Re: PATCH: PR rtl-optimization/54157: [x32] -maddress-mode=long failures

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Wed, Aug 1, 2012 at 12:14 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Wed, Aug 1, 2012 at 11:58 AM, Richard Sandiford
> <rdsandiford@googlemail.com> wrote:
>> "H.J. Lu" <hongjiu.lu@intel.com> writes:
>>> We have
>>>
>>> (gdb) r -fpreprocessed x.i -quiet -dumpbase x.i -mx32
>>> -maddress-mode=long -mtune=generic -march=x86-64 -auxbase x -O2 -version
>>> -ftree-vectorize -o x.s
>>> Starting program: /export/build/gnu/gcc-x32/build-x86_64-linux/gcc/cc1
>>> -fpreprocessed x.i -quiet -dumpbase x.i -mx32 -maddress-mode=long
>>> -mtune=generic -march=x86-64 -auxbase x -O2 -version -ftree-vectorize -o
>>> x.s
>>> GNU C (GCC) version 4.8.0 20120801 (experimental)
>>> (x86_64-unknown-linux-gnu)
>>>       compiled by GNU C version 4.7.1 20120629 (Red Hat 4.7.1-1), GMP
>>> version 5.0.2, MPFR version 3.1.0, MPC version 0.9
>>> GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096
>>> GNU C (GCC) version 4.8.0 20120801 (experimental)
>>> (x86_64-unknown-linux-gnu)
>>>       compiled by GNU C version 4.7.1 20120629 (Red Hat 4.7.1-1), GMP
>>> version 5.0.2, MPFR version 3.1.0, MPC version 0.9
>>> GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096
>>> Compiler executable checksum: 07a4e516c4e8fe4abfdafa83737d8f4a
>>>
>>> Breakpoint 1, fancy_abort (
>>>     file=0x130fe68 "/export/gnu/import/git/gcc/gcc/explow.c", line=88,
>>>     function=0x131032e <__FUNCTION__.39220> "plus_constant")
>>>     at /export/gnu/import/git/gcc/gcc/diagnostic.c:1011
>>> 1011    internal_error ("in %s, at %s:%d", function, trim_filename
>>> (file), line);
>>> (gdb) f 1
>>> #1  0x0000000000743e07 in plus_constant (mode=DImode, x=0x7ffff106a7e0,
>>>     c=99452) at /export/gnu/import/git/gcc/gcc/explow.c:88
>>> 88      gcc_assert (GET_MODE (x) == VOIDmode || GET_MODE (x) == mode);
>>> (gdb) f 2
>>> #2  0x0000000000adc4b1 in simplify_binary_operation_1 (code=PLUS,
>>> mode=DImode,
>>>     op0=0x7ffff106a7e0, op1=0x7ffff1010e80, trueop0=0x7ffff106a7e0,
>>>     trueop1=0x7ffff1010e80)
>>>     at /export/gnu/import/git/gcc/gcc/simplify-rtx.c:1956
>>> 1956          return plus_constant (mode, op0, INTVAL (op1));
>>> (gdb) call debug_rtx (op0)
>>> (symbol_ref:SI ("tmp2") <var_decl 0x7ffff0f06140 tmp2>)
>>> (gdb) call debug_rtx (op1)
>>> (const_int 99452 [0x1847c])
>>> (gdb) bt
>>> #0  fancy_abort (file=0x130fe68
>>> "/export/gnu/import/git/gcc/gcc/explow.c",
>>>     line=88, function=0x131032e <__FUNCTION__.39220> "plus_constant")
>>>     at /export/gnu/import/git/gcc/gcc/diagnostic.c:1011
>>> #1  0x0000000000743e07 in plus_constant (mode=DImode, x=0x7ffff106a7e0,
>>>     c=99452) at /export/gnu/import/git/gcc/gcc/explow.c:88
>>> #2  0x0000000000adc4b1 in simplify_binary_operation_1 (code=PLUS,
>>> mode=DImode,
>>>     op0=0x7ffff106a7e0, op1=0x7ffff1010e80, trueop0=0x7ffff106a7e0,
>>>     trueop1=0x7ffff1010e80)
>>>     at /export/gnu/import/git/gcc/gcc/simplify-rtx.c:1956
>>> #3  0x0000000000adc221 in simplify_binary_operation (code=PLUS,
>>> mode=DImode,
>>>     op0=0x7ffff106a7e0, op1=0x7ffff1010e80)
>>>     at /export/gnu/import/git/gcc/gcc/simplify-rtx.c:1904
>>
>> Things have already gone wrong by this frame: we have a DImode
>> addition of an SImode value, which isn't allowed.  Where does
>> that mismatch get introduced?
>>
>
> make_extraction in combine generates:
>
> 7474          inner = force_to_mode (inner, wanted_inner_mode,
> 7475                                 pos_rtx
> 7476                                 || len + orig_pos >= HOST_BITS_PER_WIDE_INT
> 7477                                 ? ~(unsigned HOST_WIDE_INT) 0
> 7478                                 : ((((unsigned HOST_WIDE_INT) 1 << len) - 1)
> (gdb) call debug_rtx (inner)
> (plus:SI (reg:SI 109 [ D.1765 ])
>     (const:SI (plus:SI (symbol_ref:SI ("tmp2") <var_decl 0x7ffff0f06140 tmp2>)
>             (const_int 99452 [0x1847c]))))
> (gdb) p wanted_inner_mode
> $2 = DImode
> (gdb)
>

i386,md has

(define_expand "extzv"
  [(set (match_operand:SI 0 "register_operand")
        (zero_extract:SI (match_operand 1 "ext_register_operand")
                         (match_operand:SI 2 "const8_operand")
                         (match_operand:SI 3 "const8_operand")))]
  ""

and mode_for_extraction picks word_mode for operand 1 since
its mode is VOIDmode.  This patch changes mode_for_extraction
to return the mode of operand 1 if the pattern accepts any mode.
I added *jcc_btsi_mask_2 since combine now tries a different
pattern, which leads to test failures on gcc.target/i386/bt-mask-1.c
and  gcc.target/i386/bt-mask-2.  I didn't update *jcc_btsi_mask_1
instead since I am not sure if it is used elsewhere.  Tested on
Linux/x86-64 and Linux/x32.  OK for trunk?

Thanks.

-- 
H.J.
----
gcc/

2012-08-02  H.J. Lu  <hongjiu.lu@intel.com>

	PR rtl-optimization/54157
	* combine.c (make_extraction): Pass inner_mode to mode_for_extraction
	for operand 1, otherwise pass VOIDmode.
	(simplify_comparison): Pass VOIDmode to mode_for_extraction.

	* expmed.c (mode_for_extraction): Add a machine_mode argument
	and use it instead of word_mode if it isn't VOIDmode.
	(store_bit_field_1): Pass VOIDmode to mode_for_extraction.
	(store_bit_field): Likewise.

	* expr.h (mode_for_extraction): Add a machine_mode argument.

	* config/i386/i386.md (*jcc_btsi_mask_2): New insn_and_split
	pattern.

	* recog.c (simplify_while_replacing): Pass is_mode to
	mode_for_extraction.

gcc/testsuite/

2012-08-02  H.J. Lu  <hongjiu.lu@intel.com>

	PR rtl-optimization/54157
	* gcc.target/i386/pr54157.c: New file.

Attachment: gcc-pr54157.patch
Description: Binary data