #-96]
	str	r1, [fp, #-100]
	mov	r3, #0
	str	r3, [fp, #-88]
	b	.L6
.L7:
	ldr	r3, [fp, #-96]
	str	r3, [fp, #-52]
	ldr	r3, [fp, #-52]
	vld1.32	{d16-d17}, [r3]
	vstr	d16, [fp, #-84]
	vstr	d17, [fp, #-76]
	vldr	d16, [fp, #-84]
	vldr	d17, [fp, #-76]
	vstr	d16, [fp, #-44]
	vstr	d17, [fp, #-36]
	vldr	d16, [fp, #-44]
	vldr	d17, [fp, #-36]
	vshl.i32	q8, q8, #1
	vstr	d16, [fp, #-68]
	vstr	d17, [fp, #-60]
	ldr	r3, [fp, #-96]
	str	r3, [fp, #-8]
	vldr	d16, [fp, #-68]
	vldr	d17, [fp, #-60]
	vstr	d16, [fp, #-28]
	vstr	d17, [fp, #-20]
	ldr	r3, [fp, #-8]
	vldr	d16, [fp, #-28]
	vldr	d17, [fp, #-20]
	vst1.32	{d16-d17}, [r3]
	ldr	r3, [fp, #-96]
	add	r3, r3, #16
	str	r3, [fp, #-96]
	ldr	r3, [fp, #-88]
	add	r3, r3, #1
	str	r3, [fp, #-88]
.L6:
	ldr	r3, [fp, #-100]
	mov	r2, r3, lsr #2
	ldr	r3, [fp, #-88]
	cmp	r2, r3
	bhi	.L7
	add	sp, fp, #0
	ldmfd	sp!, {fp}
	bx	lr
	.fnend
	.size	double_elements, .-double_elements
	.ident	"GCC: (Linux release 2010-06-30) 4.4.1"
	.section	.note.GNU-stack,"",%progbits

arm-v7a8-linux-gnueabi-gcc -S -mfpu=neon neon.c