From patchwork Fri Sep 30 07:03:58 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Revital Eres X-Patchwork-Id: 4448 Return-Path: X-Original-To: patchwork@peony.canonical.com Delivered-To: patchwork@peony.canonical.com Received: from fiordland.canonical.com (fiordland.canonical.com [91.189.94.145]) by peony.canonical.com (Postfix) with ESMTP id 58AEF23FB8 for ; Fri, 30 Sep 2011 07:04:02 +0000 (UTC) Received: from mail-fx0-f52.google.com (mail-fx0-f52.google.com [209.85.161.52]) by fiordland.canonical.com (Postfix) with ESMTP id 4074FA180BE for ; Fri, 30 Sep 2011 07:04:02 +0000 (UTC) Received: by fxe23 with SMTP id 23so3728353fxe.11 for ; Fri, 30 Sep 2011 00:04:02 -0700 (PDT) Received: by 10.223.94.134 with SMTP id z6mr17777346fam.8.1317366241974; Fri, 30 Sep 2011 00:04:01 -0700 (PDT) X-Forwarded-To: linaro-patchwork@canonical.com X-Forwarded-For: patch@linaro.org linaro-patchwork@canonical.com Delivered-To: patches@linaro.org Received: by 10.152.3.234 with SMTP id f10cs42803laf; Fri, 30 Sep 2011 00:04:01 -0700 (PDT) Received: by 10.236.182.135 with SMTP id o7mr10980802yhm.34.1317366239686; Fri, 30 Sep 2011 00:03:59 -0700 (PDT) Received: from mail-yi0-f50.google.com (mail-yi0-f50.google.com [209.85.218.50]) by mx.google.com with ESMTPS id n33si1914703ani.31.2011.09.30.00.03.59 (version=TLSv1/SSLv3 cipher=OTHER); Fri, 30 Sep 2011 00:03:59 -0700 (PDT) Received-SPF: neutral (google.com: 209.85.218.50 is neither permitted nor denied by best guess record for domain of revital.eres@linaro.org) client-ip=209.85.218.50; Authentication-Results: mx.google.com; spf=neutral (google.com: 209.85.218.50 is neither permitted nor denied by best guess record for domain of revital.eres@linaro.org) smtp.mail=revital.eres@linaro.org Received: by yib25 with SMTP id 25so1478660yib.37 for ; Fri, 30 Sep 2011 00:03:59 -0700 (PDT) MIME-Version: 1.0 Received: by 10.100.193.5 with SMTP id q5mr10320005anf.36.1317366238942; Fri, 30 Sep 2011 00:03:58 -0700 (PDT) Received: by 10.101.58.12 with HTTP; Fri, 30 Sep 2011 00:03:58 -0700 (PDT) In-Reply-To: References: Date: Fri, 30 Sep 2011 10:03:58 +0300 Message-ID: Subject: Re: [PATCH, SMS 1/2] Avoid generating redundant reg-moves From: Revital Eres To: Ayal Zaks Cc: gcc-patches@gcc.gnu.org, Patch Tracking Hello, > This > +      /* Skip instructions that do not set a register.  */ > +      if (set && !REG_P (SET_DEST (set))) > +        continue; > is ok. Can you also prevent !set insns from having reg_moves? (To be updated > once auto_inc insns will be supported, if they'll deserve reg_moves too.) I added a check to verify that no reg-moves are created for !set instructions. Currently re-testing on ppc64-redhat-linux (bootstrap and regtest) and arm-linux-gnueabi (bootstrap c). OK to commit once tesing completes? Thanks, Revital gcc/ * modulo-sched.c (generate_reg_moves): Skip instructions that do not set a register and verify no regmoves are created for !single_set instructions. testsuite/ * gcc.dg/sms-10.c: New file. /* { dg-do run } */ /* { dg-options "-O2 -fmodulo-sched -fmodulo-sched-allow-regmoves -fdump-rtl-sms" } */ typedef __SIZE_TYPE__ size_t; extern void *malloc (size_t); extern void free (void *); extern void abort (void); struct regstat_n_sets_and_refs_t { int sets; int refs; }; struct regstat_n_sets_and_refs_t *regstat_n_sets_and_refs; struct df_reg_info { unsigned int n_refs; }; struct df_d { struct df_reg_info **def_regs; struct df_reg_info **use_regs; }; struct df_d *df; static inline int REG_N_SETS (int regno) { return regstat_n_sets_and_refs[regno].sets; } __attribute__ ((noinline)) int max_reg_num (void) { return 100; } __attribute__ ((noinline)) void regstat_init_n_sets_and_refs (void) { unsigned int i; unsigned int max_regno = max_reg_num (); for (i = 0; i < max_regno; i++) { (regstat_n_sets_and_refs[i].sets = (df->def_regs[(i)]->n_refs)); (regstat_n_sets_and_refs[i].refs = (df->use_regs[(i)]->n_refs) + REG_N_SETS (i)); } } int a_sets[100] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99 }; int a_refs[100] = { 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, 38, 40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60, 62, 64, 66, 68, 70, 72, 74, 76, 78, 80, 82, 84, 86, 88, 90, 92, 94, 96, 98, 100, 102, 104, 106, 108, 110, 112, 114, 116, 118, 120, 122, 124, 126, 128, 130, 132, 134, 136, 138, 140, 142, 144, 146, 148, 150, 152, 154, 156, 158, 160, 162, 164, 166, 168, 170, 172, 174, 176, 178, 180, 182, 184, 186, 188, 190, 192, 194, 196, 198 }; int main () { struct df_reg_info *b[100], *c[100]; struct df_d df1; size_t s = sizeof (struct df_reg_info); struct regstat_n_sets_and_refs_t a[100]; df = &df1; regstat_n_sets_and_refs = a; int i; for (i = 0; i < 100; i++) { b[i] = (struct df_reg_info *) malloc (s); b[i]->n_refs = i; c[i] = (struct df_reg_info *) malloc (s); c[i]->n_refs = i; } df1.def_regs = b; df1.use_regs = c; regstat_init_n_sets_and_refs (); for (i = 0; i < 100; i++) if ((a[i].sets != a_sets[i]) || (a[i].refs != a_refs[i])) abort (); for (i = 0; i < 100; i++) { free (b[i]); free (c[i]); } return 0; } /* { dg-final { scan-rtl-dump-times "SMS succeeded" 1 "sms" { target powerpc*-*-* } } } */ /* { dg-final { cleanup-rtl-dump "sms" } } */ Index: modulo-sched.c =================================================================== --- modulo-sched.c (revision 179138) +++ modulo-sched.c (working copy) @@ -476,7 +476,12 @@ generate_reg_moves (partial_schedule_ptr sbitmap *uses_of_defs; rtx last_reg_move; rtx prev_reg, old_reg; - + rtx set = single_set (u->insn); + + /* Skip instructions that do not set a register. */ + if ((set && !REG_P (SET_DEST (set)))) + continue; + /* Compute the number of reg_moves needed for u, by looking at life ranges started at u (excluding self-loops). */ for (e = u->out; e; e = e->next_out) @@ -493,6 +498,16 @@ generate_reg_moves (partial_schedule_ptr && SCHED_COLUMN (e->dest) < SCHED_COLUMN (e->src)) nreg_moves4e--; + if (nreg_moves4e >= 1) + { + /* !single_set instructions are not supported yet and + thus we do not except to encounter them in the loop + except from the doloop part. For the later case + we assume no regmoves are generated as the doloop + instructions are tied to the branch with an edge. */ + gcc_assert (set); + } + nreg_moves = MAX (nreg_moves, nreg_moves4e); }