u3copy_in_user.s

该文件是rt_linux

	/* We only use the first loop if len > (7 * 64). */
	subcc		%o4, 0x1c0, %o4			! A0	Group17
	bg,pt		%icc, U3copy_in_user_loop1	! BR
	 add		%o1, 0x40, %o1			! A1

	add		%o4, 0x140, %o4			! A0	Group18
	ba,pt		%xcc, U3copy_in_user_loop2	! BR
	 srl		%o4, 6, %o3			! A0	Group19
	nop
	nop
	nop
	nop
	nop

	nop
	nop

	/* This loop performs the copy and queues new prefetches.
	 * We drop into the second loop when len <= (5 * 64).  Note
	 * that this (5 * 64) factor has been subtracted from len
	 * already.
	 */
U3copy_in_user_loop1:
	EXBLK1(ldda [%o1 + 0x008] %asi, %f2)		! MS	Group2	(%f2 results at G5)
	faligndata	%f12, %f14, %f28		! FGA		(%f28 results at G5)
	EXBLK1(ldda [%o1 + 0x010] %asi, %f4)		! MS	Group3	(%f4 results at G6)
	faligndata	%f14, %f0, %f30			! FGA	Group4	(1-cycle stall, %f30 at G7)
	EXBLK1(stda %f16, [%o0] ASI_BLK_AIUS)		! MS
	EXBLK1(ldda [%o1 + 0x018] %asi, %f6)		! AX		(%f6 results at G7)

	faligndata	%f0, %f2, %f16			! FGA	Group12 (7-cycle stall)
	EXBLK1(ldda [%o1 + 0x020] %asi, %f8)		! MS	        (%f8 results at G15)
	faligndata	%f2, %f4, %f18			! FGA	Group13	(%f18 results at G16)
	EXBLK1(ldda [%o1 + 0x028] %asi, %f10)		! MS		(%f10 results at G16)
	faligndata	%f4, %f6, %f20			! FGA	Group14	(%f20 results at G17)
	EXBLK1(ldda [%o1 + 0x030] %asi, %f12)		! MS		(%f12 results at G17)
	faligndata	%f6, %f8, %f22			! FGA	Group15	(%f22 results at G18)
	EXBLK1(ldda [%o1 + 0x038] %asi, %f14)		! MS		(%f14 results at G18)

	faligndata	%f8, %f10, %f24			! FGA	Group16	(%f24 results at G19)
	EXBLK1(ldda [%o1 + 0x040] %asi, %f0)		! AX		(%f0 results at G19)
	prefetcha	[%o1 + 0x180] %asi, #one_read	! MS
	faligndata	%f10, %f12, %f26		! FGA	Group17	(%f26 results at G20)
	subcc		%o4, 0x40, %o4			! A0
	add		%o1, 0x40, %o1			! A1
	bg,pt		%xcc, U3copy_in_user_loop1		! BR
	 add		%o0, 0x40, %o0			! A0	Group18

U3copy_in_user_loop2_enter:
	mov		5, %o3				! A1

	/* This loop performs on the copy, no new prefetches are
	 * queued.  We do things this way so that we do not perform
	 * any spurious prefetches past the end of the src buffer.
	 */
U3copy_in_user_loop2:
	EXBLK2(ldda [%o1 + 0x008] %asi, %f2)		! MS
	faligndata	%f12, %f14, %f28		! FGA	Group2
	EXBLK2(ldda [%o1 + 0x010] %asi, %f4)		! MS
	faligndata	%f14, %f0, %f30			! FGA	Group4	(1-cycle stall)
	EXBLK2(stda %f16, [%o0] ASI_BLK_AIUS)		! MS
	EXBLK2(ldda [%o1 + 0x018] %asi, %f6)		! AX
	faligndata	%f0, %f2, %f16			! FGA	Group12 (7-cycle stall)

	EXBLK2(ldda [%o1 + 0x020] %asi, %f8)		! MS
	faligndata	%f2, %f4, %f18			! FGA	Group13
	EXBLK2(ldda [%o1 + 0x028] %asi, %f10)		! MS
	faligndata	%f4, %f6, %f20			! FGA	Group14
	EXBLK2(ldda [%o1 + 0x030] %asi, %f12)		! MS
	faligndata	%f6, %f8, %f22			! FGA	Group15
	EXBLK2(ldda [%o1 + 0x038] %asi, %f14)		! MS
	faligndata	%f8, %f10, %f24			! FGA	Group16

	EXBLK2(ldda [%o1 + 0x040] %asi, %f0)		! AX
	faligndata	%f10, %f12, %f26		! FGA	Group17
	subcc		%o3, 0x01, %o3			! A0
	add		%o1, 0x40, %o1			! A1
	bg,pt		%xcc, U3copy_in_user_loop2	! BR
	 add		%o0, 0x40, %o0			! A0	Group18

	/* Finally we copy the last full 64-byte block. */
U3copy_in_user_loopfini:
	EXBLK3(ldda [%o1 + 0x008] %asi, %f2)		! MS
	faligndata	%f12, %f14, %f28		! FGA
	EXBLK3(ldda [%o1 + 0x010] %asi, %f4)		! MS	Group19
	faligndata	%f14, %f0, %f30			! FGA
	EXBLK3(stda %f16, [%o0] ASI_BLK_AIUS)		! MS	Group20
	EXBLK4(ldda [%o1 + 0x018] %asi, %f6)		! AX
	faligndata	%f0, %f2, %f16			! FGA	Group11 (7-cycle stall)
	EXBLK4(ldda [%o1 + 0x020] %asi, %f8)		! MS
	faligndata	%f2, %f4, %f18			! FGA	Group12
	EXBLK4(ldda [%o1 + 0x028] %asi, %f10)		! MS
	faligndata	%f4, %f6, %f20			! FGA	Group13
	EXBLK4(ldda [%o1 + 0x030] %asi, %f12)		! MS
	faligndata	%f6, %f8, %f22			! FGA	Group14
	EXBLK4(ldda [%o1 + 0x038] %asi, %f14)		! MS
	faligndata	%f8, %f10, %f24			! FGA	Group15
	cmp		%g1, 0				! A0
	be,pt		%icc, 1f			! BR
	 add		%o0, 0x40, %o0			! A1
	EXBLK4(ldda [%o1 + 0x040] %asi, %f0)		! MS
1:	faligndata	%f10, %f12, %f26		! FGA	Group16
	faligndata	%f12, %f14, %f28		! FGA	Group17
	faligndata	%f14, %f0, %f30			! FGA	Group18
	EXBLK4(stda %f16, [%o0] ASI_BLK_AIUS)		! MS
	add		%o0, 0x40, %o0			! A0
	add		%o1, 0x40, %o1			! A1
	membar		#Sync				! MS	Group26 (7-cycle stall)

	/* Now we copy the (len modulo 64) bytes at the end.
	 * Note how we borrow the %f0 loaded above.
	 *
	 * Also notice how this code is careful not to perform a
	 * load past the end of the src buffer just like similar
	 * code found in U3copy_in_user_toosmall processing.
	 */
U3copy_in_user_loopend:
	and		%o2, 0x3f, %o2			! A0	Group
	andcc		%o2, 0x38, %g2			! A0	Group
	be,pn		%icc, U3copy_in_user_endcruft	! BR
	 subcc		%g2, 0x8, %g2			! A1
	be,pn		%icc, U3copy_in_user_endcruft	! BR	Group
	 cmp		%g1, 0				! A0

	be,a,pt		%icc, 1f			! BR	Group
	 EX(ldda [%o1 + 0x00] %asi, %f0, add %o2, %g0)	! MS

1:	EX(ldda [%o1 + 0x08] %asi, %f2, add %o2, %g0)	! MS	Group
	add		%o1, 0x8, %o1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f0, %f2, %f8			! FGA	Group
	EX(stda %f8, [%o0 + 0x00] %asi, add %o2, 0x8)	! MS	(XXX does it stall here? XXX)
	be,pn		%icc, U3copy_in_user_endcruft	! BR
	 add		%o0, 0x8, %o0			! A0
	EX(ldda [%o1 + 0x08] %asi, %f0, add %o2, %g0)	! MS	Group
	add		%o1, 0x8, %o1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f2, %f0, %f8			! FGA
	EX(stda %f8, [%o0 + 0x00] %asi, add %o2, 0x8)	! MS	(XXX does it stall here? XXX)
	bne,pn		%icc, 1b			! BR
	 add		%o0, 0x8, %o0			! A0	Group

	/* If anything is left, we copy it one byte at a time.
	 * Note that %g1 is (src & 0x3) saved above before the
	 * alignaddr was performed.
	 */
U3copy_in_user_endcruft:
	cmp		%o2, 0
	add		%o1, %g1, %o1
	VISExitHalf
	be,pn		%icc, U3copy_in_user_short_ret
	 nop
	ba,a,pt		%xcc, U3copy_in_user_short

	/* If we get here, then 32 <= len < (6 * 64) */
U3copy_in_user_toosmall:

#ifdef SMALL_COPY_USES_FPU

	/* Is 'dst' already aligned on an 8-byte boundary? */
	be,pt		%xcc, 2f			! BR	Group

	/* Compute abs((dst & 7) - 8) into %g2.  This is the number
	 * of bytes to copy to make 'dst' 8-byte aligned.  We pre-
	 * subtract this from 'len'.
	 */
	 sub		%g2, 0x8, %g2			! A0
	sub		%g0, %g2, %g2			! A0	Group (reg-dep)
	sub		%o2, %g2, %o2			! A0	Group (reg-dep)

	/* Copy %g2 bytes from src to dst, one byte at a time. */
1:	EXNV2(lduba [%o1 + 0x00] %asi, %o3, add %o2, %g2)! MS	(Group) (%o3 in 3 cycles)
	add		%o1, 0x1, %o1			! A1
	add		%o0, 0x1, %o0			! A0	Group
	subcc		%g2, 0x1, %g2			! A1

	bg,pt		%icc, 1b			! BR	Group
	 EXNV2(stba %o3, [%o0 + -1] %asi, add %o2, %g2)	! MS	Group

2:	VISEntryHalf					! MS+MS

	/* Compute (len - (len % 8)) into %g2.  This is guarenteed
	 * to be nonzero.
	 */
	andn		%o2, 0x7, %g2			! A0	Group

	/* You may read this and believe that it allows reading
	 * one 8-byte longword past the end of src.  It actually
	 * does not, as %g2 is subtracted as loads are done from
	 * src, so we always stop before running off the end.
	 * Also, we are guarenteed to have at least 0x10 bytes
	 * to move here.
	 */
	sub		%g2, 0x8, %g2			! A0	Group (reg-dep)
	alignaddr	%o1, %g0, %g1			! MS	      (Break-after)
	EX(ldda [%g1 + 0x00] %asi, %f0, add %o2, %g0)	! MS	Group (1-cycle stall)
	add		%g1, 0x8, %g1			! A0

1:	EX(ldda [%g1 + 0x00] %asi, %f2, add %o2, %g0)	! MS	Group
	add		%g1, 0x8, %g1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group

	faligndata	%f0, %f2, %f8			! FGA	Group (1-cycle stall)
	EX(stda %f8, [%o0 + 0x00] %asi, add %o2, 0x8)	! MS	Group (2-cycle stall)
	add		%o1, 0x8, %o1			! A0
	be,pn		%icc, 2f			! BR

	 add		%o0, 0x8, %o0			! A1
	EX(ldda [%g1 + 0x00] %asi, %f0, add %o2, %g0)	! MS	Group
	add		%g1, 0x8, %g1			! A0
	sub		%o2, 0x8, %o2			! A1

	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f2, %f0, %f8			! FGA	Group (1-cycle stall)
	EX(stda %f8, [%o0 + 0x00] %asi, add %o2, 0x8)	! MS	Group (2-cycle stall)
	add		%o1, 0x8, %o1			! A0

	bne,pn		%icc, 1b			! BR
	 add		%o0, 0x8, %o0			! A1

	/* Nothing left to copy? */
2:	cmp		%o2, 0				! A0	Group
	VISExitHalf					! A0+MS
	be,pn		%icc, U3copy_in_user_short_ret	! BR	Group
	 nop						! A0
	ba,a,pt		%xcc, U3copy_in_user_short	! BR	Group

#else /* !(SMALL_COPY_USES_FPU) */

	xor		%o1, %o0, %g2
	andcc		%g2, 0x7, %g0
	bne,pn		%icc, U3copy_in_user_short
	 andcc		%o1, 0x7, %g2

	be,pt		%xcc, 2f
	 sub		%g2, 0x8, %g2
	sub		%g0, %g2, %g2
	sub		%o2, %g2, %o2

1:	EXNV2(lduba [%o1 + 0x00] %asi, %o3, add %o2, %g2)
	add		%o1, 0x1, %o1
	add		%o0, 0x1, %o0
	subcc		%g2, 0x1, %g2
	bg,pt		%icc, 1b
	 EXNV2(stba %o3, [%o0 + -1] %asi, add %o2, %g2)

2:	andn		%o2, 0x7, %g2
	sub		%o2, %g2, %o2

3:	EXNV3(ldxa [%o1 + 0x00] %asi, %o3, add %o2, %g2)
	add		%o1, 0x8, %o1
	add		%o0, 0x8, %o0
	subcc		%g2, 0x8, %g2
	bg,pt		%icc, 3b
	 EXNV3(stxa %o3, [%o0 + -8] %asi, add %o2, %g2)

	cmp		%o2, 0
	bne,pn		%icc, U3copy_in_user_short
	 nop
	ba,a,pt		%xcc, U3copy_in_user_short_ret

#endif /* !(SMALL_COPY_USES_FPU) */