On 4/16/19 10:29 AM, Steve Ellcey wrote:
> Re-ping. I know there are discussions about bigger changes to fix the
> various failures listed in PR rtl-optimization/87763 but this patch
> at least fixes one of them (gcc.target/aarch64/lsl_asr_sbfiz.c).
So I think we can address this by making the extv/extzv expander handle
both modes just like we do for insv. That's cleaner than addding new
patterns to match the subreg nonsense IMHO.
Something like this:
> (define_expand "<optab><mode>"
> [(set (match_operand:GPI 0 "register_operand" "=r")
> (ANY_EXTRACT:GPI (match_operand:GPI 1 "register_operand")
> (match_operand 2
> "aarch64_simd_shift_imm_offset_<mode>")
> (match_operand 3 "aarch64_simd_shift_imm_<mode>")))]
> ""
> {
> if (!IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]),
> 1, GET_MODE_BITSIZE (<MODE>mode) - 1))
> FAIL;
> }
> )
Note how we're using <mode>, <MODE> and the GPI iterator. With that
updated expander I get:
sbfiz32:
sbfiz w0, w0, 19, 3 // 12 [c=8 l=4] *ashiftsi_extv_bfiz
ret // 21 [c=0 l=4] *do_return
and
sbfiz64:
sbfiz x0, x0, 38, 6 // 12 [c=8 l=4] *ashiftdi_extv_bfiz
ret // 21 [c=0 l=4] *do_return
I'm going to throw that into the tester and see what we get...
jeff
