[Bug middle-end/78103] Failure to optimize with __builtin_clzl

["jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78103

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |segher at gcc dot gnu.org

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Unfortunately, it doesn't work for the #c0 testcase, after the combiner
splitter kicks in, the combiner doesn't even try that 4 insn combination. 
Haven't debugged it, but it looks like if LOG_LINKS isn't added between the 2
insns that replaced the 3 insns.
There is the successful:
Trying 9, 10 -> 12:
    9: {flags:CCZ=cmp(r85:DI,0);r86:DI=0x3f-clz(r85:DI);}
      REG_DEAD r85:DI
      REG_UNUSED flags:CCZ
   10: {r86:DI=r86:DI^0x3f;clobber flags:CC;}
      REG_UNUSED flags:CC
   12: r87:DI=sign_extend(r86:DI#0)
      REG_DEAD r86:DI
Failed to match this instruction:
(set (reg:DI 87 [ _1 ])
    (xor:DI (sign_extend:DI (minus:SI (const_int 63 [0x3f])
                (subreg:SI (clz:DI (reg/v:DI 85 [ x ])) 0)))
        (const_int 63 [0x3f])))
Splitting with gen_split_440 (i386.md:14804)
Successfully matched this instruction:
(set (reg:DI 93)
    (minus:DI (const_int 63 [0x3f])
        (clz:DI (reg/v:DI 85 [ x ]))))
Successfully matched this instruction:
(set (reg:DI 87 [ _1 ])
    (zero_extend:DI (xor:SI (subreg:SI (reg:DI 93) 0)
            (const_int 63 [0x3f]))))
allowing combination of insns 9, 10 and 12
original costs 8 + 4 + 4 = 16
replacement costs 8 + 5 = 13
deferring deletion of insn with uid = 9.
modifying insn i2    10: {r93:DI=0x3f-clz(r85:DI);clobber flags:CC;}
      REG_UNUSED flags:CC
      REG_DEAD r85:DI
deferring rescan insn with uid = 10.
modifying insn i3    12: {r87:DI=zero_extend(r93:DI#0^0x3f);clobber flags:CC;}
      REG_UNUSED flags:CC
deferring rescan insn with uid = 12.

and I was hoping that the combiner would soon try 10, 12, 13 -> 14 combination
where 10 and 12 are the above 2 new insns, 13 is r88:DI=0x40
and 14 is {r85:DI=r88:DI-r87:DI;clobber flags:CC;}.
But all it tries is 12 -> 14, 13 -> 14, 13, 12 -> 14 and all those fail.

Segher, any ideas?