Comments on: Signed Overflow

By: January 11, 2008 « Everything is Data

January 11, 2008 « Everything is Data — Sat, 02 May 2009 00:08:26 +0000

[...] Ian Lance Taylor’s blog has an interesting post on signed overflow behavior in C. According to the C standard, integer overflow results in undefined behavior, and modern versions of GCC take advantage of this to generate more efficient code. This topic was raised on -hackers by Tom a few years ago — at the time, only the -fwrapv flag was implemented by GCC. Now that GCC 4.2 provides -Wstrict-overflow, this might be worth investigating further. [...]

By: Ian Lance Taylor

Ian Lance Taylor — Fri, 11 Jan 2008 15:36:37 +0000

There is no warning about the last example from gcc mainline because gcc doesn’t actually optimize it differently based on undefined signed overflow. It just goes ahead and generated the overflow code. I wasn’t clear: I didn’t intend that to be an example of using overflow to optimize, I intended to show a case where -fwrapv differed from -fno-strict-overflow.

By: ncm

ncm — Thu, 10 Jan 2008 22:53:21 +0000

I see what happened:

$ gcc -O3 -Wstrict-overflow=5 -Wall -c t.c

is not the same as

$ gcc -O3 -Wall -Wstrict-overflow=5 -c t.c

Evidently the -Wall in the first example bumps the warning level back down to 1. I see that gcc-4.2 warns only about the first example, I suppose because it doesn’t perform that optimization on the next two. I can’t seem to get a warning about the last example from gcc-4.3. Shouldn’t I? It looks like bad code.

By: Ian Lance Taylor

Ian Lance Taylor — Thu, 10 Jan 2008 18:21:06 +0000

I think the full effects only come in on gcc mainline. At least, gcc 4.2 does not eliminate the loop, but gcc 4.3 will. All my examples were run with gcc mainline.

By: ncm

ncm — Thu, 10 Jan 2008 03:13:10 +0000

I’m running

$ gcc –version

gcc (GCC) 4.2.3 20080102 (prerelease) (Debian 4.2.2-5)

and putting all your code examples into t.c and compiling with

$ gcc -O3 -Wstrict-overflow=5 -fstrict-overflow -Wall -W -c t.c

I get no warnings at all. Should I be surprised at this?

By: Ian Lance Taylor

Ian Lance Taylor — Thu, 10 Jan 2008 02:28:25 +0000

-Wstrict-overflow should issue a warning if you compile with -O2. -Wstrict-overflow warns about cases where an optimization relies on undefined overflow; it doesn’t have any way to detect possible overflow except when an optimization is applied. In some cases you will need to use -Wstrict-overflow=5.

-Wall includes -Wstrict-overflow=1. Specifying -Wstrict-overflow with no number is equivalent to -Wstrict-overflow=2.

The docs are available at http://gcc.gnu.org/onlinedocs/ .

By: ncm

ncm — Wed, 09 Jan 2008 22:08:18 +0000

I see that -Wstrict-overflow is accepted by Debian’s gcc-4.2.3. However, it issues no warnings for the code examples presented.

Is it included in -W or -Wall? (Debian has not packaged gcc-4.2 docs yet.)