Re: [PATCH 4/5] sparse, i386: Fix boolean bit size

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Aug 26, 2011 at 8:28 AM, Pekka Enberg <penberg@xxxxxxxxxx> wrote:
> On Fri, Aug 26, 2011 at 6:59 AM, Christopher Li <sparse@xxxxxxxxxxx> wrote:
>> On Mon, Aug 22, 2011 at 6:57 AM, Pekka Enberg <penberg@xxxxxxxxxx> wrote:
>>> The value of 'ctype->bit_size' is set to 1 for booleans which confuses the i386
>>> backend:
>>>
>>>  ./compile allocate.c
>>>  compile: compile-i386.c:1406: emit_binop: Assertion `0' failed.
>>>  Aborted
>>>
>>> Looking at the code, we assume that "bit_size / 8" gives a sane result on
>>> various places. This patch fixes the problem by bumping bit_size to 8 for
>>> booleans. This also makes sizeof(_Bool) return 1 which is consistent with what
>>> GCC 4.4.3, for example, does.
>>>
>>> diff --git a/target.c b/target.c
>>> index 17b228a..6a535bc 100644
>>> --- a/target.c
>>> +++ b/target.c
>>> @@ -14,7 +14,7 @@ int max_alignment = 16;
>>>  /*
>>>  * Integer data types
>>>  */
>>> -int bits_in_bool = 1;
>>> +int bits_in_bool = 8;
>>
>> I object this part. I consider the sizeof(_Bool) == 1 as external behaviour.
>> But internally we should know that the real useful part of bool is in just one
>> bit, not any bit of that  1 byte storage.
>
> You missed the most important part of my reasoning: sparse code
> already expects "bit_size / 8" to return a non-zero number and it's
> not just compile-i386.c! So while I don't disagree with you that we
> should internally know that a bool is just one bit, I don't consider
> that to be relevant for this particular patch.

Oh, I see the same problem with sparse-llvm when generating code for OP_CAST.

This little C function:

  int sete(int x, int y)
  {
    return x == y;
  }

is compiled by GCC to:

00000170 <setne>:
 170:	55                   	push   %ebp
 171:	89 e5                	mov    %esp,%ebp
 173:	8b 45 0c             	mov    0xc(%ebp),%eax
 176:	39 45 08             	cmp    %eax,0x8(%ebp)
 179:	5d                   	pop    %ebp
 17a:	0f 95 c0             	setne  %al
 17d:	0f b6 c0             	movzbl %al,%eax
 180:	c3                   	ret

Sparse-llvm compiles the code to this which seems wrong:

00000140 <sete>:
 140:	31 c9                	xor    %ecx,%ecx
 142:	8b 44 24 08          	mov    0x8(%esp),%eax
 146:	39 44 24 04          	cmp    %eax,0x4(%esp)
 14a:	b8 ff ff ff ff       	mov    $0xffffffff,%eax
 14f:	0f 45 c1             	cmovne %ecx,%eax
 152:	c3                   	ret

However, if I bump up 'bits_in_bool' to 32:

diff --git a/target.c b/target.c
index 17b228a..2b83498 100644
--- a/target.c
+++ b/target.c
@@ -14,7 +14,7 @@ int max_alignment = 16;
 /*
  * Integer data types
  */
-int bits_in_bool = 1;
+int bits_in_bool = 32;
 int bits_in_char = 8;
 int bits_in_short = 16;
 int bits_in_int = 32;

I get this from sparse-llvm:

00000140 <sete>:
 140:	8b 44 24 08          	mov    0x8(%esp),%eax
 144:	39 44 24 04          	cmp    %eax,0x4(%esp)
 148:	0f 94 c0             	sete   %al
 14b:	c3                   	ret

I don't know sparse well enough to know what's the right thing to do
here. Linus, Jeff, Chris, someone, anyone, help!!!

                        Pekka
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Newbies FAQ]     [LKML]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Trinity Fuzzer Tool]

  Powered by Linux