> The following patch implements this optimization. It checks for a range
> that does not have the sign-bit set, and an index value that is already
> sign extended, and then does a sign extend instead of an zero extend.
>
> This has been tested with a riscv{32,64}-{elf,linux} builds and testsuite
> runs. There were no regressions. It was also tested with an x86_64-linux
> build and testsuite run.
>
> Ok?
>
> Jim
>
> gcc/
> * expr.c (do_tablejump): When converting index to Pmode, if we have a
> sign extended promoted subreg, and the range does not have the sign bit
> set, then do a sign extend.
Richard dragged me into this so I feel somewhat entitled to step up...
The patch looks OK to me, modulo:
> + /* We know the value of INDEX is between 0 and RANGE. If we have a
> + sign-extended subreg, and RANGE does not have the sign bit set, then
> + we have a value that is valid for both sign and zero extension. In
> + this case, we get better code if we sign extend. */
> + if (GET_CODE (index) == SUBREG
> + && SUBREG_PROMOTED_VAR_P (index)
> + && SUBREG_PROMOTED_SIGNED_P (index)
> + && ((width = GET_MODE_PRECISION (as_a <scalar_int_mode> (mode)))
> + <= HOST_BITS_PER_WIDE_INT)
> + && ! (INTVAL (range) & (HOST_WIDE_INT_1U << (width - 1))))
I'd use UINTVAL instead of INTVAL here.
--
Eric Botcazou
