lkml.org 
[lkml]   [2021]   [Mar]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH] MIPS: Check __clang__ to avoid performance influence with GCC in csum_tcpudp_nofold()
From
Date
On 03/09/2021 12:52 AM, David Laight wrote:
> From: Tiezhu Yang
>> Sent: 08 March 2021 12:50
>>
>> The asm code in csum_tcpudp_nofold() is performance-critical, I am sorry
>> for the poorly considered implementation about the performance influence
>> with GCC in the commit 198688edbf77 ("MIPS: Fix inline asm input/output
>> type mismatch in checksum.h used with Clang").
>>
>> With this patch, we can build successfully by both GCC and Clang,
>> at the same time, we can avoid the potential performance influence
>> with GCC.
>>
>> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
>> ---
>> arch/mips/include/asm/checksum.h | 10 ++++++++++
>> 1 file changed, 10 insertions(+)
>>
>> diff --git a/arch/mips/include/asm/checksum.h b/arch/mips/include/asm/checksum.h
>> index 1e6c135..64d353e 100644
>> --- a/arch/mips/include/asm/checksum.h
>> +++ b/arch/mips/include/asm/checksum.h
>> @@ -130,7 +130,9 @@ static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
>> __u32 len, __u8 proto,
>> __wsum sum)
>> {
>> +#ifdef __clang__
>> unsigned long tmp = (__force unsigned long)sum;
>> +#endif
> What happens if you make the above:
> #ifdef __clang__
> unsigned long tmp = (__force unsigned long)sum;
> #else
> __wsum tmp = sum;
> #endif
>
> and then leave the rest of the function the same for both compilers.
> Maybe do s/sum/sum_in/,s/tmp/sum/ to reduce the changes.

Hi David,

Thank you very much.

As you suggested, the following changes looks much better,
I will test it and then send v2 later.

diff --git a/arch/mips/include/asm/checksum.h
b/arch/mips/include/asm/checksum.h
index 1e6c135..80eddd4 100644
--- a/arch/mips/include/asm/checksum.h
+++ b/arch/mips/include/asm/checksum.h
@@ -128,9 +128,13 @@ static inline __sum16 ip_fast_csum(const void *iph,
unsigned int ihl)

static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
__u32 len, __u8 proto,
- __wsum sum)
+ __wsum sum_in)
{
- unsigned long tmp = (__force unsigned long)sum;
+#ifdef __clang__
+ unsigned long sum = (__force unsigned long)sum_in;
+#else
+ __wsum sum = sum_in;
+#endif

__asm__(
" .set push # csum_tcpudp_nofold\n"
@@ -159,7 +163,7 @@ static inline __wsum csum_tcpudp_nofold(__be32
saddr, __be32 daddr,
" addu %0, $1 \n"
#endif
" .set pop"
- : "=r" (tmp)
+ : "=r" (sum)
: "0" ((__force unsigned long)daddr),
"r" ((__force unsigned long)saddr),
#ifdef __MIPSEL__
@@ -169,7 +173,7 @@ static inline __wsum csum_tcpudp_nofold(__be32
saddr, __be32 daddr,
#endif
"r" ((__force unsigned long)sum));

- return (__force __wsum)tmp;
+ return (__force __wsum)sum;
}
#define csum_tcpudp_nofold csum_tcpudp_nofold

Thanks,
Tiezhu
>
> David
>
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)

\
 
 \ /
  Last update: 2021-03-09 03:25    [W:0.198 / U:0.312 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site