Re: gcc 4.2.3 and MMX to mem move oddity

[Uros Bizjak <ubizjak at gmail dot com>]

Hello!

>   f7:   0f 7f 5c 24 f0          movq   %mm3,-0x10(%rsp)
>   fc:   0f 7f 54 24 f8          movq   %mm2,-0x8(%rsp)
>  101:   48 8b 5c 24 f8          mov    -0x8(%rsp),%rbx
>  106:   48 89 5c 38 40          mov    %rbx,0x40(%rax,%rdi,1)
>  10b:   48 8b 5c 24 f0          mov    -0x10(%rsp),%rbx
>  110:   48 89 5c 38 48          mov    %rbx,0x48(%rax,%rdi,1)

> As you see in the intrinsic version gcc moves to mmx register to the stack, 
> reloads from the stack and writes to the destination. Why?
>
> I don't know whether earlier gcc 4.2 versions produced such stupid code.
> Compiling as 32 does similar stupidity, though gcc reloads into a mmx 
> register...

This is a variant of "Strange code for MMX register moves" [1] or its 
dupe "mmx and movd/movq on x86_64" [2]. Since touching %mm register 
switches x87 register stack to MMX mode, we penalize mmx moves severely 
in order to prevent gcc to ever allocate %mm for DImode moves, unless 
really necessary.

OTOH, in your particular case:

(insn 42 40 43 3 mmx.c:20 (set (mem:V2SI (plus:DI (reg:DI 65 [ ivtmp.43 ])
               (const_int -64 [0xffffffffffffffc0])) [0 S8 A64])
       (subreg:V2SI (reg:V4HI 90) 0)) 866 {*movv2si_internal_rex64} 
(expr_list:REG_DEAD (reg:V4HI 90)
       (nil)))

subreg in this pattern confuses RA to create:

(insn 68 36 38 3 
/usr/local/lib/gcc/x86_64-unknown-linux-gnu/4.3.0/include/mmintrin.h:389 
(set (mem/c:V4HI (plus:DI (reg/f:DI 7 sp)
               (const_int -8 [0xfffffffffffffff8])) [2 S8 A8])
       (reg:V4HI 32 mm3)) 865 {*movv4hi_internal_rex64} (nil))

(insn 70 69 42 3 mmx.c:20 (set (reg:V2SI 3 bx)
       (mem/c:V2SI (plus:DI (reg/f:DI 7 sp)
               (const_int -8 [0xfffffffffffffff8])) [2 S8 A8])) 866 
{*movv2si_internal_rex64} (nil))

(insn:HI 42 70 71 3 mmx.c:20 (set (mem:V2SI (plus:DI (reg:DI 5 di 
[orig:65 ivtmp.43 ] [65])
               (const_int -64 [0xffffffffffffffc0])) [0 S8 A64])
       (reg:V2SI 3 bx)) 866 {*movv2si_internal_rex64} (nil))


(For 32bit targets, %mm is allocated, since no integer register can hold 
64bit value.)

I'll see what can be done here to tie V4HI and V2SI together.

[1] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22076
[2] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34256

Uros.