From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 1075) id 5A1F53858D3C; Mon, 19 Jun 2023 16:28:56 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 5A1F53858D3C DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1687192136; bh=GnS6+gzLr2wzsOEuGyCbRc62iWn9R+TGz+QYTIE6rCA=; h=From:To:Subject:Date:From; b=yrFn+Xw8NH0WJq+apJicqD3em6iL5SUOGPUbiuYvJA718Fd9zcBx2cL1iUgDHl1XL xToe6eqNkTYgaSUYp48qwFVJ/sb8NyVPJPBFUVs5iiPTT2pz1eDwpnZ8i5L4hJJYHr 2454RCYGcqTeOUuMhpN/xbyjqaWnWKgytGjKRaeM= MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="utf-8" From: Jan Hubicka To: gcc-cvs@gcc.gnu.org Subject: [gcc r14-1951] optimize std::max early X-Act-Checkin: gcc X-Git-Author: Jan Hubicka X-Git-Refname: refs/heads/master X-Git-Oldrev: c72a7b849853716d94e8d313be5dce3c22850113 X-Git-Newrev: 7b34cacc5735385e7e2855d7c0a6fad60ef4a99b Message-Id: <20230619162856.5A1F53858D3C@sourceware.org> Date: Mon, 19 Jun 2023 16:28:56 +0000 (GMT) List-Id: https://gcc.gnu.org/g:7b34cacc5735385e7e2855d7c0a6fad60ef4a99b commit r14-1951-g7b34cacc5735385e7e2855d7c0a6fad60ef4a99b Author: Jan Hubicka Date: Mon Jun 19 18:28:17 2023 +0200 optimize std::max early we currently produce very bad code on loops using std::vector as a stack, since we fail to inline push_back which in turn prevents SRA and we fail to optimize out some store-to-load pairs. I looked into why this function is not inlined and it is inlined by clang. We currently estimate it to 66 instructions and inline limits are 15 at -O2 and 30 at -O3. Clang has similar estimate, but still decides to inline at -O2. I looked into reason why the body is so large and one problem I spotted is the way std::max is implemented by taking and returning reference to the values. const T& max( const T& a, const T& b ); This makes it necessary to store the values to memory and load them later and max is used by code computing new size of vector on resize. We optimize this to MAX_EXPR, but only during late optimizations. I think this is a common enough coding pattern and we ought to make this transparent to early opts and IPA. The following is easist fix that simply adds phiprop pass that turns the PHI of address values into PHI of values so later FRE can propagate values across memory, phiopt discover the MAX_EXPR pattern and DSE remove the memory stores. gcc/ChangeLog: PR tree-optimization/109811 PR tree-optimization/109849 * passes.def: Add phiprop to early optimization passes. * tree-ssa-phiprop.cc: Allow clonning. gcc/testsuite/ChangeLog: PR tree-optimization/109811 PR tree-optimization/109849 * gcc.dg/tree-ssa/phiprop-1.c: New test. * gcc.dg/tree-ssa/pr21463.c: Adjust template. Diff: --- gcc/passes.def | 2 ++ gcc/testsuite/gcc.dg/tree-ssa/phiprop-1.c | 14 ++++++++++++++ gcc/testsuite/gcc.dg/tree-ssa/pr21463.c | 4 ++-- gcc/tree-ssa-phiprop.cc | 1 + 4 files changed, 19 insertions(+), 2 deletions(-) diff --git a/gcc/passes.def b/gcc/passes.def index c9a8f19747b..faa5208b26b 100644 --- a/gcc/passes.def +++ b/gcc/passes.def @@ -88,6 +88,8 @@ along with GCC; see the file COPYING3. If not see /* pass_build_ealias is a dummy pass that ensures that we execute TODO_rebuild_alias at this point. */ NEXT_PASS (pass_build_ealias); + /* Do phiprop before FRE so we optimize std::min and std::max well. */ + NEXT_PASS (pass_phiprop); NEXT_PASS (pass_fre, true /* may_iterate */); NEXT_PASS (pass_early_vrp); NEXT_PASS (pass_merge_phi); diff --git a/gcc/testsuite/gcc.dg/tree-ssa/phiprop-1.c b/gcc/testsuite/gcc.dg/tree-ssa/phiprop-1.c new file mode 100644 index 00000000000..b448e830d52 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/phiprop-1.c @@ -0,0 +1,14 @@ +/* { dg-do compile } */ +/* { dg-options "-O1 -fdump-tree-phiprop1-details -fdump-tree-release_ssa" } */ +int max(int a, int b) +{ + int *ptr; + if (a > b) + ptr = &a; + else + ptr = &b; + return *ptr; +} + +/* { dg-final { scan-tree-dump-times "Inserting PHI for result of load" 1 "phiprop1"} } */ +/* { dg-final { scan-tree-dump-times "MAX_EXPR" 1 "release_ssa"} } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21463.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21463.c index ed0829a038c..c6f1226d683 100644 --- a/gcc/testsuite/gcc.dg/tree-ssa/pr21463.c +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21463.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O -fdump-tree-phiprop-details" } */ +/* { dg-options "-O -fdump-tree-phiprop1-details" } */ struct f { @@ -16,4 +16,4 @@ int g(int i, int c, struct f *ff, int g) return *t; } -/* { dg-final { scan-tree-dump-times "Inserting PHI for result of load" 1 "phiprop" } } */ +/* { dg-final { scan-tree-dump-times "Inserting PHI for result of load" 1 "phiprop1" } } */ diff --git a/gcc/tree-ssa-phiprop.cc b/gcc/tree-ssa-phiprop.cc index 3cb4900b6be..5dc505df420 100644 --- a/gcc/tree-ssa-phiprop.cc +++ b/gcc/tree-ssa-phiprop.cc @@ -476,6 +476,7 @@ public: {} /* opt_pass methods: */ + opt_pass * clone () final override { return new pass_phiprop (m_ctxt); } bool gate (function *) final override { return flag_tree_phiprop; } unsigned int execute (function *) final override;