From patchwork Fri Jan 12 12:26:19 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Richard Sandiford X-Patchwork-Id: 187718 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:693c:2614:b0:101:6a76:bbe3 with SMTP id mm20csp136346dyc; Fri, 12 Jan 2024 04:27:08 -0800 (PST) X-Google-Smtp-Source: AGHT+IFczgWb9zwy27Zwx5lTr4FVt3d3Dd3t/a72NGfkDW6mBponybX7PbrgvImDRJ+540U6Co8w X-Received: by 2002:a05:620a:2a15:b0:783:3304:50de with SMTP id o21-20020a05620a2a1500b00783330450demr3261813qkp.41.1705062428295; Fri, 12 Jan 2024 04:27:08 -0800 (PST) ARC-Seal: i=2; a=rsa-sha256; t=1705062428; cv=pass; d=google.com; s=arc-20160816; b=hloPJjfdMc3nVrFYCuR/KPQs+xdFbe3gas7LRxqS806dr9IGqgU+SNnBvSRbGJ0xJV qbCCx4905+B5fHG0oWwZignxyueWrbLAHISbiUps13b7K7pAvZa6WeLwswOSTF8o2mkP 4NMNbV0p1sgn8iAEMPxG3Pe0p///eXnRFVgfEV9edlABhv1uRYkglvSwkw/yKohCK2zV UQdOldtkalqONHGnWybsQE7gBu9xWscTOBvRCNrzB3mWAoAxH8VW6ldJgzvHLKSdEBR0 o+3Bod+664hMgrlf4LZB/Oa8uDTdtD31//vlOhPTAmS0N2K2AbGArf2EAin7biS/udWi /7+A== ARC-Message-Signature: i=2; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=errors-to:list-subscribe:list-help:list-post:list-archive :list-unsubscribe:list-id:precedence:mime-version:user-agent :message-id:in-reply-to:date:references:subject:cc:mail-followup-to :to:from:arc-filter:dmarc-filter:delivered-to; bh=8pDd3KY29Bp2co0fyVnlFX1dkUmIXVT2ChKyryCBsdc=; fh=nbDyqvhaKlXmvshfq55b7fCRgXpAk7mT0ZaoI41MWcE=; b=LD/COZYu55rhI+Gw0xFN9yo6J4cnoVXWi9Is0BTDK5abPCn5z1ratsZxYmYkkzbzTP YD8uEnY4zmsSsAt4Icp/5iCT+XWSP1zcjLk5OLF1/fjill50co6+ZYXCs2CC+KspsK3F cl68e9+nADwtYWZmJ1odxyVA3gQfrsTQCJACk+6pmHlb9gjfXgeNDfM+Jed6EaueQ9RY qAa9jvZTtQ2JSKWpf5ZqbtKjcW72sIwhG193/K0rzMAiWum/AcjSoLg3DbqW2Wqo9crD aLmToaVQ2Jvmi+N92TxiiI0BpTXaZ4+MQfgx6uJY3MQlogd0JZrS9qkDcnmdmlL/HjoP yNhA== ARC-Authentication-Results: i=2; mx.google.com; arc=pass (i=1); spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org"; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=arm.com Received: from server2.sourceware.org (server2.sourceware.org. [8.43.85.97]) by mx.google.com with ESMTPS id j2-20020a37ef02000000b0078314a8b5b8si2764524qkk.497.2024.01.12.04.27.08 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 12 Jan 2024 04:27:08 -0800 (PST) Received-SPF: pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) client-ip=8.43.85.97; Authentication-Results: mx.google.com; arc=pass (i=1); spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org"; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=arm.com Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 026DC3858D20 for ; Fri, 12 Jan 2024 12:27:08 +0000 (GMT) X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id D9FD03858D20 for ; Fri, 12 Jan 2024 12:26:21 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org D9FD03858D20 Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=arm.com ARC-Filter: OpenARC Filter v1.0.0 sourceware.org D9FD03858D20 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1705062384; cv=none; b=eDg4ZnIk+41fC0nRiOZRtWmfHEoNmMBd9keOzVwr2CrA244oWmL1jqNziWx5mkkHCuROCcHXjnLlqJEX/C2ULZlATtVsQo8TWmrYQLlFZJ7LaxLAA1Gs5tr9OUCQCO+mI56nmoL0V8mxFIOWNH4DMRSTX3QoCYeIHBhWYEYSh0M= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1705062384; c=relaxed/simple; bh=b757irk+wzNWQrZeD0QVM9tfGknelh4yZwBN0OOcwas=; h=From:To:Subject:Date:Message-ID:MIME-Version; b=FbBZvllNN3aHxf+cQRuZFkhtuHE8uSSmnl4Ly0PkP0NHdcZ3QOqrQv4EE9Awu9B1/kRQd1qEx/FIcUzaM/oYBFUcSWs1Wq1FXA6OqD2uaG2BwU5VZqoYMMDptvvOzrnSraqmQMtbHOkjcx6hj0Qdsmftp48F6DA57E704InyOD8= ARC-Authentication-Results: i=1; server2.sourceware.org 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 A4CBC1FB; Fri, 12 Jan 2024 04:27:07 -0800 (PST) Received: from localhost (e121540-lin.manchester.arm.com [10.32.110.72]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id E57643F73F; Fri, 12 Jan 2024 04:26:20 -0800 (PST) From: Richard Sandiford To: Andrew Pinski Mail-Followup-To: Andrew Pinski , , richard.sandiford@arm.com Cc: Subject: [PATCHv3] aarch64/expr: Use ccmp when the outer expression is used twice [PR100942] References: <20240105074330.2309587-1-quic_apinski@quicinc.com> Date: Fri, 12 Jan 2024 12:26:19 +0000 In-Reply-To: <20240105074330.2309587-1-quic_apinski@quicinc.com> (Andrew Pinski's message of "Thu, 4 Jan 2024 23:43:30 -0800") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 X-Spam-Status: No, score=-19.4 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_NONE, KAM_DMARC_STATUS, KAM_LAZY_DOMAIN_SECURITY, KAM_SHORT, SPF_HELO_NONE, SPF_NONE, TXREP, T_SCC_BODY_TEXT_LINE, URIBL_BLACK, URIBL_SBL autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.30 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1787235589346905179 X-GMAIL-MSGID: 1787887540941497669 Andrew Pinski writes: > Ccmp is not used if the result of the and/ior is used by both > a GIMPLE_COND and a GIMPLE_ASSIGN. This improves the code generation > here by using ccmp in this case. > Two changes is required, first we need to allow the outer statement's > result be used more than once. > The second change is that during the expansion of the gimple, we need > to try using ccmp. This is needed because we don't use expand the ssa > name of the lhs but rather expand directly from the gimple. > > A small note on the ccmp_4.c testcase, we should be able to get slightly > better than with this patch but it is one extra instruction compared to > before. > > Diff from v1: > * v2: Split out expand_gimple_assign_ssa so the we only need to handle > promotion once. Add ccmp_5.c testcase which was suggested. Change comment > on ccmp_candidate_p. I meant more that we should split out the gassign handling in expand_expr_real_1, since we're effectively making cfgexpand follow it more closely. What do you think about the attached version? Tested on aarch64-linux-gnu and x86_64-linux-gnu. OK for the expr/cfgexpand bits? Thanks, Richard ---- Ccmp is not used if the result of the and/ior is used by both a GIMPLE_COND and a GIMPLE_ASSIGN. This improves the code generation here by using ccmp in this case. Two changes is required, first we need to allow the outer statement's result be used more than once. The second change is that during the expansion of the gimple, we need to try using ccmp. This is needed because we don't use expand the ssa name of the lhs but rather expand directly from the gimple. A small note on the ccmp_4.c testcase, we should be able to get slightly better than with this patch but it is one extra instruction compared to before. PR target/100942 gcc/ChangeLog: * ccmp.cc (ccmp_candidate_p): Add outer argument. Allow if the outer is true and the lhs is used more than once. (expand_ccmp_expr): Update call to ccmp_candidate_p. * expr.h (expand_expr_real_gassign): Declare. * expr.cc (expand_expr_real_gassign): New function, split out from... (expand_expr_real_1): ...here. * cfgexpand.cc (expand_gimple_stmt_1): Use expand_expr_real_gassign. gcc/testsuite/ChangeLog: * gcc.target/aarch64/ccmp_3.c: New test. * gcc.target/aarch64/ccmp_4.c: New test. * gcc.target/aarch64/ccmp_5.c: New test. Signed-off-by: Andrew Pinski Co-authored-by: Richard Sandiford --- gcc/ccmp.cc | 12 +-- gcc/cfgexpand.cc | 31 ++----- gcc/expr.cc | 103 ++++++++++++---------- gcc/expr.h | 3 + gcc/testsuite/gcc.target/aarch64/ccmp_3.c | 20 +++++ gcc/testsuite/gcc.target/aarch64/ccmp_4.c | 35 ++++++++ gcc/testsuite/gcc.target/aarch64/ccmp_5.c | 20 +++++ 7 files changed, 149 insertions(+), 75 deletions(-) create mode 100644 gcc/testsuite/gcc.target/aarch64/ccmp_3.c create mode 100644 gcc/testsuite/gcc.target/aarch64/ccmp_4.c create mode 100644 gcc/testsuite/gcc.target/aarch64/ccmp_5.c diff --git a/gcc/ccmp.cc b/gcc/ccmp.cc index 09d6b5595a4..7cb525addf4 100644 --- a/gcc/ccmp.cc +++ b/gcc/ccmp.cc @@ -90,9 +90,10 @@ ccmp_tree_comparison_p (tree t, basic_block bb) If all checks OK in expand_ccmp_expr, it emits insns in prep_seq, then insns in gen_seq. */ -/* Check whether G is a potential conditional compare candidate. */ +/* Check whether G is a potential conditional compare candidate; OUTER is true if + G is the outer most AND/IOR. */ static bool -ccmp_candidate_p (gimple *g) +ccmp_candidate_p (gimple *g, bool outer = false) { tree lhs, op0, op1; gimple *gs0, *gs1; @@ -109,8 +110,9 @@ ccmp_candidate_p (gimple *g) lhs = gimple_assign_lhs (g); op0 = gimple_assign_rhs1 (g); op1 = gimple_assign_rhs2 (g); - if ((TREE_CODE (op0) != SSA_NAME) || (TREE_CODE (op1) != SSA_NAME) - || !has_single_use (lhs)) + if ((TREE_CODE (op0) != SSA_NAME) || (TREE_CODE (op1) != SSA_NAME)) + return false; + if (!outer && !has_single_use (lhs)) return false; bb = gimple_bb (g); @@ -284,7 +286,7 @@ expand_ccmp_expr (gimple *g, machine_mode mode) rtx_insn *last; rtx tmp; - if (!ccmp_candidate_p (g)) + if (!ccmp_candidate_p (g, true)) return NULL_RTX; last = get_last_insn (); diff --git a/gcc/cfgexpand.cc b/gcc/cfgexpand.cc index 1db22f0a1a3..381ed2c82d7 100644 --- a/gcc/cfgexpand.cc +++ b/gcc/cfgexpand.cc @@ -3971,37 +3971,18 @@ expand_gimple_stmt_1 (gimple *stmt) { rtx target, temp; bool nontemporal = gimple_assign_nontemporal_move_p (assign_stmt); - struct separate_ops ops; bool promoted = false; target = expand_expr (lhs, NULL_RTX, VOIDmode, EXPAND_WRITE); if (GET_CODE (target) == SUBREG && SUBREG_PROMOTED_VAR_P (target)) promoted = true; - ops.code = gimple_assign_rhs_code (assign_stmt); - ops.type = TREE_TYPE (lhs); - switch (get_gimple_rhs_class (ops.code)) - { - case GIMPLE_TERNARY_RHS: - ops.op2 = gimple_assign_rhs3 (assign_stmt); - /* Fallthru */ - case GIMPLE_BINARY_RHS: - ops.op1 = gimple_assign_rhs2 (assign_stmt); - /* Fallthru */ - case GIMPLE_UNARY_RHS: - ops.op0 = gimple_assign_rhs1 (assign_stmt); - break; - default: - gcc_unreachable (); - } - ops.location = gimple_location (stmt); - - /* If we want to use a nontemporal store, force the value to - register first. If we store into a promoted register, - don't directly expand to target. */ + /* If we want to use a nontemporal store, force the value to + register first. If we store into a promoted register, + don't directly expand to target. */ temp = nontemporal || promoted ? NULL_RTX : target; - temp = expand_expr_real_2 (&ops, temp, GET_MODE (target), - EXPAND_NORMAL); + temp = expand_expr_real_gassign (assign_stmt, temp, + GET_MODE (target), EXPAND_NORMAL); if (temp == target) ; @@ -4013,7 +3994,7 @@ expand_gimple_stmt_1 (gimple *stmt) if (CONSTANT_P (temp) && GET_MODE (temp) == VOIDmode) { temp = convert_modes (GET_MODE (target), - TYPE_MODE (ops.type), + TYPE_MODE (TREE_TYPE (lhs)), temp, unsignedp); temp = convert_modes (GET_MODE (SUBREG_REG (target)), GET_MODE (target), temp, unsignedp); diff --git a/gcc/expr.cc b/gcc/expr.cc index dc816bc20fa..f9052a6ff5f 100644 --- a/gcc/expr.cc +++ b/gcc/expr.cc @@ -11035,6 +11035,62 @@ stmt_is_replaceable_p (gimple *stmt) return false; } +/* A subroutine of expand_expr_real_1. Expand gimple assignment G, + which is known to set an SSA_NAME result. The other arguments are + as for expand_expr_real_1. */ + +rtx +expand_expr_real_gassign (gassign *g, rtx target, machine_mode tmode, + enum expand_modifier modifier, rtx *alt_rtl, + bool inner_reference_p) +{ + separate_ops ops; + rtx r; + location_t saved_loc = curr_insn_location (); + auto loc = gimple_location (g); + if (loc != UNKNOWN_LOCATION) + set_curr_insn_location (loc); + tree lhs = gimple_assign_lhs (g); + ops.code = gimple_assign_rhs_code (g); + ops.type = TREE_TYPE (lhs); + switch (get_gimple_rhs_class (ops.code)) + { + case GIMPLE_TERNARY_RHS: + ops.op2 = gimple_assign_rhs3 (g); + /* Fallthru */ + case GIMPLE_BINARY_RHS: + ops.op1 = gimple_assign_rhs2 (g); + + /* Try to expand conditonal compare. */ + if (targetm.gen_ccmp_first) + { + gcc_checking_assert (targetm.gen_ccmp_next != NULL); + r = expand_ccmp_expr (g, TYPE_MODE (ops.type)); + if (r) + break; + } + /* Fallthru */ + case GIMPLE_UNARY_RHS: + ops.op0 = gimple_assign_rhs1 (g); + ops.location = loc; + r = expand_expr_real_2 (&ops, target, tmode, modifier); + break; + case GIMPLE_SINGLE_RHS: + { + r = expand_expr_real (gimple_assign_rhs1 (g), target, + tmode, modifier, alt_rtl, + inner_reference_p); + break; + } + default: + gcc_unreachable (); + } + set_curr_insn_location (saved_loc); + if (REG_P (r) && !REG_EXPR (r)) + set_reg_attrs_for_decl_rtl (lhs, r); + return r; +} + rtx expand_expr_real_1 (tree exp, rtx target, machine_mode tmode, enum expand_modifier modifier, rtx *alt_rtl, @@ -11198,51 +11254,8 @@ expand_expr_real_1 (tree exp, rtx target, machine_mode tmode, && stmt_is_replaceable_p (SSA_NAME_DEF_STMT (exp))) g = SSA_NAME_DEF_STMT (exp); if (g) - { - rtx r; - location_t saved_loc = curr_insn_location (); - loc = gimple_location (g); - if (loc != UNKNOWN_LOCATION) - set_curr_insn_location (loc); - ops.code = gimple_assign_rhs_code (g); - switch (get_gimple_rhs_class (ops.code)) - { - case GIMPLE_TERNARY_RHS: - ops.op2 = gimple_assign_rhs3 (g); - /* Fallthru */ - case GIMPLE_BINARY_RHS: - ops.op1 = gimple_assign_rhs2 (g); - - /* Try to expand conditonal compare. */ - if (targetm.gen_ccmp_first) - { - gcc_checking_assert (targetm.gen_ccmp_next != NULL); - r = expand_ccmp_expr (g, mode); - if (r) - break; - } - /* Fallthru */ - case GIMPLE_UNARY_RHS: - ops.op0 = gimple_assign_rhs1 (g); - ops.type = TREE_TYPE (gimple_assign_lhs (g)); - ops.location = loc; - r = expand_expr_real_2 (&ops, target, tmode, modifier); - break; - case GIMPLE_SINGLE_RHS: - { - r = expand_expr_real (gimple_assign_rhs1 (g), target, - tmode, modifier, alt_rtl, - inner_reference_p); - break; - } - default: - gcc_unreachable (); - } - set_curr_insn_location (saved_loc); - if (REG_P (r) && !REG_EXPR (r)) - set_reg_attrs_for_decl_rtl (SSA_NAME_VAR (exp), r); - return r; - } + return expand_expr_real_gassign (as_a (g), target, tmode, + modifier, alt_rtl, inner_reference_p); ssa_name = exp; decl_rtl = get_rtx_for_ssa_name (ssa_name); diff --git a/gcc/expr.h b/gcc/expr.h index f04f40da6ab..64956f63029 100644 --- a/gcc/expr.h +++ b/gcc/expr.h @@ -302,6 +302,9 @@ extern rtx expand_expr_real_1 (tree, rtx, machine_mode, enum expand_modifier, rtx *, bool); extern rtx expand_expr_real_2 (sepops, rtx, machine_mode, enum expand_modifier); +extern rtx expand_expr_real_gassign (gassign *, rtx, machine_mode, + enum expand_modifier modifier, + rtx * = nullptr, bool = false); /* Generate code for computing expression EXP. An rtx for the computed value is returned. The value is never null. diff --git a/gcc/testsuite/gcc.target/aarch64/ccmp_3.c b/gcc/testsuite/gcc.target/aarch64/ccmp_3.c new file mode 100644 index 00000000000..a2b47fbee14 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/ccmp_3.c @@ -0,0 +1,20 @@ +/* { dg-options "-O2" } */ +/* PR target/100942 */ + +void foo(void); +int f1(int a, int b) +{ + int c = a == 0 || b == 0; + if (c) foo(); + return c; +} + +/* We should get one cmp followed by ccmp and one cset. */ +/* { dg-final { scan-assembler "\tccmp\t" } } */ +/* { dg-final { scan-assembler-times "\tcset\t" 1 } } */ +/* { dg-final { scan-assembler-times "\tcmp\t" 1 } } */ +/* And not get 2 cmps and 2 (or more cset) and orr and a cbnz. */ +/* { dg-final { scan-assembler-not "\torr\t" } } */ +/* { dg-final { scan-assembler-not "\tcbnz\t" } } */ +/* { dg-final { scan-assembler-not "\tcbz\t" } } */ + diff --git a/gcc/testsuite/gcc.target/aarch64/ccmp_4.c b/gcc/testsuite/gcc.target/aarch64/ccmp_4.c new file mode 100644 index 00000000000..bc0f57a7c59 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/ccmp_4.c @@ -0,0 +1,35 @@ +/* { dg-options "-O2" } */ +/* PR target/100942 */ + +void foo(void); +int f1(int a, int b, int d) +{ + int c = a < 8 || b < 9; + int e = d < 11 || c; + if (e) foo(); + return c; +} + +/* + We really should get: + cmp w0, 7 + ccmp w1, 8, 4, gt + cset w0, le + ccmp w2, 10, 4, gt + ble .L11 + + But we currently get: + cmp w0, 7 + ccmp w1, 8, 4, gt + cset w0, le + cmp w0, 0 + ccmp w2, 10, 4, eq + ble .L11 + The middle cmp is not needed. + */ + +/* We should end up with only one cmp and 2 ccmp and 1 cset but currently we get 2 cmp + though. */ +/* { dg-final { scan-assembler-times "\tccmp\t" 2 } } */ +/* { dg-final { scan-assembler-times "\tcset\t" 1 } } */ +/* { dg-final { scan-assembler-times "\tcmp\t" 1 { xfail *-*-* } } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/ccmp_5.c b/gcc/testsuite/gcc.target/aarch64/ccmp_5.c new file mode 100644 index 00000000000..7e52ae4f322 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/ccmp_5.c @@ -0,0 +1,20 @@ + +/* { dg-options "-O2" } */ +/* PR target/100942 */ +void f1(int a, int b, _Bool *x) +{ + x[0] = x[1] = a == 0 || b == 0; +} + +void f2(int a, int b, int *x) +{ + x[0] = x[1] = a == 0 || b == 0; +} + + +/* Both functions should be using ccmp rather than 2 cset/orr. */ +/* { dg-final { scan-assembler-times "\tccmp\t" 2 } } */ +/* { dg-final { scan-assembler-times "\tcset\t" 2 } } */ +/* { dg-final { scan-assembler-times "\tcmp\t" 2 } } */ +/* { dg-final { scan-assembler-not "\torr\t" } } */ +