Bubsy's code with __restrict__

	.file	"tmp1.cc"
	.section	.text.unlikely,"ax",@progbits
.LCOLDB0:
	.text
.LHOTB0:
	.p2align 4,,15
	.globl	_Z7upscalePKjPj
	.type	_Z7upscalePKjPj, @function
_Z7upscalePKjPj:
.LFB0:
	.cfi_startproc
	leaq	1228800(%rsi), %r8
	.p2align 4,,10
	.p2align 3
.L3:
	leaq	2560(%rsi), %rcx
	movq	%rsi, %rax
	movq	%rdi, %rdx
	.p2align 4,,10
	.p2align 3
.L2:
	movdqu	(%rdx), %xmm0
	addq	$32, %rax
	addq	$16, %rdx
	pshufd	$80, %xmm0, %xmm1
	pshufd	$250, %xmm0, %xmm0
	movups	%xmm1, -32(%rax)
	movups	%xmm0, -16(%rax)
	movups	%xmm1, 2528(%rax)
	movups	%xmm0, 2544(%rax)
	cmpq	%rax, %rcx
	jne	.L2
	addq	$5120, %rsi
	addq	$1280, %rdi
	cmpq	%r8, %rsi
	jne	.L3
	rep ret
	.cfi_endproc
.LFE0:
	.size	_Z7upscalePKjPj, .-_Z7upscalePKjPj
	.section	.text.unlikely
.LCOLDE0:
	.text
.LHOTE0:
	.section	.text.unlikely
.LCOLDB1:
	.text
.LHOTB1:
	.p2align 4,,15
	.globl	_Z8upscale2PKjPj
	.type	_Z8upscale2PKjPj, @function
_Z8upscale2PKjPj:
.LFB1:
	.cfi_startproc
	leaq	1228800(%rsi), %r8
	.p2align 4,,10
	.p2align 3
.L9:
	leaq	2560(%rsi), %rcx
	movq	%rsi, %rax
	movq	%rdi, %rdx
	.p2align 4,,10
	.p2align 3
.L8:
	movdqu	(%rdx), %xmm0
	addq	$32, %rax
	addq	$16, %rdx
	pshufd	$80, %xmm0, %xmm1
	pshufd	$250, %xmm0, %xmm0
	movups	%xmm1, -32(%rax)
	movups	%xmm0, -16(%rax)
	movups	%xmm1, 2528(%rax)
	movups	%xmm0, 2544(%rax)
	cmpq	%rax, %rcx
	jne	.L8
	addq	$5120, %rsi
	addq	$1280, %rdi
	cmpq	%r8, %rsi
	jne	.L9
	rep ret
	.cfi_endproc
.LFE1:
	.size	_Z8upscale2PKjPj, .-_Z8upscale2PKjPj
	.section	.text.unlikely
.LCOLDE1:
	.text
.LHOTE1:
	.ident	"GCC: (Debian 4.9.1-16) 4.9.1"
	.section	.note.GNU-stack,"",@progbits