Re: [PATCH] document optimizing macro for translating PROT_ to VM_ bits

From: Miles Bader (miles_at_lsi.nec.co.jp)
Date: 09/30/03

  • Next message: Muli Ben-Yehuda: "Re: [PATCH] document optimizing macro for translating PROT_ to VM_ bits" 
    To: Muli Ben-Yehuda <mulix@mulix.org>
Date:	30 Sep 2003 16:34:31 +0900

    Muli Ben-Yehuda <mulix@mulix.org> writes:
    > Ah, much nicer, thank you. I'll submit it, but first, what do you
    > think about this version? it keeps the optimization and enforces the
    > "bit1 and bit2 must be single bits only" rule.

    On my arch it results in huge bloated code, including mul and div insns!

    Here's what gcc 3.3.x does for both the prefix macro versions:

       #define _calc_vm_trans_2(x,bit1,bit2) \
        ((bit1) <= (bit2) ? ((x) & (bit1)) * ((bit2) / (bit1)) \
           : ((x) & (bit1)) / ((bit1) / (bit2)))
       int test2_0 (unsigned val)
       {
         return _calc_vm_trans_2 (val,0x20,0x80);
       }

    ==>

       _test2_0:
               andi 32,r6,r10 #, val, tmp43
               shl 2,r10 #, <result>
               jmp [r31]

    Here's what gcc 3.3.x does for your inlined function version:

       static inline void assert_single_bit(unsigned long bit)
       {
               extern __bug() __attribute__ ((noreturn));
               if (bit & (bit - 1))
                 __bug ();
       }
       static inline unsigned long
       _calc_vm_trans_3(unsigned long x, unsigned long bit1, unsigned long bit2)
       {
               assert_single_bit(bit1);
               assert_single_bit(bit2);

               return ((bit1) <= (bit2) ? ((x) & (bit1)) * ((bit2) / (bit1))
                       : ((x) & (bit1)) / ((bit1) / (bit2)));
       }
       int test3_0 (unsigned val)
       {
         return _calc_vm_trans_3 (val,0x20,0x80);
       }

    ==>

       _test3_0:
               movea lo(32),r0,r7 #, bit1
               movea lo(128),r0,r5 #, bit2
               divu r7,r5,r10 # bit1, bit2, tmp55
               mov r6,r10 # val, tmp52
               and r7,r10 # bit1, tmp52
               mul r5,r10,r0 # bit2, tmp52
               jmp [r31]

    [I've no idea what's up with that, I don't even think the generated code
    makes sense, but it seems like the presence of inline function is
    confusing something.]

    -Miles 

    -- 
In New York, most people don't have cars, so if you want to kill a person, you
have to take the subway to their house.  And sometimes on the way, the train
is delayed and you get impatient, so you have to kill someone on the subway.
  [George Carlin]
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
  • Next message: Muli Ben-Yehuda: "Re: [PATCH] document optimizing macro for translating PROT_ to VM_ bits"

    Relevant Pages