Hi Nikita,
> While investigating Zicond extension code generation on RISC-V, I identified
> several cases where GCC (trunk) generates suboptimal code due to premature
> if-conversion.
>
> Consider the following test case:
> CFLAGS: -march=rv64gc_zicond -mabi=lp64d -O2
>
> int test_IOR_ceqz_x (int x, int z, int c)
> {
> if (c)
> x = x | z;
> return x;
> }
>
> Before the patch:
> or a1,a0,a1
> czero.eqz a1,a1,a2
> czero.nez a0,a0,a2
> add a0,a0,a1
> ret
>
> The issue occurs when ifcvt encounters the following RTL pattern:
> (set reg1 (ior:DI (reg2:DI) (reg3:DI)))
> (set reg4 (sign_extend:DI (subreg:SI (reg1:DI))))
>
> When reg1 is no longer used, this expression could be simplified. However,
> noce_convert_multiple_sets converts the block early, preventing combine from
> optimizing the pattern.
yes, that looks not ideal indeed. I wonder if checking for sign_extension is
not too specific and we'd rather want something like
"is the single use of the current SET_DEST inside the BB we transform"?
That probably has unintended consequences as well, though...
--
Regards
Robin
