[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <DFEFCEFB42FD4CBB9CD45CD321B57F1A@H270>
Date: Wed, 4 Aug 2021 12:02:58 +0200
From: "Stefan Kanthak" <stefan.kanthak@...go.de>
To: "Szabolcs Nagy" <nsz@...t70.net>
Cc: <musl@...ts.openwall.com>
Subject: Re: [Patch] src/math/i386/remquo.s: remove conditional branch, shorter bit twiddling
"Szabolcs Nagy" <nsz@...t70.net> wrote:
>* Stefan Kanthak <stefan.kanthak@...go.de> [2021-08-01 17:59:52 +0200]:
>> Halve the number of instructions (from 12 to 6) to fetch the
>> (3-bit partial) quotient from the FPU flags C0:C3:C1, and
>> perform its negation without conditional branch.
>
> i haven't tested it but it looks good.
This is basically well-tested code I wrote about 20 years ago
for my own NOMSVCRT.LIB: I always found the bit twiddling of
J.T.Conklins code rather awful.
> i think we should not tweak x87 asm code too much though.
> it can introduce bugs and there are not many users of it.
> i think only the size saving can justify keeping any i386
> math code at all.
>From your own FAQ <http://www.musl-libc.org/faq.html>
| When will it be finished?
| When there's nothing left to remove.
The change just follows by removing 6 LOC/instructions.-)
> but i'm not against committing this.
> thanks for the patch.
regards
Stefan
>> --- -/math/i386/remquo.s
>> +++ +/math/i386/remquo.s
>> @@ -2,49 +2,44 @@
>> .type remquof,@function
>> remquof:
>> mov 12(%esp),%ecx
>> + mov 8(%esp),%eax
>> + xor 4(%esp),%eax
>> flds 8(%esp)
>> flds 4(%esp)
>> - mov 11(%esp),%dh
>> - xor 7(%esp),%dh
>> - jmp 1f
>> + jmp 0f
>>
>> .global remquol
>> .type remquol,@function
>> remquol:
>> mov 28(%esp),%ecx
>> + mov 24(%esp),%eax
>> + xor 12(%esp),%eax
>> + cwtl
>> fldt 16(%esp)
>> fldt 4(%esp)
>> - mov 25(%esp),%dh
>> - xor 13(%esp),%dh
>> - jmp 1f
>> + jmp 0f
>>
>> .global remquo
>> .type remquo,@function
>> remquo:
>> mov 20(%esp),%ecx
>> + mov 16(%esp),%eax
>> + xor 8(%esp),%eax
>> fldl 12(%esp)
>> fldl 4(%esp)
>> - mov 19(%esp),%dh
>> - xor 11(%esp),%dh
>> +0: cltd
>> 1: fprem1
>> fnstsw %ax
>> sahf
>> jp 1b
>> fstp %st(1)
>> - mov %ah,%dl
>> - shr %dl
>> - and $1,%dl
>> - mov %ah,%al
>> - shr $5,%al
>> - and $2,%al
>> - or %al,%dl
>> - mov %ah,%al
>> - shl $2,%al
>> - and $4,%al
>> - or %al,%dl
>> - test %dh,%dh
>> - jns 1f
>> - neg %dl
>> -1: movsbl %dl,%edx
>> - mov %edx,(%ecx)
>> + adc %al,%al
>> + shl $2,%ah
>> + adc %al,%al
>> + shl $5,%ah
>> + adc %al,%al
>> + and $7,%eax
>> + xor %edx,%eax
>> + sub %edx,%eax
>> + mov %eax,(%ecx)
>> ret
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
