michael@0: ;
michael@0: ; jfsseflt.asm - floating-point FDCT (SSE)
michael@0: ;
michael@0: ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0: ;
michael@0: ; Based on
michael@0: ; x86 SIMD extension for IJG JPEG library
michael@0: ; Copyright (C) 1999-2006, MIYASAKA Masaru.
michael@0: ; For conditions of distribution and use, see copyright notice in jsimdext.inc
michael@0: ;
michael@0: ; This file should be assembled with NASM (Netwide Assembler),
michael@0: ; can *not* be assembled with Microsoft's MASM or any compatible
michael@0: ; assembler (including Borland's Turbo Assembler).
michael@0: ; NASM is available from http://nasm.sourceforge.net/ or
michael@0: ; http://sourceforge.net/project/showfiles.php?group_id=6208
michael@0: ;
michael@0: ; This file contains a floating-point implementation of the forward DCT
michael@0: ; (Discrete Cosine Transform). The following code is based directly on
michael@0: ; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
michael@0: ;
michael@0: ; [TAB8]
michael@0: 
michael@0: %include "jsimdext.inc"
michael@0: %include "jdct.inc"
michael@0: 
michael@0: ; --------------------------------------------------------------------------
michael@0: 
michael@0: %macro	unpcklps2 2	; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
michael@0: 	shufps	%1,%2,0x44
michael@0: %endmacro
michael@0: 
michael@0: %macro	unpckhps2 2	; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
michael@0: 	shufps	%1,%2,0xEE
michael@0: %endmacro
michael@0: 
michael@0: ; --------------------------------------------------------------------------
michael@0: 	SECTION	SEG_CONST
michael@0: 
michael@0: 	alignz	16
michael@0: 	global	EXTN(jconst_fdct_float_sse)
michael@0: 
michael@0: EXTN(jconst_fdct_float_sse):
michael@0: 
michael@0: PD_0_382	times 4 dd  0.382683432365089771728460
michael@0: PD_0_707	times 4 dd  0.707106781186547524400844
michael@0: PD_0_541	times 4 dd  0.541196100146196984399723
michael@0: PD_1_306	times 4 dd  1.306562964876376527856643
michael@0: 
michael@0: 	alignz	16
michael@0: 
michael@0: ; --------------------------------------------------------------------------
michael@0: 	SECTION	SEG_TEXT
michael@0: 	BITS	32
michael@0: ;
michael@0: ; Perform the forward DCT on one block of samples.
michael@0: ;
michael@0: ; GLOBAL(void)
michael@0: ; jsimd_fdct_float_sse (FAST_FLOAT * data)
michael@0: ;
michael@0: 
michael@0: %define data(b)		(b)+8		; FAST_FLOAT * data
michael@0: 
michael@0: %define original_ebp	ebp+0
michael@0: %define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
michael@0: %define WK_NUM		2
michael@0: 
michael@0: 	align	16
michael@0: 	global	EXTN(jsimd_fdct_float_sse)
michael@0: 
michael@0: EXTN(jsimd_fdct_float_sse):
michael@0: 	push	ebp
michael@0: 	mov	eax,esp				; eax = original ebp
michael@0: 	sub	esp, byte 4
michael@0: 	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
michael@0: 	mov	[esp],eax
michael@0: 	mov	ebp,esp				; ebp = aligned ebp
michael@0: 	lea	esp, [wk(0)]
michael@0: 	pushpic	ebx
michael@0: ;	push	ecx		; need not be preserved
michael@0: ;	push	edx		; need not be preserved
michael@0: ;	push	esi		; unused
michael@0: ;	push	edi		; unused
michael@0: 
michael@0: 	get_GOT	ebx		; get GOT address
michael@0: 
michael@0: 	; ---- Pass 1: process rows.
michael@0: 
michael@0: 	mov	edx, POINTER [data(eax)]	; (FAST_FLOAT *)
michael@0: 	mov	ecx, DCTSIZE/4
michael@0: 	alignx	16,7
michael@0: .rowloop:
michael@0: 
michael@0: 	movaps	xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm2, XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm3, XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)]
michael@0: 
michael@0: 	; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
michael@0: 	; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
michael@0: 
michael@0: 	movaps   xmm4,xmm0		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm0,xmm1		; xmm0=(20 30 21 31)
michael@0: 	unpckhps xmm4,xmm1		; xmm4=(22 32 23 33)
michael@0: 	movaps   xmm5,xmm2		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm2,xmm3		; xmm2=(24 34 25 35)
michael@0: 	unpckhps xmm5,xmm3		; xmm5=(26 36 27 37)
michael@0: 
michael@0: 	movaps	xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]
michael@0: 
michael@0: 	; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
michael@0: 	; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
michael@0: 
michael@0: 	movaps	XMMWORD [wk(0)], xmm4	; wk(0)=(22 32 23 33)
michael@0: 	movaps	XMMWORD [wk(1)], xmm2	; wk(1)=(24 34 25 35)
michael@0: 
michael@0: 	movaps   xmm4,xmm6		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm6,xmm7		; xmm6=(00 10 01 11)
michael@0: 	unpckhps xmm4,xmm7		; xmm4=(02 12 03 13)
michael@0: 	movaps   xmm2,xmm1		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm1,xmm3		; xmm1=(04 14 05 15)
michael@0: 	unpckhps xmm2,xmm3		; xmm2=(06 16 07 17)
michael@0: 
michael@0: 	movaps    xmm7,xmm6		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm6,xmm0		; xmm6=(00 10 20 30)=data0
michael@0: 	unpckhps2 xmm7,xmm0		; xmm7=(01 11 21 31)=data1
michael@0: 	movaps    xmm3,xmm2		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm2,xmm5		; xmm2=(06 16 26 36)=data6
michael@0: 	unpckhps2 xmm3,xmm5		; xmm3=(07 17 27 37)=data7
michael@0: 
michael@0: 	movaps	xmm0,xmm7
michael@0: 	movaps	xmm5,xmm6
michael@0: 	subps	xmm7,xmm2		; xmm7=data1-data6=tmp6
michael@0: 	subps	xmm6,xmm3		; xmm6=data0-data7=tmp7
michael@0: 	addps	xmm0,xmm2		; xmm0=data1+data6=tmp1
michael@0: 	addps	xmm5,xmm3		; xmm5=data0+data7=tmp0
michael@0: 
michael@0: 	movaps	xmm2, XMMWORD [wk(0)]	; xmm2=(22 32 23 33)
michael@0: 	movaps	xmm3, XMMWORD [wk(1)]	; xmm3=(24 34 25 35)
michael@0: 	movaps	XMMWORD [wk(0)], xmm7	; wk(0)=tmp6
michael@0: 	movaps	XMMWORD [wk(1)], xmm6	; wk(1)=tmp7
michael@0: 
michael@0: 	movaps    xmm7,xmm4		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm4,xmm2		; xmm4=(02 12 22 32)=data2
michael@0: 	unpckhps2 xmm7,xmm2		; xmm7=(03 13 23 33)=data3
michael@0: 	movaps    xmm6,xmm1		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm1,xmm3		; xmm1=(04 14 24 34)=data4
michael@0: 	unpckhps2 xmm6,xmm3		; xmm6=(05 15 25 35)=data5
michael@0: 
michael@0: 	movaps	xmm2,xmm7
michael@0: 	movaps	xmm3,xmm4
michael@0: 	addps	xmm7,xmm1		; xmm7=data3+data4=tmp3
michael@0: 	addps	xmm4,xmm6		; xmm4=data2+data5=tmp2
michael@0: 	subps	xmm2,xmm1		; xmm2=data3-data4=tmp4
michael@0: 	subps	xmm3,xmm6		; xmm3=data2-data5=tmp5
michael@0: 
michael@0: 	; -- Even part
michael@0: 
michael@0: 	movaps	xmm1,xmm5
michael@0: 	movaps	xmm6,xmm0
michael@0: 	subps	xmm5,xmm7		; xmm5=tmp13
michael@0: 	subps	xmm0,xmm4		; xmm0=tmp12
michael@0: 	addps	xmm1,xmm7		; xmm1=tmp10
michael@0: 	addps	xmm6,xmm4		; xmm6=tmp11
michael@0: 
michael@0: 	addps	xmm0,xmm5
michael@0: 	mulps	xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1
michael@0: 
michael@0: 	movaps	xmm7,xmm1
michael@0: 	movaps	xmm4,xmm5
michael@0: 	subps	xmm1,xmm6		; xmm1=data4
michael@0: 	subps	xmm5,xmm0		; xmm5=data6
michael@0: 	addps	xmm7,xmm6		; xmm7=data0
michael@0: 	addps	xmm4,xmm0		; xmm4=data2
michael@0: 
michael@0: 	movaps	XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], xmm1
michael@0: 	movaps	XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)], xmm5
michael@0: 	movaps	XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
michael@0: 	movaps	XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
michael@0: 
michael@0: 	; -- Odd part
michael@0: 
michael@0: 	movaps	xmm6, XMMWORD [wk(0)]	; xmm6=tmp6
michael@0: 	movaps	xmm0, XMMWORD [wk(1)]	; xmm0=tmp7
michael@0: 
michael@0: 	addps	xmm2,xmm3		; xmm2=tmp10
michael@0: 	addps	xmm3,xmm6		; xmm3=tmp11
michael@0: 	addps	xmm6,xmm0		; xmm6=tmp12, xmm0=tmp7
michael@0: 
michael@0: 	mulps	xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3
michael@0: 
michael@0: 	movaps	xmm1,xmm2		; xmm1=tmp10
michael@0: 	subps	xmm2,xmm6
michael@0: 	mulps	xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
michael@0: 	mulps	xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
michael@0: 	mulps	xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
michael@0: 	addps	xmm1,xmm2		; xmm1=z2
michael@0: 	addps	xmm6,xmm2		; xmm6=z4
michael@0: 
michael@0: 	movaps	xmm5,xmm0
michael@0: 	subps	xmm0,xmm3		; xmm0=z13
michael@0: 	addps	xmm5,xmm3		; xmm5=z11
michael@0: 
michael@0: 	movaps	xmm7,xmm0
michael@0: 	movaps	xmm4,xmm5
michael@0: 	subps	xmm0,xmm1		; xmm0=data3
michael@0: 	subps	xmm5,xmm6		; xmm5=data7
michael@0: 	addps	xmm7,xmm1		; xmm7=data5
michael@0: 	addps	xmm4,xmm6		; xmm4=data1
michael@0: 
michael@0: 	movaps	XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
michael@0: 	movaps	XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)], xmm5
michael@0: 	movaps	XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], xmm7
michael@0: 	movaps	XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
michael@0: 
michael@0: 	add	edx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
michael@0: 	dec	ecx
michael@0: 	jnz	near .rowloop
michael@0: 
michael@0: 	; ---- Pass 2: process columns.
michael@0: 
michael@0: 	mov	edx, POINTER [data(eax)]	; (FAST_FLOAT *)
michael@0: 	mov	ecx, DCTSIZE/4
michael@0: 	alignx	16,7
michael@0: .columnloop:
michael@0: 
michael@0: 	movaps	xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm2, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 
michael@0: 	; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
michael@0: 	; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
michael@0: 
michael@0: 	movaps   xmm4,xmm0		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm0,xmm1		; xmm0=(02 03 12 13)
michael@0: 	unpckhps xmm4,xmm1		; xmm4=(22 23 32 33)
michael@0: 	movaps   xmm5,xmm2		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm2,xmm3		; xmm2=(42 43 52 53)
michael@0: 	unpckhps xmm5,xmm3		; xmm5=(62 63 72 73)
michael@0: 
michael@0: 	movaps	xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm1, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 	movaps	xmm3, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)]
michael@0: 
michael@0: 	; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
michael@0: 	; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
michael@0: 
michael@0: 	movaps	XMMWORD [wk(0)], xmm4	; wk(0)=(22 23 32 33)
michael@0: 	movaps	XMMWORD [wk(1)], xmm2	; wk(1)=(42 43 52 53)
michael@0: 
michael@0: 	movaps   xmm4,xmm6		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm6,xmm7		; xmm6=(00 01 10 11)
michael@0: 	unpckhps xmm4,xmm7		; xmm4=(20 21 30 31)
michael@0: 	movaps   xmm2,xmm1		; transpose coefficients(phase 1)
michael@0: 	unpcklps xmm1,xmm3		; xmm1=(40 41 50 51)
michael@0: 	unpckhps xmm2,xmm3		; xmm2=(60 61 70 71)
michael@0: 
michael@0: 	movaps    xmm7,xmm6		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm6,xmm0		; xmm6=(00 01 02 03)=data0
michael@0: 	unpckhps2 xmm7,xmm0		; xmm7=(10 11 12 13)=data1
michael@0: 	movaps    xmm3,xmm2		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm2,xmm5		; xmm2=(60 61 62 63)=data6
michael@0: 	unpckhps2 xmm3,xmm5		; xmm3=(70 71 72 73)=data7
michael@0: 
michael@0: 	movaps	xmm0,xmm7
michael@0: 	movaps	xmm5,xmm6
michael@0: 	subps	xmm7,xmm2		; xmm7=data1-data6=tmp6
michael@0: 	subps	xmm6,xmm3		; xmm6=data0-data7=tmp7
michael@0: 	addps	xmm0,xmm2		; xmm0=data1+data6=tmp1
michael@0: 	addps	xmm5,xmm3		; xmm5=data0+data7=tmp0
michael@0: 
michael@0: 	movaps	xmm2, XMMWORD [wk(0)]	; xmm2=(22 23 32 33)
michael@0: 	movaps	xmm3, XMMWORD [wk(1)]	; xmm3=(42 43 52 53)
michael@0: 	movaps	XMMWORD [wk(0)], xmm7	; wk(0)=tmp6
michael@0: 	movaps	XMMWORD [wk(1)], xmm6	; wk(1)=tmp7
michael@0: 
michael@0: 	movaps    xmm7,xmm4		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm4,xmm2		; xmm4=(20 21 22 23)=data2
michael@0: 	unpckhps2 xmm7,xmm2		; xmm7=(30 31 32 33)=data3
michael@0: 	movaps    xmm6,xmm1		; transpose coefficients(phase 2)
michael@0: 	unpcklps2 xmm1,xmm3		; xmm1=(40 41 42 43)=data4
michael@0: 	unpckhps2 xmm6,xmm3		; xmm6=(50 51 52 53)=data5
michael@0: 
michael@0: 	movaps	xmm2,xmm7
michael@0: 	movaps	xmm3,xmm4
michael@0: 	addps	xmm7,xmm1		; xmm7=data3+data4=tmp3
michael@0: 	addps	xmm4,xmm6		; xmm4=data2+data5=tmp2
michael@0: 	subps	xmm2,xmm1		; xmm2=data3-data4=tmp4
michael@0: 	subps	xmm3,xmm6		; xmm3=data2-data5=tmp5
michael@0: 
michael@0: 	; -- Even part
michael@0: 
michael@0: 	movaps	xmm1,xmm5
michael@0: 	movaps	xmm6,xmm0
michael@0: 	subps	xmm5,xmm7		; xmm5=tmp13
michael@0: 	subps	xmm0,xmm4		; xmm0=tmp12
michael@0: 	addps	xmm1,xmm7		; xmm1=tmp10
michael@0: 	addps	xmm6,xmm4		; xmm6=tmp11
michael@0: 
michael@0: 	addps	xmm0,xmm5
michael@0: 	mulps	xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1
michael@0: 
michael@0: 	movaps	xmm7,xmm1
michael@0: 	movaps	xmm4,xmm5
michael@0: 	subps	xmm1,xmm6		; xmm1=data4
michael@0: 	subps	xmm5,xmm0		; xmm5=data6
michael@0: 	addps	xmm7,xmm6		; xmm7=data0
michael@0: 	addps	xmm4,xmm0		; xmm4=data2
michael@0: 
michael@0: 	movaps	XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], xmm1
michael@0: 	movaps	XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], xmm5
michael@0: 	movaps	XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
michael@0: 	movaps	XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4
michael@0: 
michael@0: 	; -- Odd part
michael@0: 
michael@0: 	movaps	xmm6, XMMWORD [wk(0)]	; xmm6=tmp6
michael@0: 	movaps	xmm0, XMMWORD [wk(1)]	; xmm0=tmp7
michael@0: 
michael@0: 	addps	xmm2,xmm3		; xmm2=tmp10
michael@0: 	addps	xmm3,xmm6		; xmm3=tmp11
michael@0: 	addps	xmm6,xmm0		; xmm6=tmp12, xmm0=tmp7
michael@0: 
michael@0: 	mulps	xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3
michael@0: 
michael@0: 	movaps	xmm1,xmm2		; xmm1=tmp10
michael@0: 	subps	xmm2,xmm6
michael@0: 	mulps	xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
michael@0: 	mulps	xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
michael@0: 	mulps	xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
michael@0: 	addps	xmm1,xmm2		; xmm1=z2
michael@0: 	addps	xmm6,xmm2		; xmm6=z4
michael@0: 
michael@0: 	movaps	xmm5,xmm0
michael@0: 	subps	xmm0,xmm3		; xmm0=z13
michael@0: 	addps	xmm5,xmm3		; xmm5=z11
michael@0: 
michael@0: 	movaps	xmm7,xmm0
michael@0: 	movaps	xmm4,xmm5
michael@0: 	subps	xmm0,xmm1		; xmm0=data3
michael@0: 	subps	xmm5,xmm6		; xmm5=data7
michael@0: 	addps	xmm7,xmm1		; xmm7=data5
michael@0: 	addps	xmm4,xmm6		; xmm4=data1
michael@0: 
michael@0: 	movaps	XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
michael@0: 	movaps	XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], xmm5
michael@0: 	movaps	XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], xmm7
michael@0: 	movaps	XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4
michael@0: 
michael@0: 	add	edx, byte 4*SIZEOF_FAST_FLOAT
michael@0: 	dec	ecx
michael@0: 	jnz	near .columnloop
michael@0: 
michael@0: ;	pop	edi		; unused
michael@0: ;	pop	esi		; unused
michael@0: ;	pop	edx		; need not be preserved
michael@0: ;	pop	ecx		; need not be preserved
michael@0: 	poppic	ebx
michael@0: 	mov	esp,ebp		; esp <- aligned ebp
michael@0: 	pop	esp		; esp <- original ebp
michael@0: 	pop	ebp
michael@0: 	ret
michael@0: 
michael@0: ; For some reason, the OS X linker does not honor the request to align the
michael@0: ; segment unless we do this.
michael@0: 	align	16