On Sun, Jul 12, 2015 at 1:08 PM, Segher Boessenkool
<seg...@kernel.crashing.org> wrote:
> This rewrites all the rotate-and-mask and insert patterns.
>
> The goals are to have simpler, shorter, less error-prone code (with much
> fewer machine description patterns), as well as to get better optimised
> machine code.
>
> All "mask only" insns are now handled by a single pattern; all rotate (or
> shift) and mask by another pattern; and all insert insns by just a few
> patterns as well. Also, more patterns that can be done by a combination
> of two of these (or a load immediate and one) are now done that way.
>
> This removes a few constraints and output modifiers. I don't think any
> inline assembler code uses those (they really shouldn't), but if in fact
> it is used, it is easy to put back simpler definitions just for
> compatibility. We can use a few free letters for both constraints and
> modifiers.
>
> Bootstrapped and tested on powerpc64-linux with the usual options
> (-m32,-m32/-mpowerpc64,-m64,-m64/-mlra); no regressions.
>
> This reduces the number of generated instructions for 32-bit by about
> 1 in 2000, and for 64-bit by about 1 in 700. It always does that by
> decreasing path length so it should never make performance worse either.
>
> There are two FIXMEs for pretty harmless stuff, but the rtx_costs one
> at least probably wants fixing soonish.
>
> Oh, and this patch should be applied after the HAVE_insv patch I posted
> earlier today, for good results.
>
>
> Is this okay for trunk?
>
>
> 2015-07-12 Segher Boessenkool <seg...@kernel.crashing.org>
>
> PR target/66217
> * config/rs6000/constraints.md ("S", "T", "t"): Delete.
> * config/rs6000/predicates.md (mask_operand, mask_operand_wrap,
> mask64_operand, mask64_2_operand, any_mask_operand, and64_2_operand,
> and_2rld_operand): Delete.
> (and_operand): Adjust.
> (rotate_mask_operator): New.
> * config/rs6000/rs6000-protos.h (build_mask64_2_operands,
> includes_lshift_p, includes_rshift_p, includes_rldic_lshift_p,
> includes_rldicr_lshift_p, insvdi_rshift_rlwimi_p, extract_MB,
> extract_ME): Delete.
> (rs6000_is_valid_mask, rs6000_is_valid_and_mask,
> rs6000_is_valid_shift_mask, rs6000_is_valid_insert_mask,
> rs6000_insn_for_and_mask, rs6000_insn_for_shift_mask,
> rs6000_insn_for_insert_mask, rs6000_is_valid_2insn_and,
> rs6000_emit_2insn_and): New.
> * config/rs6000/rs6000.c (num_insns_constant): Adjust.
> (build_mask64_2_operands, includes_lshift_p, includes_rshift_p,
> includes_rldic_lshift_p, includes_rldicr_lshift_p,
> insvdi_rshift_rlwimi_p, extract_MB, extract_ME): Delete.
> (rs6000_is_valid_mask, rs6000_is_valid_and_mask,
> rs6000_insn_for_and_mask, rs6000_is_valid_shift_mask,
> s6000_insn_for_shift_mask, rs6000_is_valid_insert_mask,
> rs6000_insn_for_insert_mask, rs6000_is_valid_2insn_and,
> rs6000_emit_2insn_and): New.
> (print_operand) <'b', 'B', 'm', 'M', 's', 'S', 'W'>: Delete.
> (rs6000_rtx_costs) <CONST_INT>: Delete mask_operand and mask64_operand
> handling.
> <NOT>: Don't fall through to next case.
> <AND>: Handle the various rotate-and-mask cases directly.
> <IOR>: Always cost as one insn.
> * config/rs6000/rs6000.md (splitter for bswap:SI): Adjust.
> (and<mode>3): Adjust expander for the new patterns.
> (and<mode>3_imm, and<mode>3_imm_dot, and<mode>3_imm_dot2,
> and<mode>3_imm_mask_dot, and<mode>3_imm_mask_dot2): Adjust condition.
> (*and<mode>3_imm_dot_shifted): New.
> (*and<mode>3_mask): Delete, rewrite as ...
> (and<mode>3_mask): ... New.
> (*and<mode>3_mask_dot, *and<mode>3_mask_dot): Rewrite.
> (andsi3_internal0_nomc): Delete.
> (*andsi3_internal6): Delete.
> (*and<mode>3_2insn): New.
> (insv, insvsi_internal, *insvsi_internal1, *insvsi_internal2,
> *insvsi_internal3, *insvsi_internal4, *insvsi_internal5,
> *insvsi_internal6, insvdi_internal, *insvdi_internal2,
> *insvdi_internal3): Delete.
> (*rotl<mode>3_mask, *rotl<mode>3_mask_dot, *rotl<mode>3_mask_dot2,
> *rotl<mode>3_insert, *rotl<mode>3_insert_2, *rotl<mode>3_insert_3,
> *rotl<mode>3_insert_4, two splitters for multi-precision shifts,
> *ior<mode>_mask): New.
> (extzv, extzvdi_internal, *extzvdi_internal1, *extzvdi_internal2,
> *rotlsi3_mask, *rotlsi3_mask_dot, *rotlsi3_mask_dot2,
> *ashlsi3_imm_mask, *ashlsi3_imm_mask_dot, *ashlsi3_imm_mask_dot2,
> *lshrsi3_imm_mask, *lshrsi3_imm_mask_dot, *lshrsi3_imm_mask_dot2):
> Delete.
> (ashr<mode>3): Delete expander.
> (*ashr<mode>3): Rename to ...
> (ashr<mode>3): ... This.
> (ashrdi3_no_power, *ashrdisi3_noppc64be): Delete.
> (*rotldi3_internal4, *rotldi3_internal5 and split,
> *rotldi3_internal6 and split, *ashldi3_internal4, ashldi3_internal5
> and split, *ashldi3_internal6 and split, *ashldi3_internal7,
> ashldi3_internal8 and split, *ashldi3_internal9 and split): Delete.
> (*anddi3_2rld, *anddi3_2rld_dot, *anddi3_2rld_dot2): Delete.
> (splitter for loading a mask): Adjust.
This patch is okay with the follow-up revisions.
Thanks, David
