------- Additional Comments From cnewbold at mathworks dot com 2005-06-20
20:46 -------
Subject: Re: Optimization stomps const,
initialized local array
On Mon, 2005-06-20 at 20:39 +0000, pinskia at gcc dot gnu dot org wrote:
> I don't see the problem on the mainline, maybe I not understanding what you
> mean.
The problem is that once we're past the initialization and into the body
of the block, the array does not have the correct content. In the
example I cited, the first byte of 'encrypted' does not contain 0xf3,
but instead some random trash. Here's a break down of the assembler for
the opening of that block:
First byte of 'encrypted' (correctly) loaded with 0xf3:
2cab: c6 85 08 ff ff ff f3 movb $0xf3,0xffffff08(%ebp)
Something else dropped in here, which stomps the first element:
2cb2: ba 08 00 00 00 mov $0x8,%edx
2cb7: b9 08 00 00 00 mov $0x8,%ecx
2cbc: 89 95 08 ff ff ff mov %edx,0xffffff08(%ebp)
^^^^^^^^^^^^^^^^
OOOPS!
2cc2: 31 c0 xor %eax,%eax
2cc4: 8d 55 b8 lea 0xffffffb8(%ebp),%edx
2cc7: 89 8d 28 ff ff ff mov %ecx,0xffffff28(%ebp)
2ccd: 8d 8d 30 ff ff ff lea 0xffffff30(%ebp),%ecx
Remaining 23 bytes of 'encrypted' are loaded:
2cd3: c6 85 09 ff ff ff 09 movb $0x9,0xffffff09(%ebp)
2cda: c6 85 0a ff ff ff 62 movb $0x62,0xffffff0a(%ebp)
...
-Chris
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22129
