kernel_optimize_test/arch/arc/lib/strcmp.S
Vineet Gupta 86effd0dc6 ARC: dw2 unwind: enable cfi pseudo ops in string lib
This uses a new set of annoations viz. ENTRY_CFI/END_CFI to enabel cfi
ops generation.

Note that we didn't change the normal ENTRY/EXIT as we don't actually
want unwind info in the trap/exception/interrutp handlers which use
these, as unwinder then gets confused (it keeps recursing vs. stopping).
Semantically these are leaf routines and unwinding should stop when it
hits those routines.

Before
------

    28.52%     1.19%          9929  hackbench  libuClibc-1.0.17.so   [.] __write_nocancel
            |
            ---__write_nocancel
               |--8.95%--EV_Trap
               |           --8.25%--sys_write
               |                     |--3.93%--sock_write_iter
     ...
               |--2.62%--memset   <==== [LEAF entry as no unwind info]
                         ^^^^^^

After
-----

    29.46%     1.24%         13622  hackbench  libuClibc-1.0.17.so   [.] __write_nocancel
            |
            ---__write_nocancel
               |--9.31%--EV_Trap
               |           --8.62%--sys_write
               |                     |--4.17%--sock_write_iter
     ...
               |--6.19%--sys_write
               |           --6.19%--sock_write_iter
               |                     unix_stream_sendmsg
               |                     |--1.62%--sock_alloc_send_pskb
               |                     |--0.89%--sock_def_readable
               |                     |--0.88%--_raw_spin_unlock_irqrestore
               |                     |--0.69%--memset
               |                     |         ^^^^^^     <==== [now in proper callframe]
               |                     |
               |                      --0.52%--skb_copy_datagram_from_iter

Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
2016-09-30 14:48:22 -07:00

97 lines
2.6 KiB
ArmAsm

/*
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
/* This is optimized primarily for the ARC700.
It would be possible to speed up the loops by one cycle / word
respective one cycle / byte by forcing double source 1 alignment, unrolling
by a factor of two, and speculatively loading the second word / byte of
source 1; however, that would increase the overhead for loop setup / finish,
and strcmp might often terminate early. */
#include <linux/linkage.h>
ENTRY_CFI(strcmp)
or r2,r0,r1
bmsk_s r2,r2,1
brne r2,0,.Lcharloop
mov_s r12,0x01010101
ror r5,r12
.Lwordloop:
ld.ab r2,[r0,4]
ld.ab r3,[r1,4]
nop_s
sub r4,r2,r12
bic r4,r4,r2
and r4,r4,r5
brne r4,0,.Lfound0
breq r2,r3,.Lwordloop
#ifdef __LITTLE_ENDIAN__
xor r0,r2,r3 ; mask for difference
sub_s r1,r0,1
bic_s r0,r0,r1 ; mask for least significant difference bit
sub r1,r5,r0
xor r0,r5,r1 ; mask for least significant difference byte
and_s r2,r2,r0
and_s r3,r3,r0
#endif /* LITTLE ENDIAN */
cmp_s r2,r3
mov_s r0,1
j_s.d [blink]
bset.lo r0,r0,31
.balign 4
#ifdef __LITTLE_ENDIAN__
.Lfound0:
xor r0,r2,r3 ; mask for difference
or r0,r0,r4 ; or in zero indicator
sub_s r1,r0,1
bic_s r0,r0,r1 ; mask for least significant difference bit
sub r1,r5,r0
xor r0,r5,r1 ; mask for least significant difference byte
and_s r2,r2,r0
and_s r3,r3,r0
sub.f r0,r2,r3
mov.hi r0,1
j_s.d [blink]
bset.lo r0,r0,31
#else /* BIG ENDIAN */
/* The zero-detection above can mis-detect 0x01 bytes as zeroes
because of carry-propagateion from a lower significant zero byte.
We can compensate for this by checking that bit0 is zero.
This compensation is not necessary in the step where we
get a low estimate for r2, because in any affected bytes
we already have 0x00 or 0x01, which will remain unchanged
when bit 7 is cleared. */
.balign 4
.Lfound0:
lsr r0,r4,8
lsr_s r1,r2
bic_s r2,r2,r0 ; get low estimate for r2 and get ...
bic_s r0,r0,r1 ; <this is the adjusted mask for zeros>
or_s r3,r3,r0 ; ... high estimate r3 so that r2 > r3 will ...
cmp_s r3,r2 ; ... be independent of trailing garbage
or_s r2,r2,r0 ; likewise for r3 > r2
bic_s r3,r3,r0
rlc r0,0 ; r0 := r2 > r3 ? 1 : 0
cmp_s r2,r3
j_s.d [blink]
bset.lo r0,r0,31
#endif /* ENDIAN */
.balign 4
.Lcharloop:
ldb.ab r2,[r0,1]
ldb.ab r3,[r1,1]
nop_s
breq r2,0,.Lcmpend
breq r2,r3,.Lcharloop
.Lcmpend:
j_s.d [blink]
sub r0,r2,r3
END_CFI(strcmp)