From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 2136) id 760653858C2B; Fri, 4 Nov 2022 14:18:42 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 760653858C2B DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1667571522; bh=XpsccBRMH9QPDfhY/3nAQq1T2n3nhWPRqg/EeYOYCUk=; h=From:To:Subject:Date:From; b=hlcHNys6Vdk5fj4qGkLoqLJUY1GCyYpyAx5vwcIUlBTCqWklR2xpOEFt8QGzyhQP0 S1hUM9id2+rDfEI2K+tM+PU6SqGakHnjXVBhCcRJDhglQzcqy5njOICxZKz9IXomZM b57NXJ39lxswpvB7i/ag1wq0UcV2uH23dh4I4EG8= MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="utf-8" From: Aldy Hernandez To: gcc-cvs@gcc.gnu.org Subject: [gcc r13-3688] Set nonzero bits for multiplication and divisions by a power of 2. X-Act-Checkin: gcc X-Git-Author: Aldy Hernandez X-Git-Refname: refs/heads/master X-Git-Oldrev: 0bdf10bdf1b2c9f31e7e764dec4d56ea6044f943 X-Git-Newrev: 679be32e66428f0ba81d1c1b55f7bd47f01cb295 Message-Id: <20221104141842.760653858C2B@sourceware.org> Date: Fri, 4 Nov 2022 14:18:42 +0000 (GMT) List-Id: https://gcc.gnu.org/g:679be32e66428f0ba81d1c1b55f7bd47f01cb295 commit r13-3688-g679be32e66428f0ba81d1c1b55f7bd47f01cb295 Author: Aldy Hernandez Date: Fri Nov 4 10:20:46 2022 +0100 Set nonzero bits for multiplication and divisions by a power of 2. We're missing a lot of TLC in keeping track of nonzero bits across range-ops. It isn't an oversight, but just limited amount of hours to implement stuff. This patch keeps better track of the nonzero mask (really maybe_nonzero bits as discussed) across multiplication and division when the RHS is a power of 2. It fixes PR107342 and also touches on PR55157. In the latter, the nonzero mask is being set quite late (CCP2) but could be set by evrp time if we enhanced range-ops. I have added tests from both PRs. Tested PR tree-optimization/107342 gcc/ChangeLog: * range-op.cc (operator_mult::fold_range): New. (operator_div::fold_range): New. gcc/testsuite/ChangeLog: * gcc.dg/tree-ssa/vrp122.c: New test. * gcc.dg/tree-ssa/vrp123.c: New test. Diff: --- gcc/range-op.cc | 59 ++++++++++++++++++++++++++++++++++ gcc/testsuite/gcc.dg/tree-ssa/vrp122.c | 19 +++++++++++ gcc/testsuite/gcc.dg/tree-ssa/vrp123.c | 18 +++++++++++ 3 files changed, 96 insertions(+) diff --git a/gcc/range-op.cc b/gcc/range-op.cc index 49ee7be3d3b..25c004d8287 100644 --- a/gcc/range-op.cc +++ b/gcc/range-op.cc @@ -1742,9 +1742,13 @@ cross_product_operator::wi_cross_product (irange &r, tree type, class operator_mult : public cross_product_operator { + using range_operator::fold_range; using range_operator::op1_range; using range_operator::op2_range; public: + virtual bool fold_range (irange &r, tree type, + const irange &lh, const irange &rh, + relation_trio = TRIO_VARYING) const final override; virtual void wi_fold (irange &r, tree type, const wide_int &lh_lb, const wide_int &lh_ub, @@ -1762,6 +1766,32 @@ public: relation_trio) const; } op_mult; +bool +operator_mult::fold_range (irange &r, tree type, + const irange &lh, const irange &rh, + relation_trio trio) const +{ + if (!cross_product_operator::fold_range (r, type, lh, rh, trio)) + return false; + + if (lh.undefined_p ()) + return true; + + tree t; + if (rh.singleton_p (&t)) + { + wide_int w = wi::to_wide (t); + int shift = wi::exact_log2 (w); + if (shift != -1) + { + wide_int nz = lh.get_nonzero_bits (); + nz = wi::lshift (nz, shift); + r.set_nonzero_bits (nz); + } + } + return true; +} + bool operator_mult::op1_range (irange &r, tree type, const irange &lhs, const irange &op2, @@ -1902,10 +1932,39 @@ public: const wide_int &rh_ub) const; virtual bool wi_op_overflows (wide_int &res, tree type, const wide_int &, const wide_int &) const; + virtual bool fold_range (irange &r, tree type, + const irange &lh, const irange &rh, + relation_trio trio) const final override; private: enum tree_code code; }; +bool +operator_div::fold_range (irange &r, tree type, + const irange &lh, const irange &rh, + relation_trio trio) const +{ + if (!cross_product_operator::fold_range (r, type, lh, rh, trio)) + return false; + + if (lh.undefined_p ()) + return true; + + tree t; + if (rh.singleton_p (&t)) + { + wide_int wi = wi::to_wide (t); + int shift = wi::exact_log2 (wi); + if (shift != -1) + { + wide_int nz = lh.get_nonzero_bits (); + nz = wi::rshift (nz, shift, TYPE_SIGN (type)); + r.set_nonzero_bits (nz); + } + } + return true; +} + bool operator_div::wi_op_overflows (wide_int &res, tree type, const wide_int &w0, const wide_int &w1) const diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c new file mode 100644 index 00000000000..b2ddcda023c --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c @@ -0,0 +1,19 @@ +// { dg-do compile } +// { dg-options "-O2 -fdump-tree-evrp-details" } + +void gg(void); +int f(unsigned t) +{ + unsigned g = t*16; + if (g==0) return 1; + gg(); + gg(); + gg(); + gg(); + gg(); + gg(); + if (g<=4) return 1; + return 0; +} + +// { dg-final { scan-tree-dump "Global Exported: g_.* NONZERO 0x.*fff0" "evrp" } } diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c new file mode 100644 index 00000000000..1ad3caa4384 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c @@ -0,0 +1,18 @@ +// { dg-options "-O1 -fdump-tree-dom3-raw" } + +extern int +__attribute__((const)) +foo4b (int); + +int f4b (unsigned int r) +{ + if (foo4b (r)) + r *= 8U; + + if ((r / 2U) & 2U) + r += foo4b (r); + + return r; +} + +// { dg-final { scan-tree-dump-times {gimple_call