From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id 64548385383E for ; Mon, 26 Jul 2021 19:31:59 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 64548385383E Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 1017031B; Mon, 26 Jul 2021 12:31:59 -0700 (PDT) Received: from localhost (e121540-lin.manchester.arm.com [10.32.98.126]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 8E3323F66F; Mon, 26 Jul 2021 12:31:58 -0700 (PDT) From: Richard Sandiford To: Robin Dapp Mail-Followup-To: Robin Dapp , gcc-patches@gcc.gnu.org, richard.sandiford@arm.com Cc: gcc-patches@gcc.gnu.org Subject: Re: [PATCH 4/7] ifcvt/optabs: Allow using a CC comparison for emit_conditional_move. References: <20210625160905.23786-1-rdapp@linux.ibm.com> <20210625160905.23786-5-rdapp@linux.ibm.com> <06f8a450-4da1-b7e3-4b5b-e7085cfa822a@linux.ibm.com> Date: Mon, 26 Jul 2021 20:31:57 +0100 In-Reply-To: <06f8a450-4da1-b7e3-4b5b-e7085cfa822a@linux.ibm.com> (Robin Dapp's message of "Thu, 22 Jul 2021 14:07:18 +0200") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-12.4 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_STATUS, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 26 Jul 2021 19:32:01 -0000 > Before, we had a loop with two iterations that tried "emit_cmov (cond,=20 > op2, op3)". op2 and op3 can (or even were) If this did not succeed we=20 > would revert the condition as well as op3 and op3 in-place and try=20 > again. I found that a bit cumbersome and intended to make this explicit= =20 > but it's still kind of involved, particularly since cond may come in=20 > reversed, we additionally swap op2 and op3 and all the way back again. > > I remember from the first iteration of these patches that this (double=20 > revert) was needed for exactly one test case in the test suite on x86.=20 > When re-running it today I could not reproduce it anymore, though. Hmm, OK. Doesn't expanding both versions up-front create the same kind of problem that the patch is fixing, in that we expand (and therefore cost) both the reversed and unreversed comparison? Also=E2=80=A6 Robin Dapp writes: > @@ -4782,17 +4784,30 @@ emit_conditional_move (rtx target, enum rtx_code = code, rtx op0, rtx op1, > if (cmode =3D=3D VOIDmode) > cmode =3D GET_MODE (op0); >=20=20 > - enum rtx_code orig_code =3D code; > + /* If the first condition operand is constant and the second is not, s= wap > + them. In that case we also need to reverse the comparison. > + If both are non-constant It is possible that the conditional move > + will not expand with operands in this order, e.g. when we are about= to > + create a min/max expression and the operands do not match the condi= tion. > + In that case we also need the reversed condition code and compariso= n. */ =E2=80=A6for min/max, I would have expected this swap to create the canonic= al operand order for the min and max too. What causes it to be rejected? > + > + rtx rev_comparison =3D NULL_RTX; > bool swapped =3D false; > - if (swap_commutative_operands_p (op2, op3) > - && ((reversed =3D reversed_comparison_code_parts (code, op0, op1, = NULL)) > - !=3D UNKNOWN)) > + > + code =3D unsignedp ? unsigned_condition (code) : code; > + comparison =3D simplify_gen_relational (code, VOIDmode, cmode, op0, op= 1); > + > + if ((reversed =3D reversed_comparison_code_parts (code, op0, op1, NULL= )) > + !=3D UNKNOWN) > { > - std::swap (op2, op3); > - code =3D reversed; > - swapped =3D true; > + reversed =3D unsignedp ? unsigned_condition (reversed) : reversed; When is this needed? I'd have expected the reversed from of an unsigned code to be naturally unsigned. Thanks, Richard > + rev_comparison =3D simplify_gen_relational (reversed, VOIDmode, cm= ode, > + op0, op1); > } >=20=20 > + if (swap_commutative_operands_p (op2, op3) && reversed !=3D UNKNOWN) > + swapped =3D true; > + > if (mode =3D=3D VOIDmode) > mode =3D GET_MODE (op2); >=20=20 > @@ -4804,58 +4819,99 @@ emit_conditional_move (rtx target, enum rtx_code = code, rtx op0, rtx op1, > if (!target) > target =3D gen_reg_rtx (mode); >=20=20 > - for (int pass =3D 0; ; pass++) > + if (comparison && COMPARISON_P (comparison)) > + prepare_cmp_insn (XEXP (comparison, 0), XEXP (comparison, 1), > + GET_CODE (comparison), NULL_RTX, > + unsignedp, OPTAB_WIDEN, &comparison, &cmode); > + else > + return NULL_RTX; > + > + if (rev_comparison && COMPARISON_P (rev_comparison)) > + prepare_cmp_insn (XEXP (rev_comparison, 0), XEXP (rev_comparison, 1), > + GET_CODE (rev_comparison), NULL_RTX, > + unsignedp, OPTAB_WIDEN, &rev_comparison, &cmode); > + > + if (!swapped) > + return emit_conditional_move (target, comparison, rev_comparison, > + op2, op3, mode); > + else > + return emit_conditional_move (target, rev_comparison, comparison, > + op3, op2, mode); > +} > + > +/* Helper function for emitting a conditional move. Given a COMPARISON > + and a reversed REV_COMPARISON it will try to expand a conditional move > + with COMPARISON first and try with REV_COMPARISON if that fails. */ > + > +rtx > +emit_conditional_move (rtx target, rtx comparison, rtx rev_comparison, > + rtx op2, rtx op3, machine_mode mode) > +{ > + > + rtx res =3D emit_conditional_move (target, comparison, op2, op3, mode); > + > + if (res !=3D NULL_RTX) > + return res; > + > + return emit_conditional_move (target, rev_comparison, op3, op2, mode); > +} > + > +/* Helper for emitting a conditional move. */ > + > +static rtx > +emit_conditional_move (rtx target, rtx comparison, > + rtx op2, rtx op3, machine_mode mode) > +{ > + rtx_insn *last; > + enum insn_code icode; > + > + if (comparison =3D=3D NULL_RTX || !COMPARISON_P (comparison)) > + return NULL_RTX; > + > + /* If the two source operands are identical, that's just a move. */ > + if (rtx_equal_p (op2, op3)) > { > - code =3D unsignedp ? unsigned_condition (code) : code; > - comparison =3D simplify_gen_relational (code, VOIDmode, cmode, op0= , op1); > + if (!target) > + target =3D gen_reg_rtx (mode); >=20=20 > - /* We can get const0_rtx or const_true_rtx in some circumstances. = Just > - punt and let the caller figure out how best to deal with this > - situation. */ > - if (COMPARISON_P (comparison)) > - { > - saved_pending_stack_adjust save; > - save_pending_stack_adjust (&save); > - last =3D get_last_insn (); > - do_pending_stack_adjust (); > - machine_mode cmpmode =3D cmode; > - prepare_cmp_insn (XEXP (comparison, 0), XEXP (comparison, 1), > - GET_CODE (comparison), NULL_RTX, unsignedp, > - OPTAB_WIDEN, &comparison, &cmpmode); > - if (comparison) > - { > - class expand_operand ops[4]; > + emit_move_insn (target, op3); > + return target; > + } >=20=20 > - create_output_operand (&ops[0], target, mode); > - create_fixed_operand (&ops[1], comparison); > - create_input_operand (&ops[2], op2, mode); > - create_input_operand (&ops[3], op3, mode); > - if (maybe_expand_insn (icode, 4, ops)) > - { > - if (ops[0].value !=3D target) > - convert_move (target, ops[0].value, false); > - return target; > - } > - } > - delete_insns_since (last); > - restore_pending_stack_adjust (&save); > - } > + if (mode =3D=3D VOIDmode) > + mode =3D GET_MODE (op2); >=20=20 > - if (pass =3D=3D 1) > - return NULL_RTX; > + icode =3D direct_optab_handler (movcc_optab, mode); >=20=20 > - /* If the preferred op2/op3 order is not usable, retry with other > - operand order, perhaps it will expand successfully. */ > - if (swapped) > - code =3D orig_code; > - else if ((reversed =3D reversed_comparison_code_parts (orig_code, = op0, op1, > - NULL)) > - !=3D UNKNOWN) > - code =3D reversed; > - else > - return NULL_RTX; > - std::swap (op2, op3); > + if (icode =3D=3D CODE_FOR_nothing) > + return NULL_RTX; > + > + if (!target) > + target =3D gen_reg_rtx (mode); > + > + saved_pending_stack_adjust save; > + save_pending_stack_adjust (&save); > + last =3D get_last_insn (); > + do_pending_stack_adjust (); > + > + class expand_operand ops[4]; > + > + create_output_operand (&ops[0], target, mode); > + create_fixed_operand (&ops[1], comparison); > + create_input_operand (&ops[2], op2, mode); > + create_input_operand (&ops[3], op3, mode); > + > + if (maybe_expand_insn (icode, 4, ops)) > + { > + if (ops[0].value !=3D target) > + convert_move (target, ops[0].value, false); > + return target; > } > + > + delete_insns_since (last); > + restore_pending_stack_adjust (&save); > + > + return NULL_RTX; > } >=20=20 >=20=20 > diff --git a/gcc/optabs.h b/gcc/optabs.h > index 3bbceff92d9..f853b93f37f 100644 > --- a/gcc/optabs.h > +++ b/gcc/optabs.h > @@ -281,6 +281,7 @@ extern void emit_indirect_jump (rtx); > /* Emit a conditional move operation. */ > rtx emit_conditional_move (rtx, enum rtx_code, rtx, rtx, machine_mode, > rtx, rtx, machine_mode, int); > +rtx emit_conditional_move (rtx, rtx, rtx, rtx, rtx, machine_mode); >=20=20 > /* Emit a conditional negate or bitwise complement operation. */ > rtx emit_conditional_neg_or_complement (rtx, rtx_code, machine_mode, rtx,