Bob Plantz wrote: > How can that be? The thing that matters is the machine code. Both > syntaxes generate exactly the same machine code. Because the MSVC style of inline assembler doesn't allow for specifying any constraints such as which registers or stack slots are clobbered. This means that compiler can't assume anything about the state before and after the block, it must just throw away all dataflow information it had before the block and assume everything was clobbered, leading to tons of useless redundant loads/stores. The GNU style inline asm works within the framework of the optimizing compiler, rather than outside it by totally going behind it's back. Read the long thread that starts here: <http://gcc.gnu.org/ml/gcc/2004-05/threads.html#00070>. Brian
