C code :
    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x++)
            dst[x] = src[x] << (14 - BIT_DEPTH);
        src += srcstride;
        dst += dststride;
    }
asm :

;	r0 : *dst
;	r1 : dststride
;	r2 : *src
; 	r3 : srcstride
;	r4 : width
;	r5 : height

cglobal put_hevc_mc_pixels_8, 9, 12
	pxor		xmm0,xmm0		;set register at zero
	mov			r6,0			;height
	;8 by 8
mc_pixels_h:	;for height
	mov		r7,0			;width

mc_pixels_w:	;for width
	mov 		r9,0
	mov			r9,[r2+r7]
	shl			r9,6
	mov			[r0+r7],r9
	inc			r7
	cmp			r7, r4			;cmp width
	jl			mc_pixels_w		;width loop
	add			r0,r1			;dst += dststride
	add			r2,r3			;src += srcstride
	inc			r6
	cmp			r6,r5			;cmp height
	jl			mc_pixels_h		;height loop
    REP_RET