[Consult] tile: How to construct testsuite for tile architecture without hardware.

[Chen Gang S]

On 1/31/15 13:52, Chen Gang S wrote:
> On 1/31/15 07:00, Mike Stump wrote:
>> On Jan 30, 2015, at 2:58 PM, Chen Gang S <gang.chen@sunrus.com.cn> wrote:
>>> After finish analyzing the related warnings for tile, for me, they are
>>> neither tile's issues nor gcc common implementation issues:
>>
>> ?  They all look like they can be fixed in the port.
>>
>> Why not use :P?  From i386:
>>
>> ;; This mode iterator allows :P to be used for patterns that operate on                                                   
>> ;; pointer-sized quantities.  Exactly one of the two alternatives will match.                                             
>> (define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])
>>
>> (define_insn "*tls_global_dynamic_64_<mode>"
>>   [(set (match_operand:P 0 "register_operand" "=a")
>>         (call:P
>>          (mem:QI (match_operand 2 "constant_call_address_operand" "Bz"))
>>          (match_operand 3)))
>>    (unspec:P [(match_operand 1 "tls_symbolic_operand")]
>>              UNSPEC_TLS_GD)]
>>   “TARGET_64BIT"
>>
> 
> OK, thanks I shall try, hope I can finish within today (this month).

After try, it will report compiling error, the related information is
below:

  Modification:

    diff --git a/gcc/config/tilegx/tilegx.md b/gcc/config/tilegx/tilegx.md
    index 7ece171..fb9c097 100644
    --- a/gcc/config/tilegx/tilegx.md
    +++ b/gcc/config/tilegx/tilegx.md
    @@ -281,6 +281,8 @@
     (define_mode_iterator FI48MODE [SF DF SI DI])
     (define_mode_iterator VEC48MODE [V8QI V4HI])
     (define_mode_iterator VEC248MODE [V8QI V4HI V2SI])
    +(define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")
    +                        (HI "Pmode == HImode") (QI "Pmode == QImode")])
     
     (define_mode_attr n [(QI "1") (HI "2") (SI "4") (DI "")
                         (V8QI "1") (V4HI "2") (V2SI "4")])
    @@ -2625,7 +2627,7 @@
       [(set_attr "type" "X1")])
     
     (define_insn "indirect_jump"
    -  [(set (pc) (match_operand 0 "pointer_operand" "rO"))]
    +  [(set (pc) (match_operand:P 0 "pointer_operand" "rO"))]
       ""
       "jr\t%r0"
       [(set_attr "type" "Y1")])

  The related error:

    g++ -c  -DIN_GCC_FRONTEND -g -O2 -DIN_GCC  -DCROSS_DIRECTORY_STRUCTURE  -fno-exceptions -fno-rtti -fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings -Wcast-qual -Wmissing-format-attribute -Woverloaded-virtual -pedantic -Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -fno-common  -DHAVE_CONFIG_H -I. -Ic -I../../gcc-tile-new/gcc -I../../gcc-tile-new/gcc/c -I../../gcc-tile-new/gcc/../include -I../../gcc-tile-new/gcc/../libcpp/include  -I../../gcc-tile-new/gcc/../libdecnumber -I../../gcc-tile-new/gcc/../libdecnumber/dpd -I../libdecnumber -I../../gcc-tile-new/gcc/../libbacktrace   -o c/c-lang.o -MT c/c-lang.o -MMD -MP -MF c/.deps/c-lang.TPo ../../gcc-tile-new/gcc/c/c-lang.c
    In file included from ./tm.h:30:0,
                     from ../../gcc-tile-new/gcc/c/c-lang.c:24:
    ./insn-flags.h:114:0: warning: "HAVE_indirect_jump" redefined
     #define HAVE_indirect_jump (Pmode == DImode)
     ^
    ./insn-flags.h:113:0: note: this is the location of the previous definition
     #define HAVE_indirect_jump (Pmode == SImode)
     ^
    ./insn-flags.h:796:58: error: ‘rtx_def* gen_indirect_jump(rtx)’ was declared ‘extern’ and later ‘static’ [-fpermissive]
     static inline rtx gen_indirect_jump                  (rtx);
                                                              ^
    ./insn-flags.h:795:19: note: previous declaration of ‘rtx_def* gen_indirect_jump(rtx)’
     extern rtx        gen_indirect_jump                  (rtx);
                       ^
    ./insn-flags.h: In function ‘rtx_def* gen_indirect_jump(rtx)’:
    ./insn-flags.h:804:1: error: redefinition of ‘rtx_def* gen_indirect_jump(rtx)’
     gen_indirect_jump(rtx ARG_UNUSED (a))
     ^
    ./insn-flags.h:798:1: note: ‘rtx_def* gen_indirect_jump(rtx)’ previously defined here
     gen_indirect_jump(rtx ARG_UNUSED (a))
     ^

Do we need additional limitition/completement to follow with the related
define_insn?

And if we can really let pointer_operand has type, I guess, we also need
remove the related tile code in "gcc/config/tilegx/tilegx.c" (original
code assumed that pointer_operand is VOIDmode):

  3604           /* pointer_operand and pmode_register_operand operands do
  3605              not specify a mode, so use the operand's mode instead
  3606              (which should always be right by the time we get here,
  3607              except for constants, which are VOIDmode).  */
  3608           if (opmode == VOIDmode)
  3609             {
  3610               machine_mode m = GET_MODE (op[opnum]);
  3611               gcc_assert (m == Pmode || m == VOIDmode);
  3612               opmode = Pmode;
  3613             }


Thanks.
-- 
Chen Gang

Open, share, and attitude like air, water, and life which God blessed