2001-12-20 22:32:38

by Alex

Subject: Slight optimizations to entry.S patch

Hi,
I was familiarizing (or trying to) myself with the i386 architecture code,
and saw a few possible optimizations. I think they can save a few cycles (not
that many). Can someone comment? Are the changes worthwhile?
-Alex K.

--- entry.S.orig Thu Dec 20 16:57:39 2001
+++ entry.S Thu Dec 20 17:20:29 2001
@@ -91,9 +91,9 @@
pushl %edi; \
pushl %esi; \
pushl %edx; \
+ movl $(__KERNEL_DS),%edx; \
pushl %ecx; \
pushl %ebx; \
- movl $(__KERNEL_DS),%edx; \
movl %edx,%ds; \
movl %edx,%es;

@@ -141,13 +141,13 @@
movl EFLAGS(%esp),%ecx # and this is cs..
movl %eax,EFLAGS(%esp) #
movl %edx,EIP(%esp) # Now we move them to their "normal" places
- movl %ecx,CS(%esp) #
movl %esp,%ebx
+ movl %ecx,CS(%esp) #
pushl %ebx
andl $-8192,%ebx # GET_CURRENT
movl exec_domain(%ebx),%edx # Get the execution domain
- movl 4(%edx),%edx # Get the lcall7 handler for the domain
pushl $0x7
+ movl 4(%edx),%edx # Get the lcall7 handler for the domain
call *%edx
addl $4, %esp
popl %eax
@@ -162,13 +162,13 @@
movl EFLAGS(%esp),%ecx # and this is cs..
movl %eax,EFLAGS(%esp) #
movl %edx,EIP(%esp) # Now we move them to their "normal" places
- movl %ecx,CS(%esp) #
movl %esp,%ebx
+ movl %ecx,CS(%esp) #
pushl %ebx
andl $-8192,%ebx # GET_CURRENT
movl exec_domain(%ebx),%edx # Get the execution domain
- movl 4(%edx),%edx # Get the lcall7 handler for the domain
pushl $0x27
+ movl 4(%edx),%edx # Get the lcall7 handler for the domain
call *%edx
addl $4, %esp
popl %eax

2001-12-20 22:44:50

by Davide Libenzi

Subject: Re: Slight optimizations to entry.S patch

On Thu, 20 Dec 2001, Alex wrote:

> Hi,
> I was familiarizing (or trying to) myself with the i386 architecture code,
> and saw a few possible optimizations. I think they can save a few cycles (not
> that many). Can someone comment? Are the changes worthwhile?

The first page of your x86 book starts talking about read-after-write
pipeline stall ? Damn what a book :)




- Davide

2001-12-20 23:55:43

by Robert Love

Subject: Re: Slight optimizations to entry.S patch

On Thu, 2001-12-20 at 17:47, Davide Libenzi wrote:

> The first page of your x86 book starts talking about read-after-write
> pipeline stall ? Damn what a book :)

Haha, he actually did help w.r.t to that ... good patch. Sure, its
micro optimizations but not in any bad sense. If it works, submit it.

I'll try it out on my next reboot.

Robert Love

2001-12-21 16:40:09

by Alan

Subject: Re: Slight optimizations to entry.S patch

> I was familiarizing (or trying to) myself with the i386 architecture code,
> and saw a few possible optimizations. I think they can save a few cycles (not
> that many). Can someone comment? Are the changes worthwhile?

Measure them and see - look at the rdtsc instruction, run a million
iterations of all the various variants and see what you find