Christophe Lyon via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> @@ -19352,7 +19363,9 @@ aarch64_legitimate_constant_p (machine_mode mode, rtx
> x)
> {
> /* Support CSE and rematerialization of common constants. */
> if (CONST_INT_P (x)
> - || (CONST_DOUBLE_P (x) && GET_MODE_CLASS (mode) == MODE_FLOAT))
> + || (CONST_DOUBLE_P (x)
> + && (GET_MODE_CLASS (mode) == MODE_FLOAT
> + || GET_MODE_CLASS (mode) == MODE_DECIMAL_FLOAT)))
> return true;
>
> /* Only accept variable-length vector constants if they can be
I'd prefer to remove the mode check here, rather than extend it.
If we see a CONST_DOUBLE that doesn't have a scalar float mode
then something has gone wrong. (No need to assert for that though.
It's a fundamental assumption when TARGET_SUPPORTS_WIDE_INT.)
For consistency…
> @@ -11114,11 +11116,16 @@ aarch64_float_const_rtx_p (rtx x)
> return false;
> }
>
> -/* Return TRUE if rtx X is immediate constant 0.0 */
> +/* Return TRUE if rtx X is immediate constant 0.0 (but not in Decimal
> + Floating Point). */
> bool
> aarch64_float_const_zero_rtx_p (rtx x)
> {
> - if (GET_MODE (x) == VOIDmode)
> + if (GET_MODE (x) == VOIDmode
> + /* 0.0 in Decimal Floating Point cannot be represented by #0 or
> + zr as our callers expect, so no need to check the actual
> + value if X is of Decimal Floating Point type. */
> + || GET_MODE_CLASS (GET_MODE (x)) == MODE_DECIMAL_FLOAT)
…it would be good to drop the VOIDmode check here at the same time.
OK with those changes, thanks.
Richard
