On 05/20/2011 05:39 AM, Kirill Batuzov wrote:
> + case INDEX_op_or_i32:
> + case INDEX_op_and_i32:
> +#if TCG_TARGET_REG_BITS == 64
> + case INDEX_op_and_i64:
> + case INDEX_op_or_i64:
> +#endif
> + if (args[1] == args[2]) {
> + if (args[1] == args[0]) {
> + args += 3;
> + gen_opc_buf[op_index] = INDEX_op_nop;
> + } else {
I do wonder if it would be better to split this sort of optimization
out into a different function. You're applying identity sorts of
functions here, where you're not doing it for other operations,
such as x + 0.
Indeed, I'll argue that 0+x is more likely to happen than x|x, given
that the 0 value could have been relocation filled in by the linker.
Consider @hi16 and @lo16 relocation pairs when the symbol happens to
be linked into the low 64k of the address space.
r~
