From: Roland Dreier <[email protected]>
Date: Tue, 10 Feb 2009 17:18:49 -0800
> > > Is this required? Strength reduction optimization should do this
> > > automatically (and the code has been there for quite a while, so
> > > obviously it isn't causing problems)
>
> > GCC won't optimize that modulus the way you expect, try for yourself
> > and look at the assembler if you don't believe me. :-)
>
> Are you thinking of the case when there are signed integers involved and
> so "% modulus" might produce a different result than "& (modulus - 1)"
> (because the compiler can't know that things are never negative)?
> Because in this case the compiler seems to do what I thought it would;
> the relevant part of the i386 assembly for
>
> wqe->recv.sgl[i].to = cpu_to_be64(((u32) wr->sg_list[i].addr) %
> (1UL << (12 + page_size[i])));
>
> is
>
> movl %eax, 28(%edi,%ebx) # <variable>.length,
> <variable>.len
> movzbl 28(%esp,%esi), %ecx # page_size, tmp89
> movl $1, %eax #, tmp92
> addl $12, %ecx #, tmp90
> sall %cl, %eax # tmp90, tmp92
> movl (%esp), %ecx # wr,
> decl %eax # tmp93
> movl 12(%ecx), %edx # <variable>.sg_list, <variable>.sg_list
> andl (%edx,%ebx), %eax # <variable>.addr, tmp93
>
> ie the compiler computes the modulus, then does decl to compute
> modulus-1 and then &s with it.
>
> Or am I misunderstanding your point?
Must be compiler and platform specific because with gcc-4.1.3 on
sparc with -O2, for the test program:
unsigned long page_size[4];
int main(int argc)
{
unsigned long long x = argc;
return x % (1UL << (12 + page_size[argc]));
}
I get a call to __umoddi3:
main:
save %sp, -112, %sp
sethi %hi(page_size), %g1
sll %i0, 2, %g3
or %g1, %lo(page_size), %g1
mov 1, %o2
ld [%g1+%g3], %g2
add %g2, 12, %g2
sll %o2, %g2, %o2
mov %i0, %o1
mov %o2, %o3
sra %i0, 31, %o0
call __umoddi3, 0
mov 0, %o2
jmp %i7+8
restore %g0, %o1, %o0
I get the same with gcc-4.3.0 and -O2 on 32-bit x86:
main:
leal 4(%esp), %ecx
andl $-16, %esp
pushl -4(%ecx)
movl $1, %eax
pushl %ebp
movl %esp, %ebp
pushl %ecx
subl $20, %esp
movl (%ecx), %edx
movl page_size(,%edx,4), %ecx
movl $0, 12(%esp)
movl %edx, (%esp)
addl $12, %ecx
sall %cl, %eax
movl %eax, 8(%esp)
movl %edx, %eax
sarl $31, %eax
movl %eax, 4(%esp)
call __umoddi3
addl $20, %esp
popl %ecx
popl %ebp
leal -4(%ecx), %esp
ret
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
