Gcc with optimization flags creates incorrect machine code.

Binary package hint: gcc

Gcc version - 4.0.3 (Ubuntu 4.0.3-1ubuntu5)

Kubuntu version dapper drake - 6.06

code was compiled with the following options....

-O3 -g3 -ggdb -Wall -mtune=pentium4 -mmmx -msse -msse2 -ffast-math -funsafe-math-optimizations -ffinite-math-only -fno-trapping-math -fsingle-precision-constant -fprefetch-loop-arrays -fomit-frame-pointer -mno-push-args -DNDEBUG -D_GNU_SOURCE -D_REENTRANT

The following C code snipit was compiled incorrectly...

#define LARGEST_BIT(x,retval) __asm("bsrl %0, %1;" : "=r"(retval) : "r"(x))
#define MAX_CAHCEBLOCK_SIZE 4096
<.....>
const unsigned int numBits = sizeof( size_t ) * 8;
unsigned int largestBit;
unsigned int shiftOffset;
size_t floorOfSize;
size_t ceilOfSize;
LARGEST_BIT( size, largestBit );
floorOfSize = 1 << largestBit;
shiftOffset = ( ( floorOfSize - size ) >> ( numBits - 1 ) );
ceilOfSize = 1 << ( shiftOffset + largestBit) ;
return ( size < MAX_CAHCEBLOCK_SIZE ) ? ceilOfSize : size;
<.....>
and the resultant asm code was...

bsr %ecx,%ebx
mov $0x1, %edx
mov %edx,%eax
shl %cl,%eax
sub %ebx,%eax
shr $0x1f,%eax
add %eax,%ecx
shl %cl,%edx
cmp $0xfff,%ebx
cmvbe %edx,%ebx

If the C variable size equals 12 the C code will return 16, the asm code from gcc will return 4096.