------- Additional Comments From dorit at il dot ibm dot com  2004-11-17 11:09 
-------
(In reply to comment #1)
> PR 10469 looks like a testcase where this could improve the code generation.\

gcc.dg/vect/vect-25.c is a testcase too.

This is what is currently generated (the invariant is in r3):

        mr r9,r3
        mr r10,r3
        mr r11,r3
        mr r12,r3 
L6:
        addi r7,r1,1104
        stw r9,0(r7)
        stw r10,4(r7)
        stw r11,8(r7)
        stw r12,12(r7)
        lvx v0,0,r7
        stvx v0,0,r2
        addi r2,r2,16
        bdnz L6

instead of:

        stw r3,0(r2)
        lvewx v0,0,r2
        vspltw v0,v0,0
L4:
        stvx v0,r2,r0
        addi r2,r2,16
        bdnz L4 



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18506

Reply via email to