From: David Howells <dhowells@redhat.com>
To: linux-am33-list@redhat.com
Cc: linux-kernel@vger.kernel.org,
Akira Takeuchi <takeuchi.akr@jp.panasonic.com>,
Kiyoshi Owada <owada.kiyoshi@jp.panasonic.com>
Subject: [PATCH 32/43] MN10300: Optimise do_csum()
Date: Tue, 26 Oct 2010 03:55:52 +0100 [thread overview]
Message-ID: <20101026025552.23512.17136.stgit@warthog.procyon.org.uk> (raw)
In-Reply-To: <20101026025301.23512.24525.stgit@warthog.procyon.org.uk>
From: Akira Takeuchi <takeuchi.akr@jp.panasonic.com>
Optimise do_csum() to gang up the loads so they're less likely to get
interruptions between.
Signed-off-by: Akira Takeuchi <takeuchi.akr@jp.panasonic.com>
Signed-off-by: Kiyoshi Owada <owada.kiyoshi@jp.panasonic.com>
Signed-off-by: David Howells <dhowells@redhat.com>
---
arch/mn10300/lib/do_csum.S | 49 ++++++++++++++++++++------------------------
1 files changed, 22 insertions(+), 27 deletions(-)
diff --git a/arch/mn10300/lib/do_csum.S b/arch/mn10300/lib/do_csum.S
index e138994..1d27bba 100644
--- a/arch/mn10300/lib/do_csum.S
+++ b/arch/mn10300/lib/do_csum.S
@@ -10,26 +10,25 @@
*/
#include <asm/cache.h>
- .section .text
- .balign L1_CACHE_BYTES
+ .section .text
+ .balign L1_CACHE_BYTES
###############################################################################
#
-# unsigned int do_csum(const unsigned char *buff, size_t len)
+# unsigned int do_csum(const unsigned char *buff, int len)
#
###############################################################################
.globl do_csum
- .type do_csum,@function
+ .type do_csum,@function
do_csum:
movm [d2,d3],(sp)
- mov d0,(12,sp)
- mov d1,(16,sp)
mov d1,d2 # count
mov d0,a0 # buff
+ mov a0,a1
clr d1 # accumulator
cmp +0,d2
- beq do_csum_done # return if zero-length buffer
+ ble do_csum_done # check for zero length or negative
# 4-byte align the buffer pointer
btst +3,a0
@@ -41,17 +40,15 @@ do_csum:
inc a0
asl +8,d0
add d0,d1
- addc +0,d1
add -1,d2
-do_csum_addr_not_odd:
+do_csum_addr_not_odd:
cmp +2,d2
bcs do_csum_fewer_than_4
btst +2,a0
beq do_csum_now_4b_aligned
movhu (a0+),d0
add d0,d1
- addc +0,d1
add -2,d2
cmp +4,d2
bcs do_csum_fewer_than_4
@@ -66,20 +63,20 @@ do_csum_now_4b_aligned:
do_csum_loop:
mov (a0+),d0
- add d0,d1
mov (a0+),e0
- addc e0,d1
mov (a0+),e1
- addc e1,d1
mov (a0+),e3
+ add d0,d1
+ addc e0,d1
+ addc e1,d1
addc e3,d1
mov (a0+),d0
- addc d0,d1
mov (a0+),e0
- addc e0,d1
mov (a0+),e1
- addc e1,d1
mov (a0+),e3
+ addc d0,d1
+ addc e0,d1
+ addc e1,d1
addc e3,d1
addc +0,d1
@@ -94,12 +91,12 @@ do_csum_remainder:
cmp +16,d2
bcs do_csum_fewer_than_16
mov (a0+),d0
- add d0,d1
mov (a0+),e0
- addc e0,d1
mov (a0+),e1
- addc e1,d1
mov (a0+),e3
+ add d0,d1
+ addc e0,d1
+ addc e1,d1
addc e3,d1
addc +0,d1
add -16,d2
@@ -131,9 +128,9 @@ do_csum_fewer_than_4:
xor_cmp d0,d0,+2,d2
bcs do_csum_fewer_than_2
movhu (a0+),d0
-do_csum_fewer_than_2:
and +1,d2
beq do_csum_add_last_bit
+do_csum_fewer_than_2:
movbu (a0),d3
add d3,d0
do_csum_add_last_bit:
@@ -142,21 +139,19 @@ do_csum_add_last_bit:
do_csum_done:
# compress the checksum down to 16 bits
- mov +0xffff0000,d2
- and d1,d2
+ mov +0xffff0000,d0
+ and d1,d0
asl +16,d1
- add d2,d1,d0
+ add d1,d0
addc +0xffff,d0
lsr +16,d0
# flip the halves of the word result if the buffer was oddly aligned
- mov (12,sp),d1
- and +1,d1
+ and +1,a1
beq do_csum_not_oddly_aligned
swaph d0,d0 # exchange bits 15:8 with 7:0
do_csum_not_oddly_aligned:
ret [d2,d3],8
-do_csum_end:
- .size do_csum, do_csum_end-do_csum
+ .size do_csum, .-do_csum
next prev parent reply other threads:[~2010-10-26 2:56 UTC|newest]
Thread overview: 47+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-10-26 2:53 [PATCH 00/43] MN10300: Add SMP support and AM34 CPU core support David Howells
2010-10-26 2:53 ` [PATCH 01/43] Partially revert patch that encloses asm-offset.h numbers in brackets David Howells
2010-10-26 2:53 ` [PATCH 02/43] MN10300: BUG to BUG_ON changes David Howells
2010-10-26 2:53 ` [PATCH 03/43] bitops: Change the bitmap index from int to unsigned long [mn10300] David Howells
2010-10-26 2:53 ` [PATCH 04/43] mn10300: Use pci_claim_resource David Howells
2010-10-26 2:53 ` [PATCH 05/43] MN10300: Prevent cnt32_to_63() from being preempted in sched_clock() David Howells
2010-10-26 2:53 ` [PATCH 06/43] Typedef SMP call function pointer David Howells
2010-10-26 2:53 ` [PATCH 07/43] MN10300: Don't cast away the volatile in test_bit() David Howells
2010-10-26 2:53 ` [PATCH 08/43] MN10300: Add reads[bwl]() and writes[bwl]() David Howells
2010-10-26 2:53 ` [PATCH 09/43] MN10300: Provide the functions to fully disable maskable interrupts David Howells
2010-10-26 2:53 ` [PATCH 10/43] MN10300: Differentiate AM33_2 and AM33_3 in config David Howells
2010-10-26 2:54 ` [PATCH 11/43] MN10300: Move DMA engine control reg defs to MN103E010 processor directory David Howells
2010-10-26 2:54 ` [PATCH 12/43] MN10300: Don't hard code the cacheline size in register defs David Howells
2010-10-26 2:54 ` [PATCH 13/43] MN10300: Add CPU register bits for AM34 David Howells
2010-10-26 2:54 ` [PATCH 14/43] MN10300: Remove monitor/JTAG functions David Howells
2010-10-26 2:54 ` [PATCH 15/43] MN10300: Cache: Split cache bits out of arch Kconfig David Howells
2010-10-26 2:54 ` [PATCH 16/43] MN10300: Provide a MN10300_CACHE_ENABLED config option David Howells
2010-10-26 2:54 ` [PATCH 17/43] MN10300: Reorder asm/cacheflush.h to put primitives first David Howells
2010-10-26 2:54 ` [PATCH 18/43] MN10300: AM34: The current cacheflush routines operate by controlling tag regs David Howells
2010-10-26 2:54 ` [PATCH 19/43] MN10300: Cacheflush functions should take unsigned long addresses David Howells
2010-10-26 2:54 ` [PATCH 20/43] MN10300: SMP: Differentiate local cache flushing David Howells
2010-10-26 2:54 ` [PATCH 21/43] MN10300: AM34: Add cacheflushing by using the AM34 purge registers David Howells
2010-10-26 2:54 ` [PATCH 22/43] MN10300: Allow some cacheflushes to be avoided if cache snooping is available David Howells
2010-10-26 2:55 ` [PATCH 23/43] MN10300: Cache: Implement SMP global cache flushing David Howells
2010-10-26 2:55 ` [PATCH 24/43] MN10300: Make the boot wrapper able to use writeback caching David Howells
2010-10-26 2:55 ` [PATCH 25/43] MN10300: AM34 erratum requires MMUCTR read and write on exception entry David Howells
2010-10-26 2:55 ` [PATCH 26/43] MN10300: Rename __flush_tlb*() to local_flush_tlb*() David Howells
2010-10-26 2:55 ` [PATCH 27/43] MN10300: Make the use of PIDR to mark TLB entries controllable David Howells
2010-10-26 2:55 ` [PATCH 28/43] MN10300: Use the [ID]PTEL2 registers rather than [ID]PTEL for TLB control David Howells
2010-10-26 2:55 ` [PATCH 29/43] MN10300: SMP TLB flushing David Howells
2010-10-26 2:55 ` [PATCH 30/43] MN10300: Make the FPU operate in non-lazy mode under SMP David Howells
2010-10-26 2:55 ` [PATCH 31/43] MN10300: Implement atomic ops using atomic ops unit David Howells
2010-10-26 2:55 ` David Howells [this message]
2010-10-26 2:55 ` [PATCH 33/43] MN10300: Make various interrupt priority settings configurable David Howells
2010-10-26 2:56 ` [PATCH 34/43] MN10300: Delete idle_timestamp from irq_cpustat_t David Howells
2010-10-26 2:56 ` [PATCH 35/43] MN10300: And Panasonic AM34 subarch and implement SMP David Howells
2010-10-27 13:31 ` [Linux-am33-list] " Suzuki Takashi
2010-10-27 14:33 ` David Howells
2010-10-26 2:56 ` [PATCH 36/43] MN10300: ASB2364: Add support for SMSC911X and SMC911X David Howells
2010-10-26 2:56 ` [PATCH 37/43] MN10300: Map userspace atomic op regs as a vmalloc page David Howells
2010-10-26 2:56 ` [PATCH 38/43] MN10300: Specify an ELF HWCAP flag for MN10300 Atomic Operations Unit support David Howells
2010-10-26 2:56 ` [PATCH 39/43] MN10300: Generic time support David Howells
2010-10-26 2:56 ` [PATCH 40/43] MN10300: Update the ASB2303 defconfig David Howells
2010-10-26 2:56 ` [PATCH 41/43] MN10300: Create a defconfig for the ASB2364 board David Howells
2010-10-26 2:56 ` [PATCH 42/43] MN10300: Change "Matsushita" to "Panasonic" David Howells
2010-10-26 2:56 ` [PATCH 43/43] MN10300: Save frame pointer in thread_info struct rather than global var David Howells
2010-10-26 3:07 ` [PATCH 00/43] MN10300: Add SMP support and AM34 CPU core support David Howells
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20101026025552.23512.17136.stgit@warthog.procyon.org.uk \
--to=dhowells@redhat.com \
--cc=linux-am33-list@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=owada.kiyoshi@jp.panasonic.com \
--cc=takeuchi.akr@jp.panasonic.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®