This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tcb] Fix various aliasing and verification problems


On Tue, 2004-12-07 at 11:55 -0800, Devang Patel wrote:
> On Dec 5, 2004, at 6:52 PM, Daniel Berlin wrote:
> 
> >> New regressions in 00testsuite-linux-gnu/20041205/gcc.sum.gz:
> >>         FAIL: gcc.dg/pr17635.c (test for excess errors)
> >>         FAIL: gcc.dg/tree-ssa/ifc-20040816-1.c (test for excess 
> >> errors)
> 
> For this test case, I get following from mainline source,

> # BLOCK 1
>    # PRED: 9 [100.0%]  (fallthru) 0 [100.0%]  (fallthru,exec)
>    # ivtmp.3_12 = PHI <ivtmp.3_18(9), 16(0)>;
>    # i_24 = PHI <i_19(9), 0(0)>;
>    # A_25 = PHI <A_10(9), A_7(0)>;
> <L0>:;
>    #   VUSE <A_25>;
>    j_16 = A[i_24];
>    _ifc_.5_17 = j_16 > 41;
>    iftmp.2_4 = j_16 > 41 ? 42 : 0;
>    #   A_10 = V_MAY_DEF <A_25>;
>    A[i_24] = iftmp.2_4;
>    i_19 = i_24 + 1;
>    ivtmp.3_18 = ivtmp.3_12 - 1;
>    _ifc_.6_15 = (j_16 > 41 || !_ifc_.5_17) && ivtmp.3_18 != 0;
>    _ifc_.7_8 = ivtmp.3_18 != 0;
>    if (ivtmp.3_18 != 0) goto <L18>; else goto <L19>;
>    # SUCC: 9 [93.8%]  (dfs_back,true,exec) 10 [6.2%]  
> (loop_exit,false,exec)
> 
> Where as with tcb branch I get,
> 
>    # BLOCK 1
>    # PRED: 4 [100.0%]  (fallthru,exec) 0 [100.0%]  (fallthru,exec)
>    # ivtmp.3_2 = PHI <ivtmp.3_12(4), 16(0)>;
>    # i_9 = PHI <i_19(4), 0(0)>;
>    # A_20 = PHI <A_18(4), A_7(0)>;
> <L0>:;
>    #   VUSE <A_20>;
>    j_16 = A[i_9];
>    _ifc_.5_17 = j_16 > 41;
>    iftmp.2_4 = _ifc_.6_21 ? 42 : 0;
>    #   A_18 = V_MAY_DEF <A_20>;
>    A[i_9] = iftmp.2_4;
>    i_19 = i_9 + 1;
>    ivtmp.3_12 = ivtmp.3_2 - 1;
>    _ifc_.6_21 = (j_16 > 41 || !_ifc_.5_17) && ivtmp.3_12 != 0;
>    _ifc_.7_1 = ivtmp.3_12 != 0;
>    if (ivtmp.3_12 != 0) goto <L17>; else goto <L18>;
>    # SUCC: 4 [93.8%]  (dfs_back,true,exec) 5 [6.2%]  
> (loop_exit,false,exec)
> 
> I'll investigate this further but any idea what is causing this 
> behavior in tcb branch ?
> 

Which behavior are you asking about?
The fact that there are less blocks on the tcb branch?
Or is there some real difference in the code besides the numbering?
(i can't tell, and i've stared at it for a couple minutes :P)

> Thanks,
> -
> Devang
> 


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]