> The following testcase FAILs, because cse_local sees
> (zero_extend:TI (subreg/s/v:DI (reg:TI ...) 0))
> inside of REG_EQUAL note, and simplify-rtx.c attempts to optimize it.
> case ZERO_EXTEND:
> /* Check for a zero extension of a subreg of a promoted
> variable, where the promotion is zero-extended, and the
> target mode is the same as the variable's promotion. */
> if (GET_CODE (op) == SUBREG
> && SUBREG_PROMOTED_VAR_P (op)
> && SUBREG_PROMOTED_UNSIGNED_P (op)
> && !paradoxical_subreg_p (mode, GET_MODE (SUBREG_REG (op))))
> {
> temp = rtl_hooks.gen_lowpart_no_emit (mode, op);
> if (temp)
> return temp;
> }
This code is strange though, here's the equivalent one in convert_modes:
if (GET_CODE (x) == SUBREG
&& SUBREG_PROMOTED_VAR_P (x)
&& is_a <scalar_int_mode> (mode, &int_mode)
&& (GET_MODE_PRECISION (subreg_promoted_mode (x))
>= GET_MODE_PRECISION (int_mode))
&& SUBREG_CHECK_PROMOTED_SIGN (x, unsignedp))
x = gen_lowpart (int_mode, SUBREG_REG (x));
so can't we just pass SUBREG_REG (op) here? Same for SIGN_EXTEND above.
--
Eric Botcazou
